Spatial Monitoring of Geological Carbon Storage Progress Using Time-Lapse Satellite Images

Abstract

Abstract We propose a new Geological Carbon Storage (GCS) monitoring approach to demonstrate the potential use of satellite images for monitoring of a pilot project in Kern County, California. The scope includes identification of appropriate subsurface and surface conditions for success. This is an ideal candidate site due to its surface condition, with little vegetation for less observation noise and stable baseline measurements. We successfully detected historical land movements from 2015 to 2021 based on the satellite images with a resolution of 1 mm/year. Numerical simulation informs that land uplift ranges from 0.011 to 0.105 ft (3.27 to 31.85 mm) due to carbon dioxide injection considering geomechanical uncertainties. The spatial and vertical resolution of the observational data fulfills requirements for monitoring of GCS projects. We conduct a global sensitivity study to identify the impacting factors for land surface deformation. There are seven parameters selected from three main aspects, including rock mechanics, rock physics, and field operation factors. We use Latin hypercube to sample the parameter space for 200 simulation runs. The baseline simulation model represents the pilot site, and it couples with mechanics to compute rock deformation and land surface movements. The challenges of the complex response dataset motivate to extend the capability of a distance-based generalized sensitivity analysis (DGSA) method, using principal component analysis (PCA) and an autoencoder to extract essential features and reduce data dimensionality. The reconstructed images from both methods preserve the pattern and magnitude of land movement. PCA captures more than 96% of cumulative variance using 6 principal components (PCs), and the autoencoder allows the latent vectors (in dimension of 8) to extract necessary features and information from inputs. In general, the satellite images recover information regarding rock mechanics and field operation parameters. Complex, time-series satellite images allow us to infer a more complete set of parameters; however, the magnitude of land movement recovers limited information, that includes sandstone Young's modulus and injection rate. The two-way interaction of sandstone Young's modulus and injection rate are sensitive under all scenarios. We simulate synthetic satellite images from the baseline numerical model, and it provides additional information to history match a reservoir simulation model and reduce uncertainty while tracking the spatial evolution of stored carbon dioxide.