Highly parameterized inversion methods of groundwater model calibration, and associated sub-space methods of predictive uncertainty analysis, have become prevalent in groundwater modeling research and practice. To make these methods computationally tractable it is essential to employ parallel computing. Currently, there is a major shift in information technology towards centralized cloud computing resources. This presentation discusses the use of cloud computing for groundwater model calibration, shares insights gained from several highly-parameterized calibration projects conducted using cloud computing, and presents two different software packages for conducting parallelized PEST based groundwater model calibration, and uncertainty analysis using cloud computing.
Over the past six years several highly parameterized calibration projects have been completed using the Amazon EC2 Cloud Service. A custom HTTP protocol based Python program, was used to transfer file and initiate beoPEST processes on large distributed networks for calibration and uncertainty analysis.
To make highly parallelized, cloud computing based, calibration accessible to a wider base of groundwater modeling practitioners, a new version of PEST tailored to highly parallelized environments has been developed, PEST-HP, and integrated into a Microsoft Azure based cloud computing service named “PEST.cloud”.
Calibration of a regional-scale model of groundwater flow is provided to illustrate the methodology. Aquifer properties were calibrated to more than 1,500 static hydraulic head measurements, and a ten year monitoring period during industrial groundwater use with to 450 adjustable parameters. The PEST based model calibration was parallelized on up to 250 computing nodes located on Amazon’s EC2 servers.
Highly parameterized calibration and uncertainty analysis has been successfully conducted using cloud computing services for several years. The new developments in PEST Software will allow this type of analysis to be more widely adopted, allowing null space monte carlo uncertainty analysis to become a standard practice.