Machine Learning for Onshore Oil Seeps Detection: A Case Study in Kirkuk-Sulaimaniyah Area, Northeastern Iraq

Summary Oil seeps pose significant environmental hazards to both terrestrial and aquatic ecosystems. Traditional mapping techniques encounter logistical and political challenges, particularly in complex regions, such as Kirkuk, an area rich in oil and gas fields. These fields contribute to the proliferation of oil seeps through both natural processes and industrial activities, underscoring the need for efficient detection methods. This study introduces a novel hybrid algorithm, SAM-DT, which combines spectral angle mapping (SAM) with decision tree (DT) classification to enhance oil seep detection. By leveraging remote sensing data, including Sentinel-2A imagery, the Landsat OLI thermal band, and geomorphic and physical characteristics of oil seeps, we demonstrated the utility of integrating multisource data for this purpose. The SAM-DT algorithm’s performance was evaluated against the standard SAM algorithm, using validation from 369 sites verified through field surveys, Google Earth, PlanetScope, and QuickBird data. The results reveal that the SAM-DT algorithm achieved an accuracy of 64%, outperforming the SAM algorithm’s 35%. These findings highlight the effectiveness of the SAM-DT approach in mapping oil seeps across mountainous, semiarid, and plain regions. This study underscores the potential of SAM-DT as a robust tool that can be conducted by testing more nodes of the SAM-DT algorithm to improve the accuracy of onshore oil seep detection, paving the way for future research aimed at refining the algorithm by incorporating additional decision nodes to further enhance detection accuracy.