Hydrologic ensemble prediction - applications to streamflow and drought nowcasting and forecasting

Multi-model ensemble prediction methods have the potential to reduce forecast errors, and in applications to weather and climate forecasting, have been shown to outperform any individual model. Their application to hydrologic forecasting is complicated, however, by the correlation among forecasts that results from all forecast models using common (atmospheric) forcings. We evaluate the implications of multimodel ensemble methods for seasonal streamflow forecasting in the western U.S. in this context. We also evaluate two other hydrologic applications of multi-model ensemble prediction methods. The first is to multimodel nowcasting of drought over the continental U.S. In this application, we use four land surface models in an quasi-operational setting - the Variable Infiltration Capacity (VIC) model, the NOAH land surface model, the Community Land Model, and a grid-based version of the Sacramento Soil Moisture Accounting model. We evaluate the ability of the individual models, as well as the multi-model ensemble, to identify drought conditions in the southeastern U.S. in 2007-2008, and in the western U.S. in 2008. Finally, we evaluate seasonal streamflow forecasts at forecast points in the eastern U.S. during 2006-2008 made using a multiple ensembles from multiple climate forecast models, as compared with single model forecasts made using ensembles taken from the NCEP Climate Forecast System.