Improved sub-seasonal river water availability forecasts for southeast Asia

Predictive skill of streamflow at the subseasonal-to-seasonal (S2S) scale is controlled by rainfall forecast skill and by the initialization of land surface states. Historical rainfall is typically used prior to the forecast start date to initialize a forecast system’s soil moisture. Gauge-based rainfall measurements, however, can be problematic in regions where the measurements are of poor quality. Use of satellite data, such as the Integrated Multi-satellitE Retrievals for GPM (IMERG), can be a good alternative for rainfall correction in data poor regions, and subsequently, for improving a regional streamflow forecast. We examine such improvements with a series of experiments focused on sub-seasonal forecasts of river discharge in southeast Asia during the last two decades. The control set of forecasts utilized gauge-based precipitation data in the initialization, while the experiment set used rainfall corrected by IMERG; both sets of forecasts were performed within an offline sub-seasonal water availability forecast framework in which an offline land model is driven with bias-corrected forecast meteorology from a full S2S forecast system. Forecasts in both sets were initialized at four different times of the year (January, April, July, and October), and streamflow forecast skill was evaluated against observations at 16 locations in Myanmar, Thailand, Laos, and Cambodia. Overall, results do show significant improvement when the IMERG data are used for the initialization, especially in Myanmar, where rain gauge information is particularly uncertain and limited. However, the streamflow forecast skill remains low during the transitional period (April in Myanmar) from the dry season to the rainy season due to particularly poor S2S rainfall forecasts. Along the Mekong River, streamflow forecasts show higher skill (up to ~0.8 of correlation with observed discharge) than in Myanmar, except for the October initialization. Our results demonstrate the usefulness of satellite data for regional streamflow forecasts. Further improvement can potentially be achieved by integrating river routing and reservoir operations into the sub-seasonal water availability forecast system.