Inter-comparison of soybean models for the simulation of evapotranspiration in a humid continental climate

Abstract

Accurate simulation of evapotranspiration (ET) with crop models is essential for improving agricultural water management and yield forecasting. Few studies have evaluated multiple soybean [Glycine max (L.) Merr.] models for simulating ET under conditions of low evaporative demand that is characteristic for a warm-summer humid continental climate. Six soybean crop models, encompassing 15 different modeling approaches, were evaluated for ET simulation and compared against eddy covariance data collected over five growing seasons in Ottawa, Canada. Models were first calibrated with phenology, in-season growth, and yield data, followed by calibration with measured ET and soil water content (SWC) data during the second step. After initial calibration, simulated daily ET was higher on average than measured ET, particularly during full canopy cover (normalized bias, nBias = 17.1 to 49.2% depending on the model). Following the second calibration, simulated daily ET was closer to measured values, but bias remained (nBias = 5.9 to 52.1% during full canopy). The ensemble median reduced uncertainty in the simulation of daily ET compared to most models, but DNDC remained the top-ranking model (nRMSE = 0.7 mm d−1, nBias = 11.2%). The MONICA model was most accurate simulating cumulative ET (RMSE = 39.9 mm, nBias = 11.3%), whereas the CROPGRO models excelled simulating SWC (RMSE= 0.04 to 0.05 m³ m−3, nBias = 0.10 to 0.9% depending on soil depth). This study was instrumental in evaluating the best ET methodologies and parameters in soybean models. However, there was bias across the models compared to measured eddy covariance ET in a humid environment. The results reveal the need to further investigate possible biases in ET estimates by eddy covariance over soybean canopies, and to review the role of night-time dew contributions to ET in process-based models.