Earth-2 Studio
https://github.com/nvidia/earth2studio- Category
- AI Tools
- Rank
- No. 356Tools index
- Pricing
- Open Source
- Type
- TOOL
- Builder
- nvidia
- GitHub
- 1.1k stars
- Latest release
- 0.18.0
- Date
About
Open-source deep-learning framework for building and deploying AI weather and climate prediction workflows.
What it does
Earth-2 Studio connects forecast or diagnostic models with weather data sources, coordinate-aware tensors, statistics, and output backends. Users assemble these parts into deterministic, ensemble, or custom inference pipelines and can swap components behind shared Python interfaces.
Why it's ranked here
The appeal is breadth joined to a coherent workflow model. It supports numerous global and regional models, remote and local data sources, checkpoint retrieval, and multiple output formats. That usefulness comes with substantial environment and dependency complexity.
What's good
Components exchange explicit tensors and ordered coordinate metadata in unnormalized physical units, making intermediate results interpretable. Shared protocols let researchers replace models or data sources without rebuilding the whole pipeline. Checkpoint fetching and local caching are also handled consistently across supported stores.
Tradeoffs
The base installation does not guarantee every model or example will run. Many models require separate extras, GPU runtimes, source-pinned packages, or extensions that may compile slowly. Model checkpoints and datasets retain third-party licenses, so users must verify usage and redistribution rights themselves.
How to use it well
It best suits researchers and engineers comparing pretrained weather models, composing inference experiments, or turning those experiments into services. Start with one model-specific environment, keep coordinates explicit, and add extras selectively. Use PhysicsNeMo instead when the adjacent requirement is model training recipes.
Technical notes+
pyproject.toml defines a Hatchling-built, typed Python package for Python 3.11 through 3.14, with PyTorch, Xarray, Zarr, cloud-filesystem clients, many model-specific extras, and an optional serve dependency group containing FastAPI, Uvicorn, Redis, RQ, and Prometheus tooling. docs/userguide/about/data.md describes the internal pairing of torch.Tensor with CoordSystem, an ordered mapping of coordinate arrays, while data sources return Xarray arrays before conversion to device tensors. docs/userguide/advanced/auto.md documents AutoModelMixin and Package for checkpoint discovery, download, and Fsspec-backed caching. Makefile exposes uv-based installation, linting, typing, tox testing, coverage, documentation, and container build targets.
Observed
- License
- Apache-2.0
- Primary language
- Python
- Python support
- Python 3.11 through 3.14
- Packaging
- Hatchling build backend with pip and uv installation paths
- Interface
- Typed Python library with modular model, data, workflow, statistics, and output APIs
- Service surface
- Optional FastAPI, Uvicorn, Redis, RQ, and Prometheus dependency group
- Platform metadata
- Operating System Independent; GPU environment classifier
- Quality tooling
- Repository tasks include Ruff, mypy, Black, pytest through tox, and coverage reporting
Read from README.md, Makefile, setup.py, pyproject.toml, requirements.txt, docs/index.md, docs/modules/index.md, docs/userguide/index.md, docs/userguide/about/data.md, docs/userguide/about/index.md, docs/userguide/about/intro.md, docs/userguide/support/faq.md, docs/userguide/support/index.md, docs/userguide/about/install.md, docs/userguide/advanced/auto.md.
What it can do
Build AI weather prediction models
Weather data and model configuration → Trained deep learning weather prediction model
Deploy climate prediction workflows
Trained climate models and deployment parameters → Live climate prediction service
Generate weather forecasts
Current atmospheric data and location coordinates → Weather forecast predictions
Process climate simulation data
Raw climate datasets and simulation parameters → Processed climate simulation results
Train deep learning models on meteorological data
Historical weather datasets and neural network architecture → Trained meteorological prediction model
Create custom weather prediction pipelines
Data sources and workflow configuration → Automated weather prediction pipeline
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.