Use of satellite remote sensing in a medium

  • Slides: 22
Download presentation
Use of satellite remote sensing in a medium range global flood prediction scheme Nathalie

Use of satellite remote sensing in a medium range global flood prediction scheme Nathalie Voisin 1 , Andrew W. Wood 1 , Dennis P. Lettenmaier 1 and Eric F. Wood 2 1 Department of Civil and Environmental Engineering, University of Washington 2 Department of Civil and Environmental Engineering, Princeton University

Outline • Background and Objectives • Brief description of the prediction scheme • Use

Outline • Background and Objectives • Brief description of the prediction scheme • Use of satellite in the downscaling process of weather forecasts • Preliminary results for – Rhine Flood 1995 ( mostly rain, then snowmelt) – Danube Flood 2006 (snowmelt and rain) – Limpopo flood 2000 (tropical storm)

Background Flood prediction systems exist • in developed Countries • What about developing countries?

Background Flood prediction systems exist • in developed Countries • What about developing countries? The potential for global flood prediction system exists • Global weather models : analysis and forecasts are available • Issues: scale?

Objectives Ultimate objective: to predict streamflow and associated hydrologic variables, soil moisture, runoff, evaporation

Objectives Ultimate objective: to predict streamflow and associated hydrologic variables, soil moisture, runoff, evaporation and snow water equivalent : – At a global scale • Spatial consistency • Especially in ungauged or poorly gauged basins – medium-range time scale ( up to 2 weeks) This talk: to suggest a method to spatially downscale global weather forecasts without any local information ( gauges or radar)

The global prediction scheme (here in retrospective mode) Atmospheric inputs Daily NCEP Reanalysis 1

The global prediction scheme (here in retrospective mode) Atmospheric inputs Daily NCEP Reanalysis 1 downscaled to 0. 5 degree using linear inverse distance square interpolation. Hydrologic model spin up (0. 5 degree global simulation) Several years back Hydrology Model NCEP Reforecasts (Hamill et al. 2006) 15 ensemble members – 15 day forecast – 2. 5 degree (fixed GFS version of 1998) Bias correction at 2. 5 degree, with respect to NCEP Reanalysis (Ensures consistency between spinup and the reforecasts) Downscaling from 2. 5 to 0. 5 degree using the Schaake Shuffle ( Clark et al. 2004) with higher spatial resolution satellite GPCP 1 dd (Huffman et al. 2001) and TRMM 3 B 42 precipitations INITIAL STATE Nowcasts Hydrologic forecast simulation (0. 5 degree global simulation: stream flow, soil moisture, SWE, runoff ) Medium range forecasts ( up to 2 weeks)

The global prediction system The bias correction of GFS reforecasts 1. Bias Correction (

The global prediction system The bias correction of GFS reforecasts 1. Bias Correction ( Quantile-Quantile technique with respect to NCEP Reanalysis climatology) 2. Correction for Daily Intermittency ( with respect to NCEP Reanalysis climatology) Figure from Wood and Lettenmaier, 2004: A testbed for new seasonal hydrologic forecasting approaches in the western U. S.

Use of satellite for downscaling forecasts Limpopo Basin, 2. 5 degree grid South Africa,

Use of satellite for downscaling forecasts Limpopo Basin, 2. 5 degree grid South Africa, 2. 5 degree grid Limpopo Basin, 0. 5 degree grid

Use of satellite for downscaling forecasts Satellite Datasets → TRMM 3 B 42 –

Use of satellite for downscaling forecasts Satellite Datasets → TRMM 3 B 42 – 50 o. S-50 o. N – 0. 25 degree, 3 hourly, 2002 -present – Use 2002 -2006 ( to be updated yearly) → GPCP 1 dd – Global, but used for 50 o. N-90 o. N and 50 o. S-90 o. S – 1 degree, daily, Oct 1996 – 3 months before present – Use 1997 -2005 (to be updated yearly), interpolated to 0. 5 degree using an inverse distance square interpolation.

Use of satellite for downscaling forecasts Simplified Schaake Schuffle (Clark et al. 2004) –

Use of satellite for downscaling forecasts Simplified Schaake Schuffle (Clark et al. 2004) – to construct spatial patterns of precipitation within each 2. 5 degree cell based on observations ( here, satellite) – For each 2. 5 degree cell, for each lead time: GFS refcst, 2. 5 deg, rank ith • 15 satellite observations are randomly selected ( based on rain / no rain, specific to calendar month ) • for each ranked forecast ensemble member, it associates the corresponding ranked observation ( 15 ensemble members). → ensures that the selected highest observed precipitation event is assigned to the highest forecast Satellite, 2. 5 deg, rank ith

Use of satellite for downscaling forecasts TRMM 3 B 42, 2. 5 degree Jan

Use of satellite for downscaling forecasts TRMM 3 B 42, 2. 5 degree Jan 31 st, 2001 TRMM 3 B 42 aggregated to daily and 0. 5 degree resolution → resolution of the hydrologic model (mm/day) TRMM 3 B 42 aggregated to daily and 2. 5 degree resolution → resolution of the weather forecasts TRMM 3 B 42, 0. 5 degree Jan 31 st, 2001 →Need a downscaling method that inserts localized precipitation patterns

Use of satellite for downscaling forecasts Simplified Schaake Shuffle (2): – The corresponding observed

Use of satellite for downscaling forecasts Simplified Schaake Shuffle (2): – The corresponding observed value field at 0. 5 degree resolution gives the spatial distribution of precipitation, but NOT the magnitude Here for one ensemble member, one lead time : = / Satellite, 0. 5 deg, corresponding record to the 2. 5 degree cells Satellite, 2. 5 deg Ratio of Satellite observations, 0. 5 deg resolution

Use of satellite for downscaling forecasts Dowscaling of precipitation characteristics: - Spatial distribution from

Use of satellite for downscaling forecasts Dowscaling of precipitation characteristics: - Spatial distribution from satellite observations - Magnitude of the bias corrected GFS reforecasts - Consistency between spin up dataset and bias corrected downscaled forecasts Here for one ensemble member, one lead time : X Bias corrected GFS refcst, 2. 5 deg = Ratio of satellite obs. 0. 5 degree to 2. 5 degree Bias corrected and downscaled (0. 5 degree) GFS reforecast

Preliminary Results Rhine Flood, 1995 (Forecast of January 20 , 1995) • 5 day

Preliminary Results Rhine Flood, 1995 (Forecast of January 20 , 1995) • 5 day precipitation accumulation fields th 6 to 10 days 1 to 5 days NCEP Rean. , GFS Det. Fcst. , simple interpol. Bias corrected GFS Fcst. Ens. Avg, Downscaled

Preliminary Results Danube Flood, 2006 (Forecast of April 1 , 2006) • 5 day

Preliminary Results Danube Flood, 2006 (Forecast of April 1 , 2006) • 5 day precipitation accumulation fields st 6 to 10 days 1 to 5 days NCEP Rean. , simple interpol. Bias corrected GFS Fcst. Ens. Avg, simple interpol. Downscaled

Preliminary Results Limpopo Flood, 2000 (Forecast of February 3 , 2000) • 5 day

Preliminary Results Limpopo Flood, 2000 (Forecast of February 3 , 2000) • 5 day precipitation accumulation fields rd 6 to 10 days 1 to 5 days NCEP Rean. , GFS Det. Fcst. , simple interpol. Bias corrected GFS Fcst. Ens. Avg, Downscaled

Conclusions 1) Improvement of using this downscaling method rather than a simple inverse distance

Conclusions 1) Improvement of using this downscaling method rather than a simple inverse distance square interpolation method – E. g. representation of topography ( snowmelt) – Less obvious for tropical storm in flat and arid areas like South Eastern Africa 2) Need to compare it with more sophisticated, but local downscaling methods – Using a nested regional scale model – Equivalent downscaling techniques using high resolution datasets based on gauges, in regions where in situ network exists

Conclusions About the entire prediction scheme … 3) The scheme performance is very dependent

Conclusions About the entire prediction scheme … 3) The scheme performance is very dependent on the quality of the forecasts. 4) The full scheme, including hydrologic simulations, will be evaluated with respects to other existing flood prediction systems. Calibration will be an essential step.

Thank You! April 2006 Flood in Romania, http: //www. spiegel. de/fotostrecke/0, 5538, 13382, 00.

Thank You! April 2006 Flood in Romania, http: //www. spiegel. de/fotostrecke/0, 5538, 13382, 00. html

The global prediction scheme The hydrology model VIC - Semi-distributed model driven by a

The global prediction scheme The hydrology model VIC - Semi-distributed model driven by a set of surface meteorological data ( precipitation, wind, solar radiation derived from Tmin and Tmax, etc) - Represents vegetation, has three soil layers with variable infiltration, non linear base flow.

The global prediction scheme The river routing model - Runoff and baseflow for each

The global prediction scheme The river routing model - Runoff and baseflow for each cell is then routed toward selected locations, following directions equivalent to channels. - Routing at 0. 5 degree derived from the manually corrected global direction file from Döll and Lehner (2002) - Already calibrated and validated at 2 degree resolution over 26 basins worldwide (Nijssen et al. 2001)

Use of satellite for downscaling forecasts Scaling of precipitation 2 Corresponding record for each

Use of satellite for downscaling forecasts Scaling of precipitation 2 Corresponding record for each cell, 0. 5 degree 1 Schaake Shuffle GFS refcst, 2. 5 deg SATELLITE, 0. 5 deg SATELLITE, 2. 5 deg 4 Scale 2. 5 degree reforecast with SAT ratio Downscaled GFS reforecast 3 Ratio of SAT 0. 5 degree to 2. 5 degree

 • Link picture • www. spiegel. de/img/0, 1020, 611798, 00. jpg • http:

• Link picture • www. spiegel. de/img/0, 1020, 611798, 00. jpg • http: //www. spiegel. de/fotostrecke/0, 5538, 13382, 00. html