November 9-10, 2023
Workshop on Correctness and Reproducibility for Climate and Weather Software
Model simulations are essential tools for understanding weather and climate. As we adapt to our changing climate, simulation codes inform both our understanding and policy decisions. These complex software artifacts are often the result of multiple decades of development. And they are in a state of near-constant development as scientific capabilities advance and high-performance computing (HPC) technologies evolve.
Given the societal importance of these codes, maintaining confidence and preserving code quality and reliability is critical. Yet scientific computing applications are often developed without the use of extensive software verification tools and techniques. Instead, development practices are typically dominated by short-term concerns about performance, resources, and project timelines. Technical challenges in running and evaluating climate and weather models further complicate code verification efforts. Given the scale of these models, a thorough correctness evaluation may be prohibitively expensive. It is also customary to require regression tests to yield bitwise identical results. This requirement is often unmet due to the chaotic nature of climate and weather models and the large variety of hardware/software environments they are run on. When bitwise identical results cannot be sought, field experts are to evaluate model results in a time-consuming and subjective manner.
In short, climate and weather modeling communities are in need of practical and feasible means of ensuring correctness and reproducibility. For example, we are interested in means to easily assess whether changes to a model code result in output that is systematically different or introduce artifacts that could influence scientific conclusions. Such changes may include hardware or software stack infrastructure differences, replacing parts of the model with ML-routines, or applying data compression to the output data. In this workshop, we aim to provide a venue to discuss challenges, opportunities, and recent advances in ensuring software correctness and reproducibility for climate and weather modelers, HPC community members, and industry partners.
Topics of interest include but not limited to:
Submissions may include technical results, approaches, experiences, and opinions involving one or more of the above topics applied to:
Thursday November 9th, 2023 | ||
8:30 - 8:40 | OPENING REMARKS | |
8:40 - 9:40 | KEYNOTE: Models, Data, and Wisdom: How do we know when to trust a climate model? slides | Steve Easterbrook School of the Environment and Department of Computer Science, University of Toronto |
9:40 - 10:00 | TALK: Component Level Regression Testing in a Hierarchical Architecture. slides | Thomas Clune NASA Goddard Space Flight Center |
10:00 - 10:20 | TALK: High Performance Climate and Weather Benchmark (HPCW): a framework for reproducible benchmarks of ESM models and mini-applications. slides | David Guibert Center for Excellence in Performance Programming, Eviden |
10:20 - 10:50 | BREAK | |
10:50 - 11:10 | TALK: Correctness Challenges in HPC and ML. slides | Harvey Dam, Ganesh Gopalakrishnan Department of Computer Science, University of Utah |
11:10 - 11:30 | TALK: Reliable and reproducible Earth System Model data analysis with ESMValTool. slides | Valeriu Predoi* NCAS-CMS, University of Reading |
11:30 - 11:50 | TALK: Testing approach for porting legacy 4-mode Modal Aerosol Model (MAM4) to C++/Kokkos. slides | Balwinder Singh Atmospheric Sciences and Global Change Division, Pacific Northwest National Laboratory |
11:50 - 12:10 | TALK: Verification of the ICON model with the GT4Py dycore - challenges and insights. slides | Abishek Gopal* Institute for Atmospheric and Climate Science, ETH Zurich |
12:10 - 1:10 | LUNCH | Mesa Lab Cafeteria Included with Registration |
1:10 - 2:10 | KEYNOTE: Earth system models of the future. slides | Peter Dueben* Earth System Modelling Section, European Centre for Medium Range Weather Forecasts (ECMWF) |
2:10 - 2:30 | TALK: A Theory of Scientific Programming Efficacy. slides | Michael Coblenz Department of Computer Science, UC San Diego |
2:30 - 2:50 | TALK: An overview of the MOM6 development cycle. slides | Marshall Ward* Geophysical Fluid Dynamics Lab, NOAA |
2:50 - 3:20 | BREAK | |
3:20 - 3:40 | TALK: Challenges in Ensuring Reproducibility for Machine Learning Weather Model Training and Deployment. slides | David John Gagne Computational and Information Systems Lab, NCAR |
3:40 - 4:00 | TALK: METplus: The Long and Winding Road to Unified Verification. slides | Tara Jensen* Research Applications Lab, NCAR |
4:00 - 4:20 | TALK: Unit Testing NCEPLIBS. slides | Edward Hartnett CIRES/NOAA |
4:20 - 5:00 | OPEN DISCUSSION | |
Friday November 10th, 2023 | ||
8:30 - 9:30 | KEYNOTE: Lightweight Formal Methods: The What, Why, and How. slides | John Baugh* Civil Engineering and Operations Research, North Carolina State University |
9:30 - 9:50 | TALK: What could the next 30 years of software verification in climate science look like?. slides | Dominic Orchard* Department of Computer Science and Technology, University of Cambridge and School of Computing, University of Kent |
9:50 - 10:10 | TALK: Parallel reproducibility of the SHYFEM-MPI model. slides | Francesco Carere* Euro Mediterranean Center on Climate Change Foundation (CMCC Foundation) |
10:10 - 10:40 | BREAK | |
10:40 - 11:40 | KEYNOTE: Contained Chaos: Quality Assurance for the Community Earth System Model. slides | Dorit Hammerling Applied Mathematics and Statistics, Colorado School of Mines |
11:40 - 12:00 | TALK: Methods and Tools for the Application of UF-ECT to New Climate Models. slides | Teo Price-Broncucia Department of Computer Science University of Colorado Boulder |
12:00 - 12:20 | TALK: Ensure the correctness and reproducibility in UFS Weather Model CI. slides | Jun Wang NOAA NWS/EMC |
12:20 - 1:20 | LUNCH | Mesa Lab Cafeteria Included with Registration |
1:20 - 1:40 | TALK: Towards Ensuring Statistical Climate Reproducibility of Earth System Models in the Exascale Age. slides | Salil Mahajin Computational Earth Sciences Group, Oakridge National Laboratory |
1:40 - 2:00 | TALK: Improvements in Reproducibility Testing Through False Discovery Rate Correction. slides | Michael Kelleher Computational Earth Sciences Group, Oakridge National Laboratory |
2:00 - 3:30 | PANEL: Correctness and verification across platforms Panelists: - Ilene Carpenter, Hewlett Packard Enterprise - Karsten Peters-von Gehlen, Deutsches Klimarechenzentrum GmbH (DKRZ) - Ganesh Gopalakrishnan, University of Utah - Aaron Donahue, Livermore National Laboratory Moderator: Brian Dobbins, NCAR |
|
3:30 - 4:00 | BREAK | |
4:00 - 5:00 | CLOSING DISCUSSION |
The workshop will be held in person (with a virtual option) and at the Mesa Laboratory of the National Center for Atmospheric Research. (Helpful things to know for your visit.)
As an architectural landmark, the Mesa Laboratory of the National Center for Atmospheric Research is located atop Table Mesa at the west end of Boulder, Colorado, and is recognized as one of the major works of architect I.M. Pei.
Address: 1850 Table Mesa Dr, Boulder, CO 80305
Note: Virtual Meeting details will be announced later.