Quantifying key drivers of atmospheric methane across Pakistan using a machine learning approach

Atmospheric methane (CH₄), a potent greenhouse gas, has shown a consistent rise since the Industrial Revolution, contributing significantly to global warming and climate change. Understanding the temporal and spatial variability of methane concentrations (XCH₄) and the factors driving these changes is crucial for effective mitigation strategies. However, the complex, multidimensional, and interdependent nature of these factors poses challenges for conventional statistical and geospatial methods, which often struggle with large data volumes and imbalanced datasets. In this study, we integrate multi-source satellite datasets with environmental, meteorological, and socioeconomic variables across Pakistan for the period 2010 to 2020. We employed the random forest machine learning algorithm to analyze complex, nonlinear interactions among these variables and to map the seasonal spatial distribution of dominant CH₄ drivers. The permutation importance metric is used to identify the most influential factors affecting CH₄ concentrations. Our results show that CH₄ concentrations in Pakistan have been increasing at an average annual rate of approximately ~ 13 ppb over the study period. Random forest effectively captures the nonlinear interactions between variables, while the permutation importance metric helps identify the most influential factors. This machine learning framework offers a scalable and efficient method for interpreting complex satellite datasets, providing valuable insights for methane emission monitoring and policy development.