Remote Sensing-Based Assessment of Hayfield Productivity and Forage Quality in Mongolia’s Steppe Ecosystems

Sustainable hayfield management is critical for pastoral livelihoods and rural development in Mongolia’s semi-arid steppe ecosystems. This study integrates field measurements (n = 20) from Bornuur sub-province using Sentinel-2A and ALOS PALSAR remote sensing data to assess relationships among forage yield, nutritive quality, terrain, and vegetation indices. The Normalized Difference Vegetation Index (NDVI) shows the strongest correlation with yield (r = 0.89, R² = 0.78), followed by the Enhanced Vegetation Index (EVI) (r = 0.73, R² = 0.58), while the NDWI shows a weak negative correlation (r = −0.35, p = 0.131), suggesting a possible but inconclusive link to water scarcity. Terrain effects are weaker than vegetation indices, with slope showing a moderate negative influence (r = −0.52) and elevation a negligible effect (r = −0.05). To address the risk of overfitting given the limited sample size, we compared a full multivariate model with LASSO regression using 5-fold cross-validation and SHAP analysis. The full model achieves R² = 0.879, RMSE = 0.853, and MAE = 0.653, while LASSO yields comparable performance (R² = 0.869). The NDVI is confirmed as the dominant predictor (estimate = 18.83, p = 0.022), with SHAP analysis showing its primary contribution (mean SHAP = 1.567), far exceeding EVI (0.698) and slope (0.680). These findings demonstrate that satellite-derived vegetation indices, particularly the NDVI, provide a cost-effective tool for forage estimation, enabling evidence-based grazing management, drought risk assessment, and targeted rural development policies in data-sparse regions such as Mongolia.