Evaluation of weather based multiple regression and CERES [crop environment resource synthesis] rice models in predicting grain yield in several rice producing provinces in the Philippines

Modeling techniques in crop yield forecasting have been developed, and tested in many developed countries as part of the strategy in predicting crop yields. Such yield models also provide government decision-makers, and planners with tools to ensure food security for their people. Most existing time series records of crop yield in the Philippines, especially that of rice, were started only in 1970. An attempt to utilize the data series was embarked to develop, test, and evaluate multiple linear regression statistical weather-based rice yield models in three provinces, namely: Laguna, Nueva Ecija, and Iloilo. Data analysis was done for modern, and traditional varieties for irrigated, and upland rice during the wet, and dry seasons. Seven statistical model equations for Laguna, four in Nueva Ecija, and eight in Iloilo were developed using 1970 to 1994, 1974 to 1994, and 1975 to 1994 data series, respectively. Four weather parameters were included as independent variables, namely: solar radiation, maximum temperature, minimum temperature, and rainfall. For upland rice, the rainfall influence on crops was analyzed by the crop satisfaction index based on the concept of a simplified water balance. The important predictors for upland models are dekadal indexes during the sixth and tenth dekads of rice growth. Solar radiation totals during reproductive and maturity stages influence rice yields for irrigated and upland environments. Maximum and minimum temperatures were important during reproductive and maturity stages. Rainfall totals affect the different stages of rice growth from vegetative to maturity stage. The predicted yields from 1995 to 2000 in irrigated rice showed absolute percentage errors ranging from 3.81% to 53.33% of the actual yield. The 1995 to 1998 dry season crop predicted rice yield of the Laguna Irrigated Modern Variety Rice Yield Model (LAGIRMD) overestimated the observed yield by 10.71%, 15.29%, 14.28%, and 3.81%, respectively. In 1999, the yield was underestimated by 10.22%, in 2000 it was overestimated by 10.55%. Similarly, the other models tested, and evaluated showed absolute percentage errors ranging from a lowest value of 1.48% to a highest value of 23.84%. The results of the regression analysis of the developed multiple linear regression models indicated that coefficients of determination are significant to highly significant. However, more years of data are still needed to test the significance of the predicted as compared with the actual yields. Nevertheless, the model could be used as an operational tool that must be updated yearly for better crop yield forecast as shown in the evaluation. CERES Rice model is basically a very good tool in decision making process. However, detailed data input requirements are sometimes only available in some areas in the country. Nevertheless, exploring the model in making different experiments and scenarios are very helpful in identifying the best and suitable sites for rice production rather than doing trial and error experiments which are very expensive research and development activity.