Geospatial AI

Geospatial AI means AI models that read images taken from cars, drones, aircraft and satellites, and turn them into measurements of the ground: roads, buildings, roofs, fields, trees and water. Geospatial data is data about places on the earth. Maps used to be surveyed by people on the ground. Now they are increasingly worked out automatically, and they can update themselves as the ground changes.

Here is how it works in general. A camera or another sensor takes images from above or from the road. Each image is tagged with its exact location, using GPS. An AI model reads the images and finds objects in them, such as a new building, a damaged roof or a field of wheat. It also compares new images with old ones to find what has changed. The results are stored on a digital map, in software called a geographic information system, or GIS. People then act on the changes, or the map simply updates.

Some satellites carry radar instead of a camera. Synthetic aperture radar sends out radio waves and reads their echo, so it can see through cloud and at night, which matters during India's monsoon.

Take, for example, a bank lending to farmers. Crop monitoring reads satellite images of every field to show what is planted and how well it is growing, without anyone visiting the farm.

Consider an electricity company with thousands of kilometres of power lines. Vegetation management software finds the trees growing too close to the lines, before one falls and causes a power cut.

Say a hailstorm damages roofs across a city. Aerial imagery and roof measurement lets an insurer measure each damaged roof from aircraft photographs, before any assessor visits.

Cars use the same idea. Cameras on mapping cars feed the HD maps that tell a car which lane it is in.

How is geospatial AI different from computer vision in general? Computer vision is any AI that reads images, such as a face at a door or a defect on a factory line. Geospatial AI is computer vision in which every pixel has a place on the earth, so its results can be put on a digital map and joined to other data about the same place.