Modifying a Land Surface Model to Improve the Subseasonal Forecasting of Hydrological Variables

Past work has shown that a land surface model’s (LSM’s) implicit (not explicitly coded) relationships between soil moisture and both evapotranspiration (ET) and runoff largely determine the LSM’s mean hydrological behavior. Here we use estimates of the relationships that appear to be operating in the real world as targets for a specific LSM’s further development, focusing on improving the LSM’s performance in predicting hydrological variables (soil moisture and streamflow) at subseasonal leads. An offline hydrological forecast system serves as a testbed for evaluating potential LSM improvements. In essence, the LSM in this offline system is driven with bias-corrected meteorological forcings produced by a full subseasonal-to-seasonal (S2S) forecast system (ensuring that forecasts are fair); soil moistures and streamflow estimates produced in the offline forecast simulations are then compared to in-situ observations. In our first experiments, we find that moving the LSM’s implicit ET and runoff relationships toward the targeted relationships does indeed lead to improved hydrological predictions at the subseasonal lead, particularly for soil moisture.