This paper introduces a satellite-based machine learning (ML) model for detecting methane, a major contributor to climate change. Specifically, we present a novel approach that utilizes raw data (UnorthoDOS) without the traditional preprocessing step of geometric distortion correction (orthorectification). Our results demonstrate that ML models trained on UnorthoDOS data perform similarly to preprocessed data, and that models trained on preprocessed data outperform conventional matched filters. We also make public two ML-ready datasets (orthorectified and unorthorectified), model checkpoints, and code to enhance the reproducibility and usability of our research.
