Decline in rice grain yields with temperature: Models and correlations can give different estimates

Abstract

Based on an analysis of yield and weather data for the years 1992–2003, it has been suggested that rice crop models are inadequate because they fail to predict that rice yields decline by 15% °C −1 (mean daily air temperature) or 10% °C −1 (mean minimum temperature), temperatures averaged over the crop growth duration (about 100 days). We investigate that claim. A mechanistic and an empirical model of rice crop growth were used to make predictions of yield using the same weather data sets used in the regression analyses that supported the 15 and 10% claims. The models were used to predict yield with temperature changed by −2, −1, 0, +1 and +2 °C relative to the average for all the years (26 °C) with solar radiation held constant. Over the 4 °C temperature range, ORYZA2000 and EEQ predicted yield declines of about 0.37 and 0.71 t ha −1 °C −1 (3.5 and 7.6% °C −1 from the base yield at 26 °C). When the actual weather data for each year were used in the models, there was no significant relationship between the predictions for each year and mean daily air temperature. Even though minimum temperature was not used for the simulation of any processes in the models, predicted grain yields were significantly correlated with minimum temperature. The slope of the regression line between predicted yield and minimum temperature for the models gave a yield decline of about 1.5 t ha −1 °C −1 (which was 13.7% °C −1 from the base yield at a minimum temperature of 22.1 °C). When the weather data for the years 1992–2003 were analyzed, there was a significant negative correlation between solar radiation and minimum temperature. The lowest yields occurred in the wettest years and there was a significant negative relationship between harvest index and rainfall. We conclude that temperature responses of the models are adequate for predicting the observed results. Crop responses to temperature (below the high temperatures that cause infertility in rice) are of the order of −0.5 t ha −1 °C −1 (or about −6% °C −1 at the base yield at average mean daily temperature of 26 °C), once they are separated from the effects of other environmental factors. Yield declines calculated by regression from selected weather elements can be misleading because of correlations among the weather elements.