H196-0001
Hyperresolution Hydrology on Full River Basins: Addressing Challenges in Machine Architectures and Data Integration

Wednesday, 16 December 2020
Poster
Ethan Coon1, Julien Loiseau2, Irina Demeshko3, John D Moulton3 and Scott L Painter1, (1)Oak Ridge National Laboratory, Oak Ridge, TN, United States, (2)Los Alamos National Laboratory, Los Alamos, United States, (3)Los Alamos National Laboratory, Los Alamos, NM, United States
Abstract:
Integrated hydrologic models are powerful tools for water resource applications, answering questions regarding water availability and water quality. Recently, a combination of greatly increased computational resources and greatly improved, high-resolution datasets have suggested the opportunity for well-parameterized, hyperresolution hydrologic models on full river basins. Unfortunately, to date this has been largely aspirational, as the computers have come with architectures that have not yet been unlocked for use by most hydrologic models, and the data have come with difficulties of data-model integration, scale, coordination, and other “big data” problems.

Here we demonstrate progress toward meeting these challenges. We show how open data portals and workflow tools can be used to automate the process of acquiring and coordinating datasets for use in hyperresolution hydrologic simulations. REST APIs are used to discover and download open data products at high (10-50m) resolution anywhere in the US. A workflow management package, Watershed Workflow1 is used to leverage these APIs to generate hyperresolution hydrologic meshes for simulations on full river basins. The result is a standard simulation which can be run with reasonably good parameterizations of surface and subsurface properties and meteorological forcing datasets anywhere in the US.

Then, we discuss how novel programming models are allowing the development of code that is “performance portable” across a variety of computer architectures. Kokkos2 provides an architecture abstraction layer on which hydrologic models can be developed, ensuring efficient execution on leadership-class machines with architectures ranging from GPUs to many-core. We describe how the Advanced Terrestrial Simulator (ATS)3 was refactored to use Kokkos, and show results from simulations on Summit, a GPU-based machine that is currently the second fastest computer in the world.

  1. Watershed Workflow. https://ecoon.github.io/watershed-workflow/. 2020.
  2. C. Edwards and D. Sunderland. DOI: 10.1145/2141702.2141703
  3. E.T. Coon et al. Advanced Terrestrial Simulator. DOI: 10.11578/dc.20190911.1
Previous Abstract | Next Abstract >>