Authors: Landon Gehr, Andy Wood
Title: Scalable NextGen Model Parameter Optimization using Containerized Distributed Computing
Presentation Type: Poster Presentation
Abstract: Current hydrological models can face a challenge in the form of high computational costs and complex parameter spaces. Fitting a model is a basin-specific process relying on observational data. This process requires much set-up and computation for each individual basin, particularly when scaling to larger domains or higher-resolution simulations. Efficient and reproducible calibration workflows are therefore essential for enabling broader application of these models in research and operational settings. This research attempts to build on the existing Next Generation Water Modeling Framework (NextGen) to automate this workflow. Here we specifically optimize over the Conceptual Functional Equivalent (CFE) model and the Noah-OWP Modular parameters with the Shuffled Complex Evolution algorithm (SCE-UA). The workflow is deployed on a high-performance computing (HPC) system using SLURM to leverage parallel execution across multiple nodes, enabling the large numbers of model evaluations required. This work establishes a foundation for scalable calibration of NextGen models and supports future efforts towards facilitating comparisons between NWS Hydrologic Ensemble Forecast Service (HEFS) and the NextGen ensemble forecasts with hindcasting.