p0305 [Compatibility Mode]
Transcription
p0305 [Compatibility Mode]
Evaluating the drought monitoring capabilities of rainfall estimates for Africa Chris Funk Pete Peterson Amy McNally U.S. Department of the Interior U.S. Geological Survey What makes good rainfall estimates go bad? 1. 2. 3. 4. 5. Sparse station data Inconsistent networks Complex mean fields Biased satellite estimates Inhomogeneous satellite inputs 1 Climate Hazard Group Climatology Method Dense gauge observations Elevation, latitude, longitude Satellite mean fields Sparse gauge observations Satellite anomaly fields 1 1 FEWS NET TRMM-IR Precipitation (FTIP) Quasi-global (±60°N/S), 0.05°, pentads Built around 0.05°rainfall mean fields FEWS NET Climatology (FCLIM) Two components TRMM V6/RT as percentage (TR%) Cold cloud duration IR estimates as percentages (IR%) Global CCD models trained against TRMM 3B42 pentads CCD threshold, slope, intercept for each month FTIP = (0.5 TR% + 0.5 IR%) * FCLIM Unbiased IR precipitation (UIRP) = IR% * FCLIM Hispaniola Rainfall dekads: TRMM and FTIP Satellite % X FCLIM --------------Improved Rainfall Estimates 1 TRMM FTIP 1 Building better climatologies Use regional moving window regressions to model long term mean monthly rainfall as a function of Elevation, Slope, latitude, longitude, … TRMM, Land Surface Temperatures, Infrared brightness temperatures, …. Use standard interpolation approaches (kriging or IDW to blend in station anomalies) 1 Monthly means of TRMM v6, LST and IR Local correlations w/ satellite mean fields much better than elevation or slope 1 FCLIM Variance Explained 1 1 FCLIM Annual Means 1 Climatology Validations Region N-stns Stn Mean Model Mean MBE MAE Pct MBE Pct MAE R2 FCLIM Combined Colombia Afghanistan SE Asia Ethiopia Sahel Mexico 2216 194 22 6 76 104 1814 97 168 35 122 97 86 77 99 159 34 144 94 82 78 -1 8 1 -22 3 3 -1 20 30 9 37 10 9 23 -1 5 3 -18 3 4 -2 20 18 25 30 10 11 30 0.69 0.84 0.53 0.28 0.91 0.90 0.65 CRU Combined Colombia Afghanistan SE Asia Ethiopia Sahel Mexico 2216 194 22 6 76 104 1814 97 168 35 122 97 86 77 106 174 45 152 101 90 75 -9 -6 -10 -30 -4 -4 1 32 46 20 56 23 20 24 -9 -4 -27 -25 -4 -4 2 32 28 57 46 23 24 31 0.51 0.59 0.18 0.25 0.68 0.74 0.60 Worldclim Combined Colombia Afghanistan SE Asia Ethiopia Sahel Mexico 2216 194 22 6 76 104 1814 97 168 35 122 97 86 77 106 178 41 155 97 88 79 -9 -11 -6 -33 0 -2 -2 26 31 18 50 20 19 17 -9 -6 -17 -27 0 -3 -2 27 19 52 41 21 22 23 0.61 0.82 0.18 0.42 0.72 0.75 0.78 1 Drought monitoring validation study for Africa Modeled after previous analyses by Grimes and Dinku 0.25°block kriged monthly station data (2001-2010) Interpolated as %, multiplied by FCLIM Study looks at drought hits, misses, false alarms and correct negatives Drought event defined as a two-month period with less than 190 mm of rainfall Focused on main rainy season rainfall For each grid cell, Identify three month period with max average rainfall Build time-series of two-month rainfall combinations 1 Building time-series of main season rainfall Identify main rainy season (i.e. MAM) For each year, construct three 2-month combinations: e.g. March+April, April+May, March+May Do this for each year to obtain ~27 combinations: 2001MA, 2001AM, 2001MM, 2002MA, 2002AM, 2002MM, 2003MA, 2003AM, 2003MM, 2004MA, 2004AM, 2004MM, 2005MA, 2005AM, 2005MM, 2006MA, 2006AM, 2006MM, 2007MA, 2007AM, 2007MM, 2008MA, 2008AM, 2008MM, 2009MA, 2009AM, 2009MM, 2010MA, 2010AM, 2010MM Focus on rainfall during germination/grain filling 1 Main growing season correlation maps 1 Main growing season mean bias maps 1 Main growing season mean absolute error maps 1 Hits and misses Kriged Station Data 190 mm False Alarm Correct Negative 190 mm Hit Miss Satellite Estimate Main growing season hits 1 Main growing season correct negatives 1 Main growing season misses 1 Main growing season False Alarms 1 Conclusions 1 Incorporating climatologies improves skill FCLIM unbiasing easily reproducable with TARCAT, ECMWF, … Fitting CCD models using TRMM 3B42 seems worthwhile RFE2, FTIP, and TARCAT showed best-correlations FTIP and then the TRMM showed the smallest mean bias TARCAT, RFE2 and FTIP had lower MAE TARCAT, RFE2, FTIP, ECMWF do a good job discriminating hits and correct negatives Misses Some in the Sahel (RFE2, TRMM, RFE2) Sudan (ECMWF) False Alarms: Part of east Africa in each Sahel in ECMWF Focus on rainfall during germination/grain filling