Robert Adler Huan Wu University of Maryland, USA

Transcription

1 Evaluation of Satellite Precipitation Products Using Global Flood Calculations Robert Adler Huan Wu University of Maryland, USA Using a global hydrological model and satellite rainfall to evaluate satellite rainfall products for real-time applications

2 Global Flood Calculations Using Satellite Rainfall and Hydrological Model Rainfall input from satellite information---currently using TRMM Multi-satellite Precipitation Analysis [TMPA/3B42] (TRMM data used to adjust rain estimates from polar orbit PMW) 3-hr 0.25 Lat./Long Other available candidate rainfall products, e.g., CMORPH (NWS/CPC), GSMaP Getting ready for Global Precipitation Measurement (GPM) IMERG multi-satellite product automatic re-processing to beginning of TRMM era (1998) for consistent long record Also using global NWP output to extend flood predictions out to 5 days or beyond

3 Global Flood Monitoring System (GFMS) Global Real-time Flood Calculations Using Satellite Rainfall and Hydrological Model TMPA TRMM calibrating rainfall from other satellites as forerunner to GPM TRMM/GPM rainfall into land surface and routing models for water depth and stream flow calculations compared to flood thresholds Global Flood Detection (12 km) Indus River basin Aug Experimental Inundation Mapping at 1 km resolution Time Histories at a Point

4 Rainfall and Flooding in Japan with Typhoons Phanfone and Vongfong (October 2014) Streamflow above Flood Threshold (m 3 /s) 4 October 3-day rain ending 7 October Phanfone 3-day rain Streamflow above Flood Threshold (m 3 /s) 12 October 5 October 13 October 6 October 3-day rain ending 15 October Vongfong 14 October 7 October 15 October

5 Time Histories of Streamflow at Two rivers in Japan (3-17 Oct.) 6 October 12 km resolution Streamflow Tone River X Streamflow Flood Threshold X 5 Oct. 17 Oct. Streamflow above Flood Threshold (m 3 /s) Panels at right show time histories of calculated streamflow in two Japan rivers indicating impact of the two typhoons, with earlier storm (Phanfone) showing higher values in Tone River (near Tsukuba). Also seen is time shift from south to north as storms travel across Japan. Wu, H., R. Adler et al. (2014), Real-time global flood estimation using satellite-based precipitation and a coupled land surface and routing model, Water Resour. Res. Streamflow Kinokawa River Streamflow Flood Threshold 5 Oct. 17 Oct.

6 Streamflow and Inundation Calculations at 1km Resolution Streamflow Tsukuba 6 October Tone River Tokyo X Typhoon Phanfone swept through and produced some flooding, particularly north of Tokyo (see calculated inundation map [1 km resolution]). Inundation

7 Using IFloods Period (Spring 2013) to Evaluate Satellite Precipitation and Hydrology Iowa/Cedar Rivers 13-year Mean Annual Precipitation (Jan 1, 2002 Dec. 31, 2013) Basin (32K km 2 ) Turkey River Basin 1004 mm (highest) TMPA TMPA-RT CPC-U (gauges) (lowest) NLDAS Stage IV CMORPH-adj The reference mean annual precipitation= USGS runoff + RS ET (based on data ) Reference Precip. (925 mm) = Streamflow (298) + ET (627)

8 13-year mean annual water budget [mm] for Iowa-Cedar River Basin outlet (USGS ) Reference TMPA-RP TMPA-RT CPC-U NLDAS2 StageIV CMORPH Precipitation (Precipitation Bias) (8.5%) ET 627 (RS ET) Runoff (streamflow at outlet) 298 (USGS) 1207 (30.5%) 870 (-5.9%) 888 (-4%) 868 (-6.2%) 764 (-17.4%) Monthly NSC mm MARE (Streamflow Bias) - 34% 118% -18% -7.5% -11% -40% Streamflow at outlet calculated using 12 km runoff/routing model with various input precipitation and model ET Raingauge products best, satellite+gauge products next best, satellite only product Need to study seasonal (winter/summer) and inter-annual (dry/wet) variations in skill

9 Accumulated Rain during IFloodS period (April 1-June 30,2013) 579 mm (highest) 490 (lowest) 551 TMPA TMPA-RT CPC-U NLDAS Stage-IV CMORPH-adj Two extra rain products NMQ/Q IFC Radar Higher than highest Lower than lowest

10 Daily basin-average precipitation: Cedar River (20,000 km 2) (a large sub-basin of Iowa/Cedar basin) Cedar River basin being used for following calculations due to heavy dam influence on Iowa R. 60 Basin-averaged precipitation (mm) NLDAS2 CPC-U stageiv Q2 TMPARP CMORPH TMPART /4/1 2013/4/ /5/1 2013/5/ /5/ /6/ /6/30

11 Iowa, Cedar and Turkey River Basin Calculations for IFloods Period (April 1-June 30, 2013) 8 different rain products Bias in streamflow (model vs. obs.) for entire period vs. accumulated rainfall. Each point is for 12 different subbasins of various sizes

12 Iowa, Cedar and Turkey River Basin Calculations for IFloods Period (April 1-June 30, 2013) 9 different rain products Bias in streamflow (model vs. obs.) for entire period vs. upstream river basin size.

13 Cedar-River Basin (at outlet) Calculations for IFloods Period (April 1-June 30, 2013) 9 different rain products Bias in streamflow (model vs. obs.) for entire period vs. accumulated rainfall. Each point is for outlet of Cedar River (20K km 2 )

14 Calculated and observed daily streamflow at Cedar River outlet Rain (mm): CPC-U (575) NLDAS(570) StageIV (560) Q2 (671) NLDAS2 NSC=0.30 RE=-21% CPC-U NSC=0.25 RE=-25% StageIV NSC=0.19 RE=-26% Q2 NSC=0.28 RE=3% Q2 best? TMPARP (580) CMORPH-adj (554) TMPA-RT (509) CMORPH (612) TMPARP NSC=0.18 RE=-18% CMORPH-adj NSC=-0.17 RE=-24% TMPART NSC=-0.9 RE=-45% CMORPH NSC=-2.68 RE=-27% Satellite plus gauges better than Satellite -only

15 Peak Daily Flow (4 peaks during IFloods): Observed vs. Calculated Irrespective of Model Streamflow Phase Error

16 Summary and Future 1. The Global Flood Monitoring System (GFMS) [flood.umd.edu] is running routinely for real-time application. Includes 12 km and 1 km streamflow and 1 km inundation estimates. Also includes 5-day forecasts using NWP precipitation forecasts. 2. Significant improvement in results expected in future with improved satellite rainfall information, including GPM (IMERG) and other products. implementing a dam module to try to include the impact of manmade structures on the calculations 3. Evaluation of the GFMS at 12 km resolution using Iowa IFloods experiment period with various ground and satellite data sets show: for long-term period ( ) gauge-based products best and closest to water budget mean precipitation. Gauge-adjusted precipitation products next best, with satellite-only trailing. for 3-month Ifloods period most products produced generally underestimated mean streamflows, with great variation sub-basin to sub-basin for daily streamflow during Ifloods (in Cedar River) Q2 best, satellite/ gauge products have some skill, better than satellite-only, but underestimate; model has phase issue (early)

17 Recent (Oct.13-Nov 12, 2014) Visitors/Users of GFMS website (

18 Flood Threshold Map for Flood Detection/Intensity Parameter Routed Runoff (RR) > RR 95th Percentile + δ and Q (streamflow) > 10 m 3 /s, where δ is temporal standard deviation of RR. REFERENCE LEVEL at each grid calculated from 15-year global hydrology model run using satellite rainfall data mm Same Approach is Used for Streamflow

19 Streamflow Caluclations (1/8 th Degree ) Compared to Observations May 2013 over U.S. Obs. Streamflow Calculated Iowa River Iowa River 1 April 1 June 1 April 1 June Wabash River Illinois River 1 April 1 June 1 April 1 June

20 Flood detection verification against the Dartmouth Flood Observatory (DFO) flood database over the 38 Well Reported Areas (WRAs) for floods with duration of one or more days (not flash floods). Yellow regions are Well Reported Areas (WRA s) Dots are dams Better flood detection statistics with research (instead of RT) rain and with fewer dams (drop in FAR) Bottom line For 1+ day floods in basins with few dams using RT rainfall: POD ~ 0.9 FAR ~ 0.7 Wu et al., 2014 Water Resources Research

21 IFloodS: Mean Summer Precipitation (Jan 1, 2002 Dec. 31, 2013)

22 IFloodS: Mean Winter Precipitation (Jan 1, 2002 Dec. 31, 2013)