How to reproduce and extend¶
This project is designed to be reproduced and extended. Everything — code, notebooks, environment specification, and this documentation — lives in a single Git repository: github.com/Philip-Brohan-MO/Auto-Daily-Rainfall-MO.
If you are familiar with GitHub, fork or clone the repository. If you’d rather not, you can download the whole thing as a zip file.
Software environment¶
Everything runs in the weather-doc-extractor Conda environment specified in
environment.yml. See Installation for the one-time setup, then
activate it before doing anything else:
conda activate weather-doc-extractor
Compute¶
The local machine is used only for orchestration and validation. The
compute-intensive work — model extraction over hundreds of thousands of images,
and GPU fine-tuning — runs on Microsoft Azure ML. The training
and inference code itself is not Azure-specific and will run in any suitable Python
environment with access to a GPU, but the submission scripts and job specs under
azureml/ and scripts/ are written for an Azure ML workspace.
Running the workflow¶
The workflow is driven by the notebooks under notebooks/, run in order. Start
with the workflow overview, which lays out the stages and
links to each notebook.
The documentation¶
These web pages are built with Sphinx from the
Markdown sources in the docs/ directory, and published to
GitHub Pages automatically on every push to main.
To build them locally:
pip install sphinx myst-parser
sphinx-build -b html docs docs/_build/html
Credits and acknowledgements¶
This is a follow-on to Robot Rainfall Rescue, which established the small-VLM approach on the monthly rainfall sheets.
The real-data validation set — 64 daily rainfall images with careful, quality-controlled transcriptions — was provided by Ciara Ryan, who transcribed Irish daily rainfall sheets during her PhD. These known-good transcriptions are what let us measure how well the models are really doing.
Contact¶
Contact Philip Brohan
This document is distributed under the terms of the Open Government Licence. Source code is distributed under the terms of the BSD licence.