“PLANET IN A BOTTLE” A REALTIME OBSERVATORY FOR LABORATORY SIMULATION OF PLANETARY CIRCULATION Sai Ravela Massachusetts Institute of Technology J. Marshall, A.
Download ReportTranscript “PLANET IN A BOTTLE” A REALTIME OBSERVATORY FOR LABORATORY SIMULATION OF PLANETARY CIRCULATION Sai Ravela Massachusetts Institute of Technology J. Marshall, A.
Slide 1
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 2
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 3
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 4
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 5
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 6
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 7
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 8
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 9
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 10
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 11
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 12
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 13
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 14
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 15
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 16
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 17
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 18
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 19
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 20
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 21
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 22
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 23
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 24
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 25
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 26
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 27
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 28
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 29
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 30
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 31
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 2
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 3
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 4
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 5
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 6
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 7
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 8
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 9
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 10
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 11
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 12
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 13
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 14
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 15
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 16
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 17
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 18
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 19
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 20
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 21
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 22
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 23
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 24
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 25
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 26
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 27
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 28
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 29
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 30
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.
Slide 31
“PLANET IN A BOTTLE”
A REALTIME
OBSERVATORY
FOR
LABORATORY
SIMULATION
OF
PLANETARY
CIRCULATION
Sai Ravela
Massachusetts Institute of
Technology
J. Marshall, A. Wong, S. Stransky,
C. Hill
Collaborators: B. Kuszmaul and
C. Leiserson
Geophysical Fluids in the Laboratory
Inference from models and data is
fundamental to the earth sciences
Laboratory analogs systems
can be extremely useful
Ravela, Marshall , Wong, Stransky , 07
Planet-in-a-bottle
OBS
Z
DA
MODEL
Velocity Observations
• Velocity
measurements
using correlationbased optic-flow
• 1sec per 1Kx1K
image using two
processors.
• Resolution,
sampling and
noise cause
measurement
uncertainty
• Climalotological
temperature BC in
the numerical
model
Marshall et al., 1997
Numerical
Simulation
MIT-GCM (mitgcm.org): incompressible
boussinesq fluid in non-hydrostatic mode
with a vector-invariant formulation
• Thermally-driven System (via EOS)
• Hydrostatic mode Arakawa C-Grid
• Momentum Equations: Adams-Bashforth-2
• Traceer Equations: Upwind-biased DST with
Sweby Flux limiter
• Elliptic Equaiton: Conjugate Gradients
• Vertical Transport implicit.
Domain
1.
2.
3.
4.
5.
6.
Cylindrical coordinates.
Nonuniform discretization of the vertical
Random temperature IC
Static temperature BC
Noslip boundaries
Heat-flux controlled with anisotropic thermal
diffusivity
120x 23 x 15 (z)
{45-8 }x 15cm
Estimation from model and data
Estimate what?
1. State Estimation:
1. NWP type applications,
but also reanalysis
2. Filtering & Smoothing
2. Parameter Estimation:
1. Forecasting & Climate
3. State and Parameter
Estimation
1. The real problem.
General Approach:
Ensemble-based, multiscale methods.
Schedule
Producing state estimates
Ensemble-methods
Reduced-rank Uncertainty
Key questions
Where does the ensemble come
from?
Statistical sampling
How many ensemble members are
necessary?
What about the computational cost
of ensemble propagation?
Does the forecast uncertainty contain
truth in it?
What happens when it is not?
What about spurious longrange
correlations in reduced rank
representations?
Tolerance to nonlinearity
Model is fully nonlinear
Dimensionality
Square-root representation
via the ensemble
Variety of approximte
filters and smoothers
Ravela, Marshall, Hill, Wong and Stransky, 07
Approach
Snapshots capture
flow-dependent
uncertainty
(Sirovich)
BC+IC
Deterministic update:
5 – 2D updates
5 – (Elliptic) temperature
Nx * Ny – 1D problems
P(Yt|Xt) P(Xt|Xt-1):
Deterministic update
P(X0|T): IC
Perturbation
1
P(T ): Thermal BC
Perturbations 4
P(Xt|Xt-1): Snapshots
in time
10
E>e0
?
P(Yt|Xt) P(Xt|Xt-1):
Ensemble update
Ravela, Marshall, Hill, Wong and Stransky, 07
EnKF revisited
The analysis ensemble is a
(weakly) nonlinear
combination
of the forecast ensemble.
This form greatly facilitates
interpretation of smoothing
Evensen 03, 04
Ravela and McLaughlin, 2007
Ravela and McLaughlin, 2007
Next Steps
Lagrangian Surface
Observations : Multi-Particle
Tracking
Volumetric temperature
measurements.
Simultaneous state and
parameter estimation.
Targeting using FTLE &
Effective diffusivity
measures.
Semi-lagrangian schemes for
increased model timesteps.
MicroRobotic Dye-release
platforms.
Ravela et al. 2003, 2004, 2005, 2006, 2007
THE AMPLITUDEPOSITION FORMULATION
OF DATA ASSIMILATION
With thanks to
K. Emanuel, D. McLaughlin and W. T. Freeman
Many reasons for
position error
There are many sources of position error: Flow and
timing errors, Boundary and Initial Conditions,
Parameterizations of physics, sub-grid processes,
Numerical integration…Correcting them is very
difficult.
Thunderstorms
Solitons
Hurricanes
Amplitude
assimilation of
position errors
is nonsense!
3DVAR
EnKF
Distorted analyses are optimal, by definition. They are also
inappropriate, leading to poor estimates at best, and blowing
the model up, at worst.
Key Observations
Why do position errors occur?
Flow & timing errors, discretization and numerical
schemes, initial & boundary conditions…most prominently
seen in meso-scale problems: storms, fronts, etc.
What is the effect of position errors?
Forecast error covariance is weaker, the estimator is both
biased, and will not achieve the cramer-rao bound.
When are they important?
They are important when observations are uncertain and
sparse
Joint Position Amplitude Formulation
Question the standard
Assumption; Forecasts
are unbiased
Bend, then blend
Improved control of solution
Flexible Application
Data Assimilation
Hurricanes , Fronts & Storms
In Geosciences
Reservoir Modeling
Alignment a better metric
for structures
Super-resolution simulations
texture (lithology) synthesis
Students
Ryan Abernathy:
Scott Stransky
Flow & Velocimetry
Robust winds from GOES
Classroom
Fluid Tracking
Under failure of brightness
constancy
Bend,
then Blend
Cambridge 1-step
(Bend and Blend)
Variational solution to
jointly solves for diplas and
amplitudes
Expensive
Cambridge 2-step
(Bend, then blend)
Approximate solution
Preprocessor to 3DVAR
or EnKF
Inexpensive
Key Observations
Why is “morphing” a bad idea
Kills amplitude spread.
Why is two-step a good idea
Approximate solution to the joint inference
problem.
Efficient O(nlog n), or O(n) with FMM
What resources are available?
Papers, code, consulting, joint prototyping etc.
Adaptation to multivariate
fields
Ravela & Chatdarong, 06
Velocimetry, for Rainfall Modeling
Aligned time sequences of cloud fields are used to
produce velocity fields for advecting model
storms.
Velocimetry derived this way is more robust than
existing GOES-based wind products.
Other applications
Magnetometry Alignment (Shell)
Super-resolution
Example-based
Super-resolved Fluids
Ravela and Freeman 06
Next Steps
Fluid Velocimetry: GOES & Laboratory, release
product.
Incorporate Field Alignment in Bottle project DA.
Learning the amplitude-position partition function.
The joint amplitude-position Kalman filter.
Large-scale experiments.