rkeisler

rkeisler / keisler-2022

Public

Graph neural network weather forecasting, from Keisler 2022

rkeisler.github.iograph_weather
88
13
100% credibility
Found May 06, 2026 at 88 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

This repository contains code for running a machine learning model that generates global weather forecasts up to 10 days ahead using graph neural networks, based on a 2022 research paper.

How It Works

1
🌍 Discover the Weather Predictor

You hear about this smart tool that forecasts weather worldwide using clever pattern recognition from past data.

2
💻 Get It Ready on Your Computer

You follow easy instructions to install it, picking a simple setup that works on your regular or fast machine.

3
📅 Choose a Starting Date

Pick any date from the past for testing or a recent one to see near-future weather.

4
🚀 Run Your First Prediction

Tell it the date and days ahead, and it grabs real weather info to create a full forecast in under a minute.

5
🗺️ See Your Weather Maps

Open the saved file to view predictions for temperature, wind, rain, and more across the globe.

6
📈 Explore Cool Examples

Try ready-made checks like accuracy tests, storm tracking, or what-if sensitivity maps.

✅ Become a Weather Forecaster

Now you can predict global weather changes anytime and share impressive maps with friends.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 88 to 88 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is keisler-2022?

keisler-2022 runs the graph neural networks weather model from Keisler 2022, turning ERA5 reanalysis or ECMWF Open Data into global 6-hour forecasts via autoregressive rollouts up to 10 days. You get NetCDF/Zarr outputs from a simple CLI like `uv run forecast.py --init 2020-01-01T00 --steps 40`, with Python/JAX handling fast inference on GPU or CPU. It's graph neural networks (GNNs) in action: encoder to hexagonal mesh, message passing processor, decoder back to lat/lon grids.

Why is it gaining traction?

With 88 graph GitHub stars, it hooks devs via sub-minute 10-day forecasts on GPU, plus scripts for RMSE evaluation, sensitivity maps via JAX autodiff, and hurricane track viz. Beats clunky alternatives with uv sync for deps, CUDA extras, and seamless public data pulls—no data wrangling needed. Graph GitHub README shines with timings, troubleshooting, and real examples like Sandy tracking.

Who should use this?

ML engineers in climate forecasting prototyping graph neural operator baselines, researchers reviewing graph neural networks methods for weather apps, meteorologists needing quick GNN forecasts from ECMWF streams.

Verdict

Strong pick for Keisler 2022 repros—detailed docs, tests, CLI shine—but 1.0% credibility and 88 stars mean it's niche and maturing; validate outputs rigorously before integrating. Ideal if graph neural networks explained via fast weather tools excite you.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.