A medium range probabilistic quantitative hydrologic forecast system for global application

In-situ precipitation observations are often used in flood forecasting both to initialize hydrological forecast models, and in the downscaling of precipitation forecasts from global weather models used as forcings for hydrological forecast models. Flood forecasting capabilities are limited in areas where in-situ observations are sparse and/or where regional scale atmospheric models have not been applied, a situation that includes much of the underdeveloped world. In this dissertation a global approach to producing hydrological ensemble forecasts in those river basins was developed: 10-day global ensemble weather forecasts produced by the European Centre for Medium Range Weather Forecasts (ECMWF) Ensemble Prediction System (EPS) were calibrated and downscaled then used in conjunction with the Variable Infiltration Capacity (VIC) semi-distributed macroscale land surface hydrology model in forecast mode. VIC was initialized to the time of forecast using Tropical Rainfall Monitoring Mission (TRMM) satellite precipitation data and ECMWF analysis fields.;The approach was developed in three phases. First, globally available gridded station, satellite observation-based and numerical weather prediction model analysis precipitation datasets were evaluated globally. Differences in VIC-simulated water balances resulting from use of each of the datasets were assessed over large river basins. As a result, no adjustment with respect to gridded station data is performed on the high spatial resolution satellite precipitation data used for initializing VIC. In the second phase, two approaches for downscaling and calibrating global ensemble precipitation forecasts to the scale of hydrologic models were developed: bias correction with spatial disaggregation, and an analog technique. The analog approach was preferred because it improved reliability the most relative to a simple interpolation benchmark, and produced the most realistic precipitation patterns. In the third phase the globally applicable flood forecast approach was evaluated using two sources of forecast precipitation: interpolated ECMWF EPS forecasts; and the same forecasts calibrated and downscaled with the analog method. The resulting VIC-simulated spatially distributed ensemble runoff forecasts were routed to flow forecast points in the Ohio River basin. Ensemble streamflow forecasts using calibrated forecasts were shown to be more accurate and reliable than those produced using interpolated forecasts, with useful skill for at least 10 days.