TRMM Multi-satellite Precipitation Analysis (TMPA) (sometimes known as 3B42/43, TRMM product numbers) R. Adler, G. Huffman, D. Bolvin, E. Nelkin, D. Wolff NASA/Goddard Laboratory for Atmospheres with key input from F. Su (NASA/MSFC/U. of Washington)
Input Data Varying Over Time During TRMM Era 1998 2000 2002 2004 2006 TMI,PR SSM/I F13 SSM/I F14 SSM/I F15 SSMIS F16 SSMIS F17 AMSR-E AMSU-B N15 AMSU-B N16 AMSU-B N17 MHS N18 MHS MetOp GPCP IR Histograms CPC Merged IR
TRMM Multi-Satellite Precipitation Analysis (TMPA or 3B42 [TRMM product number]) [Adler/Huffman] 3-hr window with passive microwave (gaps filled with Geo- IR) calibrated by TRMM Research product uses TRMM radar information and monthly gauges Huffman et al., 2007, J. Hydromet. 9+ years ( 98-07 )of 3-hr analysis available. Most requested TRMM product from NASA archive
TMPA Design Calibrate microwave sensors with TRMM Best - TRMM Combined Instrument (2B31) in V.6 - TMI in real-time (RT) - Histogram matching -Calibrate IR with combined microwave - colder clouds rain more histogram matching - use month-long match-ups for stability Combination implemented as IR values filling gaps in microwave coverage - Data boundaries remain [V.6; non-real-time] Rescale combination fields to the monthly Satellite-Gauge combination - Reproduce large-area bias of wind-corrected gauge analysis
Validation using Kwajalein Radar TRMM has provided upgrades to the 10-cm weather radar on the U.S. Army base at Kwajalein Island and collected data since 1999 for validation - essentially the only long record of research-grade radar over tropical ocean - multiple attempts to solve problems in the data record The Relative Calibration Adjustment (RCA) technique has improved data - accounts for both gain and pointing angle 10 N changes - based on daily and hourly 95 th percentile signal from fixed list of ground clutter targets Radar precipitation estimates are gaugecalibrated - only use data within 100 km to minimize range effects - gauges not wind-corrected 9 N 8 N QuickTime and a TIFF (Uncompressed) decompressor are needed to see this picture. 167 E 168 E 169 E
Kwajalein Results 100 km, 3 hour (No averaging for satellite; average of 15-min data from radar) Average over 100-km radius (to avoid radar range effects) - computed from 3-hr accumulations - RT calibrated to V.6 with trailing month calibration Correlation 3h 0.84 Daily 0.89 5-Daily 0.90 1-Daily 0.91 Monthly 0.92 Seasonal 0.94 Bias seems relatively constant RMS decreases steadily - distribution pulls in toward the 1:1 mm/hr 100 km 3 hour
Kwajalein Results 100 km, Daily (No averaging for satellite; average of 15-min data from radar) Average over 100-km radius (to avoid radar range effects) - computed from 3-hr accumulations - RT calibrated to V.6 with trailing month calibration Correlation 3h 0.84 Daily 0.89 5-Daily 0.90 1-Daily 0.91 Monthly 0.92 Seasonal 0.94 Bias seems relatively constant RMS decreases steadily - distribution pulls in toward the 1:1 mm/hr 100 km Daily
Testing of TMPA (V6 and RT) in Hydrological Model Scattergrams of daily basin-averaged precipitation from gauged and satellite TMPA V6 mm/day over basin precipitation estimates for sub-basin 3802 in La Plata basin (189,000 km 2 )forthe period Jan 2003 to August 2006. Su (MSFC/U. of Wash.) TMPA-RT
Daily Simulated Streamflow for Basin 3802 (Area: 189, 300km 2 ) (Jan 2003 - Aug 30 2006) [Using VIC model] Simulated with satellite precip. Simulated with gauge precip. Su (MSFC/U. of Wash.) Improvement in TMPA-RT in 2005 leads to improved streamflow simulations as compared to pre-2005 Change in TMPA-RT
TMPA FUTURE Upgraded TMPA-RT is in beta test (will be initiated in Jan. 2008) - adding MHS (NOAA18 and MetOp) - RT calibrated to V.6 with trailing monthly gauge calibration (this will reduce or eliminate warm season positive biases over land due to PMW techniques) Future improvements - recalibrated F15 SSM/I - F16 (although issues) and later SSMIS - reprocessed AMSU-B record in V.7 TMPA (or sooner?) - Early RT run (~ 4 hr) - shift from SGI Unix to Linux platform Major Improvements: Interpolation/morphing scheme Extension to high latitudes
Expanding TMPA into a Globally Complete Product Utilize AMSU-B, SSMI/S algorithms/products being developed by other investigators Develop/adapt AIRS-based precipitation estimates (adapting, expanding from Susskind) Adapt AIRS-based algorithm to ATOVS Current Microwave coverage in N.H. winter
Example of AMSR and AIRS match-up and potential ATOVS coverage As a first step, we calibrated Susskind et al. (1997) AIRS to AMSR-E for June 2006 - compare AIRS, AMSR-E, calibrated AIRS for one 3-hr period - qualitative agreement, but not quantitative - added coverage by 3 ATOVS (gray) is essentially complete in 3 hr at high latitudes AIRS AMSR
Deriving Relations and Validating Results with CloudSat Radar Data Thanks to the A-Train: - CloudSat provides a curtain of cloud/precip data at all latitudes - AMSR-E provides 2D maps of precip AMSR-E B A AIRS B A - Here, sfc-based CloudSat echo matches AMSR-E rain area C C - CloudSat echo based above the sfc shows up in AIRS, but not AMSR-E CloudSat C B A Reflectivity Low High
Example of AIRS filling in a feature over snow where AMSR cannot reliably estimate AMSR-E Land precip feature 16 January 2004 mm/d Calibrated AIRS 16 January 2004 mm/d
Existing High Time Resolution Global Precipitation Product: GPCP One-Degree Daily Uses TOVS (AIRS after 2005) in High Latitudes January 1, 2005 Validating using data from Finland, and working on analyzing data from BALTRAD, and Canada
Conclusions TMPA was first example of multi-satellite precipitation analysis--now almost 10 years of 3-hr data Both research version (V6) and real-time (RT) versions being used in various applications (see upcoming flood and landslide talk) and to study statistics of extremes (our research) Minor changes will improve current real-time product in near future; major improvements will move us toward a global satellite analysis and provide more detailed time/space resolution Robert.F.Adler@nasa.gov http://precip.gsfc.nasa.gov
TMPA (3B42/43 ) COMPARISON TO GPCP AND TCI (2B31) V.6 3B43 differs from GPCP TCI calibrator for tropical ocean has less interannual variability than GPCP The average 3B43 tracks with the TCI, but: A negative bias develops in 2000 due to the start of NESDIS AMSU-B estimates - the issue is light precip detection over ocean - Aug. 2003 algorithm has a bigger detection bias - June 2007 algorithm has smaller detection bias - AMSU-B reprocessed before June 2007, but 3B43 hasn t been 7% TMPA not a climate variation product!
5. HIGH LAT. Prior Work (cont.) The alternative we chose is working with satellite soundings - Susskind et al. (1997) developed a calibrated cloud volume proxy from TOVS Precip = revised cloud depth * cloud fraction * ƒ ( latitude, season ) cloud top ht. ( scaled RH + scaled cloud fraction ) 0 = sat. sfc 500 mb 9 = dry 0 = overcast 4.5 = clear - the calibration is TOVS swath data vs. daily FGGE station precip data - results show low precip rates, very high fractional occurrence done as a regression uses instantaneous data as a proxy for daily data has only one sample for the day