Satellite-Based Evapotranspiration in Hydrological Model Calibration

Lulu Jiang,John S Kimball,Jing Tao,Xiuwan Chen,Huan Wu,Lorenzo Alfieri

doi:10.3390/rs12030428

Abstract

Hydrological models are usually calibrated against observed streamflow (Qobs), which is not applicable for ungauged river basins. A few studies have exploited remotely sensed evapotranspiration (ETRS) for model calibration but their effectiveness on streamflow simulation remains uncertain. This paper investigates the use of ETRS in the hydrological calibration of a widely used land surface model coupled with a source–sink routing scheme and global optimization algorithm for 28 natural river basins. A baseline simulation is a setup based on the latest model developments and inputs. Sensitive parameters are determined for Qobs and ETRS-based model calibrations, respectively, through comprehensive sensitivity tests. The ETRS-based model calibration results in a mean Kling–Gupta Efficiency (KGE) value of 0.54 for streamflow simulation; 61% of the river basins have KGE > 0.5 in the validation period, which is consistent with the calibration period and provides a significant improvement over the baseline. Compared to Qobs, the ETRS calibration produces better or similar streamflow simulations in 29% of the basins, while further significant improvements are achieved when either better ET or precipitation observations are used. Furthermore, the model results show better or similar performance in 68% of the basins and outperform the baseline simulations in 90% of the river basins using model parameters from the best ETRS calibration runs. This study confirms that with reasonable precipitation input, the ETRS-based spatially distributed calibration can efficiently tune parameters for better ET and streamflow simulations. The application of ETRS for global scale hydrological model calibration promises even better streamflow accuracy as the satellite-based ETRS observations continue to improve.

Highlights

Physically-based hydrological models are an essential tool for water resources planning, drought, and flood prediction, climate change assessment, and many other precipitation-runoff related applications
Such regionalization methods tend to lose efficiency during the transfer process because the assumptions of spatial proximity and physical similarity are not always valid [4,5,6] Secondly, Qobs-based calibration has long been criticized for its equifinality problem, i.e., different parameter sets can satisfy the objective functions during the calibration period, while the model performance deteriorates significantly in the validation period [7,8]
The ETRS based spatially distributed calibration takes a further step than the lumped calibration by deriving an optimized spatially distributed model parameter set that accounts for the ETRS spatial patterns [9]

Summary

Introduction

Physically-based hydrological models are an essential tool for water resources planning, drought, and flood prediction, climate change assessment, and many other precipitation-runoff related applications. Streamflow prediction in ungauged basins often uses the regionalization approach which transfers hydrological parameters from Qobs-based calibrations developed from donor gauged basins. Such regionalization methods tend to lose efficiency during the transfer process because the assumptions of spatial proximity and physical similarity are not always valid [4,5,6] Secondly, Qobs-based calibration has long been criticized for its equifinality problem, i.e., different parameter sets can satisfy the objective functions during the calibration period, while the model performance deteriorates significantly in the validation period [7,8]. To accomplish the “goodness-of-fit” between simulated and observed streamflow, Qobs calibration models are computationally time-consuming in routing spatially distributed runoff through complex drainage networks to produce estimated streamflow at the river basin outlet for each model time step [10]

Objectives

Methods

Results