Vegetation restoration prediction model

Understanding the coupled dynamics of wildfire spread and post-fire vegetation recovery is critical for effective ecosystem management, particularly in the Wildland\u2013Urban Interface (WUI). Despite extensive research on fire behavior and post-fire regrowth, gaps remain in capturing the joint influence of topography, fuel structure, climate, and burn severity on recovery trajectories. In this study, we integrate a physically informed fire-spread model with a machine-learning\u2013based vegetation recovery model to provide a unified framework for evaluating both fire expansion and post-fire EVI dynamics. The fire-spread model, validated against CAL FIRE and NASA observations of the Eaton Fire, reproduces large-scale fire patterns with reasonable accuracy, though early-stage spread is conservative in areas with complex urban\u2013forest interfaces and limited meteorological data. The post-fire EVI recovery model, trained on 15 land-cover types, achieves high predictive performance for large-sample categories (R\u00b2, NSE, and KGE > 0.8), capturing land-cover-specific responses to burn severity, precipitation, temperature, and topography. Shrublands and pastures recover rapidly, often within two years, whereas Mixed and Evergreen Forests require 4\u20135 years due to slower tree establishment and early competition from herbaceous vegetation. Partial dependence analyses indicate that pre-fire EVI, precipitation, and minimum temperature are primary drivers of recovery, while burn severity and slope exert secondary influences. Across all land-cover types, EVI increments decrease over time and stabilize near pre-fire levels, reflecting asymptotic recovery. This integrated modeling approach provides a quantitative tool for anticipating vegetation dynamics and informing post-fire management under increasingly variable climatic and landscape conditions.