PAR2 : Parallel Random Walk Particle Tracking Method for solute transport in porous media

Computational modeling of solute migration in groundwater systems is a fundamental component in water resources management and risk analysis. Therefore, it is imperative to have fast and reliable computational tools to simulate solute transport in groundwater systems. In this work we present PAR2, a GPU-accelerated solute transport simulator based on the Random Walk Particle Tracking (RWPT) technique, a Lagrangian method particularly suited for parallelization. PAR2 is able to run on any computing platform equipped with an NVIDIA GPU, such as common desktops and High-Performance Computing (HPC) nodes. The program is developed in C++/CUDA. In our illustration, groundwater flow is simulated on a structured grid using MODFLOW, which can be linked to PAR2 using the LMT package. Simulation parameters can be defined through a convenient YAML configuration file. Additionally, we propose an analytical treatment of the dispersion tensor that allows the RWPT to be effectively implemented using GPU parallelization. The speedup gained with the parallelization drastically reduces the total simulation time, allowing the application of computationally expensive algorithms (e.g., Monte-Carlo simulation) on large-scale stochastic hydro-systems.

地下水系统中溶质迁移的计算模拟是水资源管理和风险评估的基础性组成部分。因此，具备快速且可靠的计算工具以模拟地下水系统中的溶质运移至关重要。在本研究中，我们推出了PAR2，这是一款基于随机游走粒子追踪（Random Walk Particle Tracking，简称RWPT）技术的GPU加速溶质运移模拟器，RWPT技术是一种特别适合并行化的拉格朗日方法。PAR2能够在配备NVIDIA GPU的任何计算平台上运行，例如普通桌面和高性能计算（High-Performance Computing，简称HPC）节点。该程序采用C++/CUDA语言开发。在我们的示例中，利用MODFLOW在结构化网格上模拟地下水流动，MODFLOW可通过LMT包与PAR2连接。模拟参数可通过便捷的YAML配置文件进行定义。此外，我们提出了一种对扩散张量的解析处理方法，该方法使得RWPT能够通过GPU并行化得以有效实施。通过并行化获得的加速效果显著缩短了总模拟时间，使得在大型随机水文系统中应用计算成本高昂的算法（如蒙特卡洛模拟）成为可能。