Modeling the Relationship between Extreme Values of Rainfall and Rice Yields using Machine Learning in the Lower Northern Provinces

This research presents the development of a model for the relationship between extreme values of rainfall and rice yield in the lower northern provinces. Data on annual rice yield and total annual rainfall from 1990 to 2019 of six lower northern provinces which included Nakhonsawan, Phichit, Phitsanulok, Kamphaengphet, Phetchabun and Sukhotha were used to develop a relationship model for the six lower northern provinces. Three machine learning methods were used for the development of a model. The methods comprised multiple regression analysis, random forest and support vector machine. In this research, the performance of each model in each province was evaluated from the mean absolute error (MAE), the root mean square error (RMSE), the mean square error (MSE) and R-squared value. The results showed that a model created using the random forest method, with extreme rainfall values as independent variable, outperformed the models built using the multiple regression analysis and support vector matching methods. This was particularly true in the provinces of Nakhon Sawan, Phichit, Phitsanulok, and Kamphaeng Phet, where it had both higher R-squared values and lower root mean squared error values. In addition, it was found that using the extreme values of rainfall as an input variable for modeling, the model performance is better than using total annual rainfall. The extreme values of rainfall can explain 64-79% of the variation in rice yield in the lower northern provinces.