Upload
marybeth-beasley
View
222
Download
0
Tags:
Embed Size (px)
Citation preview
AMSR-E Ocean RainfallAlgorithm Status
AMSR-E Science Team MeetingHuntsville, AL2-3 June, 2010
C. KummerowColorado State University
~ 10 km
TB observed
TB model #1
TB model #2
TB model #3
Bayesian Inversion
~10 km
TB observed
TB model #3
TB model #2
TB model #1
Cloud Resolving Model Database
)|()()|( RTRTR bb PPP ×∝
TRMM PR/TMI & Model Database
GPROF_2008
GPROF2008 Algorithm (Ocean)
Create Geophysical Data Base (TRMM)
Start with non-raining background (1-D Var or OE)
Add observed PR rain profiles to non-raining background
Compute Tb at TMI channels and resolution and compare to observations
Adjust rain profiles to be consistent with PR and TMI
Create Rainfall/Tb Data Base (AMSR-E)
Use adjusted 4 km rain and non-raining profiles to compute Tb for AMSR-E
Create Database (raining and non-raining) pixels in 1K SST and 2 mm TPW bins.
Cluster profiles in each SST/TPW bin to 1000 self-similar Tbs for computational efficiency
Run Retrieval
Determine SST (Reynolds) & TPW (Sensor)
Compare observed Tb to dbase entries within ±1K (SST) and ±2 mm (TPW)
Weight profiles [observed difference vs uncertainty in channel from dbase]
Cloud profile database is partitioned into separate SST and TPW bins
Net change is a 2% increase from PR. Biggest change in colder SST with high TPW
Original Tb differences
Tb differences after adding
rain below PR threshold
Rain is added below threshold of PR and DSD assumption in PR is modified
Tb differences after DSD
modification
Tb from Background and Radar rain structureOptimal combined retrieval would leave all points on the 1:1 line
Database Preparation Steps
All pixels - raining and non-raining are retained for database (11,000,000 per month. Up to 32,000 in each SST/TPW bin)
For retrieval to run, all entries in SST and TPW bin are clustered based upon similar Sfc rainfall and Tb1-9. Use 1000 clusters for raining and 200 for non-raining. Numbers based on trial and error. Seek fewest clusters with no change ( < 1%) in final retrieval.
To account for colder environments (e.g. AMSR-E, SSMI), SST and TPW from existing profiles are reduced artificially by lowering SST while removing lowest layer of atmosphere corresponding to 6°/km. GPM will solve properly.
GPROF2008 continues to be a Bayesian algorithm consisting of:
P(R|Tb) = P(Tb|R)•P(R)/P(Tb)
Pixels are classified only by background SST and TPW. Database entries within ±1K in SST and ±2 mm in TPW are searched for potential solutions. Error covariances are established from fit between observed and simulated Tb in the a-priori database.
No more rain screens. All pixels are compared to database. Bayes’ theorem determines rain or no rain. Consequence: almost all pixels have a very small probability of rain.
No more convective/stratiform separation. Only necessary because CRM database was skewed to convective pixels. Entire code is now exceedingly simple.
GPROF 2008
GPROF 2008
+ TRMM radar
GPROF 2008
Hurricane FloydSeptember 13, 1999
GPROF 2008 example
GPROF2008
Hurricane Floyd Sep 3, 1999 TMI Algorithm Comparison 0.25 x 0.25 gridded
1.28 mm/day
RSS
GSmap
0.96 mm/day
0.99 mm/day
Surface Precipitation Jan 2007
CMORPH Diurnal Precipitation Variability
Coincident Overpass TPW Comparison
Coincident Overpass Comparison
TMI / AME TPWCoincident Overpass February, 8th, 2007
Pre-Processors
TMI_L1CTMI_1B11AME_L1CAME_L2ASSMI_L1C
GPROF08 input file
Spacecraft position Pixel Location Pixel Time, Tbs, EIA
Elevation Land masks Sea Surface Temp Clustered Profiles
GPROF2008
Native Binary Output
Land - All Coast – SSM/I Ocean - Each Sensor
HDF HDF-EOS
NetCDF
Profile Databases Ancillary Datasets
Format Converter*
Sensor Data
Temporary Files
Programs
Input Data
Output Files
SST directory spec Ancillary dir spec Sat, Sensor Info Channel Freqs, Errs Profile dbase names
Ancillary Info
GPROF 2008 Processing Algorithm
Output from GPROF 2008
• Basic DiagnosticsPixelStatusQualityFlagOceanSearchRadiusChiSquared
• Surface PrecipitationSurface PrecipitationSurface RainConvective Fraction
• Precipitation StructureProfile number [1-100] per hydrometeor type and freezing level
• Probability of PrecipitationProbability of Precipitation