Ichnos: An efficient node based stochastic particle tracking algorithm

Abstract:

Particle tracing is a common simulation method to study groundwater transport. Typically, mesh-based numerical tools are used to solve for a 3D flow velocity field, then particle tracking identifies the trajectories of virtual particles injected in the medium. Particle tracking has been used extensively in groundwater hydrology for plume characterization, advective transport, visualization, stochastic simulations etc. In groundwater hydrology, the numerical models calculate the flow velocity using either finite difference or finite element numerical schemes. The velocity interpolation scheme depends on the numerical method, so that each numerical model is often associated with each own particle tracking tool (e.g. Modflow-Modpath).

Here we present a framework for particle tracking that can be used for all types of velocity fields. We use a node-based method, where the velocity field is represented as a point cloud and stored in a data structure that allows rapid distance queries. The proposed framework also allows for splitting the velocity field into multiple sub-point clouds so that the workload of the particle tracking can be split into many processors. This allows us to trace particles within multi-million cloud point velocity fields.

The proposed framework is applied to two 3D groundwater hydrologic models of the Central Valley aquifer in California. The CVHM (Central Valley Hydrologic Model) is based on finite difference scheme and the C2VSim (California Central Valley Simulation) is based on a semi-finite element numerical method. Both models are transient, yet the time span of the models does not allow to trace the particles in an insightful way. To this end we propose a stochastic particle tracking approach, where the velocity is sampled at each time step from a pool of velocities of the transient model. This stochastic analysis significantly increases the reliability of the model predictions without the need to increase the spatial resolution of the numerical models.