the evaluation of a passive microwave-based satellite rainfall estimation algorithm with an ir-based...
TRANSCRIPT
The Evaluation of a Passive Microwave-Based Satellite Rainfall Estimation Algorithm with an IR-Based Algorithm at Short time Scales
Robert Joyce RS Information Systems
John Janowiak Climate Prediction Center/NCEP/NWS
Phillip Arkin ESSIC – University of Maryland
Pingping Xie Climate Prediction Center/NCEP/NWS
2nd International Precipitation Working Group October 25-28, 2004
*CPC Morphing Technique
1. CMORPH concept
2. CMORPH methodology
3. Validation
4. Improvement potential and future work
5. Conclusions
Outline
“CMORPH” is not a precipitation estimation technique but rather a method that creates spatially & temporally complete information using existing precipitation products that are derived from passive microwave observations.
At present, precipitation estimates are used from various passive microwave sensor types on 8 platforms:
• AMSU-B (NOAA 15, 16, 17)• SSM/I (DMSP 13, 14, 15)• TMI (TRMM – NASA/Japan)
• AMSR-E (Aqua)
NOAA/NESDIS (Ferraro et al)
TMI rainfall estimates from NASA’s 2A12 algorithm (Kummerow et al., 1996) Goddard Profiling (GPROF) version 5, soon version 6
AMSR-E precipitation estimates from GPROF-6 rainfall algorithm run at NOAA/NESDIS/ORA.
SSMI precipitation estimates from NOAA/NESDIS/ORA GPROF-6 SSMI rainfall algorithm.
AMSU-B rainfall estimates from new NESDIS/ORA AMSU-B rainfall algorithm (Weng et al., 2003)
Half hourly, 0.0727 lat/lon (8 km at equator) resolution arrays (separate for each sensor type) are assigned the nearest rainfall estimate within swath regions
Averaging of retrieval estimates within same grid points (AMSR-E and TMI only)
Anomalous microwave estimated rainfall screened with NESDIS Satellite Services Division (SSD) daily Interactive Multi-sensor Snow and Ice Mapping System (IMS) product
• PMW rainfall gridded to PMW rainfall gridded to 8km resolution8km resolution
3-hr mosaic: good coveragebut time of obs. varies by 3 hrs
MANY thanks to NESDIS/OSDPD MANY thanks to NESDIS/OSDPD & R. Ferraro (NESDIS/ORA)& R. Ferraro (NESDIS/ORA)
• PMW rainfall gridded to PMW rainfall gridded to 8km resolution8km resolution
3-hr mosaic: good coveragebut time of obs. varies by 3 hrs
IR Data
• All 5 geostationary meteorological satellites• Obtained via McIDAS• Merged into ½ hr global (60N-60S), ~ 4 km maps
• Corrections for limb darkening & parallax applied
Refs: Janowiak et al., Bull. Amer. Meteor. Soc., Feb 2001) Joyce et al., J. Appl. Meteor., Apr 2001
Most recent 8 days (each ½ hr period) available at:
ftpprd.noaa.gov: pub/precip/global_full_res_IR
Propagated forward
t = 0 t + 0.5 hr t + 1hr t + 1.5 hr
0330 GMT 0400 GMT 0430 GMT 0500 GMT
t = 0 t + 0.5 hr t + 1 hr t + 1.5 hr
0330 GMT 0400 GMT 0430 GMT 0500 GMT
Propagated backward
a
b
Observed
c 0330 GMT 0400 GMT 0430 GMT 0500 GMT
0.67 0.33
0.670.33
Time InterpolationWeights
Propagated & “Morphed”
The microwave precip. features are propagated forward and backward in time.
Then the features are “morphed” by linearly interpolating in time as depicted in panel “c”
each grid value for the 0400 & 0430 GMT images is obtained by:
P0400= 0.67 x P0330 + 0.33 x P0500
P0430= 0.33 x P0330 + 0.67 x P0500
Propagated forward
t = 0 t + 0.5 hr t + 1hr t + 1.5 hr
0330 GMT 0400 GMT 0430 GMT 0500 GMT
t = 0 t + 0.5 hr t + 1 hr t + 1.5 hr
0330 GMT 0400 GMT 0430 GMT 0500 GMT
Propagated backward
a
b
Observed
c 0330 GMT 0400 GMT 0430 GMT 0500 GMT
0.67 0.33
0.670.33
Time InterpolationWeights
Propagated & “Morphed”
The microwave precip. features are propagated forward and backward in time.
Then the features are “morphed” by linearly interpolating in time as depicted in panel “c”
each grid value for the 0400 & 0430 GMT images is obtained by:
P0400= 0.67 x P0330 + 0.33 x P0500
P0430= 0.33 x P0330 + 0.67 x P0500
Propagated forward
t = 0 t + 0.5 hr t + 1hr t + 1.5 hr
0330 GMT 0400 GMT 0430 GMT 0500 GMT
t = 0 t + 0.5 hr t + 1 hr t + 1.5 hr
0330 GMT 0400 GMT 0430 GMT 0500 GMT
Propagated backward
a
b
Observed
c 0330 GMT 0400 GMT 0430 GMT 0500 GMT
0.67 0.33
0.670.33
Time InterpolationWeights
Propagated & “Morphed”
The microwave precip. features are propagated forward and backward in time.
Then the features are “morphed” by linearly interpolating in time as depicted in panel “c”
each grid value for the 0400 & 0430 GMT images is obtained by:
P0400= 0.67 x P0330 + 0.33 x P0500
P0430= 0.33 x P0330 + 0.67 x P0500
Satellite - CPC gauge analysisSatellite - CPC gauge analysis
Merged PMW – only & RadarMerged PMW – only & Radar
Difference from gauge analysisDifference from gauge analysis
Satellite - CPC gauge analysisSatellite - CPC gauge analysis
CMORPH & RadarCMORPH & Radar
Difference from gauge analysisDifference from gauge analysis
• DJF 2003-2004 statistics using DJF 2003-2004 statistics using Australian Bureau of Australian Bureau of Meteorology 0.25 degree Meteorology 0.25 degree lat/lon daily rain gauge lat/lon daily rain gauge analyses for validationanalyses for validation
• Red line = CMORPHRed line = CMORPH
• Blue line = Merged PMWBlue line = Merged PMW
• Black = gauge analyses Black = gauge analyses
Limitations
• Present estimation algorithms cannot retrieve precip. over snow orice covered surfaces
- New algorithms being developed (Liu, Ferraro)
• Data Latency: ~ 18 hours past real-time
• Will not presently detect precip. that develops, matures & decays between microwave scans
• Limits on how far back data can be processed … early 1990’s?
• Hourly, 0.25 degree lat/lon CMORPH Hourly, 0.25 degree lat/lon CMORPH timestamp = 1 (30 minutes from timestamp = 1 (30 minutes from nearest PMW scan) correlation with nearest PMW scan) correlation with Stage II radar rainfall (top panel)Stage II radar rainfall (top panel)
• Hourly, 0.25 degree lat/lon IR-based Hourly, 0.25 degree lat/lon IR-based PMW/IR combined frequency PMW/IR combined frequency matching rainfall estimation matching rainfall estimation (IRFREQ) correlation with Stage II (IRFREQ) correlation with Stage II radar rainfall (2radar rainfall (2ndnd from top) from top)
• CMORPH – radar rainfall correlation CMORPH – radar rainfall correlation minus IRFREQ minus IRFREQ
• Correlation pair counts Correlation pair counts
• Hourly, 0.25 degree lat/lon Hourly, 0.25 degree lat/lon CMORPH timestamp = 2 (60 CMORPH timestamp = 2 (60 minutes from nearest PMW scan) minutes from nearest PMW scan) correlation with Stage II radar correlation with Stage II radar rainfall (top panel)rainfall (top panel)
• Hourly, 0.25 degree lat/lon Hourly, 0.25 degree lat/lon IRFREQ correlation with Stage II IRFREQ correlation with Stage II radar rainfall (2radar rainfall (2ndnd from top) from top)
• CMORPH – radar rainfall CMORPH – radar rainfall correlation minus IRFREQ correlation minus IRFREQ
• Correlation pair counts Correlation pair counts
• Hourly, 0.25 degree lat/lon Hourly, 0.25 degree lat/lon CMORPH timestamp = 3 (90 CMORPH timestamp = 3 (90 minutes from nearest PMW scan) minutes from nearest PMW scan) correlation with Stage II radar correlation with Stage II radar rainfall (top panel)rainfall (top panel)
• Hourly, 0.25 degree lat/lon Hourly, 0.25 degree lat/lon IRFREQ correlation with Stage II IRFREQ correlation with Stage II radar rainfall (2radar rainfall (2ndnd from top) from top)
• CMORPH – radar rainfall CMORPH – radar rainfall correlation minus IRFREQ correlation minus IRFREQ
• Correlation pair counts Correlation pair counts
• Hourly, 0.25 degree lat/lon Hourly, 0.25 degree lat/lon CMORPH timestamp = 4 (120 CMORPH timestamp = 4 (120 minutes from nearest PMW scan) minutes from nearest PMW scan) correlation with Stage II radar correlation with Stage II radar rainfall (top panel)rainfall (top panel)
• Hourly, 0.25 degree lat/lon Hourly, 0.25 degree lat/lon IRFREQ correlation with Stage II IRFREQ correlation with Stage II radar rainfall (2radar rainfall (2ndnd from top) from top)
• CMORPH – radar rainfall CMORPH – radar rainfall correlation minus IRFREQ correlation minus IRFREQ
• Correlation pair counts Correlation pair counts
• The cumulative percentage of half The cumulative percentage of half hourly periods sampled for an hourly periods sampled for an eight day period, in 30 minute eight day period, in 30 minute increments to nearest past/future increments to nearest past/future scan, instantaneous (timestamp = scan, instantaneous (timestamp = 0, top) 0, top)
• cumulative % sampled within 30 cumulative % sampled within 30 minutes of half hourly frame minutes of half hourly frame (timestamp <= 1, middle)(timestamp <= 1, middle)
• cumulative % sampled within 60 cumulative % sampled within 60 minutes of half hourly frame minutes of half hourly frame (timestamp <= 2)(timestamp <= 2)
• Half hourly, 0.25 degree lat/lon Half hourly, 0.25 degree lat/lon CMORPH correlation against CMORPH correlation against withheld MWCOMB rainfall 23 June – withheld MWCOMB rainfall 23 June – 6 August 2004. Temporal distance of 6 August 2004. Temporal distance of CMORPH to nearest PMW scan = 30 CMORPH to nearest PMW scan = 30 minutes (timestamp = 1, top) minutes (timestamp = 1, top)
• Half hourly, 0.25 degree lat/lon Half hourly, 0.25 degree lat/lon IRFREQ correlation against withheld IRFREQ correlation against withheld MWCOMB rainfall (timestamp = 1, 2MWCOMB rainfall (timestamp = 1, 2ndnd from top) from top)
• CMORPH correlation minus IRFREQ CMORPH correlation minus IRFREQ (3(3rdrd from top) from top)
• # of correlation pairs (bottom)# of correlation pairs (bottom)
• Half hourly, 0.25 degree lat/lon Half hourly, 0.25 degree lat/lon CMORPH correlation against CMORPH correlation against withheld MWCOMB rainfall, temporal withheld MWCOMB rainfall, temporal distance of CMORPH to nearest distance of CMORPH to nearest PMW scan = 60 minutes (timestamp PMW scan = 60 minutes (timestamp = 2, top) = 2, top)
• Half hourly, 0.25 degree lat/lon Half hourly, 0.25 degree lat/lon IRFREQ correlation against withheld IRFREQ correlation against withheld MWCOMB rainfall (timestamp = 2, MWCOMB rainfall (timestamp = 2, 22ndnd from top) from top)
• CMORPH correlation minus IRFREQ CMORPH correlation minus IRFREQ (3(3rdrd from top) from top)
• # of correlation pairs (bottom)# of correlation pairs (bottom)
• Half hourly, 0.25 degree lat/lon Half hourly, 0.25 degree lat/lon CMORPH correlation against CMORPH correlation against withheld MWCOMB rainfall, temporal withheld MWCOMB rainfall, temporal distance of CMORPH to nearest distance of CMORPH to nearest PMW scan = 90 minutes (timestamp PMW scan = 90 minutes (timestamp = 3, top) = 3, top)
• Half hourly, 0.25 degree lat/lon Half hourly, 0.25 degree lat/lon IRFREQ correlation against withheld IRFREQ correlation against withheld MWCOMB rainfall (timestamp = 3, MWCOMB rainfall (timestamp = 3, 22ndnd from top) from top)
• CMORPH correlation minus IRFREQ CMORPH correlation minus IRFREQ (3(3rdrd from top) from top)
• # of correlation pairs (bottom)# of correlation pairs (bottom)
• Half hourly, 0.25 degree lat/lon Half hourly, 0.25 degree lat/lon CMORPH correlation against CMORPH correlation against withheld MWCOMB rainfall, temporal withheld MWCOMB rainfall, temporal distance of CMORPH to nearest distance of CMORPH to nearest PMW scan = 120 minutes PMW scan = 120 minutes (timestamp = 4, top) (timestamp = 4, top)
• Half hourly, 0.25 degree lat/lon Half hourly, 0.25 degree lat/lon IRFREQ correlation against withheld IRFREQ correlation against withheld MWCOMB rainfall (timestamp = 4, MWCOMB rainfall (timestamp = 4, 22ndnd from top) from top)
• CMORPH correlation minus IRFREQ CMORPH correlation minus IRFREQ (3(3rdrd from top) from top)
• # of correlation pairs (bottom)# of correlation pairs (bottom)
Daily times series of correlation Daily times series of correlation comparing IRFREQ, CMORPH, comparing IRFREQ, CMORPH, and CMORPH-IR 0.25 degree and CMORPH-IR 0.25 degree lat/lon rainfall estimates with lat/lon rainfall estimates with same high-quality rain gauge and same high-quality rain gauge and radar analyses over the U.S. for radar analyses over the U.S. for the 7 May – 27 July 2004 period.the 7 May – 27 July 2004 period.
IRFREQ = blue linesIRFREQ = blue lines
CMORPH = green linesCMORPH = green lines
CMORPH-IR = red linesCMORPH-IR = red lines
• Improve accuracy of CMORPH PMW rainfall vectors
•Continue to investigate model winds – GMORPH
• Continue investigation into the development of CMORPH – IR
• •
FUTURE WORK
Further Information
Satellite Estimated Rainfall Validation over United States:http://www.cpc.ncep.noaa.gov/products/janowiak/us_web.shtml Australia:http://www.bom.gov.au/bmrc/wefor/staff/eee/SatRainVal/dailyval.html CMORPHWeb: http://www.cpc.ncep.noaa.gov/products/janowiak/cmorph.html (includes data access info.)
E-mail: [email protected] or [email protected]
Paper: Joyce, R. J., J. E. Janowiak, P. A. Arkin and P. Xie, 2004: CMORPH: A method that produces global precipitationestimates from passive microwave and infrared data at high spatial and temporal resolution. J. Hydromet. Vol. 5, No. 3,
pages 487-503.
The End – Thank You