Forecasting Yield of Winter Rice Using SMLR and ANN Techniques in Central Brahmaputra Valley Zone of Assam, India

Abstract

The present study was undertaken to develop pre-harvest yield forecast models for winter rice (Sali rice) at the F1 (vegetative) and F2 (mid-season) stages for two districts of the Central Brahmaputra Valley Zone (CBVZ), using yield data and weekly weather data from 1994 to 2019, applying a modified Hendrick and Scholl technique. The models were validated using an independent dataset for two years (2020–21). Model fitting was performed using stepwise multiple linear regression (SMLR) and artificial neural network (ANN) techniques. Model performance was evaluated based on the coefficient of determination (R²), root mean square error (RMSE), and normalized RMSE (RMSEn), while model accuracy was assessed through percent error (PE) between observed and predicted yields. The ANN models demonstrated higher efficiency, with significantly improved R² values and lower RMSE and PE compared to SMLR models, indicating a stronger ability to capture nonlinear relationships between weather variables and yield. This reflects the ANN model’s robustness in handling complex interactions among predictors, leading to more accurate and reliable yield forecasts. Temperature (both maximum and minimum) and relative humidity (morning and afternoon) emerged as the most influential weather parameters affecting winter rice yield in the CBVZ.