WRF Massive

This project provides tooling to orchestrate large-scale WRF (Weather Research and Forecasting) simulation using Slurm. The main goal is to make it easy to define, stage, submit, and monitor many WRF runs (including pre/post processing steps). The framework is designed to be modular, which allows you to test the entire pipeline on a laptop or workstation and scale to HPC clusters with minimal changes.

Key points:

• Orchestration: the code manages job creation, staging of input/output, submission to Slurm.
• Extensible: the project is structured so that general-purpose code lives in the wrf_massive package, while project-specific configuration and templates live in separate workspaces.

More specifically,

• The orchestration code and core CLI are implemented in wrf_massive.
  • wrf_massive is designed to be as project/workspace-agnostic as possible. It contains the main logic to configure runs, produce Slurm job scripts, submit them and collect outputs.
• Concrete projects/simulation runs should live in workspaces
  • An example workspace is provided in workspace/example with sample configuration, job templates and instructions.
• Optionally, WRF and WPS are included as submodules in submodules. You can pull them using git submodule update --init --recursive and build them according to their own instructions.

Submodules vs external WRF/WPS

• Using the included submodules (optional) to build WRF/WPS inside this repository:

  • Clone the NCAR repos using git submodule update --init --recursive
  • Compile and setup WRF as per NCAR's instructions.

• Using an external WRF/WPS installation (recommended for reproducible or managed HPC installs):

  • If you already have a system-wide or separately-maintained WRF/WPS installation, you do not need the submodules. Point this orchestration tool to your existing installation using the configuration mechanism in your workspace (see workspace/example/README.md for an example configuration). This repository only orchestrates runs and does not require WRF/WPS to be compiled inside it.

Workspace examples

The repository contains a workspace/example folder with a self-contained example workspace and an example README.md. Use that as your primary reference for how to structure a workspace that uses this tool.

Environment setup

A conda environment file is provided as environment.yml at the project root to create a reproducible Python environment. It contains the Python packages typically needed to run the orchestration tools, utility scripts and tests.

Create the environment (conda/miniconda/anaconda required):

conda env create -f environment.yml
# or to update an existing env
conda env update -f environment.yml

Activate it:

conda activate wrf-massive  # or the name defined in environment.yml

If you prefer pip or another environment manager, use the package list in environment.yml as a reference to recreate the environment. The project also contains pyproject.toml files that provide metadata for the Python packages in the workspace.

Quick start (high level)

1. Create and activate the Python environment using environment.yml.
2. Prepare WRF/WPS:
3. Inspect workspace/example to copy example configuration and job templates into your workspace.
4. Use the CLI in wrf_massive to create and submit evaluations (see wrf_massive/cli.py for the available commands). Typical invocation is via the Python CLI in your activated environment.