# Where to Find WeatherNext Documentation: Complete Guide to Google's AI Weather Models

> Find WeatherNext documentation within the google-deepmind/weathernext GitHub repo. Explore the README, Colab notebooks, and code comments for a complete guide to Google's AI weather models.

- Repository: [Google DeepMind/weathernext](https://github.com/google-deepmind/weathernext)
- Tags: getting-started
- Published: 2026-08-16

---

**WeatherNext documentation is located entirely within the google-deepmind/weathernext GitHub repository, spanning the root README, model-specific READMEs in `docs/`, runnable Colab notebooks, and inline code comments in the architecture files.**

- **Root README.md** – high-level overview, model family summary, quick-start instructions, and licensing
- **`docs/` directory** – model-specific guides for WeatherNext 1 Graph (GraphCast), WeatherNext 1 Gen (GenCast), and WeatherNext 2
- **Colab notebooks** – end-to-end runnable demos in `docs/weathernext2/`, `docs/weathernext1_graph/`, and `docs/weathernext1_gen/`
- **Source code** – extensively documented architecture files and utility modules

## Primary Entry Points

### Root README: Your Starting Point

The repository's [[`README.md`](https://github.com/google-deepmind/weathernext/blob/main/README.md)](https://github.com/google-deepmind/weathernext/blob/main/README.md) serves as the central hub. It contains:
- Summary of all three WeatherNext model variants
- Installation instructions via `pip install git+https://github.com/google-deepmind/weathernext.git`
- Links to pretrained weights hosted on Google Cloud Storage
- Citation information and license details

### Model-Specific READMEs in `docs/`

Each WeatherNext variant maintains dedicated documentation:

| Model | Location |
|-------|----------|
| **WeatherNext 2** (latest) | [[`docs/weathernext2/README.md`](https://github.com/google-deepmind/weathernext/blob/main/docs/weathernext2/README.md)](https://github.com/google-deepmind/weathernext/blob/main/docs/weathernext2/README.md) |
| **WeatherNext 1 Graph** (GraphCast) | [[`docs/weathernext1_graph/README.md`](https://github.com/google-deepmind/weathernext/blob/main/docs/weathernext1_graph/README.md)](https://github.com/google-deepmind/weathernext/blob/main/docs/weathernext1_graph/README.md) |
| **WeatherNext 1 Gen** (GenCast) | [[`docs/weathernext1_gen/README.md`](https://github.com/google-deepmind/weathernext/blob/main/docs/weathernext1_gen/README.md)](https://github.com/google-deepmind/weathernext/blob/main/docs/weathernext1_gen/README.md) |

These READMEs detail architecture specifics, checkpoint file names, and hardware requirements.

## Interactive Documentation: Colab Notebooks

The WeatherNext documentation includes three fully runnable notebooks that demonstrate end-to-end workflows:

- **`docs/weathernext2/wn2_demo.ipynb`** – Complete WeatherNext 2 demo covering data loading, model initialization, autoregressive rollout, and visualization
- **`docs/weathernext1_graph/graphcast_demo.ipynb`** – GraphCast deterministic forecast walkthrough
- **`docs/weathernext1_gen/gencast_mini_demo.ipynb`** – Probabilistic ensemble generation with GenCast

Each notebook can be opened directly in Google Colab from the GitHub interface.

## Architecture Documentation in Source Code

### Core Model Implementations

The most detailed WeatherNext documentation lives in the architecture source files themselves:

| File Path | Purpose |
|-----------|---------|
| [`weathernext/weathernext2/architecture.py`](https://github.com/google-deepmind/weathernext/blob/main/weathernext/weathernext2/architecture.py) | **FGN (Forecast-Generation Network)** – WeatherNext 2's fully-convolutional graph architecture |
| [`weathernext/weathernext1_graph/graphcast.py`](https://github.com/google-deepmind/weathernext/blob/main/weathernext/weathernext1_graph/graphcast.py) | One-step GraphCast GNN implementation |
| [`weathernext/weathernext1_gen/gencast.py`](https://github.com/google-deepmind/weathernext/blob/main/weathernext/weathernext1_gen/gencast.py) | Diffusion-based ensemble forecasting engine |

These files contain extensive docstrings explaining:
- Tensor shapes and dimensions
- Mesh transformation operations
- Autoregressive step functions

### Shared Utilities (`weathernext/utils/`)

Common functionality used across all models is documented in:

- **[`rollout.py`](https://github.com/google-deepmind/weathernext/blob/main/rollout.py)** – Autoregressive inference loop with `run_autoregressive()` function
- **[`normalization.py`](https://github.com/google-deepmind/weathernext/blob/main/normalization.py)** – Input/output scaling for physical variables
- **[`icosahedral_mesh.py`](https://github.com/google-deepmind/weathernext/blob/main/icosahedral_mesh.py)** – Spherical mesh construction and indexing
- **[`autoregressive.py`](https://github.com/google-deepmind/weathernext/blob/main/autoregressive.py)** – Core stepping logic for multi-forecast generation

## External Developer Guides

Google provides supplementary WeatherNext documentation at [developers.google.com/weathernext](https://developers.google.com/weathernext/guides/models):

- **Model Guide & API Documentation** – cloud deployment patterns
- **Forecast Data Feed Access** – Google Cloud, WeatherLab, and OpenMeteo integration

These external resources focus on operational usage rather than research implementation.

## Code Example: Navigating the Documentation

```python

# Verifying your WeatherNext installation and locating documentation

import weathernext

# Inspect the package structure to find available modules

print(dir(weathernext))

# Expected output includes: 'weathernext1_graph', 'weathernext1_gen', 

#                          'weathernext2', 'utils'

# Accessing WeatherNext 2 architecture with inline documentation

from weathernext.weathernext2 import architecture as wn2_arch

# View docstring for the main FGN class

help(wn2_arch.FGN)

# Load pretrained weights (paths documented in README)

params = wn2_arch.load_weights(
    "gs://dm_graphcast/WeatherNext2_<model_version>.npz"
)

# Run inference using documented utilities

from weathernext.utils import rollout

forecast = rollout.run_autoregressive(
    model=wn2_arch.FGN(params),
    init_state=initial_conditions,  # ERA5 or HRES data

    steps=10,                       # 60-hour forecast (10 × 6hr)

    dt_hours=6
)

```

## Key Documentation Files Reference

| Documentation Type | File Path | URL |
|-------------------|-----------|-----|
| Root overview | [`README.md`](https://github.com/google-deepmind/weathernext/blob/main/README.md) | [GitHub](https://github.com/google-deepmind/weathernext/blob/main/README.md) |
| WeatherNext 2 guide | [`docs/weathernext2/README.md`](https://github.com/google-deepmind/weathernext/blob/main/docs/weathernext2/README.md) | [GitHub](https://github.com/google-deepmind/weathernext/blob/main/docs/weathernext2/README.md) |
| GraphCast guide | [`docs/weathernext1_graph/README.md`](https://github.com/google-deepmind/weathernext/blob/main/docs/weathernext1_graph/README.md) | [GitHub](https://github.com/google-deepmind/weathernext/blob/main/docs/weathernext1_graph/README.md) |
| GenCast guide | [`docs/weathernext1_gen/README.md`](https://github.com/google-deepmind/weathernext/blob/main/docs/weathernext1_gen/README.md) | [GitHub](https://github.com/google-deepmind/weathernext/blob/main/docs/weathernext1_gen/README.md) |
| WeatherNext 2 demo | `docs/weathernext2/wn2_demo.ipynb` | [GitHub](https://github.com/google-deepmind/weathernext/blob/main/docs/weathernext2/wn2_demo.ipynb) |
| FGN architecture | [`weathernext/weathernext2/architecture.py`](https://github.com/google-deepmind/weathernext/blob/main/weathernext/weathernext2/architecture.py) | [GitHub](https://github.com/google-deepmind/weathernext/blob/main/weathernext/weathernext2/architecture.py) |

## Summary

- **WeatherNext documentation is repository-native** – all essential guides, examples, and API references live in `google-deepmind/weathernext`
- **Start with the root README** for installation, model selection, and checkpoint access
- **Consult model-specific READMEs** in `docs/` for architecture details and hardware requirements
- **Run Colab notebooks** for hands-on learning without local setup
- **Read source code directly** – the architecture files ([`architecture.py`](https://github.com/google-deepmind/weathernext/blob/main/architecture.py), [`graphcast.py`](https://github.com/google-deepmind/weathernext/blob/main/graphcast.py), [`gencast.py`](https://github.com/google-deepmind/weathernext/blob/main/gencast.py)) contain definitive implementation details
- **Reference `weathernext/utils/`** for shared inference patterns and data handling

## Frequently Asked Questions

### Where is the official WeatherNext API documentation?

The official WeatherNext documentation resides entirely within the GitHub repository. There is no separate documentation site. The root [[`README.md`](https://github.com/google-deepmind/weathernext/blob/main/README.md)](https://github.com/google-deepmind/weathernext/blob/main/README.md) provides installation and setup, while model-specific guides in `docs/` cover detailed usage. For cloud deployment patterns, Google Developers hosts supplementary guides at [developers.google.com/weathernext](https://developers.google.com/weathernext/guides/models).

### How do I run WeatherNext examples without installing anything?

Open the Colab notebooks directly from GitHub. Navigate to [`docs/weathernext2/wn2_demo.ipynb`](https://github.com/google-deepmind/weathernext/blob/main/docs/weathernext2/wn2_demo.ipynb) or the GraphCast/GenCast equivalents, then click the "Open in Colab" badge. These notebooks run in Google's free tier and include pip installation cells, so no local environment is required.

### What hardware do I need to run WeatherNext models?

Requirements vary by model as documented in their respective READMEs. WeatherNext 2 (FGN) runs on TPU v4 or v5e according to [[`docs/weathernext2/README.md`](https://github.com/google-deepmind/weathernext/blob/main/docs/weathernext2/README.md)](https://github.com/google-deepmind/weathernext/blob/main/docs/weathernext2/README.md). GraphCast requires less memory and can run on single GPUs for short rollouts. The Colab notebooks include fallback configurations for limited hardware.

### Where are the pretrained WeatherNext model weights documented?

Checkpoint paths and filenames are listed in each model's README within `docs/`. The root README provides the base Google Cloud Storage bucket (`gs://dm_graphcast/`) and naming conventions. The [[`weathernext2/architecture.py`](https://github.com/google-deepmind/weathernext/blob/main/weathernext2/architecture.py)](https://github.com/google-deepmind/weathernext/blob/main/weathernext/weathernext2/architecture.py) `load_weights()` function accepts these paths directly and handles download caching.