Optimizing of Phenological Crop Model Parameters for Rice (Oryza sativa).

Abstract

Crop phenology models are important components of crop growth models. In the case of phenology models, generally only a few parameters are calibrated and default cardinal temperature values are used. A recent study shows that this can lead to a temperature-dependent systematic phenology prediction error. We expanded on this study to include another model, more rice phenological growth stages, and different optimization approaches. We used phenological models from two rice growth models (Oryza2000 and CERES-Rice) to assess the importance of optimizing cardinal temperatures on model performance, and hence systematic error. We used two optimization approaches: the typical single-stage (planting to heading) and three-stage model optimization (for planting to panicle initiation (PI), PI to heading (HD), and HD to physiological maturity (MT)) to simultaneously optimize all model parameters: cardinal temperatures, development rate constants, and photoperiod sensitivity parameters. Data for this study came from eight California rice cultivars collected over three years and six locations. A temperature dependent systematic error was found for all cultivars and stages, however it was generally small and insignificant. Both optimization approaches in both models resulted in only small changes in cardinal temperature relative to the default values and thus optimization of cardinal temperature did not reduce this error. Compared to single stage optimization, the three-stage optimization had little effect on determining time to panicle initiation or heading but significantly improved the precision in determining the time from heading to maturity: the RMSE reduced from an average of 7.8 to 3.0 in Oryza2000 and from 8.6 to 3.2 in CERES-Rice. 

Keywords: rice, phenology stages, model optimization, temperature, sensitivity analysis.