An improved model to simulate rice yield

Rice is the staple food for about half of the world’s population. Although global production has more than doubled in the last 40 years, food security problems still persist and need to be managed based on early and reliable forecasting activities. This is especially true since the frequency of extreme weather events is forecasted to increase by the intergovernmental panel on climate change (IPCC). The most advanced crop yield forecasting systems are based on simulation models. However, examples of operational systems implementing models which are suitable for reproducing the peculiarities of paddy rice, especially on small scales, are missing. The rice modelWARM is used within the crop yield forecasting system of the European Commission. In this article we evaluated the WARM model for the simulation of rice growth under flooded and unflooded conditions in China and Italy. The WARM model simulates crop growth and development, floodwater effect on the vertical thermal profile, blast disease, cold-shock induced spikelet sterility during the pre-flowering period and hydrological peculiarities of paddy soils. We identified the most relevant model parameters through sensitivity analyses carried out using the Sobol’ method and then calibrated using the simplex algorithm. Data from 11 published experiments, covering 13 locations and 10 years, were used. Two groups of rice varieties were identified for each country. Our results show that the model was able to reproduce rice growth in both countries. Specifically, the average relative root mean square error calculated on aboveground biomass curves was 21.9% for the calibration and 23.6% for validation. The parameters of the linear regression equation between measured and simulated values were always satisfactory. Indeed, intercept and slope were always close to their optima and R2 was always higher than 0.79. For some of the combinations of country and simulated variable, the indices of agreement calculated for the validation datasets were better then the corresponding ones computed at the end of the calibration, indirectly proving the robustness of the modeling approach. WARM’s robustness and accuracy, combined with the low requirements in terms of inputs and the implementation of modules for reproducing biophysical processes strongly influencing the year-to-year yield variation, make the model suitable for forecasting rice yields on regional, national and international scales.