FlareSat: A Benchmark Landsat 8 Dataset for Gas Flaring Segmentation in Oil and Gas Facilities

Gas flaring is widely used in oil and gas facilities to prevent pressure buildup and ensure operational safety. However, large-scale flaring releases substantial amounts of greenhouse gases, contributing to climate change. Detecting and monitoring gas flaring is therefore essential for mitigation efforts. Satellite imagery provides key advantages for these tasks, including open data availability, global coverage, and broad spectral information. Despite the existence of datasets for flame and smoke detection, as well as hyperspectral data for methane emissions, there remains a lack of open satellite datasets specifically designed for gas flaring detection using deep learning. This dataset aims to address this gap by providing a specialized resource for gas flaring segmentation using Landsat 8 imagery. It includes 7,337 labeled image patches (256 × 256 pixels) covering 5,508 oil and gas facilities across 94 countries, spanning both onshore and offshore sites. To improve model robustness and reduce false positives, the dataset also incorporates patches containing visually similar sources such as wildfires and urban areas with high solar reflectance. The temporal coverage of the dataset is 2019, focusing on the months that recorded the highest increases in pollution from gas flaring (World Bank Global Gas Flaring Data) (American Petroleum Institute) and to the VIIRS Nightfire study—March, August, September, and December—during which 8,098 unique flare locations were observed (links to be added). A corresponding research paper describing the dataset and baseline segmentation experiments is currently under review at Journal of the Brazilian Computer Society (JBCS) and will be added as a reference once published.