Integrating meteorological insights and machine learning for sugarcane yield prediction in South Gujarat

Abstract

Accurate crop yield forecasting is essential for sustainable agricultural management and food security. This study leverages meteorological parameters and machine learning techniques to develop a robust yield prediction model for sugarcane in the major sugarcane growing district of Gujarat. This study evaluates the performance of Stepwise Multiple Linear Regression (SMLR) and three machine learning (ML) models: "Artificial Neural Networks (ANN)," "Random Forest Regression (RFR)," and "Support Vector Regression (SVR)" for predicting sugarcane yield in four districts of Gujarat, India (Navsari, Bharuch, Surat, and Tapi). Historical yield data (2001–2019) and weather variables were used to train and test the models, with validation performed on a holdout dataset (2020–2022). Results indicate that ANN outperformed other models in most districts, achieving the lowest errors and highest predictive accuracy. Specifically, in Bharuch, ANN achieved an RMSE of 2491.28 t/ha and MAPE of 3.57%; in Surat, the RMSE was 8139.02 t/ha and MAPE 9.92%; in Tapi, the RMSE was 3630.44 t/ha and MAPE 4.43%. In Navsari, the model also performed well with an RMSE of 5388.97 t/ha and MAPE of 8.55%. SMLR demonstrated strong performance in Navsari but required further optimization in other regions. RFR and SVR showed mixed results, with significant errors in Surat and Tapi, highlighting challenges in capturing regional variability. Feature importance analysis revealed that weather variables, such as relative humidity and rainfall, were critical predictors across all districts. The study underscores the importance of integrating remote sensing data with meteorological variables to enhance model accuracy, particularly for SMLR. ANN is recommended for yield forecasting in Bharuch, Surat, and Tapi, while SMLR is suitable for Navsari. These findings provide valuable insights for improving sugarcane yield prediction models, supporting sustainable agricultural practices, and aiding policymakers in resource allocation and risk management.