Distances Between 3D Models Based on Topological Characterization of Reservoir Rocks: Application to CO2 Storage in Fluvial Reservoirs
Wednesday, 17 December 2014
Modelling subsurface structures is mainly performed in under-constrained contexts as only sparse and indirect data are available. Stochastic modelling is the common strategy to solve such problems, especially since such subsurface 3D models serve as support for quantitative studies (e.g.: reservoir characterizations, etc.). Multiple 3D models are generated from the dataset and represent a sampling of plausible subsurface structure representations. From this model sampling, statistical analysis on targeted parameters (e.g.: reserve estimations, flow behaviours, etc.) and a posteriori uncertainties are performed to assess risks. However, on one hand, uncertainties may be huge, which requires many models to be generated for scanning the space of possibilities. On the other hand, some computations performed on the generated models are time consuming and cannot, in practice, be applied on all of them. This issue is particularly critical in CO2 storage studies as many scales of investigations are required, from meter to regional ones, to estimate storage capacities and associated risks. Recent petroleum studies propose to define distances between models to allow sophisticated multivariate statistics to be applied on the space of uncertainties so that only sub-samples, representative of initial set, are investigated for dynamic time-consuming studies. In this work, it is proposed to define such distances in the particular case of CO2 storage as differences exist between the petroleum and CO2 storage dynamic behaviours and problematic. Moreover, particular interests are also paid to the characterization of the topology of the reservoir rocks. The skeleton of the reservoir rock is studied through graph spectral analysis and topological indices. A synthetic data set of fluvial models with different geological characteristics was generated and studied using the defined distance. Relationships between model settings and distances are shown and discussed.