Reducing vulnerability of rainfed agriculture through seasonal climate predictions: A case study on the rainfed rice production in Southeast Asia

Abstract

Rainfed rice production needs to contribute more to the current and future world food security due to the increasing competition for limited water supplies including irrigation water. However, it is vulnerable to climate variabilities and extremes hence the utilization of climate predictions is crucial. In this study, the predictive accuracy and applicability of a seasonal climate predictions (SINTEX-F) were evaluated for rainfed rice areas where climate uncertainties are main constraints for a stable and high production. Outputs from SINTEX-F such as daily rainfall, maximum and minimum air temperatures, and wind speed were tested for Indonesia and Lao PDR through the cumulative distribution function-based downscaling method (CDFDM), which is a simple, flexible and inexpensive bias reduction method through removing bias from the empirical cumulative distribution functions of the GCM outputs. The CDFDM outputs were compared with historical weather data. Obtained results showed that discrepancies between SINTEX-F and the historical weather data were significantly reduced through CDFDM for both sites. ORYZA, an ecophysiological rice growth model that simulate agroecological rice growth processes, was used to evaluate the applicability of the SINTEX-F for grain yield predictions. Obtained results from on-farm field validation showed that the predicted grain yield was close to the actual grain yield that was obtained through optimum sowing timing given by the predictions. A normalized root mean square error between predicted and actual grain yield showed satisfactory model fit in predictions. This implies that SINTEX-F was applicable for improving rainfed rice production through CDFDM. However, CDFDM has a limitation in orographic precipitation, the high-resolution daily weather data or a sophisticated special interpolation method should be considered in order to improve the representation of the geographical pattern for the parameters derived from CDFDM.