Calibration of land surface model using remotely sensed evapotranspiration and soil moisture predictions

Abstract

Calibration is the process of estimating the optimal parameters for a model to accurately reflect the real system, using historical records of system data. The model calibration, however, is frequently limited by availability, quality, quantity and the nature of the ground observations. Lack of streamflow observations in the vast majority of the world, for example, constrains the calibration of hydrologic and land surface models. In this study, an attempt is made to calibrate a land surface model by using satellite retrievals of soil moisture and evapotranspiration (ET), without relying on streamflow measurements. This paper examines the capability of using satellite measurements for the calibration of hydrologic/land surface models for ungauged locations. The Australian Water Resources Assessment Landscape model (AWRA-L) modified to have single hydrological response unit (HRU) per each grid cell is chosen, as a simple land surface model that requires minimum forcing variables. Initial parameters for the control case are generated based on the fraction of trees and the Budyko's dryness index. Microwave soil moisture retrievals from the Advanced Microwave Scanning Radiometer-EOS (AMSR-E) and daily estimates of ET from the Moderate Resolution Imaging Spectroradiometer (MODIS) are adopted to calibrate a selection of the AWRA-L parameters. Shuffled complex evolution uncertainty algorithm (SCE-UA) is employed to perform local calibration at 25-km grid cell in the Kyeamba catchment, southeastern Australia. Multiple criteria objective function for calibration is selected based on the AWRA-L output behavior of evapotranspiration and soil moisture compared to respective remote measurements. It considers the bias and the correlation between observed and simulated evapotranspiration and the correlation between observed and simulated soil moisture. Calibration experiments are carried out in daily time step from 2003 to 2007 and validated from 2008 to 2010. The optimum parameters obtained are employed to calculate the monthly average runoff ratio and is evaluated against the runoff ratio derived from streamflow observations at Kyeamba catchment. The results show that the calibration of AWRA-L using remotely sensed evapotranspiration and soil moisture can improve the predictions of evapotranspiration and runoff. Validation conducted in a separate period also exhibit improvements in the prediction of evapotranspiration, whereas the improvement in soil moisture is trivial during both calibration and validation periods. The monthly runoff ratio estimated after calibration is improved compared to the runoff ratio in the control case. This indicates the potential of calibration with evapotranspiration and soil moisture in improving streamflow predictions. Further research is warranted to increase efficiency in prediction of runoff ratio, so that the calibration scheme can be applied in the regions with sparse or no gauging stations.