Data and Compute Architecture for Effective Use of Cloud Computing to Perform Benchmark Applications of the NextGen Framework.

Model selection, configuration, and calibration exercises are tedious, error prone, and difficult to generalize across models and modeling domains. Robust workflows often require bespoke solutions to handle model and domain specific input data, creation of configuration files, and evaluation of model results. The Next Generation Water Resources Modeling Framework (NextGen) , configured as the National Water Model, encounters these challenges at each of the approximately 800,000 3–10 km2 catchments across a continental-scale modeling domain. This complexity is compounded by the various model and module options available to use with NextGen. Manually creating model configuration files, formatting data for model applications, estimating parameters, and evaluating output are prohibitive in this case. NOAA’s National Weather Service Office of Water Prediction (OWP) is addressing these challenges through the development of software and workflows that apply modern data and compute practices, allowing for flexible implementations and deployments of a NextGen model benchmark application.

In this presentation, we will discuss the development of workflows to reduce manual intervention and input required from modelers during the benchmarking process. We will explain how such workflows simplify many of the steps involved, including selecting model domains, establishing initial model parameters, aligning forcing data, configuring and performing parameter estimation, and utilizing results. We will then show how cloud native data formats for hydrofabric and forcing are combined with cloud compute resources to effectively and efficiently execute benchmarking of formulation performance. These efforts, developed alongside the modular NextGen framework, will also provide fundamental support for, and integration with, other work such as OWP’s Formulation Selection Decision Support (FSDS) tool. These combined systems will ultimately empower a continental scale modeling ecosystem which can apply the most performant models together in a single simulation.