Transcript Predicting Hurricanes and Improving Climate Models using Ensemble Data Assimilation Jeffrey Anderson, NCAR DAReS The National Center for Atmospheric Research is sponsored by the.

Predicting Hurricanes and Improving Climate
Models using Ensemble Data Assimilation
Jeffrey Anderson, NCAR DAReS
The National Center for Atmospheric Research is
sponsored by the National Science Foundation.
How an Ensemble Filter Works for
Geophysical Data Assimilation
1. Use model to advance ensemble (3 members here)
to time at which next observation becomes available.
Ensemble state
estimate after using
previous observation
(analysis)
Ensemble state
at time of next
observation
(prior)
How an Ensemble Filter Works for
Geophysical Data Assimilation
2. Get prior ensemble sample of observation, y = h(x), by
applying forward operator h to each ensemble member.
Theory: observations
from instruments with
uncorrelated errors can
be done sequentially.
How an Ensemble Filter Works for
Geophysical Data Assimilation
3. Get observed value and observational
error distribution from observing system.
How an Ensemble Filter Works for
Geophysical Data Assimilation
4. Find the increments for the prior observation ensemble
(this is a scalar problem for uncorrelated observation errors).
Note: Difference between
various ensemble filter methods
is primarily in observation
increment calculation.
How an Ensemble Filter Works for
Geophysical Data Assimilation
5. Use ensemble samples of y and each state variable to linearly
regress observation increments onto state variable increments.
Theory: impact of
observation increments on
each state variable can be
handled independently!
How an Ensemble Filter Works for
Geophysical Data Assimilation
6. When all ensemble members for each state variable
are updated, there is a new analysis. Integrate to time
of next observation …
Research with DART
• Public domain software for Data
Assimilation
used at -
– Well-tested, portable,
extensible, free!
•
Models
– Toy to HUGE
• Observations
– Real, synthetic, novel
• An extensive Tutorial
– With examples, exercises,
explanations
• People
– You don’t have to go it alone!
and many more …
DART is:
• Education
• Exploration
• Research
• Operations
Basic Capability: Ensemble Analyses and
Forecasts in Large Geophysical Models
20 of 80
members
6-hour forecast
500 hPa height
18Z 14 Jan 2007
Forecast from CAM (Community Atmosphere Model)
Diagnosis of Noise in the CAM
Finite Volume core using DART
Kevin Raeder*
Jeff Anderson*
Peter Lauritzen+
Tim Hoar*
*NCAR/CISL/IMAGe/DAReS
+NCAR/ESSL/CGD/AMPS
The National Center for Atmospheric Research is sponsored by the
National Science Foundation.
CAM & DART
CAM = 3.5.xx, Finite Volume core, 1.9x2.5, 30 min ∆t.
DART = Data Assimilation Research Testbed, an ensemble
Kalman filter data assimilation system.
Assimilate observations used in operational forecasting:
U, V, and T from radiosondes, ACARS, and aircraft,
U and V from satellite cloud drift winds,
every 6 hours to bring CAM as close to the atmosphere as
possible, balancing the obs and model errors.
This system is competitive with operational weather centers’
data assimilation systems.
“Houston, we have a Problem.”
CAM FV core - 80 member mean - 00Z 25 September 2006
Suspicions turned to the polar filter (DPF)
CAM FV core - 80 member mean - 00Z 25 September 2006
Using a continuous polar filter
(alt-pft) does not show this effect.
Ensemble Mean V @ 266hPa - 00Z 25 Sep 2006 - CAM FV core
The differences are minimal except at the
transition region of the default polar filter.
Ensemble Mean V @ 266hPa - 00Z 25 Sep 2006 - CAM FV core
Three adjacent E-W cross-sections from the
region of the discontinuity reveal more detail.
m/
s
m/s
m/s
East Longitude
Ensemble Mean V @ 266hPa - 00Z 25 Sep 2006 - CAM FV core
That wasn’t so bad!
• The use of DART diagnosed a problem that had been
unrecognized (or at least undocumented).
• The problem can be seen in ‘free runs’ - it is not a
data assimilation artifact.
• Without assimilation, can’t get reproducing
occurrences to diagnose.
• Could have an important effect on any physics in
which meridional mixing is important.
• The alternate polar filter ‘fixes’ this problem,
but . . .
More suspicious patterns, not fixed by ALT_PFT
2 ∆y noise in ensemble average V
Ensemble Mean V @ 266hPa CAM FV core 00Z 25 September 2006
North-South cross sections
46º East
206º East
Polar filter
noise (fixed)
Residual
Noise
Residual
Noise
Ensemble Mean V @ 266hPa CAM FV core 00Z 25 September 2006
Another instance of noise from real-time use of DARTCAM in a chemistry field campaign (ARCTAS)
6 hour forecast of a single ensemble member
Ensemble Member 10 V @ 266hPa CAM FV core 06Z 13 April 2008
Noise not restricted to V winds …
suspicious
Ensemble Member 10 T @ 266hPa CAM FV core 06Z 13 April 2008
suspicious
Ensemble Member 10 U @ 266hPa CAM FV core 06Z 13 April 2008
Doubling the dynamical time splitting reduced the noise;
implicates model as opposed to assimilation.
Ensemble Mean V @ 266hPa CAM FV core 00Z 25 September 2006
Notes and Conclusions
The noise here may seem small and transient,
but since it had not been recognized by any of the labs
which are using this FV core, the effects on climate
runs had not been explored.
Spurious mixing is happening.
Parameterizations may have been mistuned.
More time may need to be spent fixing the remaining
noise and looking at other unexamined pieces of the
code.
Evaluating the atmospheric forcing
on recent Arctic sea ice loss
Jennifer E. Kay
National Center for Atmospheric Research (NCAR)
Colorado State University (CSU)
Collaborators: Julienne Stroeve (NSIDC),
Andrew Gettelman, Kevin Raeder, Jeff Anderson (NCAR),
Graeme Stephens, Tristan L’Ecuyer, Chris O’Dell (CSU)
Special Thanks: Cecile Hannay (NCAR)
March 10, 2008 MODIS image of the Alaska coastline
New Tool: Data Assimilation
DART = Data Assimilation Research Testbed
Fig. 1 from Rodwell and Palmer (2007)
Lots of science and model assessment can be done!
- Do climate models capture observed atmospheric processes?
- Do analysis increments reveal the underlying mechanisms for persistent model biases?
1. New observations and tools
2. Mechanisms for recent sea ice loss
3. Arctic CAM-DART project
The 2007 record minimum
extent was 4.13 million km2.
The 2008 minimum extent
was 4.52 million km2.
DART-CAM Assimilations
July 2007 minus July 2006
Sea Ice Fraction
Month
Surface boundary condition
July 2006
observed (Hurrell et al., 2008)
July 2007
observed (Hurrell et al., 2008)
Research Questions:
- Does
CAM capture changes in atmospheric forcing important for sea ice loss?
- Does the surface affect the atmospheric forcing on sea ice loss in CAM?
From CAM forecasts to
monthly averages…
Average all 12hour forecasts.
CAM monthly mean SLP
July06 vs. July07
CAM forecasts show large differences in mean sea level pressure fields.
CAM-forecasted clouds
July 2007 had cloud decreases under high SLP,
but cloud increases over the ice-free ocean.
CAM-forecasted shortwave radiation
CAM downwelling and net surface solar radiation responded to
cloud changes and surface albedo decreases.
CAM-forecasted longwave radiation
Surface downwelling LW radiation changes related to low cloud changes.
CAM-forecasted clouds and radiation
July07 minus July06
Arctic Ocean
Western Pacific
Eastern Pacific
70-90 N
70-90 N, 180-240 E
70-75 N,150-180 E
Sea ice area fraction
-0.03
-0.11
-0.49
Total cloud cover
-10%
-13%
+17%
Low cloud cover
-10%
-12%
+20%
FSDS (Wm-2 )
+28
+33
-13
FSNS (Wm-2 )
+11
+32
+32
FLDS (Wm-2 )
-8
-7
+19
Overall, July 2007 had fewer clouds, more downwelling and absorbed
shortwave radiation, and less downwelling longwave radiation.
Over open water, 2007 had more clouds, less downwelling shortwave radiation,
more absorbed shortwave radiation, and more downwelling longwave radiation.
Modeled vs. observed cloud changes
July 2007 minus July 2006
CAM Total Cloud Changes
MODIS Terra Cloud Changes
Sea Ice Area
Fraction Changes
Unlike CAM, MODIS shows variability in the cloud response over open water.
Summary
- New satellite data and model-observation comparison tools are
improving our understanding of atmospheric processes.
- While 2007 was a ‘perfect storm’ for ice loss, 2008 had the 2nd
lowest ice extent with relatively ‘normal’ atmospheric forcing.
- The timing of ice loss matters, and can be used to understand
ice loss forcing mechanisms.
- Comparing CAM forecasts from July 2006 and July 2007
revealed ubiquitous low cloud increases over open water. This
negative feedback on sea ice loss was not seen in observations.
Application of Radio Occultation Data in
Analyses and Forecasts of Tropical Cyclones
Using an Ensemble Assimilation System
Hui Liu, Jeff Anderson, and Bill Kuo
Joint Statistical Meeting
August 2008
An Example of using assimilation to
evaluate the impact of novel observations.
GPS Radio Occultation (RO)
Basic measurement principle:
Deduce atmospheric water vapor and temperature
based on measurement of GPS signal phase delay.
Limb sounding of atmosphere
as LEO satellite receivers rise
or set with respect to GPS
satellites
Global observations are related to:
Temperature,
Humidity,
Ionospheric stuff.
COSMIC GPS RO Research Mission (2006 - 2011)
A set of six mini-satellites
in Low Earth Orbit (LEOs)
with GPS receivers were
launched on
15 April 2006
Vandenberg AFB
15 April 2006.
COSMIC launch picture provided by Orbital Sciences Corporation
Global coverage including oceans and polar areas!
7 Dec 2007 … 1878 soundings
GPS Radio Occultation Refractivity
• Has accurate measurements of both water
vapor and temperature with high vertical
resolution
• Minimally affected by clouds and precipitation
• Has great potential to improve weather
analyses and forecasts over data-sparse and
cloudy areas like tropical oceans
So, RO is especially useful for
tropical cyclone forecasts
Challenges for Assimilation of RO
Refractivity
• RO refractivity is a function of both water
vapor and temperature
• Retrieval of water vapor and temperature
requires accurate estimate of covariance
between RO data, temperature, and moisture
• These covariances are highly time-varying
and not well known
Ensemble Kalman Filter Assimilation
• Covariance of RO refractivity with water
vapor and temperature is computed
from online ensemble forecasts
• The error covariance is time-varying,
related to weather patterns
Typhoon Shanshan
(Sep 10-17, 2006)
Central SLP
pressure
Operational forecasts
using variational
assimilation failed to
predict the curving of
the typhoon.
COSMIC RO soundings
RO soundings,
randomly distributed
over the domain,
provide large-scale
information.
101 profiles on 13 September 2006
Assimilation experiments
•
WRF/DART ensemble assimilation at 45km resolution
•
8-14 September 2006 (typhoon develops on the 10th)
•
32 ensemble members.
Control/NoGPS run:
•
Assimilate operational datasets including radiosonde,
cloud winds, land and ocean surface observations, SATEM
thickness, and QuikSCAT surface winds.
GPS run:
•
Assimilate the above observations + RO refractivity.
Typhoon central pressure in analyses
Sep 10
Sep 14
Intensity of the typhoon is enhanced with RO data.
Typhoon Maximum surface wind in analyses
Sep 10
Sep 14
Intensity of the typhoon is enhanced with RO data.
Impact of RO refractivity on Ensemble forecasts
(16 members, with a finer nested grid of 15km)
initialized at 00UTC 13 and 14 Sept 2006.
Forecast from 00UTC 13 Sep 2006
Ensemble Forecasts of central sea level pressure
Ensemble
Mean
Observed
Observed
Ensemble
mean
with RO data
Sep 13
Sep 16
Sep 13
Intensity of the typhoon is increased with RO data
Sep 16
Ensemble Forecasts of maximum surface wind
Ensemble
Mean
Observed
Observed
Ensemble
mean
with RO data
Intensity of the typhoon is increased with RO data
Forecast Probability of Rainfall >60mm/24h,
12Z 14-15 Sep
Observed
Ensemble
mean
with RO data
OBS
Probability = Rainy members/total
members
Rainfall probability is increased with RO data
Ensemble Forecasts of Typhoon Track
Ensemble
mean
Observed
Ensemble
mean
Observed
NOGPS
with RO data
GPS
Curving of the Typhoon is well predicted in both cases.
Ensemble Forecasts of Typhoon Track Error
Ensemble
mean
Curving of the Typhoon is well predicted in both cases.
Summary
• Forecasts of the typhoon intensity and
rainfall probability are improved by using
RO refractivity observations with the
WRF/DART ensemble system.
• The curving path of the typhoon is well
predicted.
Mesoscale WRF Surface-Data Assimilation:
Spring 2007 Experiments at the
National Severe Storms Laboratory
David Dowell
David Stensrud
NCAR, Boulder, CO
NSSL, Norman, OK
Nusrat Yussouf
Mike Coniglio
CIMMS, Norman, OK
NSSL, Norman, OK
Jeff Anderson
Chris Snyder
NCAR, Boulder, CO
NCAR, Boulder, CO
Acknowledgments: Nancy Collins, Tim Hoar, Greg Carbin
Motivation
Investigate the value of assimilating surface observations for
mesoscale NWP
• predictions of surface boundaries, convective storm environments
• probabilistic precipitation forecasts
Using surface obs to update the model state can be difficult
• strong gradients near the surface
• situation-dependent background-error covariances needed
Recent work provides encouragement
• Hacker and Snyder 2005 -- significant correlations between state variables
at sfc and those at heights up to several km AGL, even at night
• Fujita et al. 2007 -- improvement in 6-12 hour MM5 ensemble forecasts
through assimilating surface obs for only 6 hours
Mesoscale Ensemble Forecasting (WRF-ARW 2.1)
• CONUS grid
– 30-km horizontal grid spacing, 31 vertical levels
– Mean initial and boundary conditions from NAM
• 30-member ensemble
– Initial and boundary condition perturbations (from WRF-Var)
– Parameterization diversity
•
•
•
•
•
Microphysics: Lin et al. (6 class), WSM (3 class)
Shortwave radiation: Dudhia, Goddard
PBL: YSU, Mellor-Yamada-Janjic, NCEP GFS
Surface layer: MM5 similarity, Eta similarity (Janjic)
Cumulus: Kain-Fritsch, Betts-Miller-Janjic, Grell-Devenyi
Observations
• Hourly observations from approximately 1500 sites over USA, Mexico,
and Canada
• Horizontal wind components (u and v) at 10 m AGL (2.0 m s-1 error)
• Potential temp. () and dewpoint (Td) at 2 m AGL (2.0 K error)
• All model state variables updated
– 300-km (20-level) localization radius around each observation
• Observations in model diagnosed through PBL and surface-layer
schemes (“U10”, “V10”, “T2”, “Q2”)
Daily Experiments (March-June 2007)
• Hourly mesoanalyses
12Z
18Z
0Z
6Z
hourly assimilation
• Ensemble forecasts with surface-data assimilation
12Z
18Z
0Z
assimilation
6Z
forecast
• Ensemble forecasts without surface-data-assimilation
– NAM 18Z analysis + i.c. and b.c. perturbations + parameterization
diversity
18Z
0Z
6Z
forecast
March 28 Tornado Outbreak
May 4 (Greensburg, KS) Tornado Case
Impact of Surface-Data Assimilation on Forecasts:
RMS Difference between Obs and Ensemble Mean
v wind
component
ob - model (m/s)
v at 10 m
12Z initialization,
no assimilation
12Z initialization,
6 hr assimilation
time (UTC), 4-5 May
temperature
ob - model (K)
 at 2 m
12Z initialization,
no assimilation
12Z initialization,
6 hr assimilation
time (UTC), 4-5 May
Probability (1-hr convective precip. > 1 mm)
0300 UTC 5 May 2007
%
9-hr forecast without assimilation
(18Z initialization)
%
9-hr forecast with assimilation
(12Z initialization + 6 hr assimilation)
Future Work
• More analysis of spring 2007 cases
– Verification at Oklahoma Mesonet sites
– Sounding verification
– Statistics stratified by ensemble-member
characteristics (e.g., PBL scheme)
• Higher-resolution ensemble forecasting
• Longer assimilation windows
Projects making use of
ensemble statistics…
Hurricane Katrina Sensitivity Analysis
Ryan Torn, SUNY Albany
Contours are ensemble
mean 48h forecast of deeplayer mean wind.
Color indicates change
in the longitude of
Katrina.
MOPITT CO
assimilation prototype
(CAM/CHEM model)
Ave Arellano, NCAR/ACD
Support for
ARCTAS field
experiment
Other ongoing projects
Doppler radial velocity assimilation
Radar reflectivity assimilation
WRF column model for boundary layer
using ARM intensive obs.
Mesoscale reanalysis for T-Parc typhoons
Prediction with AM2, GFS, COAMPS
Other ongoing projects
OSSEs for chemical remote sensing in
CAM/chem and WRF/chem
Assimilation of cloud -moisture, -ice,
-fraction
Gulf of Mexico mesoscale eddies with MIT
ocean GCM
Quasi-operational ensemble prediction for
Taiwan
Other ongoing projects
Space weather, ionosphere, magnetosphere
prediction
Solar cycle prediction using helioseismology
Martian OSSEs and assimilation with
WRF/MARS
Maintaining Ensemble Diversity
Adaptive Inflation in
DART can nearly
eliminate tuning!
We’re looking for interesting partnerships.
Contact: [email protected]
Or see the DART web-site at:
www.image.ucar.edu/DAReS/DART