Vegetation-Based Early Warning System for Post-Mining Reclamation to Reduce Disaster Risk and Achieve SDGs

Abstract

Post-mining land reclamation in Indonesia faces critical operational challenges, particularly in continuously monitoring long-term vegetation recovery trajectories and proactively preventing slope failures and landslide disasters across expansive, remote concession areas. Traditional field-based monitoring protocols are often constrained by high labor costs, limited spatial coverage, and severe logistical difficulties, making the early detection of localized land degradation difficult. To address these limitations, this study develops and evaluates an operational vegetation-based early-warning framework designed to optimize post-mining land management and directly align with Indonesia's commitment to Sustainable Development Goals (SDGs)—specifically SDG 15 (Life on Land), SDG 13 (Climate Action), and SDG 12 (Responsible Consumption and Production). The proposed system utilizes multitemporal Normalized Difference Vegetation Index (NDVI) and Soil Adjusted Vegetation Index (SAVI) time series derived from quarterly Sentinel-2 Level-2A satellite imagery. These remote sensing spectral indices were integrated with key contextual attributes, including detailed planting-age datasets and topographic parameters derived from Digital Elevation Models (DEMs), such as slope steepness and elevation profiles. This feature space was subsequently used to train a Multilayer Perceptron (MLP) neural network classifier using 462 ground-verified sample pixels collected from an active post-mining reclamation site in Central Sulawesi, Indonesia. The MLP classification model achieved a robust 76% overall predictive accuracy across all designated reclamation status categories. Crucially for disaster risk reduction applications, the model showed exceptional sensitivity to high-risk degradation zones, achieving 98% recall in identifying areas vulnerable to canopy loss, vegetation stress, and potential geotechnical slope instability. To translate these predictive outputs into actionable management tools, the trained model and decision-support pipeline were fully integrated into a web-based Geographic Information System (GIS) dashboard, enabling concession managers and regulatory authorities to visualize spatial risk zones interactively in near real time. The findings demonstrate that combining multitemporal satellite vegetation indices with machine learning classifiers provides a scalable, cost-effective methodology for monitoring post-mining land restoration. By facilitating early identification of site-specific vegetation decline before severe slope failure occurs, this early-warning framework strengthens evidence-based decision-making, mitigates environmental disaster risks, and advances regulatory compliance within Indonesia’s national environmental governance and ecological restoration policy frameworks.