GeoStack
Welcome to the documentation for GeoStack.
Overview¶
GeoStack is a vertically integrated, JAX-native research software stack for geophysical modeling, inference and data assimilation. We suggest you start with the GeoStack overview, which covers the problem, the three organising axes, the package map and an end-to-end walkthrough. Its reading list links every report, the GeoModels pages and the benchmarks gallery.
geostack has no package of its own. The scripts and notebooks here compose the libraries it builds on (pipekit, geotoolz, xrtoolz and georeader), which are pinned as git submodules under references/. So any reusable code goes into those libraries.
The map of the stack: three organising axes, the tiers, a methane walkthrough, open questions and the plan.
Code sketches for every pattern in the overview, from operators to reproducibility artifacts.
Every public name in the libraries GeoStack builds on. Search it before writing new code.
Features¶
Pixi environment management with conda-forge packages
DVC for data versioning and pipeline management
Hydra + hydra-zen for configuration management
MyST for scientific documentation
JupyterLab and Marimo notebook environments
Agent-ready:
AGENTS.mdandCODE_REVIEW.mdfor coding agents, and a capability index of the libraries geostack builds on
Getting Started¶
See the Getting Started tutorial to set up your environment.