Projecting Rice Yield in the Mekong River Delta under Climate Change

Abstract

The Mekong River Delta (MRD) and the Red River Delta (RRD) in Vietnam are the most important rice production and export regions in the world. Predicting rice yield in the MRD and the RRD under climate change is crucial for the global rice supply and food security. This paper aims to use a machine learning method (Random Forest) and a deep learning algorithm (Long Short-Term Memory) to predict rice yield in the two deltas. The Long Short-Term Memory (LSTM) method is specifically designed for sequential predictions. We found that LSTM outperformed Random Forest and Multilinear Regression in terms of Root Mean Square Error and R-squared. The LSTM model we built also beat most of the traditional crop simulation models in extant literature. The predictions show that the rice yields between 2025 and 2100 will increase in the MRD with volatility under SSP2-4.5, SSP3-7.0, and SSP5-8.5 climate change scenarios. However, the rice yield between 2025 and 2100 will not be higher than the peak in 2015. The rice yield in the RRD will decrease in the next 80 years under SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5. We find that climate change has spatially heterogeneous effects on rice yield. Our results also suggest that policy makers should prepare for possible food shortages under climate change.