Unraveling Spatiotemporal Heterogeneity and Driving Mechanism of Vegetation Recovery in Co-Seismic Landslide Areas via Nonlinear Trajectories and Bayesian Networks.

Earthquakes often trigger extensive landslides that strip large areas of vegetation, leaving barren surfaces and posing significant challenges to ecological restoration. A notable example is the 2008 Wenchuan earthquake, which induced numerous co-seismic landslides and created widespread bare areas that have hindered vegetation recovery. Understanding post-disruption vegetation recovery patterns and their climatic drivers is therefore essential for targeted restoration efforts. This study establishes an integrated framework combining nonlinear trajectory analysis with Bayesian probabilistic inference to quantify vegetation recovery dynamics and identify key climatic controls. Using Landsat time-series data (2008-2023), we applied a Generalized Additive Model (GAM)-based nonlinear trajectory approach to extract recovery trajectories, identify trajectory-derived recovery inflection points marking the transition from accelerated to decelerated recovery, and quantify spatial variations in recovery rates. A Bayesian Network (BN) was then used to quantify the influence of ecological memory and environmental drivers on recovery regimes, revealing critical temperature-moisture-radiation configurations and enabling scenario-based evaluation of climatic constraints on regional recovery potential. GAM results indicate that most areas reached a critical recovery inflection point within 5-7 years, while delayed recovery was mainly observed in high-elevation, rugged terrain. The BN further identified that a synergistic "warm-moist-moderate light" regime is necessary for fast recovery. Scenario simulations highlighted clear spatial differences in limiting factors: solar radiation is the dominant constraint in rugged highlands, whereas temperature and precipitation exert stronger controls in low-elevation or moderately disturbed areas. This spatial variability suggests that restoration interventions should be tailored to local conditions to maximize efficiency. Overall, this study provides a transferable framework for diagnosing environmental constraints and guiding restoration strategies to enhance ecosystem recovery capacity in disaster-prone mountainous landscapes.