The Science Behind ASTRA's Map Information

Why we built our heatmap on Risk Terrain Modelling · ASTRA Deep Dives, No. 1

Here’s a question we spent months on before writing a single line of code: what should a safety map actually show you? The obvious idea is to show where crime happens. But the more you dig into that answer, the complicated it becomes. The map we ended up building is based on a method called Risk Terrain Modeling (RTM) from our trusted partner MineCrime, and once you see how it works, you’ll understand every colour on your screen. This is the story of how it works and why we think it’s the best tool for the job.

The problem with most safety maps

Most safety apps show you one of two things. The first is dots: pins marking where incidents were reported. Dot maps feel informative, but when alone they’re past indicators; they tell you where something happened once, not where conditions make anything likely again. A pickpocketing pin from eight months ago tells you almost nothing about tonight. The second is crowdsourced alerts: whatever users happened to report, wherever they happened to be. Those feeds skew towards wherever the most users live, amplify fear more than fact, and go quiet exactly where fewer people are watching. Both approaches share the same blind spot: they map events, when what you actually want to know about is places. The best method for the most accurate data is to combine reported events with actual statistics and data.

A smarter question: what makes a place risky?

In the late 2000s, criminologists Joel Caplan and Leslie Kennedy at Rutgers University flipped the question. Decades of research had already shown that street crime isn’t randomly scattered: it clusters around specific environmental features. Poorly lit transit stops. Dense clusters of late-night venues. Vacant properties. These features don’t cause crime, but they create the conditions where it becomes more likely, the way dry brush and wind create the conditions for fire. Caplan and Kennedy’s insight was that if you could map the influence of every one of those features, you could see risk before it turns into incidents. They called the method Risk Terrain Modeling — RTM — and published it openly for anyone to use and test.

How a risk terrain is built

Picture Milan divided into thousands of small cells, like a fine mesh laid over the city. Now take one research-backed risk factor; let’s use late-night venues. Science shows their influence isn’t confined to their front doors: in one landmark study in Irvington, New Jersey, only one of sixty shootings in a year happened at a bar, club or liquor store, but thirty-one happened within a block of one. So the venue becomes a layer on the map, a glow of influence radiating about a block outward. Every factor gets its own layer like this, each shaped by what the research says: some matter by their presence, some by how densely they cluster, some by how close you are to one. Stack all the layers, and each cell of the city sums up everything influencing it. That stacked map is the risk terrain: a picture of where conditions converge.

And it works

This is what won us over: RTM doesn’t ask to be believed, it asks to be tested. In the Irvington study, researchers built a risk terrain from just four evidence-backed factors, then compared it against where shootings actually occurred. More than 42% of a year’s shootings fell inside the 10% of the city the model rated highest risk. Each additional risk factor converging on a place more than doubled the odds of an incident there. Results like these are why RTM has since been applied by police departments and researchers across crime types from burglary to robbery; and most of all why it beat every alternative we considered. Dot maps describe the past. Crowdsourcing describes attention. RTM describes conditions. And conditions are the thing that stays true when you walk out your door tonight.

Reading the colours on your screen

ASTRA translates this science into the heatmap you see. Milan is divided into thousands of small zones, each scored by RTM-based analysis of environmental and incident data, then coloured by band: the warmer the colour, the more risk-relevant conditions converge there. One tip for reading it well: think of the ASTRA map as a weather forecast, not prophecy. A warm zone means keep your awareness up, the way clouds mean carry an umbrella; a cool zone means conditions are calm, not that anything is guaranteed. The map reflects the character of places, never a judgment about the people in them, and it updates as the underlying model does rather than reacting to every headline.

Go deeper if you like

Everything above comes from published, peer-reviewed work — one of our favourite things about RTM is that its paper trail is completely public. The RTM Manual and the Risk Terrain Modelling Compendium (Caplan & Kennedy, 2011, Rutgers Center on Public Security) are free to download at riskterrainmodeling.com. We’d love it if you read them! the best users we can imagine are the ones who check our homework.