Enhanced remote sensing framework for early detection and quantification of underground spontaneous combustion in reclaimed coal waste dumps

The spontaneous combustion of coal waste dumps after reclamation presents a major challenge for post-mining ecological restoration. Early detection of combustion risk is essential for effective management. Traditional ground-based and unmanned aerial vehicle (UAV) monitoring are limited by cost, technical constraints, and observation frequency. Focusing on this issue, this study introduces an enhanced framework for identifying combustion areas and assessing their intensity using Sentinel−2 time-series satellite imagery. The Breaks for Additive Seasonal and Trend algorithm is applied to five satellite-based vegetation index (VI) time series, and combustion areas are classified based on a random forest model. Moreover, a novel Spontaneous Combustion Intensity Index (SCII) is developed to quantify intensity. Validation using deep (25 cm) soil temperature data confirms the effectiveness of the SCII, with an overall accuracy of 95.7%. The SCII strongly agrees with the soil temperature data (R² 0.54–0.71; root mean square error (RMSE) 3.87–8.65 °C; mean absolute error (MAE) 1.77–6.86 °C). Furthermore, temporal SCII changes align with the subsurface temperature variation (Spearman’s ρ = 0.78). Compared to traditional methods, this approach enables large-scale, continuous monitoring, capturing early-stage and dynamic processes while reducing reliance on ground or UAV observations.