Reports
Monitoring
Comparsion and solution

HeatWise AI

Inspiration

Urban heat is not distributed equally.

In the same city, one neighbourhood may have trees, open spaces and better ventilation, while another is surrounded by concrete roofs, asphalt roads and dense buildings. Yet most heat dashboards only show a red hotspot on a map.

That leaves municipal authorities with unanswered questions:

Why is this area hot? What can be done there? How will we know if the solution worked?

We built HeatWise AI to answer those questions—not as another heat map, but as a complete decision-support system.

What It Does

A user selects a city and hotspot. HeatWise AI first estimates its land-surface temperature and classifies its heat risk.

It then explains the likely drivers behind that result, such as low vegetation, high impervious cover, surface albedo, dense construction or atmospheric conditions.

Once the problem is understood, users can compare cooling strategies such as:

Tree-cover expansion
Cool-roof coatings
Shade infrastructure
Reflective or permeable pavement

The platform compares the selected strategy with a no-action scenario, showing the modelled temperature and risk difference.

HeatWise AI also examines nearby OpenStreetMap assets—including buildings, roads, parks and water features—to provide local implementation context.

Finally, it generates a municipal action brief and supports before-and-after monitoring so that simulated impact is kept separate from real measured results.

How We Built It

We use two XGBoost models:

A regressor estimates land-surface temperature.
A classifier calculates hotspot probability and risk.

The models use environmental, seasonal and urban-form features such as vegetation, canopy cover, impervious surfaces, albedo, temperature, humidity, wind and building characteristics.

Our reasoning is guided by the urban surface-energy balance:

$$ R_n = H + LE + G $$

This connects the model to real heat processes. Vegetation supports cooling through evapotranspiration, reflective surfaces reduce absorbed radiation, and concrete stores heat.

We built the application using Next.js, React, TypeScript, MapTiler, OpenStreetMap and Tailwind CSS, with a contextual Groq-powered AI assistant.

Challenges We Faced

Our first version focused too much on model predictions. We realised that telling a municipal officer a location is hot does not tell them what to do next.

We redesigned the platform around a complete journey:

Detect → Explain → Compare → Implement → Monitor

Another challenge was avoiding misleading claims. A simulated temperature reduction is not a real measured outcome. We therefore clearly separate observations, model predictions, counterfactual scenarios and post-implementation measurements.

We also had to make technical model features understandable. Instead of showing only raw parameters, we translate them into clear heat drivers while preserving the underlying evidence.

What We Learned

We learned that an effective AI product needs more than an accurate model. It needs transparent assumptions, understandable explanations, safe limitations and a way to measure real outcomes.

Sometimes the most responsible answer an AI system can give is:

This is a model estimate and requires local verification.

What Makes It Different

HeatWise AI does not stop at identifying hotspots. It helps users understand why they exist, compare possible responses, inspect local feasibility and monitor what happens after implementation.

It turns urban heat data into a practical and accountable decision workflow.

What’s Next

We plan to expand satellite-data ingestion, validate the models across more cities and seasons, add uncertainty estimates, integrate municipal GIS systems and support long-term intervention monitoring.