Improving Runoff Prediction Using Remotely Sensed Actual Evapotranspiration During Rainless Periods

Remotely sensed actual evapotranspiration (RS-ET) has been used extensively for multiobjective calibration of rainfall-runoff (RR) models to improve runoff prediction in ungauged basins. This study aims to implement RS-ET during periods when actual evapotranspiration (ET) simulation in RR models contains larger uncertainty and RS-ET is of relatively higher accuracy. Two calibration schemes are proposed and compared over 208 basins in the United States using the Xinanjiang model: (1) calibration against both observed streamflow and RS-ET over the entire period of record and (2) calibration simultaneously against observed streamflow over the entire period of record and RS-ET only during rainless periods. The 208 basins are then regarded as ungauged basins in turn and modeled with parameters transferred from a donor basin (i.e., parameter regionalization) using the spatial proximity method. Results show that compared with conventional calibration solely against observed streamflow, (1) both multiobjective calibration methods degrade streamflow simulation but improve ET simulation in the calibration and validation periods (however, the simulation efficiencies of both variables are improved in regionalization); (2) greater improvements of regionalization results are achieved by using RS-ET only during rainless periods; and (3) the efficacy of using RS-ET is superior for originally poorly simulated basins. Therefore, RS-ET is recommended to be used for multiobjective calibration during rainless periods. (C) 2019 American Society of Civil Engineers.