Bottom-hole Pressure Data Integration for CO2 Sequestration in Deep Saline Aquifers

AbstractWe describe a novel approach to update geological models using bottom-hole pressure data from injection and observation wells during CO2 sequestration. Our proposed history matching workflow, in conjunction with compositional simulations of supercritical CO2 injection into a saline formation, involves: (a) inversion of zeroth order frequency of the injection-well pressure to modify the spatial permeability field around the injection well, (b) transient pressure arrival time inversion of observation well pressures to modify the inter-well spatial permeability field, and (c) gradient-based minimization of pressure mismatch at all wells using a global permeability multiplier. The proposed approach has been demonstrated on 3-D synthetic models generated using well-log data from the Weaber-Horn well in the Illinois basin. In this case supercritical CO2 is injected for 11 months into a centrally located well, with pressure response monitored during injection and 1 month of shut-in at three observation wells. Injection is assumed to take place in a high-permeability layer close to the bottom of the model, with pressure responses monitored at all three observation wells only in the injection layer. Forecasts of pressure response at injection/observation wells over a three-year injection period produced with the inverted model also agree well with those from the reference model. Inversion of the 3-D model is more challenging and suffers from non-uniqueness, unless the condition of proximity to the prior model is imposed. In all cases, improved forecast of the CO2 plume evolution was observed after the pressure history matching. We also show how the integration of time-lapse seismic data into the inversion process results in further improvement in gas saturation forecast. A systematic and efficient approach to integration of pressure data from CO2 injection operations is presented, offering improved CO2 plume prediction - especially when time-lapse seismic data is not available