Strong earthquakes often trigger large-scale secondary geological disasters, causing significant damage to local ecosystems. Among these, landslides represent one of the most severe types of disasters, leading to substantial vegetation loss. Consequently, post-disaster vegetation recovery (Vegetation Productive Capability, VPC) has received widespread attention. While some studies have assessed post-disaster vegetation recovery, they often lack predictions for landslides VPC and fail to fully explore the intricate nonlinear relationships between landslide vegetation recovery and environmental factors. To address these gaps, this study comprehensively evaluates post-disaster vegetation recovery and subsequently conducts modeling analyses on the evaluation results using data-driven techniques. Specifically, we employ Moderate Resolution Imaging Spectroradiometer (MODIS) Net Primary Productivity (NPP) data and a Generalized Additive Model (GAM) to assess and predict VPC for 25,945 landslides triggered by the 2008 Wenchuan earthquake. Experimental results demonstrate the GAM model’s effectiveness in predicting vegetation recovery in landslides, achieving a good fit (AUC=0.798). Random cross-validation (mean AUC=0.797) and spatial hold-out cross-validation (mean AUC=0.752) further validate its efficacy. Moreover, This study also explores complex nonlinear relationships between recovery and environmental factors, revealing significant impacts of population density, slope, temperature, precipitation, and residual vegetation. However, the impacts of terrain factors, specifically profile and plan curvature, show considerable uncertainty. This study marks a significant step towards developing robust methodologies for landslides VPC prediction, paving the way for enhanced post-disaster ecological restoration strategies.