Locally adaptive methods for KDE-based random walk models of reactive transport in porous media
Abstract
Random Walk Particle Tracking (RWPT) coupled with Kernel Density Estimation (KDE) has been recently proposed to simulate reactive transport in porous media. KDE provides an optimal estimation of the area of influence of particles which is a key element to simulate nonlinear chemical reactions. However, several important drawbacks can be identified: (1) the optimal KDE method is computationally intensive and thereby cannot be used at each time step of the simulation; (2) it does not take advantage of the prior information about the physical system and the previous history of the solute plume; (3) even if the kernel is optimal, the relative error in RWPT simulations typically increases over time as the particle density diminishes by dilution. To overcome these problems, we propose an adaptive branching random walk methodology that incorporates the physics, the particle history and maintains accuracy with time. The method allows particles to efficiently split and merge when necessary as well as to optimally adapt their local kernel shape without having to recalculate the kernel size. We illustrate the advantage of the method by simulating complex reactive transport problems in randomly heterogeneous porous media.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2017
- Bibcode:
- 2017AGUFM.H32D..07S
- Keywords:
-
- 0414 Biogeochemical cycles;
- processes;
- and modeling;
- BIOGEOSCIENCES;
- 1009 Geochemical modeling;
- GEOCHEMISTRY;
- 1830 Groundwater/surface water interaction;
- HYDROLOGY;
- 1831 Groundwater quality;
- HYDROLOGY