Effective trail management is essential for preventing environmental degradation and promoting sustainable recreational use. This study proposes a GIS-based, explainable machine learning framework for predicting forest trail degradation using exclusively environmental variables, eliminating the need for costly visitor monitoring data that remains unavailable in most operational forest settings. Field surveys conducted in Geumjeongsan, South Korea, classified trail segments as degraded or non-degraded based on physical indicators such as erosion depth, trail width, and soil hardness. Environmental predictors—including elevation, slope, trail slope alignment (TSA), topographic wetness index (TWI), vegetation type, and soil texture—were derived from spatial analysis. Three machine learning algorithms (Binary Logistic Regression, Random Forest, and Gradient Boosting) were systematically compared using confusion matrix metrics and AUC-ROC (Area Under the Receiver Operating Characteristic Curve). Random Forest (RF) was selected for its strong performance (AUC-ROC = 0.812) and seamless integration with SHAP (SHapley Additive exPlanations) for transparent interpretation. Spatial block cross-validation achieved an AUC-ROC of 0.729, confirming robust spatial generalization. SHAP analysis revealed vegetation type as the most significant predictor, with hardwood forests showing higher degradation susceptibility than mixed forests. A susceptibility map generated from the RF model indicated that 40.7% of the study area faces high to very high degradation risk. This environmental-only approach enables proactive trail management across data-limited forest systems globally, providing actionable insights for sustainable trail maintenance without requiring visitor use data.

Accurate prediction of land use and land cover (LULC) change is essential for sustainable development and climate change adaptation planning. This study projects LULC changes across 17 administrative regions of South Korea from 2020 to 2050 using the Future Land Use Simulation (FLUS) model under four integrated SSP-RCP scenarios: SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5. The model was calibrated with land cover data for 2000–2010 and validated against observations for 2010–2020 using socioeconomic variables together with CMIP6 climate projections. In practical terms, FLUS produces scenario-based maps of future land patterns that inform land regulation, infrastructure planning, and climate adaptation. Across all scenarios, urban areas expanded by 488,000–585,000 ha, mainly through the conversion of agricultural land, which accounted for 10–24% of transitions in high-growth regions. Agricultural land decreased by 124,000–174,000 ha, and forests declined by 473,000–572,000 ha. Transformation intensity peaked around 2030 and then slowed in later decades. Urban expansion was greatest under SSP5-8.5, followed by SSP3-7.0, SSP1-2.6, and SSP2-4.5. Gyeonggi Province exhibited the most pronounced spatial change, whereas Seoul showed limited additional growth consistent with its already saturated urban structure. Validation results indicated an overall accuracy range of 57–83% with metropolitan areas generally outperforming provincial regions. These findings reveal spatial and temporal hotspots of land cover change and provide region-specific information that can guide urban development, land and ecosystem management, climate adaptation policy, and progress toward carbon neutrality.

Korean local governments promote industrial parks as a response to depopulation, yet their effect on existing rural settlement space remains unmeasured. Transferring the road-effect-zone framework to industry, we separated industrial occupation, the exposure of settlement land (residential land and farmland) within 100 m of industry, and the contiguity reduction (a scenario measure: shrinkage of the largest contiguous settlement patch when settlement land within that zone is excluded, not actual loss) in 1232 townships of 121 non-capital counties. Unlike roads, industry rarely subdivided settlement space; it encroached on it. In 2025 the median park township had 2.0% of settlement land exposed, against 0.4% elsewhere, and a contiguity reduction of 2.1%, against 0.2%; it was 20% or more in 10.1% of park townships, against 2.9%. Per percentage point of township area, industry dispersed outside park boundaries was associated with a 4.6-point contiguity reduction, whereas industry within parks showed no detectable association. In 2000–2007, half of new industrial land replaced settlement land. Of 116 county ordinances retrieved and reviewed, only three set a distance for factories in general, against 100 for solar power facilities. Park-related industrialisation is associated with spatial encroachment on existing settlement that population or employment outcomes alone do not capture.

Grid-level population projection is essential for spatial planning under demographic decline, particularly for ensuring that population allocation accounts for grid extinction risk. This study develops a two-stage machine learning framework to predict residential grid transitions across South Korea’s 1 km grid system and assess how spatial policies shape depopulation outcomes through 2050. Stage 1 employs Random Forest classification to predict grid state transitions (macro-averaged F1 score = 0.694), while Stage 2 applies LightGBM regression for population prediction (coefficient of determination = 0.950). The extinction probability map from Stage 1 is incorporated into scenario simulations to adjust population allocation based on predicted residential viability. Feature importance analysis reveals that baseline population, household count, and demographic composition are key determinants of grid-level residential transitions. Five spatial development scenarios simulated through 2050 reveal substantial policy sensitivity. Cumulative extinction rates range from 3.1% under extreme dispersion to 24.5% under extreme concentration, representing a 25 percentage point divergence attributable to spatial allocation policy. Provincial heterogeneity is pronounced, with rural provinces facing extinction rates up to 39.9% while metropolitan areas remain largely unaffected. Comparing scenario outcomes enables pre-identification of policy-sensitive grids (19.5%) where allocation choices determine residential survival. These grids are predominantly located in areas with high forest cover and greater spatial isolation compared to stable grids, but differ in demographic profiles. Aging-Vulnerable grids (14.0%) exhibit high aging ratios with limited economic base, while Moderate-Vulnerability grids (5.5%) show younger demographics with relatively higher economic activity. These differential characteristics provide a spatially explicit basis for differentiated policy responses. Beyond depopulation planning, the spatial outputs of this framework can inform related planning domains such as land use transition planning, carbon management, and infrastructure prioritization under demographic decline.