ORYZA (v3) rice crop growth modeling for MR269 under nitrogen treatments: Assessment of cross-validation on parameter variability

Abstract

Complex rice crop models are sometimes evaluated with limited data. A one-round validation approach is typically used, in which data is arbitrarily divided into two or more mutually exclusive sets. Some sets are used for calibration, while others are used for validation. It is unknown whether a more structured cross-validation approach would result in variations in the calibrated parameters when applied to the same data sets. The objectives of this paper are (i) to calibrate and evaluate the performance of ORYZA (v3) for simulation of high-yielding MR269 rice variety physiological traits grown in Malaysian rice fields with limited data using a cross-validation approach; and (ii) to assess crop genetic parameter variability that resulted from the cross-validation approach and explore the benefits of the approach with limited data. The cross-validation approach produces six calibrated crop parameter sets (three calibration–validation combinations for two parameter cohorts). Further validation with independent field data sets revealed that two of the six calibrated crop parameter sets produced satisfactory to good fits for the crop dry biomass of green leaves, panicles, and stems, as well as the dry total aboveground biomass of MR269 (NSE ≥ 0.5). This study implies that the plausibility of multiple feasible parameter sets must be acknowledged, and a more robust calibration approach must be considered when working with a complex crop model with limited data. The systematic cross-validation approach as demonstrated in this study allows for a more extensive model evaluation given small data sets.