Improvements on Scalable and Reproducible Cloud Implementation of Numerical Groundwater Modeling

In the past decade the groundwater modeling industry has trended toward more computationally intensive methods that necessarily require more parallel computing power due to the number of model runs required for these methods. Groundwater modeling that requires many parallel model runs is often limited by numerical burden or by the modeler's access to computational resources. Over the last 15 years the evolution of the cloud in accelerating groundwater model solutions has progressed; however, there are no apparent literature reviews of MODFLOW and PEST cloud implementation, specifically with regards to open-source and efficient scalable solutions. Here we describe infrastructure as code used to develop the architecture for running PEST++ in parallel on the cloud using Docker containers and open-source software to allow simple and repeatable cloud execution. The architecture utilizes Amazon Web Services and Terraform to facilitate cloud deployment and monitoring. A publicly available MODFLOW-6 model was used to evaluate parallel performance locally and in the cloud. Local model runs were found to have a linear 12 s increase in model run time per agent on a typical office computer compared to the cloud implementation's 0.02 s per model, indicating near perfect scaling even at up to 200 concurrent model runs. A consulting groundwater model was calibrated with the cloud infrastructure, which enabled acceleration of project completion at minimal cost.