THE POTENTIAL FOR MEDIUM-RANGE GLOBAL FLOOD PREDICTION Nathalie Voisin1, Andrew W. Wood1, Dennis P. Lettenmaier1, and Eric F. Wood2 1Department of Civil and Environmental Engineering 2Department of Civil and Environmental Engineering University of Washington, Seattle, WA Princeton University, Princeton, NJ H53C-0644 1 Introduction 3 Step 1: GFS Reforecasts Bias Correction 4 Step 2: Downscaling from 2.5 to 0.5 degree Precipitation is downscaled using the Schaake Shuffle (Clark et al. 2004) with the satellite product 1997-2005 GPCP1dd (Huffman et al. 2001) used as observations. GFS reforecast precipitation is scaled using the ratio of GPCP1dd values at 0.5 degree over the value at 2.5 degree: The daily temperature average is downscaled via an inverse square distance interpolation. The daily temperature range is assigned using a Schaake Shuffle type of selection using Adam et al. (2006) global temperature dataset: While weather and climate forecast methods have advanced greatly over the last two decades, this capability has yet to be evidenced in mitigation of water-related natural hazards (primarily floods and droughts), especially in the developing world (Lettenmaier et al, 2006). For instance, Mozambique experienced major droughts in 2005 and 2002 which resulted in widespread food shortages and major floods in 2000 and 2001 which affected large parts of the country. In Southeast Asia, early monsoon rains that began in July 2000 resulted in flooding of the Mekong River and its tributaries in Cambodia, Vietnam, Laos and Thailand. It was the worst flooding in several decades and affected more than 4.5 million people and killed several hundreds. Mitigation of these events through advance warning was at best modest; despite the above noted improvement in weather and climate forecasts, there is at present no system for forecasting of floods and droughts globally, notwithstanding that the potential clearly exists. We describe development of a methodology that is eventually intended to generate global flood and drought predictions routinely. It draws heavily from the experimental North American Land Data Assimilation System (NLDAS) and the companion Global Land Data Assimilation System (GLDAS) for development of nowcasts, and the University of Washington Experimental Hydrologic Prediction System to produce ensemble hydrologic forecasts based on the NCEP Global Forecast System for lead times from seven days to six months using the University of Washington/Princeton University Variable Infiltration Capacity (VIC) macroscale hydrology model. In the prototype (tested using retrospective data), VIC is driven globally up to the time of forecast with daily ERA40 precipitation (rescaled on a monthly basis to a station-based global climatology), ERA40 wind, and ERA40 average surface air temperature (with temperature ranges adjusted to a station-based climatology). In the retrospective forecasting mode, VIC is driven by global NCEP ensemble 15-day reforecasts provided by Tom Hamill (NOAA/ERL), bias corrected with respect to the adjusted ERA40 data and further downscaled spatially using higher spatial resolution Global Precipitation Climatology Project (GPCP) 1dd daily precipitation. Downward solar and longwave radiation, surface relative humidity, and other model forcings are derived from relationships with the daily temperature range during both the retrospective (spinup) and forecast period. The initial system is implemented globally at one-half degree spatial resolution. We evaluate model performance retrospectively for predictions of major floods for the Rhine and Meuse in 1995. The similar quantile-quantile method as used in (Wood et al. (2005) is applied directly to the GFS reforecast values. This step ensures that the forecasts are consistent ( signal characteristics) with the dataset used during the spinup period. ERA40 cumulative distribution function (CDF) GFS reforecast CDF GFS control run ( deterministic forecast) bias corrected GFS reforecast value GFS reforecast value 15 ensembles Jan 20th, 1995 GFS reforecasts at cell (50oN,2.5oW) – observed precipitation events sorting as part of the Schaake Shuffle Jan 20th, 1995 GFS reforecast of one ensemble, at one 2.5 degree cell (50oN,2.5oW) Precipitation downscaling Temperature downscaling Bias correction applied to cell (50oN,2.5oW) for the GFS reforecast of Jan 20th, 1995 Precipitation Daily Average Temperature Original GPCP 1dd precipitation value used for downscaling the corresponding bias corrected 2.5 degree GFS reforecast 25 GPCP 1dd values used for downscaling the corresponding bias corrected 0.5 degree GFS reforecast 2 Approach Daily average 2.5 degree bias corrected GFS reforecast temperature 25 daily temperature averages ( at 0.5 degree) obtained by interpolation 25 values ( at 0.5 degree) of Tmin as a function of Tmax This schematic is similar experimental procedure as used by Wood and Lettenmaier (2006) West-wide seasonal hydrologic forecast system. Hydrologic simulations are performed using the Variable Infiltration Capacity model developed at the University of Washington and Princeton University. 15 (ensembles) 2.5 degree GPCP1dd precipitation events assigned to the 15 ensemble GFS reforecast value In this poster Daily ECMWF ERA40 Weather input into the hydrologic model NCEP Reforecasts (Hamill 2006) 15 ensemble members – 15 day forecast – 2.5 degree (fixed GFS version of 1998) 5 Hydrologic Simulations , January 1995 Rhine and Meuse Floods ( GFS reforecast as of Jan 20th, 1995 ) 5-day accumulation Runoff 5-day change in soil moisture Daily streamflow forecasts 5-day accumulation Precipitation Adjusted daily ECMWF ERA40 Daily ERA40 downscaled to 0.5 degree using linear inverse square distance interpolation. Monthly ECMWF ERA40 has been rescaled to match observations based Adam et al. (2006) precipitation and maximum and minimum temperatures. GFS ens. Avg GFS det. Fcst ERA40 GFS ens. Avg GFS det. Fcst ERA40 GFS ens. Avg GFS det. Fcst ERA40 Fcst of Jan 20th, 1995 Section 3 Bias correction with respect to adjusted ERA40 (Ensures consistency between spinup and the reforecasts) Fcst of Jan 22th, 1995 Later on: Update of initial conditions Section 4 Downscaling from 2.5 to 0.5 degree using higher spatial resolution satellite (GPCP1dd, Huffman et al. 2001) precipitation and the Schaake Shuffle ( Clark et al. 2004) GFS ens. Avg : average of the 15 GFS reforecast ensembles GFS det. Fcst : GFS reforecast control run – GFS deterministic forecast ERA40 : ECMWF 40 year reanalysis, surrogate for observations Climatology monthly flow Climatology daily flow Section 5 6 Ongoing Improvements Hydrologic model spin up (0.5 degree global simulation) INITIAL STATE Hydrologic forecast simulation References Adam, J.C., E.A. Clark, D.P. Lettenmaier, and E.F. Wood, 2006: Correction of Global Precipitation Products for Orographic Effects .J. Climate,19 (1), 15-38. Clark, M.P., S. Gangopadhyay, L.E. Hay, B. Rajagopalan, and R.L. Wilby (2004): The Schaake Shuffle: A method to reconstruct the space-time variability of forecasted precipitation and temperature fields. Journal of Hydrometeorology, 5, 243-262. Huffman, G.J., R.F. Adler, M. Morrissey, D.T. Bolvin, S. Curtis, R. Joyce, B McGavock, J. Susskind, 2001: Global Precipitation at One-Degree Daily Resolution from Multi-Satellite Observations. J. Hydrometeor., 2, 36-50. Lettenmaier, D.P., A. de Roo, and R. Lawford, 2006. Towards a capability for global flood forecasting, WMO Bulletin 55, 185-190. Wood, A.W. and D.P. Lettenmaier, 2006, A testbed for new seasonal hydrologic forecasting approaches in the western U.S., Bulletin of the American Meteorological Society (in press). CONTACT: nathalie@hydro.washington.edu dennisl@u.washington.edu (0.5 degree global simulation: stream flow, soil moisture, SWE, runoff ) - Improve VIC calibration (presently essentially uncalibrated) - Apply to more events: Mozambique floods 2000 and 2001, Danube floods 2000, 2002 and 2006, Mississippi floods 1993, Mekong floods 2000, Yangtze floods 1998, Oder floods 1997,etc. Medium range forecasts ( up to 2 weeks) Nowcasts Several years back The hydrologic model is essentially uncalibrated, see ongoing improvement section.