Inspiration

2025 was Singapore's hottest year on record, and 2026 is the year we become a super-aged society, with about 1 in 5 residents aged 65 or older. NEA already publishes heat stress levels from its WBGT stations, but those describe the weather at one point. They don't show how many elderly residents live nearby or how far the nearest place to cool down is. We wanted to join those pieces up.

What it does

HeatGuard gives every planning area a heat risk score based on three multiplicative factors:

$$\text{Heat Risk Score} = \text{Heat Now} \times \text{Elderly Density} \times \text{Access Gap}$$

Heat Now: Air temperature and WBGT readings, averaged per planning area over the last hour.
Elderly Density: Residents aged 65+ per planning area, scaled from 0 to 1.
Access Gap: Spatial distance from an area's elderly population to the nearest Senior Activity Centre or Active Ageing Centre (acting as cooling shelters), scaled from 0 to 1.

Each area is categorized into Low, Medium, or High risk:

High: Triggers door-to-door outreach calls/visits and extended opening hours at nearby Active Ageing Centres.
Medium: Sends targeted advisories to nearby centres and Town Councils.
Low: Maintained under routine automated monitoring.

Planners can also query the underlying data in plain English through Databricks Genie.

How we plan to build it

We pull live weather feeds and static CSV/GeoJSON datasets from data.gov.sg into Delta tables on Databricks using Python notebooks (Auto Loader/Lakeflow), organized under Unity Catalog. Databricks SQL cleans the data, maps weather stations to planning areas, and calculates the heat risk score. Results are presented via an AI/BI dashboard and an interactive Genie conversational analytics space.

Challenges we expect

Spatial Resolution: Outdoor weather station readings are sparse, making planning-area thermal values estimates.
Granularity: HDB elderly population counts are aggregated by planning area rather than block/flat level, and air-conditioning ownership is unobserved.
Weight Calibration: Component weights and risk thresholds reflect our preliminary modeling judgment, so we plan to back-test and calibrate them against historical high-WBGT days.

What's next

In the two-week build phase, we will:

Days 1–5: Load temperature, WBGT, elderly demographic, and centre coordinate feeds into Delta Lake; map stations to planning areas; compute baseline risk scores.
Days 6–10: Back-test scores against historical extreme heat days, tune threshold cut-offs, build the AI/BI dashboard and Genie space, and rehearse the final presentation.

Beyond the sprint: The framework can integrate real-time indoor sensor networks, H3 hexagonal spatial indexing, and direct alert integration with Town Council and Active Ageing Centre dispatch systems.