Skip to article frontmatterSkip to article content
Site not loading correctly?

This may be due to an incorrect BASE_URL configuration. See the MyST Documentation for reference.

Project Overview

What is weather nowcasting?

Weather nowcasting predicts conditions over the next few minutes to a few hours — much shorter-range than standard numerical weather prediction (NWP). It’s especially valuable for fast-moving, high-impact events: heavy rainfall, thunderstorms, lightning, snowfall, fog, and strong wind.

Machine learning is a natural fit here: instead of solving the full physical equations of the atmosphere, models learn spatial and temporal patterns directly from historical radar observations. This is cheaper to run than NWP and can capture localized, rapidly evolving features — complementing rather than replacing physical models.

How the pieces fit together

Source radar data
      ↓
Validation      — checked against the MLCast Zarr spec (see Validator)
      ↓
Intake catalog  — versioned, documented datasets (see Datasets)
      ↓
Sampling        — turns source data into training-ready indices (see Datasets)
      ↓
Training        — ConvGRU ensemble models, via CLI or Python API (see Usage)
      ↓
Evaluation      — shared metrics and benchmarks (see Evaluation)

See Model Architecture for how the model itself works.

Where to go next