GPU for Noah-MP Refactoring #357
Description
1. Requester Information:
This should include the name and contact information of the person making the request.
- PI's Full Name: Gou-Yue Niu
- PI's Affiliated Institute: University of Arizona
- PI's Affiliated Email Address: niug@arizona.edu
2. Project Information:
Provide the CIROH project name associated with this research work along with a brief description of the project and its goals. This can help the infrastructure team understand the context and purpose of the requested resources. Please highlight how this project will be benefit from and/or provide benefit to other resources on the shared infrastructure.
CIROH Project Name:
Development and Evaluation of a Differentiable, Learnable, Parsimonious but Functionally Equivalent Version of Noah-OWP-Modular (DFE-NOM) for Use in NextGen NWM
Project Description and Goals:
This CIROH project aims to refactor and evolve the current Noah-OWP-Modular (NOM) land surface model used in NOAA’s NextGen National Water Model into a Differentiable, Functionally Equivalent NOM (DFE-NOM). The DFE-NOM will integrate physics-based hydrologic modeling with machine learning to enable differentiable parameter learning (dPL), improved physical consistency, and enhanced predictive performance for evapotranspiration (ET), snow water equivalent (SWE), and streamflow. By modularizing the codebase using the Basic Model Interface (BMI) and implementing it in PyTorch, the project will make the model interoperable, extensible, and computationally efficient—facilitating model swapping and data-driven experimentation across the CIROH consortium.
Relevance to Shared Infrastructure:
The project will heavily utilize CIROH’s shared computational infrastructure (e.g., GPU-enabled HPC resources) for model training, large-scale evaluation, and multi-basin testing across the conterminous U.S. The DFE-NOM will directly benefit other CIROH efforts by:
Providing BMI-compliant, differentiable modules for vegetation, snow, and soil that can be easily integrated with other NextGen components or alternative process models developed by CIROH partners.
Sharing trained parameter sets, process modules, and data assimilation tools that improve the efficiency and scalability of community hydrologic modeling workflows.
Serving as a benchmark framework for future ML-augmented process modeling studies across CIROH institutions.
In turn, DFE-NOM development will benefit from shared CIROH resources—including common datasets, cloud infrastructure (AWS/Google Cloud), and collaborative development pipelines—ensuring interoperability, reproducibility, and broad adoption within the NextGen community.
3. Project Description:
If your project involves developing software or scripts, briefly describe the software you plan to develop.
This project will develop a Differentiable, Functionally Equivalent Noah-OWP-Modular (DFE-NOM)—a refactored, PyTorch-based version of the current Noah-OWP-Modular (NOM) land surface model used in NOAA’s NextGen National Water Model (NWM). The new software will feature modular, BMI-compliant components for vegetation, snow, and soil processes, enabling easy swapping and integration of alternative process modules developed across the CIROH community.
The software will incorporate differentiable parameter learning (dPL) capabilities, allowing machine learning algorithms to be trained jointly with physical model equations for optimizing parameters governing evapotranspiration, snow, and streamflow. Supporting scripts will be created for model training, benchmarking, and evaluation across multiple basins, leveraging HPC and GPU resources for large-scale testing.
All model modules, training utilities, and example workflows will be shared through open-source CIROH repositories, ensuring interoperability, reproducibility, and reusability across CIROH partners and NextGen NWM developers.
4. Resource Requirements:
Specify the compute, storage, and network resources needed for the project. Be as specific as possible about the number of resources required, and any specific configurations or capabilities needed. This information will help the infrastructure team determine the appropriate resources to allocate.
Options:
-
HPC or VM
Jetstream2 with GPU and large memory -
vCPU
Standard -
Memory
128 GB -
Disk Space
20 TB -
Pantarhei/NSF ACCESS Allocation (specify one)
- More info about Pantarhei: CIROH Pantarhei Documentation
- More info about NSF ACCESS: ACCESS Resource Allocations
I have a NSF credentials
ACCESS ID/username is mfarmani
5. Working Group
Working Group 1/2/3/4 (select one):
6. Timeline
Project start date: 7/1/2025
Project end date: 3/31/20207
In addition, please indicate the expected timeline for the project and when the resources will be needed. This information can help the infrastructure team plan and allocate resources accordingly.
7. Security and Compliance Requirements:
If there are any specific security or compliance requirements for the project, please state them clearly below. This will help ensure that the necessary security measures are in place for the project.
8. Approval:
Indicate the necessary approval processes or sign-offs required for the request.