Forecasting of wheat yield through statistical models for different districts of Himachal Pradesh, India

Abstract

Aim of study: The objective was to construct weather-based forecasting models that can estimate wheat yield well in advance at pre-harvest stages. Area of study: This study primarily focused on diez wheat-growing areas in Himachal Pradesh, India. Material and methods: The forecast model was developed using long-term data on yield and meteorological parameters spanning from 1990 to 2020. The models were then validated using data from the remaining two years, 2021 and 2022. The yield projection for wheat was derived for ten districts at two growth stages, namely vegetative stage (F1) and pre-harvest stage (F2). The stepwise regression analysis was conducted using a trial and error strategy to determine the optimal combination of predictors that were statistically significant at a 5% level. Multiple regression techniques were employed to fit the model, with the best fit determined by the highest coefficient of determination (R2) value and the lowest percentage error. Main results: The findings indicated that the constructed forecast model successfully accounted for 43 to 93% of the variability in wheat output. The crop yield forecasting models provided accurate projections, with a percentage deviation ranging from 1.2 to 9.8 at the F1 stage and from 0.2 to -9.1 at the F2 stage. These deviations fell within the allowed limit of ±10%. Throughout both years of validation, the observed yield closely matched the anticipated wheat output for all ten districts. Research highlights: The characteristics that had the greatest impact on wheat yield were temperature and relative humidity. The crop forecasting models provided accurate estimates of wheat yield throughout all stages of the forecast for the majority of districts.