Paddy Yield Modelling Using Selected Physical and Socio-economic Parameters: A Synergy Between Geospatial and Machine Learning Approaches

Paddy is the staple food crop of South East Asia. Due to a fast-rising global population and scarce agricultural resources, there is a pressing demand to increase rice production to address food scarcity, escalating costs, and poverty. In this regard, remote sensing and GIS gained popularity as cost-effective tools for monitoring crop health, growth, and yield estimation, enabling timely harvesting decisions. The present research focuses on the Purba Bardhaman district, WB, India, aiming to uncover the potential physical and socio-economic factors influencing paddy yield. A blend of geospatial techniques and machine learning methods were applied for this estimation. Multi-source geospatial data were considered for mapping 17 selected factors and categorizing them. The Kharif paddy yield was forecasted using the random forest regression method. The training R2; mean errors like MAE, MSE, and RMSE; and explained variance score came as 0.99, 0.021143, 0.000798, 0.03, and 0.99. The predicted average yield value remained almost the same from 5.15 tonnes/ha in 2018 to 5.18 tonnes/ha in 2022.