Authors: Anthony Preucil – RTI International
Title: Large-Scale calibration workflows within the SYMFLUENCE Modeling Platform
Presentation Type: Poster Presentation
Abstract: Calibrating and regionalizing parameters for large samples of basins is a growing need in the hydrologic modeling community to support data-driven machine learning, traditional modeling, and operational forecasting. While such methods have been established and documented, many studies reflect the difficulty of setting up related end-to-end model workflows citing challenges with computation-intense data ingest, data preprocessing, large spatial scales, and experiment reproducibility. This study aims to demonstrate that calibration of models for a large number of basins across the Continental United States (CONUS) is not only possible, but practical using SYMFLUENCE (SYnergistic Modeling Framework for Linking and Unifying Earth-system Nexii for Computational Exploration). SYMFLUENCE is a new open-source platform with a vision to simplify hydrologic modeling experiments by providing model agnostic workflows for all steps of the modeling pipeline including data acquisition, preprocessing, and calibration. Our research focuses on producing a set of calibrated parameters for various implementations of the SAC-SMA model coupled to SNOW-17. The models were calibrated with several repeated optimization runs using alternative algorithms. This study evaluates computational performance of SYMFLUENCE under different configurations when calibrating at scale. The experimental setup demonstrates the need for this type of software infrastructure when performing large-scale calibration experiments. More importantly, the calibrated parameter set across CONUS can be considered a baseline for both research applications and operations. As the community moves towards the next generation of water modeling, SYMFLUENCE will provide the tools needed to be able to perform reproducible experiments with ease and drive advances in parameter estimation for traditional and experimental hydrologic models alike.