Hydrologic model predictability improves with spatially explicit calibration using remotely sensed evapotranspiration and biophysical parameters

Abstract

Listen

Translate article

A hydrologic model, calibrated using only streamflow data, can produce acceptable streamflow simulation at the watershed outlet yet unrealistic representations of water balance across the landscape. Recent studies have demonstrated the potential of multi-objective calibration using remotely sensed evapotranspiration (ET) and gaged streamflow data to spatially improve the water balance. However, methodological clarity on how to “best” integrate ET data and model parameters in multi-objective model calibration to improve simulations is lacking. To address these limitations, we assessed how a spatially explicit, distributed calibration approach that uses (1) remotely sensed ET data from the Moderate Resolution Imaging Spectroradiometer (MODIS) and (2) frequently overlooked biophysical parameters can improve the overall predictability of two key components of the water balance: streamflow and ET at different locations throughout the watershed. We used the Soil and Water Assessment Tool (SWAT), previously modified to represent hydrologic transport and filling-spilling of landscape depressions, in a large watershed of the Prairie Pothole Region, United States. We employed a novel stepwise series of calibration experiments to isolate the effects (on streamflow and simulated ET) of integrating biophysical parameters and spatially explicit remotely sensed ET data into model calibration. Results suggest that the inclusion of biophysical parameters involving vegetation dynamics and energy utilization mechanisms tend to increase model accuracy. Furthermore, we found that using a lumped, versus a spatially explicit, approach for integrating ET into model calibration produces a sub-optimal model state with no potential improvement in model performance across large spatial scales. However, when we utilized the same MODIS ET datasets but calibrated each sub-basin in the spatially explicit approach, water yield prediction uncertainty decreased, including a distinct improvement in the temporal and spatial accuracy of simulated ET and streamflow. This further resulted in a more realistic simulation of vegetation growth when compared to MODIS Leaf-Area Index data. These findings afford critical insights into the efficient integration of remotely sensed “big data” into hydrologic modeling and associated watershed management decisions. Our approach can be generalized and potentially replicated using other hydrologic models and remotely sensed data resources – and in different geophysical settings of the globe.