Nearmap converting decades of imagery into AI-powered insights

Aerial imagery and AI company Nearmap is seeking to create a 'single source of truth' about the changing relationship between the built and natural environments, using decades of aerial data and custom AI models to track everything from damaged roofs to urban tree cover.

The company's ambition is to bring together groups that often rely on different datasets and perspectives.

Its aerial photography and AI technology provides multiple use cases across property insurance, policymaking, planning, news reporting and real estate data tracking.

High-resolution, detailed imagery gives these groups insight into the changing environment at both a local and national level over months, years and even decades, said Senior VP AI & Computer Vision at Nearmap, Michael Bewley.

"That's what we're building, and it's what we've been building in some ways for about 20 years," he said.

The company began as an aerial imagery business and has developed custom camera systems designed to capture highly detailed imagery from aircraft flying at altitudes of between 10,000 and 20,000 feet.

While drones can be useful for capturing imagery of individual properties, they are not practical for repeatedly surveying entire cities, Bewley explained. Aircraft equipped with specialised camera systems allow Neamap to capture large areas at a scale and frequency that would be difficult to achieve with drones.

He argues that aerial imagery can provide greater detail than satellite imagery for certain applications.

Instead of simply identifying whether a building exists, its technology is capable of identifying highly specific features and conditions, such as the materials used on a roof, damage requiring repair or deteriorating equipment such as air-conditioning systems.

"It's all about this built environment and how the natural environment interacts with it," Bewley said.

It operates regular capture programs across four countries, collecting imagery between one and six times a year, depending on the location and use case. Nearmap also deploys cameras after major natural disasters, seeking to capture affected areas within days of an event.

Australia has been a key part of the company's operations, with its imagery currently covering approximately 90 per cent of the nation's population. Its coverage also extends to New Zealand, the US and Canada.

The sheer scale of its archive has become a significant asset for the business. It recently surpassed 100 petabytes of imagery, creating a historical record that can be used to analyse how properties, cities and landscapes have changed over time.

AI adoption significantly expands functionality 

About a decade ago, the company began investing heavily in artificial intelligence, developing custom-built models capable of recognising specific objects, conditions and states within aerial imagery.

Those models are designed for particular industry applications, allowing Nearmap to expand beyond collecting images to extracting information from them.

"We're basically turning pixels into meaning and then assigning attributes to those pixels," Bewley said.

The technology can also be used to generate risk assessments, including estimates of what damage will be sustained by properties should a hurricane or bushfire hit.

Repeated imagery collection allows changes to be monitored over time. In construction, for example, capturing an area up to six times per year can provide a regular visual record of how a major new skyscraper, sports stadium or data centre is progressing.

The same approach can be applied after natural disasters to track damage and recovery across a town or city in the weeks and months following the event.

Bewley cited a major hailstorm in which imagery and AI models were used to identify properties that had received temporary repairs. The system was able to detect coverings placed over damaged roofs and map affected properties shortly after the event.

By continuing to capture the same areas in subsequent months, Nearmap tracks whether temporary repairs remain in place and track how communities are recovering.

Implications for environmental analysis

A national tree study conducted in Australia drew interest from scientists because of the consistency and scale of its urban tree coverage data. Such datasets can provide a detailed view of vegetation in populated areas and enable researchers to examine how tree cover changes over time.

On a broader scale, Nearmap wants to combine its physical observations of the environment with demographic information to create a more comprehensive picture of how Australian communities are changing.

Bewley showed a visualisation described as 'the nation on a page', with each suburb a small dot, mapping tree cover, building cover and population across Australian suburbs using its imagery in concert with government demographic data, showing the intersection of population density and the natural environment.

The goal is not simply to collect more imagery, but to use long-term, regularly updated data to generate new conversations about how cities and communities are developing over time.

Its broader ambition is to provide a common dataset that can support decisions across various industries - from insurers assessing property risk, planners examining urban development and researchers tracking environmental change.