Hydraulic Fracture Test Site (HFTS)

  The purpose of this study is to understand the role of hydraulic fractures during oil/gas production. A coupled multiphase flow (TOUGH+) and geomechanics (Millstone) simulator has been developed at the Lawrence Berkeley National Lab, which will be combined with GEOS simulation results (used to model stimulation process) to predict production. Field data available at the test site will be used to provide initial conditions, model parameters, as well as to evaluate model performance. This project is a collaboration among multiple institutes and research groups. Current duties of Dr. Li include investigating possible methods to couple TOUGH+, Millstone and GEOS, building the simulation scenarios according to the field data available, performing numerical tests and analyzing simulation results with respect to the data collected. The video below shows an example of gas production via a horizontal well in a fractured reservoir.

Subgrid Topography Model

    A porosity-type subgrid model is developed to simulate effects of subgrid-scale topographical features on coarse grids. The main purpose of developing subgrid model is to allow fast simulation of hydrodynamics and scalar transport over large domains (e.g. large coastal wetlands, floodplains, etc.). Compare to existing subgrid models, the proposed model automatically preserves high-resolution surface connectivity during grid-coarsening, which makes it well-adapted for general topography with complex river networks or topographical features of various scales. The figure below shows modeled salinity in the Nueces Delta (TX). The left/right columns are simulations without/with the subgrid model. It can be seen that without subgrid model, simulation result deteriorates when grid resolution increases from 5m to 30m (fig c).

More info:  

                     Li, Z. and Hodges, B.R., 2019, Modeling subgrid-scale topographic effects on shallow marsh     

                     hydrodynamics and salinity transport, Advances in Water Resources, (129) 1-15, 









Surface-subsurface Flow Exchange

    When investigating the fate of flow in the Trinity River Delta (TX), evidence revealed possible contribution to the surface water budget from the subsurface. Thus the FrehdC model is being further developed to enable simulation of surface-subsurface flow exchange. As a first step, a novel numerical solution to the 1D Richards equation is proposed to achieve conservative, efficient and robust simulations for variably-saturated subsurface flow. The new method applies a predictor-corrector type numerical scheme with a post-allocation procedure to enforce mass conservation. This method is still being improved to enable 3D simulation with coupled surface hydrodynamics. The figure below shows validation of the proposed method (PCA) against Hydrus-1D and Warrick's analytical solution using an infiltration problem.

Model Salinity in Shallow Marshes

    Using the Fine Resolution Environmental Hydrodynamic Model (Frehd), the surface elevation and salinity transport in the Nueces Delta (TX, USA) was modeled and validated with field data. Several novel schemes were tested to maintain model stability at shallow wetting/drying front under wind stress. The simulation results highlighted the necessity of preserving high-resolution surface connectivity at practical coarse-grid scale. The predicted impact of freshwater inflow provided guidance for adjusting existing freshwater-pumping strategy to alleviate the hyper-saline condition in the delta.

    More info:  

                     Li, Z. and Hodges, B.R., 2018, Model instability and channel connectivity for 2D coastal marsh                           simulations,Environ Fluid Mech, https://doi.org/10.1007/s10652-018-9623-7

Automatic Channel Identification

    Bathymetry is an important input of hydrodynamic models. Although high-resolution lidar data can be used to create model bathymetry, it is not yet a cost-friendly data collection procedure. When building the hydrodynamic model for the Trinity Delta in Texas, we received the lidar data that did not penetrate water, resulting noises in the water regions. The kmeans clustering algorithm was applied to separate water from land based on their colors. The major channels and lakes were successfully extracted. Further improvement of the identification procedure could be achieved via supervised learning methods such as the Convolutional Neural Network (CNN), which remains an ongoing research topic.

  • LinkedIn Clean Grey