UAV-thermal driven machine learning framework for predicting methane emissions in tropical landfills

• Developed an UAV-thermal driven machine learning framework to predict CH₄ flux in tropical landfills. • K-means clustering revealed CH₄ emissions driven by thermal zones using 3D modeling, with prediction accuracy (R² > 0.9) achieved in both spatial and temporal models. • Achieved high prediction accuracy using random forest models (R² = 0.801 spatially, 0.851 temporally) with only temperature as input. • Identified UAV-derived temperature as the most influential predictor through SHAP analysis. • Demonstrated practical potential for UAV-assisted, low-cost CH₄ monitoring without direct gas measurements. Accurate quantification of methane (CH₄) emissions from landfills is critical for effective greenhouse gas mitigation. Conventional ground-based monitoring techniques, however, often lack spatial and temporal resolution—particularly in tropical environments where thermal and microbial dynamics are complex. This study proposes an integrative framework that combines UAV-based thermal imaging, ground-based temperature sensing, and in situ CH₄ flux measurements to model emissions from an active tropical landfill. Spatial and temporal datasets were analyzed using K-means clustering to identify thermal-emission zones, and centroid-based regression models from clustering demonstrated strong explanatory power (R² > 0.9), supporting temperature-driven CH₄ prediction. Random forest regression was employed to evaluate predictive performance, yielding R² values of 0.865 (training), 0.853 (testing), and 0.801 (internal validation) for spatial analysis, and 0.851, 0.826, and 0.802, respectively, for temporal data. SHapley Additive exPlanations (SHAP) were applied to interpret model predictions, identifying UAV-derived temperature as the most influential predictor. External evaluation using independent data from an old landfill (OLF) demonstrated consistent model behavior under contrasting waste conditions while revealing the influence of unmeasured factors such as waste composition and microbial activity. This framework enables real-time, non-invasive CH₄ hotspot detection and offers a scalable monitoring solution for landfills, particularly in resource-limited settings, providing a practical tool for emission screening, compliance assessment, and strategic landfill planning aligned with international CH₄ mitigation goals.