ehw24-hail-photogrammetry

This repository contains the code used for the upcoming publication Performance assessment of drone-based photogrammetry coupled with machine-learning for the estimation of hail size distributions on the ground in the special issue Outcome of the 4th European Hail Workshop 2024: Opening the discussion .

Content

• detection: Scripts used to run the model to detect hail in image tiles
• DHAPpy: Prototype of a python package used in notebooks for further development
• training: Scripts used to train the RCNN models with detectron2
• notebooks: Notebooks used to prepare and analyze the data
• zenodo: Data placeholder for experiments and event (see zenodo/README.md)

Usage

First, clone this repository, navigate into it and install the package DHAPpy (python >= 3.11) with pip:

python -m pip install -e DHAPpy

If you use anaconda, you can use pip inside it.

You can then run commands like this for example:

python -m dhappy.cli split-tiles PATH_TO_IMAGES

Detection

To detect hail in tiles, install detectron2 and run the following:

python detection/detect_hail_lom.py /store_new/mch/msrad/msradv/drone/models/model_20220628/events/20220628 /store_new/mch/msrad/msradv/drone/hail_events/20250523_JS/20250523-ortho-clean -n jtm lai por -t 0.55 0.85 0.6 -f .tif

Data

Model base

For botn experiments and event, we used the R50-FPN 3x baseline for COCO Instance segmentation from the detectron2 model zoo.

References:

• https://github.com/mlainer/hail-photogrammetry
• Portmann, J., Lainer, M., Brennan, K. P., Jourdain de Thieulloy, M., Guidicelli, M., and Monhart, S. (2025). Performance assessment of drone-based photogrammetry coupled with machine-learning for the estimation of hail size distributions on the ground. Front. Environ. Sci. 13, 1602917. doi:10.3389/fenvs.2025.1602917