Weather-Based Prediction of Sorghum Yield in Hamirpur District

Abstract

The study was conducted during 2024-25. Five multivariable models were tested for the sorghum yield prediction in Hamirpur district of the Bundelkhand region, India namely Artificial Neural Network (ANN), PCA with Artificial Neural Network (PCA-ANN), Least Absolute Shrinkage and Selection Operator (LASSO), Ridge Regression, and Stepwise Multiple Linear Regression (SMLR). Average meteorological data of different crop phenological stages were used as predictor variables to build the models. The coefficient of determination (R²), root mean square error (RMSE) and mean absolute error (MAE) were used to evaluate the model's performance. The five models were compared, and the PCA-ANN model was found to have the highest prediction accuracy, suggesting the dimensionality reduction process of PCA improved predictive ability of the ANN model. The PCA-ANN model had the highest predictive performance, with low prediction errors (MAE = 0.33 q ha⁻¹ and RMSE = 0.50 q ha⁻¹) during the calibration process. The evaluation metrics ranked the forecasting performance of the models in the order: PCA-ANN>ANN>SMLR>LASSO>Ridge Regression. PCA, along with ANN, has emerged as a powerful and reliable technique for sorghum yield prediction under the influence of changing climate and can prove useful for farmers in the region of Bundelkhand for decision support.