Evaluating the impact of satellite data streams on simulated watershed processes using streamflow prediction models (Invited Presentation)

Integrating the most advanced science into hydrologic forecasting is critical to assure that decision makers have access to robust and reliable information when managing water resources. The degree to which new data streams can improve forecast skill in light of model structural uncertainty needs to be better understood to progress forecasting methodologies in an informed way. Model evaluations using multiple variables is a necessary step in identifying and resolving model deficiencies, particularly as hydrologic forecast systems evolve to incorporate more spatially explicit and physically realistic models.

A limitation in the current National Weather Service (NWS) streamflow forecast system is the reliance on climatological potential evapotranspiration (PET) estimates derived from historical evaporation pan measurements dating back to 1956. As a result, critical temporal and spatial changes in evapotranspiration (ET) due to such things as climatic and land cover changes are not reflected in current forecasts. This can lead to errors in the modeled water balance and subsequent over- or under-estimation of watershed discharge. Through several studies we are exploring the use of satellite-based PET estimates, referred to as MODIS-PET, as input to NWS hydrologic forecast models. Evaluating the impact of satellite data streams is challenging given the lack of validation data, other than basin discharge, in most watersheds. We have found that evaluating a model using only basin discharge gives a limited picture of the impact of the MODIS-PET on the model behavior, particularly when the model has been recalibrated for the new input data. The MODIS-PET tends to produce slightly degraded discharge simulations, on average, when compared to using the climatological PET data. However, ET simulations better reflect the magnitude and variability of latent heat flux observations when evaluated for basins where flux tower data is available.

We expand upon these previous evaluation efforts to assess simulations of discharge, soil moisture and ET in basins located in north-central Iowa and south-central Mississippi within which multiple sub-basin observations are available. Simulated watershed processes from lumped and distributed NWS forecast models are evaluated using available ground- and satellite-based data. Point-to-grid and spatially averaged data-model comparisons are made.