Abstract
Aman rice, a major staple crop in Bangladesh, is cultivated during the monsoon season and is highly dependent on climatic conditions such as rainfall and temperature. This study aims to identify the most effective machine learning models for predicting Aman rice yields by leveraging 52 years of historical data (1970-2022). Data preprocessing included outlier correction, statistical imputation, and aggregation of monthly averages for varibales like rainfall, temperature, humidity and others during the monsoon (June-September). Various machine learning models - Random Forest, Neural Network, Decision Tree, Linear Regression, and Gradient Boosting - were employed to capture yield trends under changing climatic conditions. Each model was evaluated based on Root Mean Squared Error (RMSE), R-squared (R(2)), and Mean Absolute Error (MAE). Random Forest emerged as the most accurate model showing robustness to climate variability through sensitivity analysis. While Gradient Boosting also performed well, though with slightly higher error margins. Linear Regression provided reasonable outputs, but it struggled with non-linear patterns. In contrast, Neural Networks and Decision Trees showed less accuracy in capturing intricate relationships between climate variables and rice yields. The Random Forest model predicts Aman rice yields to reach 133.31 metric tonnes by 2030 (34.11 % of total rice production) and 140 metric tonnes by 2050 (32.86 %). Climate projections suggest a rise in temperatures from 26.5°C to 37.41 °C in 2030 to 27.33°C-38.26 °C by 2050, with monsoon rainfall increasing slightly from 302.37 mm to 305.7 mm. These changes in climatic conditions could place additional stress on rice production, especially due to higher temperatures. The findings align with international studies highlighting the challenges that rising temperatures and fluctuating rainfall pose to crop yields. These findings emphasize the need for adaptive agricultural techniques and policies to mitigate climate change impacts on rice production, supporting food security and sustainable development in Bangladesh.