Rice (Oryza sativa L.) is the primary staple crop of Uttar Pradesh, occupying approximately 36.5 % of the state's total agricultural area. Reliable pre-harvest yield forecasting is essential for agricultural planning, food security management and climate-resilient decision-making. This study aimed to develop pre-harvest forecasting models for rice yield in Varanasi district by assessing the individual, joint and composite effects of weather variables using statistical modelling. Long-term rice yield time series spanning 38 years (1986–2023) were used along with weekly weather parameters such as maximum and minimum temperature, rainfall and wind velocity-covering standard meteorological data from weeks 24–42 every year. Weather indices (weighted and unweighted) and their relationships were generated using correlation-based methods. Stepwise multiple regression was applied to fit models incorporating a temporal trend (T) to account for technological advancements. Among individual weather variables, rainfall had the strongest explanatory power (R² = 41.47 %), followed by maximum temperature (R² = 29.53 %), wind velocity (R² = 22.81 %) and minimum temperature (R² = 19.88 %). Joint effect models improved explanatory power, with the maximum temperature–rainfall combination achieving the highest R² (44.43 %). The composite model, combining individual indices and interaction terms, explained 44.08 % of yield variation. The T was consistently positive and significant across all models, reflecting the contributions of improved crop varieties and management practices. Combining meteorological indices with stepwise regression modelling provides a reliable approach for pre-harvest rice yield forecasting.
