Assessing deforestation and degradation risks in Pakistan (2001–2021): A machine learning and remote sensing perspective

Deforestation and forest degradation in semi-arid regions drive critical environmental challenges. This study analyzed deforestation and fragmentation dynamics in Pakistan (2001–2021) using multi-source remote sensing data, Google Earth Engine (GEE), and machine learning (Random Forest (RF) and Extreme Gradient Boosting (XGBoost)). RF achieved 94.81% overall accuracy in land use/land cover (LULC) classification, while XGBoost demonstrated 90.17% accuracy in spatial fragmentation analysis. The net forest cover increased by 3.48% (from 43,799.72 km² to 45,522.45 km²); however, structural fragmentation also intensified, as previously contiguous forest areas became more divided reflected in a 3.4% increase in edge forests and 2.9% rise in isolated islets. High-risk deforestation zones (9.09% of the landscape) were concentrated near urban-agricultural frontiers, driven by road proximity and elevation (SHAP analysis). Urban areas expanded by 20.2% (2010–2021), displacing cropland and barren land, which declined by 1.94%. Morphological Spatial Pattern Analysis (MSPA) revealed persistent connectivity loss, with core forests representing only 3.51% of the study area. These findings underscore the need for conservation strategies prioritizing landscape connectivity and mitigating connectivity-based degradation in semi-arid regions. • RF and XGBoost models achieved 94.81% and 90.17% accuracy in LULC and fragmentation. • Forest cover increased by 3.48%, while edge and islet fragmentation worsened. • High-risk deforestation zones (9.09%) linked to urban-agricultural expansion and elevation. • Urban growth (20.2%) reduced cropland and barren land, affecting forest connectivity.