<p>Accurate quantification of atmospheric methane column concentration (XCH₄) is essential for monitoring greenhouse gas emissions and supporting climate change mitigation. However, current satellite observations are frequently disrupted by cloud cover and aerosol interference, resulting in substantial data gaps and retrieval errors that restrict their long-term applicability. In this study, an OOA-LightGBM framework integrating the Osprey Optimization Algorithm with the Light Gradient Boosting Machine was developed to reconstruct XCH₄ over China from March 2020 to February 2025. By incorporating meteorological, topographical, vegetation, and anthropogenic factors, the framework effectively addresses the limitations of Sentinel-5P TROPOMI data. Comparative evaluation against established optimizers confirmed the superior predictive performance of OOA-LightGBM (R² = 0.85, RMSE = 9.96 ppb, MAE = 6.02 ppb for the full period), with robust accuracy across multiple temporal scales. The reconstructed dataset achieves seamless spatiotemporal coverage, substantially reducing striping and missing data, and shows strong consistency with TCCON ground-based observations. Spatiotemporal analyses reveal pronounced spatial gradients—with higher XCH₄ levels in the southeast and lower in the northwest—along with a steady upward trend and marked seasonal variability. SHAP results further show that both natural and anthropogenic factors jointly shape the observed methane patterns. This approach demonstrates the potential of machine learning—enhanced satellite reconstruction to improve methane monitoring, support source attribution, and inform emission mitigation strategies, underscoring its value for data-driven climate action and sustainable development.</p> Graphical Abstract <p></p> <p>This study presents an innovative workflow for seamless reconstruction of atmospheric methane (XCH₄) across China, addressing the limitations of satellite observations. Methane emissions arise from the combined influence of natural and anthropogenic processes, motivating the need for satellite-based, large-scale, and long-term monitoring. However, TROPOMI observations over China suffer from substantial spatial gaps and striping artifacts, limiting their utility for environmental analysis. To address these deficiencies, we integrate multisource datasets—including vegetation indices (EVI), nighttime lights (NTL), digital elevation (DEM), and meteorological variables from ERA5—to represent key drivers that shape methane variability. These datasets are incorporated into an optimized machine learning framework, OOA-LightGBM, which couples LightGBM with the Osprey Optimization Algorithm to enhance parameter tuning, model stability, and predictive efficiency. The resulting model generates high-resolution, spatially continuous XCH₄ fields, effectively removing gaps and noise. The reconstructed maps reveal pronounced spatial heterogeneity and temporal dynamics of methane across China, demonstrating the capability of OOA-LightGBM to support rapid, accurate, and scalable methane monitoring for environmental assessment and sustainable development applications.</p>