The DInSAR Analysis with Machine Learning for Delineating Geothermal Sites at the Brady Geothermal Field

Abstract

The availability of high temporal resolution and different types of data acquired with advanced SAR techniques allow researchers to compute and analyze the displacement values in a region of interest for various purposes accurately and cost-effectively. In line with that, geothermal activities, such as production and injection operations, create direct and indirect deformations in and around the geothermal area that are in the form of subsidence or uplift movements. These movements are also strongly related to the structural geology of the site. To this end, identifying deformation patterns and understanding their temporal and spatial changes provides invaluable information that can be related to the operational activities in the geothermal site to characterize the geothermal reservoir and understand the operational impacts in the region. In this study, we present the machine learning algorithms integrated with Differential Interferometric Synthetic Aperture Radar (DInSAR) technique to delineate subsidence and uplift movements in the Brady Geothermal Site located in Nevada. For this purpose, we used DInSAR to identify vertical displacements in millimetric accuracy with a pair of Sentinel-1A SAR data (Jan.10th & Dec.24th, 2019). Then, we implemented K-means clustering as the machine learning algorithm to delineate the patterns of movements in the geothermal field and to compare the identified classes with earlier studies. According to the findings, average displacement was calculated as 9 mm/year while the maximum subsidence and maximum uplift were measured as -13 mm/year and 21 mm/yr, respectively. For validation purposes, The DInSAR results are compared and verified with earlier studies focusing on displacement trends for Brady’s geothermal site. We apply the k-means algorithm for the validation of patterns created by geothermal activities. Besides, the visual investigation of the spatial distribution of subsidence and uplift reveals that the calculated deformation patterns overlap with the wells' distribution, fault zone, and geothermal activities in the region, pointing out the operational sequences in the site. The findings indicate the usability of the k-means clustering algorithm to highlight and enclose the deformation clusters closely related to geothermal activities.