Incorporating Satellite Remote Sensing Data into Hydrologic Models: Towards Improved Performance in Modeling the Past and Reduced Uncertainty in Predicting the Future

Abstract:

In many regions of the worlds, studies of past hydrological variability have to rely on hydrological models either because river gauge measurement is not available or because measurements do not reflect the natural flow due to water diversion or reservoir regulation. However, results from these studies are subject to major uncertainty related to the challenges in quantifying vegetation conditions and evapotranspiration, both of which are important for surface water and energy budgets. This study incorporates satellite remote sensing data for ET and vegetation into the VIC model to improve the model performance in simulating the surface water budget, hydrological seasonality, and timing of hydrological extremes. Using the Connecticut River Basin as an example, and driven with the NASA NLDAS-2 meteorological forcing data, the VIC model has been modified to read in LAI and ET data derived from MODIS among others. The MODIS LAI data provides VIC with the inter-annually varying seasonal cycle of vegetation, and the MODIS ET data replaces the model simulated ET. The data-enhanced model performs significantly better in simulating river discharge, its magnitude, seasonality, timing, soil moisture and its temporal variation. Incorporation of the ET data led to an increase of stream flow correlations between model and observations on the daily and biweekly temporal scales, and the seasonality is better represented on a monthly scale with particular magnitude improvements during the summer when ET is greatest. Incorporation of the LAI data led to improved simulation of inter-annual variability. This joint application of remote sensing and modeling helps quantify the extent to which remote sensing data improves model performance, facilitates a more accurate understanding and attribution of past hydrological variability/changes, and helps characterize the range of model-related uncertainties in future predictions.