<p>Landslides are major ecological disturbances in mountainous environments and play a critical role in shaping forest structure and long-term carbon storage. Understanding how aboveground biomass (AGB) recovers after landslides disturbances is essential for evaluating forest resilience. However, current knowledge remains limited because most studies rely on small scale field observations or simplified recovery metrics. Remote sensing now allows landscape scale assessment of recovery, but the mechanisms controlling variation in post landslide biomass accumulation remain unclear. To address this gap, we developed an explainable machine learning framework to investigate drivers of AGB recovery within 608 mapped landslide scars in the Kosñipata Valley of the Peruvian Andes. We compared four machine learning models and selected XGBoost as the best performing model, achieving a Pearson’r of 0.713, an RMSE of 17.25, and an MAE of 12.16. SHapley Additive exPlanations were then applied to quantify the influence of environmental, geomorphic and disturbance related variables on biomass patterns and to evaluate nonlinear and interactive responses. The results show that landslide age, elevation and residual vegetation are the dominant variables shaping biomass recovery. Older scars contain substantially more biomass, and the effect is especially strong when residual vegetation is present. Interactions further indicate that residual vegetation plays a more important role in accelerating recovery in older and lower elevation scars. This study demonstrates the value of combining interpretable machine learning with multitemporal remote sensing to uncover mechanisms of post landslide ecological change. The findings provide quantitative insight needed for hazard recovery planning, biodiversity conservation, and carbon assessment in mountainous tropical landscapes.</p>