An Experimental seasonal hydrological forecast system for East Africa

This presentation describes the implementation and preliminary verification of an experimental hydrologic forecast system for East Africa that uses the Variable Infiltration Capacity (VIC) model, a hydrology model driven by downscaled Climate Forecast System (CFSv2) seasonal precipitation and temperature forecasts to provide hydrological forecasts. East Africa ranks among the most food and water insecure regions in the world. The agriculture in this region is mostly rainfed and hence is sensitive to interannual variability in rainfall. The increased food and water demands of a growing population place further stress on the water resources of this region. The region fulfills many of its energy needs via hydropower and relies on the economic benefits of environmental tourism, forcing water managers to reconcile the demands of many sectors. Skillful seasonal hydrological forecasts (mainly using soil moisture and runoff) for this region can inform timely water and agricultural management decisions, support the proper allocation of the region's water resources, and help mitigate socio-economic losses. CFSv2 is a fully coupled dynamical seasonal forecast system that provides forecasts going out 9 months. Raw CFSv2 forecasts, however, are not suitable for driving a hydrologic model (such as VIC) due to their coarse resolution and bias. In this study we used Bias Correction Constructed Analog, a statistical approach to downscale CFSv2 seasonal forecasts to the scale of the hydrologic model (0.5° longitude and latitude). In this method, first raw CFSv2 forecasts are bias-corrected using a quantile mapping approach (similar to Bias Correction Spatial Downscaling), then the Constructed Analog method is applied to downscale the bias corrected CFSv2 forecasts. We used CFSv2 precipitation and temperature forecasts over the Indian and Pacific Oceans as predictors of East Africa precipitation and temperature fields. In doing so, we benefit from the known relationship between the precipitation and temperature over East Africa and the Indian and Pacific Ocean, as well as the relatively high skill of CFSv2 forecasts over oceans, as opposed to land areas. We also report the results of a verification of this approach done by comparing the hydrological forecasts generated using the downscaled CFSv2 hindcasts as described above with a reference dataset of the hydrologic variables (soil moisture and runoff) generated by driving the VIC model with the observed gridded atmospheric forcings.