Climatic determinants of lowland rice development

Abstract Accurate modelling of plant development is the basis for any assessment of climate change impact on crop yields. Most rice models simulate development (phenology) based on temperature and photoperiod, but often the reliability of these models is reduced beyond the environment they were calibrated for. In our study, we tested the effects of relative air humidity and solar radiation on leaf appearance rate in greenhouse experiments and analysed data sets from field studies conducted in two extremely different rice‐growing environments in Nepal and Senegal. We also analysed environmental effects on duration to flowering of one popular IRRI material (IR64) for eight different sites covering the entire temperature range where rice is widely cultivated. Both low relative air humidity and low solar radiation significantly decreased leaf appearance rate. Mean air temperature explained 81% of the variation in duration to flowering across sites, which was furthermore significantly influenced by relative air humidity. Across all sites, a simple linear regression approach including mean air temperature and mean relative humidity in the calculation of duration to flowering led to a root mean square error (RMSE) of 10 days, which was slightly lower than the RMSE of 11 days achieved with an automated calibration tool for parameter optimization of cardinal temperatures and photoperiod sensitivity. Parameter optimization for individual sites led to a much smaller prediction error, but also to large differences in cardinal temperatures between sites, mainly lower optimum temperatures for the cooler sites. To increase the predictive power of phenological models outside their calibration range and especially in climate change scenarios, a more mechanistic modelling approach is needed. A starting point could be including relative air humidity and radiation in the simulation procedure of crop development, and presumably, a closer link between growth and development procedures could help to increase the robustness of phenological models.