Negev Shelter Access Analysis

Interactive map for analyzing bomb shelter accessibility for Bedouin communities in the Eastern Negev. The app identifies underserved areas and proposes model-suggested new shelter locations using precomputed DBSCAN + K-means clustering. Covered in Yediot Ahronot / Ynet.

Live demo: negevurbanresearch.github.io/shelter_access

License

This project is released under the MIT License.

Data & analysis

Input data

• Shelters
  • Distributed mobile shelters and permanent shelters in educational institutes from Eshkol Negev Mizrach
  • Formal and informal shelter locations collected by Bimkom field workers and community submissions
• Buildings: footprints from MS Planetary Computer

Precomputed optimizer outputs ship under data/optimal_locations/. Building and administrative GeoJSON used by the map are included in data/.

Site optimization algorithm (Python)

scripts/shelter_optimizer_ensemble.py runs an offline DBSCAN + K-means ensemble:

1. DBSCAN: natural building clusters across 10 eps multipliers (0.1–1.0) relative to coverage radius
2. K-means: systematic centroids at k=750 and k=1500 (2 seeds each)
3. Selection: choose non-overlapping candidates that maximize coverage, accounting for existing shelters

Assumptions encoded in the optimizer include ~7 people per building footprint and a 500-shelter planning target (see constants at the top of the script).

Quick start (web app)

npm install
npm start

Then open http://localhost:3000. For a live-reload server: npm run dev.

The site is static (HTML/CSS/JS + GeoJSON). No backend is required at runtime.

Reproducing the analysis (Python)

Scripts under scripts/ are one-off preprocessing jobs; outputs are already stored in data/. Re-run only when updating source assets.

python -m pip install -r requirements.txt
python scripts/shelter_optimizer_ensemble.py

Other utilities:

Script	Role
shelter_optimizer_ensemble.py	DBSCAN + K-means shelter siting
create_lightweight_data.py	Lightweight buildings GeoJSON for the map
calculate_accessibility_heatmap.py	Accessibility heatmap JSON
generate_shelter_statistics.py	Summary charts (writes to output/)
filter_geospatial_data.py	Spatial filters for study area
simplify_statistical_areas.py	Simplify statistical-area polygons

Alert analysis submodule

alert-analysis/ scrapes and filters historical rocket-alert places. Large alert GeoJSON files are gitignored; regenerate with the Node export/filter scripts documented in alert-analysis/README.md.

Technical architecture

Layer	Stack
Visualization	deck.gl (WebGL), Turf.js, vanilla JS
Analysis	Python (numpy, scikit-learn, geopandas, shapely)
Deploy	GitHub Pages (static site)

Project structure

shelter_access/
├── index.html                 # Main application
├── css/styles.css
├── js/
│   ├── app.js                 # UI / map controller
│   └── spatial-analysis-simple.js
├── data/                      # GeoJSON + precomputed optimizer outputs
├── scripts/                   # Offline Python analysis (run once to refresh data)
├── alert-analysis/            # Optional alert scrape / filter tooling
├── CITATION.cff               # Citation metadata (GitHub + Zenodo)
├── LICENSE                    # MIT
├── requirements.txt           # Python deps for scripts/
└── package.json

Authors

1. Noam J. Gal (corresponding) — Department of Spatial Sciences, The Hebrew University of Jerusalem
2. Artem Nikitin — The Center for Urban Innovation, The Hebrew University of Jerusalem
3. Michael Drogochinsky — The Center for Urban Innovation, The Hebrew University of Jerusalem
4. Yonatan Cohen — Negev Urban Research Lab, Ben Gurion University
5. Merav Battat — Negev Urban Research Lab, Ben Gurion University
6. Talia Kaufmann — The Center for Urban Innovation, The Hebrew University of Jerusalem
7. Ariel Noyman — Media Lab, Massachusetts Institute of Technology

Project partners also included local partners at Bimkom and East Negev / Civix, as well as consultation and data provided by Arch. Lobna Alsana.

How to Cite

To add with details by paper publication

Zenodo archive

Instructions for adding before publication:

1. Merge this citation-ready metadata to main.
2. On Zenodo, enable GitHub integration for NegevUrbanResearch/shelter_access.
3. Create a GitHub Release (e.g. tag v1.0.0). Zenodo will archive the release and mint a DOI, using fields from CITATION.cff.
4. Add the DOI under identifiers in CITATION.cff and update the BibTeX block above (and optionally a DOI badge in this README).

Do not add a competing .zenodo.json unless you need Zenodo-only fields; if both exist, Zenodo prefers .zenodo.json and ignores CITATION.cff for deposit metadata.