A live heat index dashboard across the weather station network
5 km
Latest data reading—
Next scheduled update—
Singapore Time · page checks every 5 minutes
HeatGuard SG
See Singapore’s heat now.Plan for tomorrow.
Singapore at its best, with care at its heart.
Local micro-environment
Estimate possible heat from the surroundings.
Local environment heat estimate
Click a planning area to compare vegetation, buildings, coastline exposure and location.
5 km
Selected planning area
Planning area —
Possible local heat index—°C500 m surroundings
Select a planning area to view its local environment estimate.
Green cover proxy—
Building footprint—
Building density—
Nearest coast—
Spatial downscaling under a fixed network background heat condition. Research estimate only—not an observation, next-day forecast or official alert.
Planning-area Heat Risk Score
See where protection is needed first.
ML-assisted Heat Risk Score by planning area
2026 elderly density + heat + nearby eldercare access. Zoom in for subzones and HDB blocks.
5 km
Selected planning area
Area —
—/ 100
Heat Risk Score
Heat Risk Score
- Heat
- —
- Residents 65+
- —
- Eldercare access
- —
- HDB buildings
- —
Heat signal—
Elderly vulnerability—
Cooling-access gap—
Automated classification
Tiered alert recommendation
LowRoutine monitoring
MediumPrepare local support
HighTargeted outreach
Recommended response
Preparing community guidance…
Planning guidance with human review built into the response workflow.
Priority queue
Highest-risk populated areas
— residents aged 65+ mapped from official June 2026 planning-area statistics · risk score comes from the validated planning-area classifier.
PriorityPlanning area65+ residentsLocal heatRisk
Open planning-risk model and cooling scenarios · policy-calibrated random forest · Low / Medium / High score cut-offs 29 and 76 · model weights and inference code are available in the project repository.
Cooling infrastructure simulation
Test interventions before the heat arrives.
Highest-risk zones and simulated cooling impact
Explore real parks, cycling paths and park connectors, then test targeted additions.
5 km
Scenario controls
Intervene in —
Counterfactual result
Estimated planning impact
Before——
→
After——
Risk-score change—
Heat proxy change—
Cooling points—
Existing green score—
Mapped park cover—
Cycling + PCN—
Select a planning area to test a cooling scenario.
Start where need is highest
Priority zones
Choose a row to move the scenario to that area.
Scenario effects are planning assumptions, not measured causal effects. Existing parks and active-mobility networks come from mapped infrastructure; added shade has a smaller marginal effect where the green baseline is already strong.
End-user data collection
Help make your block cooler.
Drop a pin, answer a few structured questions and add a photo. Each contribution enters a council-reviewed queue before it can affect the public map.
01Web formPin · dropdowns · photo
02On-site QRScan · one tap · no sign-up
03AAC-assistedStaff-supported tablet entry
Location
Pin the place
Click the map to drop a pin
Latitude —Longitude —
Summary mode
Review your contribution
Check the structured record below before it enters the council-review queue.
Structured answers
Contribution received
Thank you for helping your neighbourhood adapt.
Your structured report has been added to the review queue.
Estimated local valueReview pendingReceipt —
Model & evidence
From local surroundings to possible heat.
The local-environment model downscales a network heat condition using nearby greenery, HDB building form, coastline exposure and geographic position. It estimates local variation; it does not predict tomorrow's weather.
Every fold excludes one complete location
2,210 hourly station observations
Validation grouped by weather station
Spatial support is shown and guarded at inference
500m
default local radius
11
training stations
2,210
hourly examples
3
published radii
Peer-station background → local spatial features → regularised residual model → spatial-support guard → calibrated heat-index interval
Weather background
The same-hour station network supplies the background heat condition, so the model learns local differences rather than pretending weather is static.
Local environment
Parks, HDB footprints and height, building density, coastline proximity, latitude, longitude and CBD distance describe each neighbourhood.
Guarded inference
Predictions far from known station profiles are conservatively pulled toward the network background instead of extrapolating without bound.
Completed B1 implementation
Six integrated capabilities, connected end to end.
The backend combines current station heat with mapped local environment features, then refreshes local estimates and planning-area views with explicit uncertainty.
Live station heat index
Latest temperature and relative-humidity readings are paired into a current heat index across the station network.
Local environment heat estimate
A separate model estimates possible heat from nearby vegetation, buildings, coastline exposure and location at selectable radii.
Planning-area Heat Risk Score
A separately trained classifier combines heat conditions, official 2026 age structure and mapped eldercare access across 55 planning areas.
Tiered recommendations
Low, Medium and High classes drive a ranked review queue and response prompts.
Weather + spatial inputs
Public station observations provide the background while mapped environment layers describe local conditions.
Cooling intervention simulator
High-risk areas can be tested with added cooling points and a green-corridor shade proxy.
01NEA + demographic ingestion
02Delta quality + feature tables
03MLflow local heat model
04Planning-area risk scoring
05Human-reviewed action queue
Responsible use & data lineage
Evidence, provenance and human review built in.
Research local estimate
The output combines public station and spatial data into a possible local heat estimate while official public guidance remains with Singapore's responsible agencies.
Demographics are contextual
Planning-area and subzone age structure uses official 2026 statistics. HDB-building population values are modelled estimates and are labelled accordingly.
Mapped support access
MOH eldercare points provide a consistent indicator of nearby community support for planning review.
Human review stays in the loop
Risk tiers prioritise inspection and support decisions alongside official public guidance.
DAISI B1 participant guide
URA planning areas
SingStat resident population · June 2026
MOH eldercare services
Models · local-environment heat estimator + policy-calibrated planning-area risk classifier