Integrating remote sensing and machine learning to evaluate environmental drivers of post-fire vegetation recovery in the Mount Kenya forest

In recent decades, the increasing frequency and severity of wildfires have been linked to climate change and human activities. Understanding the dynamics of post-fire vegetation recovery (PVR) is therefore critical for forest ecosystem restoration and management. The present study analysed burn severities and investigated the impact of environmental variables on post-fire recovery (PVR) in the Mount Kenya Forest Ecosystem (MKFE). The Random Forest (RF) regression model was employed to predict PVR and identify factors that significantly contribute to PVR in the Mount Kenya Forest ecosystem. Landsat satellite imageries from 2011 to 2021 were used to classify burn severity into seven classes based on the differenced Normalized Burn Ratio (dNBR) index. Climate data, soil organic carbon, and topographic variables were integrated into the RF model to predict trends in PVR. The RF model achieved excellent accuracy with a coefficient of determination (R²) of 0.9013 and a Root Mean Square Error (RMSE) of 0.0280 on the training dataset, and R² of 0.8753 and RMSE of 0.0406 on the validation set. The model further revealed a strong positive relationship between temperature and Land Surface Temperature (LST), as well as vegetation recovery. On the other hand, topographic variables demonstrated a strong negative relationship with vegetation recovery. The combined influence of topographic and temperature condition variables highlights the heterogeneous nature of recovery processes, hence the need for spatially targeted management strategies. These findings have significant implications for adaptive management strategies in tropical montane ecosystems facing increasing wildfire risks.