Imaging Pressure Distribution in Geological Reservoirs from Surface Deformation Data

Abstract

Geological reservoirs are widely used for storing or disposing of various fluids and gases, including groundwater, wastewater, carbon dioxide, air, gas, and hydrogen. Monitoring these sites is essential due to the stored assets’ economic value and the disposed materials’ hazardous nature. Reservoir pressure monitoring is vital for ensuring operational success and detecting integrity issues, but it presents challenges due to the difficulty of obtaining comprehensive pressure distribution data. While direct pressure measurement methods are costly and localized, indirect techniques offer a viable alternative, such as inferring reservoir pressure from surface deformation data. This inversion approach integrates a forward model that links pressure distribution to deformation with an optimization algorithm to account for the ill-posed nature of the inversion. The application of forward models for predicting subsidence, uplift, and seismicity is well-established, but using deformation data for monitoring underground activity through inversion has yet to be explored. Previous studies have used various analytical, semi-analytical, and numerical models integrated with optimization tools to perform efficient inversions. However, analytical or semi-analytical solutions are impractical for complex reservoirs, and advanced numerical models are computationally expensive. These studies often rely on prior information, which may only sometimes be available, highlighting the need for innovative approaches. This study addresses these challenges by leveraging advanced numerical models and genetic algorithms to estimate pressure distribution from surface deformation data without needing prior information. The forward model is based on a discrete Green matrix constructed by integrating the finite element method with Python scripting. This matrix encapsulates the influence of reservoir properties and geometry on the displacement field, allowing for the rapid evaluation of displacement due to arbitrary pressure distributions. Precomputing Green’s matrix reduces computational load, making it feasible to apply advanced optimization methods like GA, which are effective for solving ill-posed problems with fewer observation points than unknown parameters. Testing on complex reservoir cases with synthetic data showed less than 5% error in predicted pressure distribution, demonstrating the approach’s reliability.