Satellite-derived Rainfall Estimates over the Western U.S.: Fact or Fiction? John Janowiak Bob Joyce Pingping Xie Phil Arkin Mingyue Chen Yelena Yarosh
OUTLINE 1.Brief review of IR & passive microwave info. 2.Describe “CMORPH” 3.Validation (US & Australia) 4.Simple gauge vs. satellite sampling study 5.A look at western US precip 6.Conclusions & on-going work
Surface Infrared Geostationary & Polar
Surface Passive Microwave “Emission” Detects thermal emission from raindrops - most physically direct - over ocean only - polar platform only
Surface Freezing Level Passive Microwave “Scattering” Upwelling radiation from Earth’s surface Upwelling radiation is scattered by ice particles in the tops of convective clouds - land & ocean - polar platform only
IR: great sampling / provides poor estimate of rainfall MW: poor sampling / provides good estimate of rainfall >>>>> Combine them to meld strengths of each Others have done this – IR used to produce precip. estimate when MW data unavailable - Turk (NRL, Monterey), - Adler & Huffman (GSFC), - Gao, Hsu,Sarooshian (U. AZ)
3-hr mosaic: good coverage but time of obs. varies by 3 hrs
CPC Morphing Technique “CMORPH” Spatial Grid: o lat/lon (8 km at equator) Temporal Res’n: 30 minutes Domain: Global (60 o N - 60 o S) Period of record: Dec present “CMORPH” uses IR only as a transport vehicle. Underlying assumption is that error in using IR to transport percip. features is < error in using IR to estimate precip. Bob Joyce! Paper (Joyce et al.) submitted to J. Hydrometeor.
“CMORPH” is NOT a precipitation estimation technique but rather a technique that creates spatially & temporally complete information using existing precipitation products that are derived from passive microwave observations. ` At present, precipitation estimates are used from 3 sensor types on 7 platforms: AMSUB (NOAA 15, 16, 17 ) SSM/I (DMSP 13, 14, 15) TMI (TRMM ) Soon: AMSR (“ADEOS-II”) & AMSR-E (“Aqua”)
Remote Sensing Errors & Limitations - Indirect estimates inferred radiometrically - Instrument calibration - Conversion from retrieval to rain rate (algo.) - Temporal sampling ERRORS: LIMITATIONS: - Measurements not temporally continuous - Depending on instrument only convective (“scattering’) precip. may be sensed
Raingauge Errors & Limitations - Wind & gauge exposure effects - Human element (time, accuracy) - Automated (calibration, maintenance) - Biological contamination ERRORS: LIMITATIONS: - Representativeness of area (point value) - Spatially incomplete - Available frequency (daily, 6-hr)
0.50 Box Mean Precip: 0.15” Std. Dev. : 0.22 “ Min. precip : 0 Max. precip. : 0.95” o x1 o box in s-central TN (July14, 2003) Distribution of Rainfall by Amount
BIAS RMSE CORR 1 gauge gauge Radar Cmorph
40% 57% 66% 50% 62% 70%
Gauge 40% 44%
: Synthetic Data Sampling Study Question: Are there situations when an estimate from satellite is ‘better’ for assessing area-mean precipitation than a measurement from gauge(s)? Design: - randomly assign precip to 169 locations (13 x 13 array) - 50% of locations have “0” precip. - Repeat for 1000 ‘days’ - Daily “truth” is the 169 value mean
: Assumptions: - gauge measurement is perfect - gauge values are totally representative of the area sampled by satellite ie. area avg. - multiple gauges in an area are distributed optimally
Approach (overly?) simplistic: - ‘real-world’ nonzero rainfall distribution characteristics not modeled - on average, the % of locations with rain over an area is < 50% used here - rainfall ‘generators’ exist that more nearly duplicate the statistics of actual rainfall over time-space - much work on aspects of this topic done in hydro. & satellite sampling communities (Bell et al. :1990, 1996, 2003)
Samples of synthetic precipitation within a 1 o x 1 o lat/lon box at satellite resolution Precip amounts of 0 to 1 chosen randomly; impose condition that 50% are = “0”
1 o x 1 o lat/lon box containing 169 satellite pixels X
X X 2 Gauges
X XX 3 Gauges
X X XX 4 Gauges
X X XX X 5 Gauges
X X XXX X XX X 9 Gauges
X X XXX X XX X XX XX 13 Gauges
X X XXX X XX X XX XX XX XXX XX XXX XX 25 Gauges
1 gauge 2 gauges 9 gauges Time series of absolute error (1 st 100 days) Light blue: satellite with 200% positive bias Dark blue: satellite with 100% positive bias Green: satellite with 50% positive bias Red: satellite with 10% positive bias
Light blue: 9 gauges Dark blue: 5 gauges Green: 3 gauges Red: 1 gauge 200% error 10%/ error (satellite) 50% error 100% error (90%) “Perfect” Gauge 531 Point of 50% error accumulation 9
1 gauge 2 gauges 9 gauges Light blue: sat.ellite with 0-200% pos. random error Dark blue: satellite with 0-100% pos. random error Green: satellite with 0- 50% pos. random error Red: satellite with 0-10% pos. random error Time series of absolute error
(90%) 0-10% error 0-50% error 0-100% error 0-200% error Light blue: 9 gauges Dark blue: 5 gauges Green: 3 gauges Red: 1 gauge Error at 50% point “Perfect” Gauge
RMSE satellite (200% + bias) gauge ~10% of earth satellite (100% + bias) satellite (0-200% random) gauges gauges gauges gauges satellite (50% + bias) satellite (0-100% random) gauges gauges satellite (0-50% random) gauges satellite (10% + bias)
Number of HADS/RFC stations per ¼ degree lat/lon grid box (9/10/2003)
10% of earth (60N-60S) 29% of land area ( “ ) “CAMS” - 1 or more gauges per 1 o Grid 10% of earth (60N-60S); 29% of land “CAMS” - 2 or more gauges per 1 o Grid
Crude RH Adjustment to CMORPH (Aug 2003) Scofield, 1987 Rosenfeld and Mintz (1988) McCollum et al. (2000)
CMORPH vs. gauge over ‘NAME’ zones
CMORPH with RH adjustment vs. gauge over ‘NAME’ zones
Time series of statistics over 9 NAME Zones Evap. adjusted
Conclusions Fact or Fiction? 1.CMORPH estimates compare quite favorably to raadar estimates over the US and to gauge analyses over the US and Australia. 2.Satellite estimates of rainfall can be useful over the western U.S. (and elsewhere) – perhaps better than gauge data in some situations 3.Many satellite techniques overestimate rainfall considerably in semi-arid regions during the warm season, but an RH adjustment is promising.
Work in Progress 1.Refine & implement evaporation adjustment 2.Investigate use of model winds to advect rainfall - more accurate results? - allow reprocessing to early 1990’s - reduce processing time substantially 3.Incorporate microwave rainfall estimates from new instruments (AMSR, AMSR-E) 4. Investigate derivation of advection vectors from microwave data - temporal resolution to 10 minutes?
‘0’ precip 0 < precip < 1 Long-term Box Mean = 0.25 So, 200% error = % error = % error = 0.125
Correlation with MW availability