Multiscale attribution of atmospheric methane variability in China using satellite observations and interpretable machine learning.

Methane is one of the primary anthropogenic greenhouse gases influencing near-term climate warming. Atmospheric concentrations of methane continue to rise during the 21st century despite ongoing mitigation efforts. Understanding how environmental drivers regulate methane across time and space remains challenging. In China, diverse, overlapping emission sources complicate this regulation, requiring scale-dependent analytical frameworks. Consequently, current mitigation strategies remain constrained and less effective. We develop an integrated multiscale analytical framework to investigate the seasonal and regional variability of atmospheric methane across China. Using Sentinel-5P/TROPOMI observations from 2019 to 2025, the study evaluates the primary environmental controls on methane. We validated satellite retrievals against ground measurements and implemented a multi-model framework, incorporating geographically weighted regression (GWR), grey relational analysis (GRA), and ensemble machine learning, applied to attribute driver-induced methane variability. The results reveal pronounced seasonal variability in methane concentrations, with temperature emerging as the dominant control factor across most regions, particularly during summer, and consistent with the thermal sensitivity of microbial methane production. Vegetation productivity and precipitation have secondary, seasonally dependent effects, with hydrological controls gaining significance in winter. Ensemble machine learning analyses independently confirm the hierarchy of driver importance identified by GRA, demonstrating that the inferred relationships are mathematically robust across different algorithmic architectures. By resolving nonlinear, scale-dependent methane-environment interactions at seasonal and regional scales, this study provides a high-resolution, data-driven framework for process-informed attribution and provides regional screening evidence for seasonal and region-specific methane management, while requiring local validation for facility-level mitigation decisions.