Packages

neural-lam

Graph-based neural weather prediction models for limited area modelling, built on PyTorch and PyTorch Geometric. Includes LAM versions of the Keisler (2022) graph model, GraphCast and the hierarchical model from Oskarsson et al. (2023), for training and running forecasts.

mllam-data-prep

Prepares training datasets for data-driven weather models from a declarative YAML config: data sources, variables, transformations and how they map onto the model's inputs. Produces training-optimised zarr datasets.

Roadmap

The community roadmap shows the work planned and proposed for upcoming releases of the packages above. It is read live from GitHub: issues and pull requests on a vX.Y.Z milestone are on the roadmap, and those on a vX.Y.Z (proposed) milestone are discussed at the next community dev meeting.

View the roadmap