An Open-Source Tool for Mapping War Destruction at Scale in Ukraine using Sentinel-1 Time Series

Olivier Dietrich^*,1, Torben Peters¹, Vivien Sainte Fare Garnot², Valerie Sticher², Thao Ton-That Whelan³, Konrad Schindler¹, Jan Dirk Wegner²

¹ ETH Zurich ² University of Zurich ³ International Committee of the Red Cross

^*Corresponding author: Olivier Dietrich

🇺🇦 Results

The results, including heatmaps and building footprints, can be found here: ⏳ COMING SOON⏳

🎮 Dashboards

We provide two Google Earth Engine dashboards:

Damage Mapping Tool	Ukraine Damage Explorer

🌲 Repository Structure

notebooks/                     # Jupyter notebooks to reproduce results
  ├── classification.ipynb
  ├── country_stats.ipynb
  └── evaluation.ipynb

src/                            # Source code for data processing, model training, and inference
  ├── classification/           # Classification script to train and evaluate models
  │     ├── dataset.py
  │     ├── main.py
  │     ├── metrics.py
  │     ├── models.py
  │     ├── reducers.py
  │     └── utils.py
  │
  ├── data/                     # Scripts for data processing and handling
  │     ├── overture/
  │     ├── sentinel1/
  │     ├── sentinel2/
  │     ├── quadkeys.py
  │     ├── unosat.py
  │     └── utils.py
  │
  ├── inference/                # Script for inference over entire country
  │     ├── dense_inference.py
  │     └── full_ukraine.py
  │
  ├── postprocessing/           # Download from drive and postprocess results
  │     ├── drive_to_results.py
  │     └── utils.py
  │
  ├── utils/                    # Utils functions
  |     ├── gdrive.py
  |     ├── gee.py
  |     ├── geo.py
  |     └── time.py
  │
  ├── constants.py
  └── __init__.py

🛠️ Setup

This was tested on Ubuntu 22.04 LTS with Python 3.12.7 and 64GB RAM

🐑 Clone the repository

git clone https://github.com/odietric/ukraine-mapping-tool.git
cd ukraine-mapping-tool

🐍 Python Environment

To create the environment, run the following command:

conda create -p ./ukraine_env python=3.12 geopandas gdal -c conda-forge --strict-channel-priority --yes
conda activate ./ukraine_env
python -m pip install uv
python -m uv pip install -r requirements.txt

or simply:

make env

💾 Data

All project data is publicly available:

• UNOSAT

  Preprocessed UNOSAT labels and AOIs are stored in data/unosat_labels.geojson and data/unosat_aois.geojson, respectively.

• Sentinel-1

  Sentinel-1 images are processed in the cloud via Google Earth Engine.

• Overture Maps Building Footprints

  To obtain the preprocessed Overture Maps Building Footprints in Ukraine, you can either download the file ukraine_buildings.parquet from the link above and place it in the data/overture_buildings folder, or compute it yourself by running the following command:

  bash src/data/overture/download_and_preprocess.sh

💽 Google Drive Access Configuration

To download the results from Google Drive, you need to configure the Google Drive API and add the files pydrive_settings.yaml and pydrive_credentials.json to the secrets folder.

1. Create Google API Credentials:

   • Go to Google Cloud Console: API & Services > Credentials > Create Credentials > OAuth client ID > Desktop app.

2. Download and Configure Credentials:

   • Download the .json credentials file.

   • Store the client_id and client_secret values from this file in pydrive_settings.yaml as follows:

     client_config_backend: 'settings'
     
     client_config:
       client_id: <your_client_id>
       client_secret: <your_client_secret>
     
     save_credentials: True
     save_credentials_backend: 'file'
     save_credentials_file: </path/to/project/>secrets/pydrive_credentials.json
     
     get_refresh_token: True
     oauth_scope:
       - "https://www.googleapis.com/auth/drive"

   • Replace <your_client_id> and <your_client_secret> with values from your credentials JSON file.

   • Replace /path/to/project/secrets/pydrive_credentials.json with the full path to the pydrive_credentials.json file within your project.\

   • The pydrive_credentials.json file will be automatically created upon the first authentication attempt.

3. Example usage in code:

   from pydrive2.auth import GoogleAuth
   from pydrive2.drive import GoogleDrive
   from src.constants import SECRETS_PATH
   
   # Initialize authentication
   GoogleAuth.DEFAULT_SETTINGS["client_config_file"] = SECRETS_PATH / "pydrive_credentials.json"
   gauth = GoogleAuth(settings_file=SECRETS_PATH / "pydrive_settings.yaml")
   drive = GoogleDrive(gauth)

Citation

@article{dietrich2024ukraine,
  title={An Open-Source Tool for Mapping War Destruction at Scale in Ukraine using Sentinel-1 Time Series},
  author={Olivier Dietrich and Torben Peters and Vivien Sainte Fare Garnot and Valerie Sticher and Thao Ton-That Whelan and Konrad Schindler and Jan Dirk Wegner},
  journal={arXiv preprint arXiv:/2406.02506},
  year={2024},
  url={https://arxiv.org/abs/2406.02506},
}

License

This project is licensed under the MIT License - see the LICENSE file for details.