Global efforts to reduce emissions are hampered by incomplete and unreliable data from coal power plants regarding their release of greenhouse gases. This inability to acquire data, particularly relevant in regions lacking robust reporting mechanisms, hinders evidence-based policy-making and proper action against prominent emissions contributors. To address this issue, we propose using a transformer-based approach to determine the coal plant’s operational status – entirely through satellite images. This method leverages the readily available global coverage of satellite imagery to provide independent, real-time monitoring of coal plant operations, even in areas where traditional data collection methods are challenging. The model learns to identify visual indicators, such as cooling tower plumes in unseen test images. Our transformer-based approach outperforms the best prior work with statistical significance, reaching an 80% accuracy on a held-out set of unseen coal plant satellite images. These results have far-reaching implications for climate monitoring and regulations. By providing more accurate and independently sourced information on coal plant operations worldwide, our approach can support more effective emissions tracking, inform climate negotiations, and introduce greater transparency into the process of climate regulation.
Loading....