Adaptive and Differentiated Land Governance for Sustainability: The Spatiotemporal Dynamics and Explainable Machine Learning Analysis of Land Use Intensity in the Guanzhong Plain Urban Agglomeration

Urban agglomerations underpin regional economic growth and sustainability transitions, yet the spatial heterogeneity and drivers of land use intensity (LUI) remain insufficiently resolved in inland settings. This study develops a high-resolution framework&mdash:combining a 1 km hexagonal grid with XGBoost-SHAP&mdash:to (i) map subsystem-specific LUI evolution, (ii) identify dominant drivers and nonlinear thresholds, and (iii) inform differentiated, sustainable land governance in the Guanzhong Plain Urban Agglomeration (GPUA) over 2000&ndash:2020. Composite LUI indices were constructed for human settlement (HS), cropland (CS), and forest (FS) subsystems: eleven natural, socioeconomic, urban&ndash:rural, and locational variables served as candidate drivers. The results show marked redistributions across subsystems. In HS, the share of low-intensity cells declined (86.54% to 83.18%) as that of medium- (12.10% to 14.26%) and high-intensity ones (1.22% to 2.56%) increased, forming a continuous high-intensity corridor between Xi&rsquo:an and Xianyang by 2020. CS shifted toward medium-intensity (32.53% to 50.57%) with the contraction of high-intensity cells (26.62% to 14.53%), evidencing strong dynamism (55.1% net intensification: 38.5% net decline). FS transitioned to low-intensity dominance by 2020 (59.12%), with stability and delayed growth concentrated in conserved mountainous zones. Urban&ndash:rural gradients were distinct: HS rose by &gt:20% (relative to 2000) in cores but remained low and stable in rural areas (mean &lt: 0.20): CS peaked and stayed stable at fringes (mean &asymp: 0.60): FS shifted from an inverse gradient (2000&ndash:2010) to core-area recovery by 2020. Explainable machine learning revealed inverted U-shaped relationships for HS (per capita GDP) and CS (population density) and a unimodal peak for FS with respect to distance to urban centers: model performance was strong (HS R2 up to 0.82) with robust validation. Policy recommendations are subsystem-specific: enforce growth boundaries and prioritize infill/polycentric networks (HS): pair farmland redlines with precision agriculture (CS): and maintain ecological redlines with differentiated conservation and afforestation (FS). The framework offers transferable, data-driven evidence for calibrating thresholds and sequencing interventions to reconcile land use intensification with ecological integrity in rapidly urbanizing contexts.