Estimating daily evapotranspiration in the agricultural-pastoral ecotone in Northwest China: A comparative analysis of the Complementary Relationship, WRF-CLM4.0, and WRF-Noah methods.

Accurate estimation of evapotranspiration (ET) over regional scale is essential for ecohydrological research, agricultural production, and water resources management. However, few studies have been done to estimate regional ET in data lacking, highly heterogeneous arid areas such as the Agricultural-Pastoral Ecotone in Northwest China (APENC). In this study, we compared three actual ET-estimation methods driven by Weather Research and Forecasting (WRF) model in a semi-arid region. We selected the state of the art Weather Research and Forecasting-Community Land Model 4.0 (WRF-CLM4.0) model, the widely used WRF-Noah model and an empirical Complementary Relationship (CR) model to compare their model structures and mechanisms of estimating daily ET in the study region. The WRF model was chosen to address the problem of data scarcity in the study region and to derive model input for ET estimation with high spatial resolution. The seasonal and pooled performances of the three models were verified with in situ observations. Results indicate that the WRFCLM4.0 model shows a better applicability in the study region, with a superior performance for the pooled datasets (Pearson correlation coefficient [r] = 0.89, root-mean-square error [RMSE] = 0.66 mm/d and NashSutcliffe efficiency coefficient [NSE] = 0.90), while the CR model has a comparable performance (r = 0.91, RMSE= 0.86mm/d andNSE= 0.85) and theWRF-Noahmodel shows theworst performance (r= 0.82, RMSE= 0.94 mm/d and NSE = 0.81). The differences are mainly caused by different representations of the land surface characteristics and hydrology of the study region by the three different models. Our analysis shows that theWRFCLM4.0 model and the CR model are more applicable to the APENC than the WRF-Noah model. For regional applications, the CR model, with fewer parameters and simpler structure, is able to capture the local characteristic and well-suited for data lacking, highly heterogeneous landscapes such as the APENC. (C) 2020 Elsevier B.V. All rights reserved.