Automatic detection of surface deformation features from a large volume of Interfero16 metric Synthetic Aperture Radar (InSAR) data is challenging because the magnitude of InSAR measurement noise varies substantially in both space and time. In this work, we present a computer vision algorithm based on Laplacian of Gaussian (LoG) filtering for detecting the size and location of unknown surface deformation features. Because our algorithm detects spatially coherent features, tropospheric noise artifacts that share similar spatial characteristics may also be detected. We estimate the tropospheric noise spectrum directly from data, which allows us to simulate new instances of noise that resemble the actual InSAR observations. Based on these simulations, we quantify the likelihood that a detected feature is a real deformation signal. We demonstrate the performance of our algorithm using Sentinel-1 data acquired between 2014 and 2019 over the ∼ 80,000 km2 oil-producing Permian Basin in West Texas, one of the most productive oil fields in the world. We detect clusters of deformation features associated with oil production, wastewater injection, and fault activities. The number of detected deformation features increases substantially over the study period, which is consistent with the over-all rise in oil production within the Permian Basin since 2014. Our algorithm is robust and flexible, and can be integrated to various multi-temporal InSAR time series methods for detecting a broad range of deformation features.

Since 2008, the rate of seismic events within the Central United States has dramatically increased, which is likely associated with wastewater injection from nearby oil and gas operations. Surface deformation measurements derived from spaceborne interferometric synthetic aperture radar (InSAR) data can be used to quantify the magnitude and spatial extent of the injection-related stress perturbation, which are critical for understanding the complex interaction between the injected fluid and the earth’s subsurface. In this study, we processed Sentinel-1 InSAR data over Central and West Texas using a recently developed processing framework that performs topography/geometry phase corrections prior to the interferogram formation (Zebker 2017). We streamlined the creation of upsampled digital elevation maps (DEMs) from NASA Shuttle Radar Topographic Mission (SRTM) data, as well as the collection of Sentinel-1 precise orbit data. We developed a tool for InSAR time-series analysis and data visualization. To detect unknown deformation signatures from large volumes of InSAR data, we employed computer vision ideas for feature detection independent of scale, well known through their success in the Scale Invariant Feature Transform (SIFT). We used multi-scale Laplacian-of-Gaussian (LoG) filters to find local maxima and minima in a coarse deformation solution, corresponding to “bowls” of uplift and subsidence, respectively. This allowed us to drastically cut down processing time of high-resolution InSAR products. As a validation, our method successfully detected all sinkhole locations, injection-related uplift signals and production-related subsidence signals as reported in Kim and Lu (2017) over a 100 km x 100km search area without the need for manual inspection. We then examined the Dallas Fort Worth Basin area for evidence of deformation near wastewater injection and oil/gas production sites. We begin to quantify the uncertainty from common noise sources to produce more confident time-series results.

The expansion in spatial coverage and data volume of Interferometric Synthetic Aperture Radar (InSAR) is prompting the need for automated InSAR processing. To be useable by stakeholders, deformation maps derived from InSAR must come with estimates of reliability. In this study, we develop a new computer vision algorithm for automatic detection of surface deformation features in InSAR deformation maps. We estimate the atmospheric noise power spectrum directly from interferograms, which we use to generate realistic synthetic noise instances. This allows us to calculate a likelihood that features in a real deformation map came from atmospheric artifacts. Because the procedure only focuses on the probability of false alarm for candidate features, it does not require any geophysical model for the signals of interest. Our method is agnostic to the computer vision algorithm used, and it can be embedded within InSAR processing frameworks to quantify the uncertainty of machine learning detection results. We demonstrate our algorithm using 80 Sentinel-1 SAR images covering 80,000 km2 of the Permian Basin in West Texas, where oil and gas production activities have led to a rise in the number of low magnitude earthquakes. Our algorithm reliably detects millimeter-to-centimeter deformation features related with oil and gas production, groundwater pumping, wastewater injection, and the M5.0 earthquake west of Mentone, Texas. Our method provides guidance on the minimum number of Sentinel-1 acquisitions needed for interferogram stacking to confidently detect the subtle deformation. A decrease in uncertainty can be achieved by detecting and removing SAR images corrupted by tropospheric noise, which reduces the number of required acquisitions for mitigating tropospheric noise.

The Permian Basin has become the United States’ largest producer of oil over the past decade. Along with the rise in production, there has been an increase in the rate of low magnitude earthquakes, some of which have been associated with hydrocarbon extraction and wastewater injection. A detailed knowledge of changes to the subsurface can aid in understanding the causes of seismicity, and these changes can be inferred from InSAR surface deformation measurements. In this study, we show that both cm-level cumulative deformation, as well as mm-level coseismic deformation signals, are detectable in West Texas. In a region west of Mentone, TX, we reconstructed the subtle coseismic deformation signal on the order of ~5 mm associated with the recent M4.9 earthquake. Over ~100,000 km2 of the Permian Basin, we created annual cumulative LOS deformation maps, decomposing into vertical and eastward components where overlapping data are available. These maps contain numerous subsidence and uplift features near active production and disposal wells. The most important deformation signatures are linear streaks that extend tens of kilometers near Pecos, TX, where a cluster of increased seismic events was cataloged by TexNet. As validated by independent GPS data, our InSAR processing strategy achieved millimeter-level accuracy. A careful treatment of the InSAR tropospheric noise, which can be as large as 15 cm in West Texas, is required to detect surface deformation signals with such low signal-to-noise ratio. We developed an outlier removal technique based on robust statistics to detect the presence of strong, non-Gaussian noise. We compared the surface deformation solutions of multiple InSAR time series methods, and all of them produced more accurate and consistent deformation trends after removing outlier InSAR measurements. We are exploring a Bayesian generalization of SBAS velocity estimation by including probabilistic data rejection to determine which pixels should be excluded from the model fitting. This technique provides a full posterior distribution of the model parameters along with the best-fit surface velocity.