Modelling Vegetation Patterns in Southern Kaduna Nigeria using XGBoost and SHAP

Abstract

Understanding the drivers of vegetation dynamics is critical for sustainable land management and ecological conservation. This study applied the XGBoost machine learning algorithm to model NDVI-based vegetation patterns in Southern Kaduna, Nigeria, using hydrological, climatic, topographic, and anthropogenic predictors within the R statistical computing environment. The model demonstrated excellent predictive performance, with training R² = 0.9978, RMSE = 0.0022, MAE = 0.0016; spatial cross-validation R² = 0.9054, RMSE = 0.0142, MAE = 0.0100; and independent testing R² = 0.9523, RMSE = 0.0101, MAE = 0.0076, highlighting its ability to capture complex nonlinear interactions. Global and local SHAP analyses revealed that NDWI, representing surface moisture and hydrological conditions, was the most influential predictor, followed by NDBI, reflecting urbanization and land-use change. Climatic variables, slope, elevation, evapotranspiration, and population density exerted secondary but meaningful effects. SHAP dependence plots further illustrated the directional relationships and interactions among predictors, confirming that vegetation greenness is highly sensitive to moisture availability and urban development pressures. These findings emphasize the dominant role of hydrological and anthropogenic factors in controlling vegetation variability in the region. The study provides actionable insights for ecological management, urban planning, and sustainable development in Southern Kaduna.