Authors: David Baude, Rebecca Diehl, Ijaz Ul Haq, Kristen Underwood, Alexander Prescott, and Beverley Wemple – University of Vermont; Belize Lane and Colin Phillips – Utah State University
Title: Scalable Preprocessing and Visualization of NWM Retrospective Data for Basin-Scale Streamflow Modeling
Presentation Type: Poster Presentation
Abstract: The growing availability of large-scale hydrologic datasets such as the National Water Model (NWM) retrospective dataset enables new opportunities for data-driven streamflow modeling, but their practical use is often limited by the complexity of transforming gridded data into basin-scale inputs. In this work, we present an end-to-end preprocessing and visualization framework that converts 44 years of NWM retrospective forcings into structured, basin-level datasets for Alabama and CAMELS-linked regions. Our approach maps USGS gages to NWM hydrofabric catchments, delineates upstream contributing areas, and aggregates gridded meteorological variables using zonal statistics.
To efficiently process long temporal records and large spatial domains, we design a scalable workflow based on chunked NetCDF operations and parallel execution, incorporating memory-aware strategies to ensure stable performance. The resulting datasets integrate dynamic forcings, static basin attributes, and retrospective streamflow observations, enabling direct use in sequence-based models such as LSTMs for streamflow reconstruction in ungaged basins. An interactive visualization interface further supports exploration of basin connectivity, temporal forcing patterns, and streamflow dynamics, enabling rapid validation and diagnostic analysis. This work provides a reproducible and scalable foundation for hydrologic modeling and streamflow reanalysis using multi-decadal NWM data.