Inspiration
Singapore is becoming both hotter and older. Existing heat monitoring can tell planners where temperatures are high, but it does not directly answer a more operational question:
When cooling and outreach resources are limited, where should intervention happen first?
HeatShield Planner is designed to bridge that gap by combining environmental exposure, elderly vulnerability, and cooling accessibility into a decision-support system for community planners.
What it does
HeatShield Planner combines live heat signals, elderly population data, and access to cooling facilities into an explainable Heat Risk Score:
Heat Risk = Heat Exposure × Elderly Vulnerability × Cooling Access Gap
Instead of stopping at a heat-risk map, the system adds a resource-constrained intervention layer.
A planner can specify a limit such as:
“We can deploy only 3 additional cooling interventions.”
HeatShield then ranks candidate locations and estimates which interventions would provide the greatest reduction in exposure-weighted vulnerability.
The goal is to move from:
“Where is heat risk highest?”
to:
“Given limited resources, what should we do about it?”
Target users
The proposed users are:
- Town councils
- Active Ageing Centre coordinators
- Community planners
- Public-sector teams responsible for heat resilience and elderly support
Data
The MVP will integrate public Singapore datasets including:
- NEA air temperature data
- NEA rainfall data
- HDB elderly population data
- MSF Senior Activity Centre / Active Ageing Centre locations
Only area-level public data will be used. The system does not perform individual-level profiling.
How we plan to build it
The proposed Databricks pipeline is:
Open data sources → Lakeflow / notebook ingestion → Delta Bronze/Silver → Spark SQL cleaning and geospatial joins → Heat Risk Score → resource-constrained intervention optimiser → MLflow tracking → Unity Catalog → Databricks App / AI-BI dashboard
The Round 1 concept focuses on an end-to-end MVP that can:
- Ingest live environmental data and demographic datasets.
- Generate explainable Low / Medium / High heat-risk tiers.
- Identify the main risk drivers for each area.
- Allow users to set a resource constraint
K. - Simulate before/after cooling coverage.
- Rank interventions by expected improvement in exposure-weighted vulnerability.
Key challenge
The main challenge is that public datasets have different spatial and temporal resolutions.
Rather than presenting model outputs as precise clinical predictions, HeatShield will clearly expose its assumptions and treat intervention effects as planning proxies.
Expected impact
The primary KPI is:
High-risk elderly residents brought within effective cooling coverage per intervention.
Secondary measures include:
- Reduction in exposure-weighted risk
- Percentage of high-risk areas covered
- Cooling coverage gained per unit of intervention resource
What's next
If selected for the build phase, the first priority will be a complete end-to-end Databricks prototype. Stretch goals include next-day heat-risk forecasting and additional intervention scenarios.
Log in or sign up for Devpost to join the conversation.