YIELD OF MAIN AGRICULTURAL CROPS IN THE PENZA REGION: ANALYSIS OF PREDICTORS AND MEDIUM-TERM FORECAST

Against the backdrop of increasing climatic instability and import substitution policies, medium-term crop yield forecasting is acquiring strategic importance for regional agricultural planning. Penza Oblast, which holds a leading position in sugar beet production and several other crops within the Volga Federal District, lacks verified systematic forecasting research. The aim of this study is to develop and validate a suite of predictive yield models for five major crops in Penza Oblast — grains and legumes, sugar beet, sunflower, potato, and vegetables — over a forecasting horizon of up to three years. The models are built on Ridge regression with L2 regularisation, exponential observation weighting, and an expanding window validation scheme. For each crop, an extensive grid search was conducted (12000–25000 configurations) across a space of 135 predictors comprising regional statistical data, MODIS NDVI satellite indices, agroclimatic indicators, and ERA5-Land reanalysis data for the period 1990–2025 (n = 36). Automated feature selection produced crop-specific predictor sets of 17–31 variables with low cross-crop overlap (Jaccard coefficient 0.09–0.26) and agronomically interpretable composition. Backtesting yielded MAPE values ranging from 6,52% (sugar beet) to 15,18% (sunflower). Ridge regression outperformed Random Forest and XGBoost across all crops by 4,5–9,7 percentage points, confirming the advantage of regularised linear models under small sample conditions. Point and interval forecasts for 2026–2028 (bootstrap, N = 2000) indicate moderate yield growth for grains and sugar beet, with stabilisation projected for the remaining crops. The developed forecasting framework achieves high to good accuracy (per the Lewis scale) for four out of five crops and is recommended for integration into regional medium-term agricultural planning systems — primarily for regional agricultural authorities and large agro-industrial holdings.