SATELLITE-DRIVEN METHANE EMISSION PREDICTION IN MUBI CATTLE MARKET: A MACHINE LEARNING APPROACH FOR CLIMATE MITIGATION

Abstract

Methane (CH4) emissions from livestock activities in sub-Saharan Africa are a critical component in regional greenhouse gas inventories. Despite their importance, these emissions have not been extensively studied. This study employs Sentinel-5P TROPOMI and ERA5 reanalysis data to predict methane concentrations in the Mubi cattle market, Adamawa State, Nigeria, using an XGBoost model. By integrating temporal lags, seasonal features, and environmental variables, the model achieves an R² of 0.7517 and MAE of 4.69 ppb on an interpolated dataset (1785 daily records). XGBoost outperforms LSTM, TCN and Transformer models with R² of -0.52, -0.79, and -1.68, respectively, demonstrating its efficacy in capturing methane dynamics. SHAP analysis reveals that lagged methane values contributed up to 12.3 ppb with wet season conditions (via month sine) as primary drivers. A spatial heatmap highlights emission hotspots within the market, supporting targeted mitigation. This scalable framework provides a robust tool for methane monitoring in data-scarce regions, offering insights for Nigeria’s climate policy and global methane reduction efforts.