Joint Data Compression and Model Reduction for Conditional Stochastic Modeling of Subsurface Flow and Transport Processes

Well and tracer tests are important experimental tools that help to reduce the uncertainty of predictions in heterogeneous and uncertain subsurface systems. Yet, they yield time series data with very limited information per individual data point.

In part (I) of this project, time series data will be compressed down to a low number of characteristic features, while minimizing the information lost by the compression. The compression will be designed to fit with a simultaneous model reduction from transient to steady-state equations. This will drastically reduce the computational costs of the stochastic modeling and data assimilation task, and hence make stochastic approaches applicable to larger and more complex problems.

In part (II) of this project, the techniques developed and tested in part (I) are applied towards stochastic multiscale modeling of transport in fractured porous media or heterogeneous media. The method is based on flow-aligned blocks and uses multi-rate mass transfer models to parameterize unresolved sub-block heterogeneity. The use of temporal characteristics make the scale transition of parameters swift and simple. This allows for the application to larger and more complex problems.