Earth System Model Skill Packages: Deep Knowledge Bundles for Noah-MP, CLM, CAM, MOM6, WRF, E3SM, and More
Published:
Earth system models are some of the most complex scientific software ever written, and they are also some of the worst-documented for newcomers. I have been building a series of “skill packages” — structured, progressive-disclosure knowledge bundles — for the major Earth system and land surface models, designed to be used by both new graduate students and AI coding agents.
Author: Koutian Wu; GitHub: ktwu01
What Is a Skill Package
Each repo in this series is built around a SKILL.md routing hub that points to a reference/ directory of deep-dive documents. The structure is designed so that:
- A new graduate student can find the install path, the case workflow, and the debugging section without reading every paper in the bibliography.
- An AI coding agent can route to the right reference doc on demand instead of stuffing the entire model manual into context.
- Contributors know exactly where new content belongs.
The format borrows from progressive disclosure: high-signal index at the top, full depth one click down.
The Skill Packages
| Model | Domain | Repo |
|---|---|---|
| Noah-MP | Land surface (NCAR/noahmp + HRLDAS) | noahmp-skill |
| CTSM / CLM | Land surface (Community Terrestrial Systems Model) | ctsm-skill |
| JULES | Land surface (Joint UK Land Environment Simulator) | jules-skill |
| SUMMA | Land surface (Structure for Unifying Multiple Modeling Alternatives) | summa-skill |
| ParFlow | Watershed flow (parallel) | parflow-skill |
| VIC | Macroscale hydrology (Variable Infiltration Capacity) | vic-skill |
| CAM | Atmosphere (Community Atmosphere Model) | cam-skill |
| WRF | Atmosphere (Weather Research and Forecasting) | wrf-skill |
| MOM6 | Ocean (Modular Ocean Model 6) | mom6-skill |
| E3SM | Coupled (Energy Exascale Earth System Model) | e3sm-skill |
What Each Skill Package Covers
The exact contents vary by model, but every package has the same shape:
- Architecture — the data flow and the major code units, in a few diagrams.
- Physics options — what knobs exist, what they actually do.
- Case workflow — how to set up, compile, and run a real case.
- Output and diagnostics — the formats and how to inspect them.
- Coupling — how the model talks to its neighbors (atmosphere, ocean, land).
- Debugging — the failure modes that bite every new user.
- Contributing — how to send a PR upstream.
Why I Built These
Three reasons:
- Personal use. I work on land surface models for a living. I needed the notes anyway.
- AI-agent compatibility. When AI agents try to modify Earth system code, they fail in revealing ways. A well-structured skill package improves their hit rate dramatically. (Related: ESM-bench, my benchmark for measuring exactly this.)
- Onboarding. Most of these models would benefit from one good “first 30 days” document. The skill packages are an attempt at that document.
Who These Are For
- New graduate students starting on a specific model.
- AI coding agents working on Earth system code.
- Researchers cross-checking how a similar problem is handled in another model.
- Anyone writing tutorials or documentation for these communities.
Contributing
Each repo is independent. If you spot something wrong in a specific skill package, open an issue or PR there. If you want to start a skill package for a model that isn’t on this list, the noahmp-skill repo is the closest thing to a template.
Earth system models are one of the few software ecosystems where a strong index is more valuable than a clever feature. These skill packages are my attempt at that index.