Mapping critical thresholds and dynamic drivers of net primary productivity response to shrub encroachment in Xilin Gol grasslands using multi-source remote sensing

Shrub encroachment is a global ecological phenomenon altering grassland structure and function, yet its spatially explicit relationships with net primary productivity (NPP) remain poorly quantified. We integrated multi-source remote sensing data, including MODIS Normalized Difference Vegetation Index (NDVI), Landsat classifications, and localized meteorological inputs, to assess NPP dynamics across the Xilin Gol grasslands from 2000 to 2020. We utilized the refined Carnegie-Ames-Stanford Approach (CASA) model to obtain NPP estimates that were consistent with MODIS standard products (R2 ranging from 0.66 to 0.92) and showed favorable agreement with field-based NPP samples (R2 = 0.71). Using piecewise regression, we identified a critical shrub cover threshold (15–27%), revealing a threshold-dependent nonlinear response where moderate encroachment coincided with enhanced productivity via resource-island effects, while excessive dominance was associated with a decisive decline. Furthermore, a Random Forest (RF) model coupled with SHapley Additive exPlanations (SHAP) unveiled a non-stationary hierarchy of environmental controls. We documented a fundamental temporal shift in predictor importance: from early topographic dependency and thermal regulation to a contemporary regime primarily dominated by precipitation dynamics. While hydroclimatic variability remains the primary constraint, site-specific soil and topographic factors play a supplementary but emerging role in modulating realized productivity. Our study provides a transferable geospatial framework for identifying ecological thresholds and offers evidence-based targets for adaptive management strategies in dryland ecosystems navigating climate change.