Generative AI on top of AWS-hosted Earth Observation data, helps urban planners plan for liveable cities

One of our earth observation partners, NEO, processes and analyzes ?geospatial data, satellite imagery and aerial photography and offers their customers actionable insights from this data.? Municipalities, for example, use this information to improve a city’s quality of life based on the number, distribution and vicinity of trees: the 3-30-300 rule.

Scaling eyes in the sky

Neo’s customers regularly request new insights, for which data pipelines are then built by NEO engineers. The insights are then accessible via Signaleyes: a specialized application. This required a lot of effort and lead time.

To significantly scale this service, Itility built a self-service user interface supported by Generative AI. Users can select (NEO’s) geodata sources, zoom in on specific areas of interest, and perform an analysis based on a preconfigured library of concepts.?We have built a back end, using Fast API, AWS Cognito, and the Open Street Map service, and connected this back end to several geospatial databases containing images and metadata on buildings, trees, green spaces, and solar panels.

In a matter of minutes, a user can e.g. count the number of trees within a certain radius around buildings, filter for specific tree dimensions, and translate this insight into urban planning.

Some analyses are pre-configured. An example of such an analysis is the ‘3-30-300 rule’. This rule is used for sustainable urban planning. It checks whether there are at least 3 trees within a certain distance from a home, the neighborhood has at least a 30% tree canopy cover, and the nearest high-quality public green space is within 300 meters. Municipalities use the solution to plan for greener and more livable cities and neighborhoods. The user friendly GenAI (chat) interface allows them to analyze geospatial data by interacting in natural language. No more need for assistance by a highly technical domain expert, anyone can turn data into value.