Spatiotemporal dynamics and environmental drivers of fractional vegetation cover in a Semi-Arid region using machine learning

Understanding the spatiotemporal dynamics of vegetation in semi-arid regions is critical for effective land management and climate adaptation. This study investigates long-term changes in fractional vegetation cover (FVC) across Dera Ismail Khan (D.I. Khan), Pakistan, from 2000 to 2024 using multi-sensor Landsat imagery, climatic datasets, and advanced modeling approaches. FVC was derived using the NDVI-based Pixel Dichotomy Model (PDM), while temporal trends were assessed using the Trend-Free Prewhitened Mann–Kendall (TFPW-MK) test and Theil–Sen slope estimation. Results indicate a significant greening trend, with an average annual increase of 0.37% in FVC, though spatial heterogeneity was evident, riverine and southern zones experienced higher vegetation gains, while the northern arid zones remained low in cover. Geographically Weighted Regression (GWR) revealed spatial non-stationarity in the relationships between FVC and environmental drivers, with temperature, vapor pressure deficit (VPD), and elevation exerting locally varying effects. Complementary Extreme Gradient Boosting (XGBoost) modeling, interpreted through SHapley Additive exPlanations (SHAP) values, confirmed the dominant influence of elevation and temperature and provided nonlinear insights into vegetation-climate interactions. Despite high model accuracy (R² = 0.93), anthropogenic factors such as land-use change were not explicitly modeled, representing a key limitation. This integrated framework underscores the utility of combining satellite-based FVC estimation with interpretable machine learning and spatial regression to identify ecologically vulnerable zones in data-scarce drylands. The findings support targeted interventions for land restoration and adaptive resource planning in semi-arid regions.

Text 120645.pdf - Published Version Restricted to Repository staff only Download (8MB) Official URL or Download Paper: https://link.springer.com/article/10.1007/s00704-0...

Actions (login required)

View Item