Fluid Flow in Evolving Sedimentary Deposits
Matthew Wolinsky, Duke University
Sedimentary basins are regions of the Earth where subsidence creates space for sediment to accumulate and form stratigraphy. Porous sand reservoirs within these basins house essential energy resources, such as oil and natural gas. However, offshore resource recovery is complicated by the presence of pressurized pore water, which causes well blowouts and slope failures. Here I consider the problem of predicting fluid overpressure in passive continental margins, where overpressure is generated by loading of low permeability deposits with rapidly deposited sediment. Continental margin deposits can be kilometers thick, but typically span 100 or more kilometers in width. This large aspect ratio has led researchers to predict overpressure using quasi-1D models, ignoring horizontal fluid flow. Here I show that these models are inappropriate, by coupling a fully 2D fluid flow model to a sedimentary process model which generates stratigraphy.
Two factors complicate 2D modeling of overpressure. First, over long timescales, fluid flow occurs in a domain which evolves due to erosion and deposition, forcing frequent remeshing. Second, the stratigraphy within this high aspect ratio domain is complex, with thinly interbedded sands and muds interspersed between large scale sand and mud bodies. This creates rapid vertical variations in permeability, imposing a tradeoff between mesh size, discretization error, and conditioning of the discretized fluid flow equations. I introduce a dynamic semi-structured meshing algorithm to overcome the first difficulty, and a sub-grid embedding scheme to overcome the second, resulting in an efficient, robust computational scheme.
Stratigraphic simulations show that typical deposits include significant heterogeneity due to interfingering sand and mud layers. In coupled simulations, this heterogeneity leads to flow anisotropy, with sandy layers acting as horizontal flow conduits. These simulations confirm that quasi-1D approaches are inappropriate for modeling, but that fully 2D approaches are feasible and can be made computationally efficient.
Abstract Author(s): Matthew A. Wolinsky