HeatGuard SG

HeatGuard SG is DeltaNexus's heat-planning dashboard for Singapore. It combines current station heat index, a planning-area Heat Risk Score, cooling-infrastructure scenarios and published local micro-environment heat estimates in one map-led interface.

Open the live dashboard · Download the one-page concept poster

Contributors

DeltaNexus: Chen Xuanhong and Zhang Jiaheng, National University of Singapore, Chemistry, 2025 cohort.

@fromzerotoinfinityplus

Full project credits are recorded in CONTRIBUTORS.md.

Implemented modules

• Live Heat Index — current heat-index conditions across the weather-station network, with observation and refresh times.
• Area Heat Risk — a policy-calibrated random-forest score that combines forecast heat, official 2026 age structure and mapped eldercare access across 55 planning areas.
• Cooling Simulation — selectable parks, cycling routes and park connectors, area-specific green baselines, adjustable cooling points and corridor shade.
• Local Heat Estimate — published spatial estimates combining a station-network background with nearby vegetation, HDB building form, coastline exposure and geographic position at 250 m, 500 m or 1 km.
• Community Data — structured, privacy-minimised collection for green-pocket proposals, build confirmation, monthly condition checks and on-site comfort feedback.

The public site also includes planning-area and subzone demographics, close-zoom HDB building detail, a ranked priority queue, model evidence and the completed DAISI B1 delivery flow.

Repository layout

backend/          FastAPI service, weather ingestion and model runtime
backend/model/    V4 TFT and planning-area risk-model weights
dist/             GitHub Pages website, spatial layers and published model results
docs/             Concept poster and project visuals
reports/          Held-out validation evidence
scripts/          Forecast publication, model validation and data-build tools

Run locally

Use Python 3.10 or newer. The project was developed in the local cueq environment.

conda activate cueq
python -m pip install -r backend/requirements.txt
cp backend/.env.example .env
python run_live.py

Open http://localhost:8000. The service exposes /api/health, /api/heatguard-data.json and /api/community-submissions, and serves the website from the same origin. Community records and photos are private backend runtime data and are excluded from Git.

For a website-only preview:

python -m http.server 8001 --directory dist

Open http://localhost:8001. HTTP serving enables the lazy-loaded HDB and green-infrastructure layers.

Local environment heat results

dist/data/local-heat-results.json contains the public outputs for 55 planning areas at three analysis radii. The validation summary is published in reports/local-heat-model-validation.md: 2,210 hourly observations across 11 stations, leave-one-station-out MAE 1.011 °C and RMSE 1.482 °C.

The fitted local-environment model, its private weights and training implementation are not included in this public repository. The public site contains only the interface, validation summary and prediction results. The result is spatial downscaling under a specified network heat background—not a next-day weather forecast, direct observation or official alert.

Reproduce inference

The repository includes both released model artefacts:

• backend/model/best_model.pt — V4 history-only full Temporal Fusion Transformer.
• backend/model/risk_model.joblib — policy-calibrated planning-area random forest.
• backend/model/v4_runtime_metadata.json — feature order and training normalisation used by the V4 runtime.

Install CPU PyTorch and the inference dependencies, then publish a fresh result:

python -m pip install torch==2.6.0 --index-url https://download.pytorch.org/whl/cpu
python -m pip install -r requirements-inference.txt
python scripts/publish_forecast.py

DATA_GOV_SG_API_KEY is optional and belongs in a local .env file or GitHub Actions secret. No credentials are stored in this repository. The scheduled workflow recomputes forecasts from current public observations and commits only the updated result file.

Data sources

• data.gov.sg / NEA: air temperature, relative humidity and rainfall
• URA: planning-area and subzone boundaries
• SingStat and HDB: 2026 age structure and building context
• MOH: eldercare-service locations
• NParks: parks, nature reserves and Park Connector Network
• LTA: cycling network

Detailed spatial provenance is recorded in dist/data/spatial/metadata.json. HeatGuard SG is designed for planning review with human oversight; public health actions should continue to follow official Singapore guidance.

License

Released under the MIT License. Source datasets remain subject to their respective providers' terms.