Impacts of High-Resolution Land and Ocean Surface Initialization on Local
Download
Report
Transcript Impacts of High-Resolution Land and Ocean Surface Initialization on Local
Impacts of High-Resolution Land and
Ocean Surface Initialization on Local
Model Predictions of Convection
Jonathan L. Case
ENSCO, Inc./Short-term Prediction Research and Transition (SPoRT) Center
Huntsville, Alabama
Talk Outline
Experiment objectives
NASA Data and Tools
Goddard’s Land Information System (LIS)
Moderate Resolution Imaging Spectroradiometer (MODIS)
Simulation methodology
Preliminary results
Future work
NSSTC Data Assimilation Workshop
5 May 2009
Slide 1
Hypothesis and Experiment Objectives
Hypothesis: High-resolution land and water
datasets from NASA utilities can lead to
improvements in simulated summertime
pulse-type convection over the S.E. U.S.
Experiment objectives
Use NASA LIS to provide high-resolution
land surface initializations
Incorporate SPoRT MODIS composites for detailed
representation of sea surface temperatures (SSTs)
Demonstrate proof of concept in using these
datasets in local model applications with the
Weather Research and Forecasting (WRF) model
Quantify possible improvements to WRF
simulations
NSSTC Data Assimilation Workshop
5 May 2009
Slide 2
NASA Land Information System (LIS)
High-performance land surface modeling and data
assimilation system
Runs a variety of Land Surface Models (LSMs)
Integrates satellite, ground, and reanalysis data
to force LSMs in offline mode
Can run coupled to Advanced Research WRF
Data assimilation capability (EnKF) built-in
Modular framework enables easy substitution of
datasets, LSMs, forcings, etc.
Adopted by AFWA for operational use in WRF
Previous SPoRT work with LIS
Case et al. (2008) manuscript in J. Hydrometeor.
• Quantified positive impacts to WRF forecasts over Florida by
initializing model with LIS land surface output
• Focused on verification of primary meteorological variables
NSSTC Data Assimilation Workshop
5 May 2009
Slide 3
Land Surface Modeling with LIS
Inputs
Physics
Topography,
Soils
Land Surface Models
(e.g. Noah, VIC, SIB, SHEELS)
Land Cover,
Vegetation
Properties
NSSTC Data Assimilation Workshop
Soil
Moisture &
Temp
Evaporation
Weather/
Climate
Water
Resources
Homeland
Security
Meteorology
(Atmospheric
Forcing)
Snow
Soil Moisture
Temperature
Outputs Applications
Runoff
Data Assimilation Modules
Snowpack
Properties
5 May 2009
Military
Ops
Natural
Hazards
Slide 4
MODIS SST Product
RTG
Once daily
1/12 degree resolution
OSTIA
Once daily
5-km resolution
MODIS
Four times daily
1-km resolution
MODIS provides superior resolution
Quality check with the latency product
MODIS
Current weakness is high latency in areas
with persistent cloud cover
Collaboration with Jet Propulsion Laboratory
to improve product with AMSR-E data
Latency Product
NSSTC Data Assimilation Workshop
5 May 2009
Slide 5
Experiment Design
Run parallel WRF simulations
Once daily 27-h simulations, initialized at 0300 UTC
Control: Initial / boundary conditions from NCEP 12-km NAM model
Experimental: Same as Control, except
• Replace land surface data with LIS output fields
• Replace SSTs with SPoRT MODIS composites
Evaluation and Verification
Graphical and subjective comparisons
Verification using Meteorological Evaluation Tools (MET) package
• Developed by WRF Development Testbed Center
• Standard point/grid verification statistics
• Method for Object-Based Diagnostic Evaluation (MODE)
– Object-oriented, non-traditional verification method
– Summer convective precipitation verification
NSSTC Data Assimilation Workshop
5 May 2009
Slide 6
WRF Model Configuration
Model domain over Southeastern U.S.
Advanced Research WRF v3.0.1.1
4-km horizontal grid spacing
39 sigma-p levels from surface to 50 mb
Positive definite advection of scalars
Model physics options
•
•
•
•
•
Radiation: Dudhia SW and RRTM LW
Microphysics: WSM6
Land Surface: Noah LSM (same as LIS)
PBL: MYJ scheme
Cumulus parameterization: None
NSSTC Data Assimilation Workshop
ny = 311
• Min. spacing near surface of 0.004 sigma
• Max. spacing of 0.034 sigma
nx = 309
5 May 2009
Slide 7
LIS Offline “Spin-up” Run
LIS/Noah LSM run from 1 Jan 2004 to 1 Sep 2008
Same soil and vegetation parameters as in WRF
Atmospheric forcing
• 3-hourly Global Data Assimilation System analyses
• Hourly Stage IV radar + gauge precipitation products
Run long enough to allow soil to reach equilibrium state
Output GRIB-1 files to initialize WRF land surface variables
Incorporation of LIS data into WRF initial condition
Slight modifications to WRF Preprocessing System (WPS)
• Created Vtable.LIS & added LIS fields into METGRID.TBL file
• Soil moisture/temperature, skin temp, snow-water, land-sea mask
LIS data over-write NAM land surface data
Similarly, MODIS SSTs over-write NAM / RTG SSTs in WPS
NSSTC Data Assimilation Workshop
5 May 2009
Slide 8
Precip Verification with MET / MODE
Traditional grid point verification
Bias, threat score, HSS at various accumulation intervals / thresholds
Neighborhood precipitation verification
• Occurrence of precipitation threshold in a “box” surrounding a grid point
• Relaxes stringency and determines model skill at distance thresholds
MODE object classification
Resolves objects through convolution thresholding:
• Filter function applied to raw data using a tunable radius of influence
• Filtered field thresholded (tunable parameter) to create mask field
• Raw data restored to objects where mask meets/exceeds threshold
For our study, MODE is run with:
• 1-h, 3-h accumulated precipitation
• 5 mm, 10 mm, and 25 mm thresholds
• Radius of influence = 12 km (produced best object “matching”)
NSSTC Data Assimilation Workshop
5 May 2009
Slide 9
1 June 2008 Sensitivity Example:
0-10 cm Soil Moisture Differences
Control
(NAM)
LIS
LIS – NAM
NSSTC Data Assimilation Workshop
5 May 2009
Slide 10
1 June 2008 Sensitivity Example:
SST Differences
MODIS
Control
(RTG)
MODIS – RTG
NSSTC Data Assimilation Workshop
5 May 2009
Slide 11
1 June 2008 Sensitivity Example:
WRF 3-h Precip Diffs (06z to 06z)
Control
LIS – CNTL
NSSTC Data Assimilation Workshop
LIS
Stage IV
5 May 2009
Slide 12
1 June 2008 Sensitivity Example:
MODE 5-mm / 3-h precip “Objects”
Control
LISMOD
Grid
Grid
Grid
Grid
Fcst
Area
Area
Area
Area
hour
UnUnMatch
Match
match
match
3
0
1389
0
1389
6
0
169
0
206
9
0
1408
0
1421
12
0
2092
0
2066
15
54
1415
185
1351
18
3611
4415
3212
3864
21
5197
7602
6124
7034
24
1005
6362
1437
5868
27
1160
1013
102
2000
NSSTC Data Assimilation Workshop
Control
LISMOD
5 May 2009
Slide 13
Precip Verification: Jun-Aug 2008
3-hour Precipitation Bias
Control-5mm
LISMOD-5mm
1
0.9
LISMOD-10mm
1.8
0.6
3
6
9
Bias Score
15
18
21
27
1.2
1
0.8
3-hour Precipitation Bias
0.6
3
6
9
12
15
18
21
Control-25mm
24
27
• WRF generally has low
skill (Heidke SS, right)
1.5
1
6
3
6
9
12
15
18
21
10 mm
25 mm
2
0
3
6
9
12
15
18
21
-4
-6
-8
Forecast Hour
NSSTC Data Assimilation Workshop
24
27
0.12
0.1
0.08
0.06
0.04
0.2
0.02
0.18
LISMOD-10mm
0
0.16
0.14
3
6
9
12
15
18
21
24
27
Forecast Hour
0.12
0.1
0.08
0.06
3-hour Precipitation HSS
0.04
0.16
3
6
9
12
15
Control-25mm
LISMOD-25mm
18
21
24
27
Forecast Hour
0.14
0.12
0.1
0.08
0.06
0.04
0
3
6
9
12
15
18
21
24
27
9
12
15
18
21
24
27
LISMOD 3-h Precip
HSS Improvement
Forecast Hour
24 5 mm
27
Forecast Hour
0.14
0.02
LISMOD 3-h Precip Bias Improvement
4
-2
Control-10mm
0.16
0.02
0.2
0
0.18
LISMOD-25mm
Forecast Hour
2
80
Bias % Improvement
24
Forecast Hour
1.4
Bias Score
12
• LISMOD reduces
bias some, esp.
during daylight
hours (12-24 h)
H eidke Skill Score
3-hour Precipitation Bias
0.7
Control-10mm
H eidke Skill Score
0.8
• LISMOD incrementally
improves skill, esp. at
higher thresholds
25
HSS % Improvement
Bias Score
1.1
H eidke Skill Score
• WRF has high bias
1.2
0.5
3-hour Precipitation HSS
LISMOD-5mm
0.18
1.3
0.4
3
0.2
2.5
Control-5mm
0.2
1.4
1.6
3-hour Precipitation HSS
5 mm
10 mm
25 mm
20
15
10
5
0
-5
3
6
-10
5 May 2009
Forecast Hour
Slide 14
Neighborhood Precipitation Verification
Neighborhood Precip HSS: 5 mm (3 h) -1
0.4
0.35
Control: 20 km box
Control: 50 km box
Control: 80 km box
Neighbor Mean HSS % Improve: 20 km box
LISMOD: 20 km box
LISMOD: 50 km box
LISMOD: 80 km box
30
25
0.25
0.2
0.15
0.1
0.05
0
3
0.4
0.35
Mean HSS
0.3
6
9
12
15
18
NeighborhoodForecast
Precip
HSS:
Hour
Control: 20 km box
Control: 50 km box
Control: 80 km box
21
10 mm
24
27
(3 h) -1
% Improvement
Mean HSS
0.3
20
10 mm
25 mm
15
10
5
0
-5
3
6
9
12
15
18
21
24
27
-10
LISMOD: 20 km box
LISMOD: 50 km box
LISMOD: 80 km box
-15
Forecast Hour
0.25
0.2
0.15
Neighbor Mean HSS % Improve: 80 km box
0.1
0.05
30
0
0.4
0.35
0.3
6
9
12
15
18
NeighborhoodForecast
Precip
HSS:
Hour
Control: 20 km box
Control: 50 km box
Control: 80 km box
21
25 mm
24
27
(3 h) -1
LISMOD: 20 km box
LISMOD: 50 km box
LISMOD: 80 km box
0.25
0.2
% Improvement
3
Mean HSS
5 mm
5 mm
25
10 mm
20
25 mm
15
10
5
0.15
0
0.1
-5
3
6
9
12
15
18
21
24
27
0.05
0
3
6
9
12
15
18
Forecast Hour
NSSTC Data Assimilation Workshop
21
24
27
Forecast Hour
5 May 2009
Slide 15
MODE Precip Object Verification
3-h Accumulated Precip Objects
1-h Accumulated Precip Objects
MODE: 10-mm Precip (Un)Matched Area
200000
70000
60000
# of grid points
# of grid points
250000
MODE: 10-mm Precip (Un)Matched Area
Control Match
LISMOD Match
Control Unmatch
LISMOD Unmatch
150000
100000
50000
50000
40000
30000
20000
10000
0
0
3
6
9
12
15
18
21
24
27
12 13 14 15 16 17 18 19 20 21 22 23 24
Forecast Hour
Forecast Hour
MODE: 10-mm (Un)Matched % Change
MODE: 10-mm (Un)Matched % Change
10
Matched
% Change
0
3
6
9
12
15
18
-10
-15
-20
Forecast Hour
NSSTC Data Assimilation Workshop
21
24
27
% Change
Unmatched
5
-5
Control Match
LISMOD Match
Control Unmatch
LISMOD Unmatch
30
25
20
15
10
5
0
-5
-10
-15
-20
Matched
Unmatched
12 13 14 15 16 17 18 19 20 21 22 23 24
Forecast Hour
5 May 2009
Slide 16
Summary / Future Work
Simulation methodology using NASA data and tools
Land Information System land surface data
MODIS SST composites
Provides high-resolution representation of land/water surface,
consistent with local & regional modeling applications
Ongoing / Future efforts
Conduct rigorous model verification
• Use MET to generate objective statistics and object-oriented
output for precipitation
Evaluate how combined NASA surface datasets can lead to
improved short-term local model forecasts of convection
NASA / SPoRT website: http://weather.msfc.nasa.gov/sport/
NSSTC Data Assimilation Workshop
5 May 2009
Slide 17
Backups
NSSTC Data Assimilation Workshop
5 May 2009
Slide 18
LIS High-Level Overview
Uncoupled or
Analysis Mode
Coupled or
Forecast Mode
Station Data
Global, Regional
Forecasts and
(Re-) Analyses
Satellite Products
ESMF
LSM Physics
(e.g. Noah, VIC, SIB,
SHEELS)
Data Assimilation
• Soil moisture
• LST, Snow cover
NSSTC Data Assimilation Workshop
WRF
LSM First Guess /
Initial Conditions
5 May 2009
Slide 19
SPoRT MODIS SST Composites
Real-time, 1-km SST product
Composites available up to four times per day
• 0400, 0700, 1600, and 1900 UTC
Primarily over Gulf of Mexico, western Atlantic waters, and larger
lakes (e.g. Florida’s Lake Okeechobee)
GRIB-1 files posted to publicly available ftp site
• Sub-sampled to 2-km spacing for model applications
Compositing technique
Build complete SST composite with multiple Earth Observing System
satellite passes (both Aqua and Terra)
At each pixel, examine 5 most recent readings:
• Take average of 3 warmest readings
• This method helps to eliminate cloud contamination
NSSTC Data Assimilation Workshop
5 May 2009
Slide 20
Tropical Storm Fay:
Rainfall and Dramatic Soil Moistening
NSSTC Data Assimilation Workshop
5 May 2009
Slide 21
Tropical Storm Fay:
Rainfall and Dramatic Soil Moistening
NSSTC Data Assimilation Workshop
5 May 2009
Slide 22