Earth2Studio

Next-generation AI
weather modeling

Access a leading collection of weather and climate AI models, production-ready data sources, composable inference APIs, and GPU-accelerated workflows in one Python package.

Get Started

Run an AI forecast with just a few lines of code

UV pip

$uv pip install "earth2studio[fcn]"

from earth2studio import run; from earth2studio.data import GFS
from earth2studio.io import ZarrBackend; from earth2studio.models.px import FCN
model = FCN.load_model(FCN.load_default_package())
run.deterministic(["2024-01-01"], 10, model, GFS(), ZarrBackend("fcn.zarr"))

Open platform

AI weather and climate tooling for every sector

Earth2Studio gives research groups, agencies, enterprises, developers, and classrooms a shared Python surface for models, data, verification, and operational workflows.

Scientists & researchers

Benchmark models on identical data through one API, design ensemble experiments, and verify outputs with built-in deterministic and probabilistic metrics.

Met services & agencies

Run, fine-tune, and deploy forecasting capability on infrastructure you control, from medium-range global guidance to rapid regional workflows.

Enterprise

Build ensemble risk workflows for energy, insurance, logistics, agriculture, and climate resilience with reproducible AI forecasts.

Developers

Build weather-aware APIs, dashboards, agents, and decision products on a composable SDK that keeps models, data, IO, and workflows separate.

Educators & students

Teach Earth system AI with an open on-ramp that can fetch data, load models, run forecasts, and store outputs using familiar scientific Python tools.

Open Source Integrated

Built on the scientific Python ecosystem

ZaZarr XrXarray CuCuPy PaPyArrow RpRAPIDS PtPyTorch ObObstore Fsfsspec

ZaZarr XrXarray CuCuPy PaPyArrow RpRAPIDS PtPyTorch ObObstore Fsfsspec

AI for weather and climate does not need to feel unfamiliar. If you already work with the scientific Python and PyData ecosystem, Earth2Studio gives you familiar building blocks for running modern AI weather models.

Model interfaces

Forecast with the largest collection of AI models in the community

FourCastNet 3AFNO-based medium-range forecasting

AIFS 2.0ECMWF AI forecast model workflows

StormScopeSatellite and radar-conditioned forecast workflows

HEAL-DAData assimilation and analysis correction

Pangu-WeatherOperational-style global forecast rollouts

AuroraFoundation-model forecasting and analysis

StormCast-CONUSRegional CONUS forecasting workflows

CorrDiffDiffusion downscaling workflows

DLESyMCoupled Earth-system model inference

GraphCastGlobal graph neural weather forecasts

ACE-2AI2 climate and weather model interface

AtlasGenerative medium-range forecast workflows

and more

Data connectors

Connect to weather and climate data from around the globe

NOAA

ECMWF

NASA

EUMETSAT

EarthMover

Dynamical

Planetary Computer

Copernicus CDS

NCEP

NCAR

AWS Open Data

NNJA

and more

Explore the API

Modular components for your use case

PrognosticMedium range models Global forecast rollouts and ensemble workflows PrognosticNowcasting models Rapid-update precipitation, satellite, and radar workflows AssimilationData assimilation models Observation-informed analysis and correction workflows PrognosticClimate models Seasonal, climate, and coupled Earth-system models ReanalysisAnalysis data ERA5, HRRR, IFS, and analysis-ready sources NumericalForecast data Operational forecasts for initialization and conditioning DataFramesObservation data Conventional, satellite, radar, and station datasets DiagnosticDownscaling models Super-resolution, correction, and diagnostic downscaling tools

Agent ready

Automate setup discovery and first forecasts

Install Earth2Studio skills, then ask your coding agent to recommend a model, configure an environment, fetch data, or launch a deterministic forecast.

skills.sh

$ npx skills add NVIDIA/skills --skill earth2studio-install $ npx skills add NVIDIA/skills --skill earth2studio-discover $ npx skills add NVIDIA/skills --skill earth2studio-data-fetch $ npx skills add NVIDIA/skills --skill earth2studio-deterministic-forecast

DiscoverRecommend data, models, IO, and docs for a workflow.

InstallSet up Earth2Studio and model-specific dependencies.

RunCreate a forecast with GFS, FourCastNet3, and Zarr output.

Start here

Run a forecast, then make it your own

Earth2Studio keeps the pieces separate: data sources fetch initial states and observations, models transform state, IO stores results, and workflows compose them. Easy to get started, easy to extend.

Tutorial Install Examples API Reference