Multimodal AI-Driven Global Mapping of Anthropogenic Methane Emitters

Methane is the second most important driver of climate change, and its mitigation is considered a critical near-term strategy for slowing global warming. Anthropogenic activities constitute the dominant source of methane emissions, a significant portion of which originates from facility-scale point sources. While bottom-up inventories are valuable, they often suffer from spatial and temporal gaps. Top-down methods using remote sensing can effectively complement these limitations. The complex characteristics of different methane emitters pose significant challenges for feature-annotation-based object detection algorithms. Here, we present a novel remote sensing detection framework based on vision-language multimodal AI and introduce a Multi-scale Adaptive Sliding Window (MASW) strategy for global-scale identification of anthropogenic methane emitters. Applying this approach globally to satellite imagery, we construct a top-down global inventory comprising 51,076 precisely geolocated facilities across all major emission categories, which far exceeding the coverage of existing bottom-up inventory. We further analyze the spatial distribution of these sources and reveal distinct, development-stage-dependent patterns in emission profiles across economies, thereby elucidating the fundamental distribution patterns of global methane emitters. This inventory and its derived insights provide a critical evidence base for targeted monitoring, verification, and mitigation efforts, supporting the implementation of global climate goals.