A Surface-Adaptive Framework for Methane Point Source Detection and Quantification Using Sentinel-2 and Google Earth Engine

Methane point sources are spatially sparse, temporally intermittent, and strongly affected by surface heterogeneity, posing significant challenges for large-scale and continuous monitoring. While hyperspectral sensors provide high retrieval accuracy, their limited spatial and temporal coverage motivates the use of global, open-access multispectral satellite data for scalable identification and quantification of methane emissions. In this study, we present a systematic methane point source detection and quantification framework built upon Sentinel-2 imagery and the Google Earth Engine (GEE) platform, enabling scalable and operational analysis across diverse land surface types and emission sources. The framework incorporates adaptive plume detection and segmentation strategies tailored to different land surface conditions by exploiting characteristic methane signatures in the spatial, spectral, and temporal domains. Dedicated data-driven models are employed to segment methane plumes over homogeneous oil and gas regions, spectrally challenging environments such as vegetated and offshore areas, and heterogeneous sources including landfills and coal mining sites. Detected plumes are subsequently quantified using wind-informed emission estimation to derive point-source emission rates directly from Sentinel-2 observations. The proposed framework is evaluated across multiple representative land surfaces, successfully identifying the majority of high-emission sources as well as several previously unreported ones, and demonstrating improved detection consistency and generalization compared to conventional single strategy approaches. By leveraging the global coverage of Sentinel-2 and the computational scalability of GEE, this work provides a practical pathway toward near-global screening and monitoring of methane point sources, supporting climate mitigation and emission inventory improvement efforts.