Understanding Hydrologic Model Structure through Satellite Observations

Improving the spatial and temporal representation of the surface water balance in streamflow forecasting models is critical for increasing the accuracy and utility of water supply, flood and drought predictions. Satellite remote sensing applications can play a vital role in assuring that forecast products reflect rapid and ongoing changes in the physical system, such as from land cover and climatic changes. We are investigating the application of satellite remote sensing data in operational streamflow prediction. Specifically, the consequences of the role of hydrologic model structure accuracy on streamflow simulations through the use of satellite data. We want to understand how various fluxes and states differ among forecast models when using spatially and temporally representative data. In a series of prior studies, we investigated the use of a daily satellite-derived potential evapotranspiration (PET) estimate as input to the lumped National Weather Service (NWS) streamflow forecast model for watersheds in the Upper Mississippi and Red river basins. The spatial PET product appears to represent the day-to-day variability when compared with ground-based estimated PET in point to grid scale comparisons more realistically than current climatological methods used by the NWS. Evaluation statistics of the overall streamflow simulations, however, show little variation even while analysis of the model states indicates the model progresses differently between simulations with the baseline PET and the satellite-derived PET input. For instance, the upper zone states, responsible for high flows, show a profound difference, while peak flow simulations tend to show little variation in the hydrograph timing and magnitude. Using the spatial PET input, the lower zone states show improvement with simulating the recession limb and the baseflow portions of the hydrograph. We anticipate that through a better understanding of the relationship between model structure, model states, and simulated streamflow through model validation we will be able to diagnose why simulations of discharge from the forecast model have failed to improve when provided seemingly more representative input data. Estimating realistic inputs of PET from remote sensing platforms and identifying model structure that limits accurate forecasting are critical to demonstrating the full benefit of a satellite data for operational use.