Modelling grain yield under the influence of agrotechnological and climatic factors: a regularised regression analysis

Purpose. The purpose of the study is to model the yield of cereal and legume crops at a regional level (using the example of the Odesa region) based on a complex of agrotechnological and climatic factors using regularised regression methods (Ridge, Lasso, ElasticNet). Methodology. The study applies economic and mathematical analysis using official statistics for 1995–2024. For modelling, 11 independent variables were selected, reflecting agrotechnological and climatic factors. The assessment was carried out using classical linear regression and its regularised modifications. Multicollinearity diagnostics, cross-validation, and analysis of statistical significance of coefficients were performed. Results. The study proved that regularised regression models (Ridge, Lasso, ElasticNet) significantly outperform classical OLS regression in yield forecasting accuracy under multicollinearity conditions. Ridge regression demonstrated the best predictive performance (R² = 0.606; RMSE = 4.46), effectively stabilising coefficient estimates for correlated predictors. Testing of the research hypothesis confirmed the statistically significant positive impact (p < 0.05) of agrotechnological factors on yield: sown area, mineral fertilisers, and pesticides. Among climatic factors, precipitation and the hydrothermal coefficient were identified as key positive drivers of productivity. Conversely, the hypothesis regarding negative stressors was confirmed: air moisture saturation deficit and wind speed exert a significant negative impact (p < 0.05) on yield, highlighting the vulnerability of the region to arid conditions. Originality. The novelty lies in combining agrotechnological and climatic indicators within the framework of regularised models. This approach accounts for multicollinearity and at the same time identifying the most significant yield factors. The obtained results provide deeper insights into relationships between grain yield and agrotechnological and climatic factors in the conditions of the Southern Steppe of Ukraine. Practical implications. The results of the study can be used by agricultural enterprises and regional authorities to optimise fertiliser systems, plan crop areas, select adaptive varieties, and increase the resilience of agricultural production to climate risks. The proposed models provide tools for predicting yields and making management decisions in agriculture.