earth2studio-test

testing out nvidia earth2studio

Basic Inference (no GPU required)

PYTHON_VERSION=3.12
EARTH2STUDIO_VERSION=0.13.0

uv init --python=$PYTHON_VERSION
uv add "earth2studio @ git+https://github.com/NVIDIA/earth2studio.git@$EARTH2STUDIO_VERSION"

# For examples/ notebooks in vscode
uv add "ipykernel"

# For easiest (inference only) examples/getting-started/
uv add "earth2studio[dlwp] @ git+https://github.com/NVIDIA/earth2studio.git@$EARTH2STUDIO_VERSION"
uv add cartopy

Data Assimilation examples (GPU required)

These examples combine N-D gridded and sparse in-situ tabular data as inputs to forecast models.

1. 01_stormcast_sda.ipynb: Combines gridded HRRRR input with NOAA’s Integrated Surface Database (ISD) is a global database that consists of hourly and synoptic surface observations compiled from numerous sources

Skypilot setup

Since many examples require a GPU we'll use SkyPilot to privision a GPU-enabled VM in Azure.

NOTE: this takes a while because it installs lots of large dependencies like PyTorch and CUDA.

# Install skypilot if you haven't already
pixi shell
sky check azure
sky launch vscode-earth2studio-T4.yaml
# Once launched, ssh or connect via VSCode Remote SSH to the VM to run notebooks in examples/ folder