A robust parameter approach for estimating CERES-Rice model parameters for the Vietnam Mekong Delta

Rice crop models, such as CERES-Rice, are useful tools to predict rice growth and understand the effects of climatic and management changes on rice yield. As these models involve complex processes, non-linearity and interdependence of parameters may result in high uncertainty in model results. In this study, we tested the CERES-Rice model in the Vietnam Mekong Delta for two rice cultivars (Jasmine 85 and VD20) in four experimental sites. We applied a two-stage calibration procedure. In the first stage, initial sets of good performing parameters were identified. In the second stage, the good performing parameter sets and their minimum-maximum ranges were refined using the Robust Parameter Estimation (ROPE) approach. We found that in the first calibration stage, the majority of the parameter sets that performed well in the calibration sites performed poorly in the validation sites. For example, the simulated rice yields were within +/−5% of the observed yields in the calibration sites. But in the validation sites, the same parameter sets resulted in the differences of −77% to 19% for Jasmine 85 and −47% to 9.5% for VD20. This is because these parameter sets had a broad range of values, which compensated each other well in calibration but were less successful to do so in validation. When the ROPE was applied, the value ranges of the parameters narrowed down and the model performance in validation improved. In general, the parameter ranges identified by ROPE are considered more robust and have higher probability to perform better. Our positive results show a good potential of the ROPE approach for calibration of the CERES-Rice model, which can also be applied with similar other crop growth models. The robust parameter value ranges derived in this study may be used as reference parameter values for future applications of the model in the region.