Earth Observation

What makes nature wild?
We use freely available satellite data and explainable AI to map wilderness areas on earth. The objective of this work is to develop an automatic procedure for assessing the near-natural or wild character of each land area of the earth. Apart from that, however, additional information is gained on what exactly makes nature wild.

Reconstruction of urban topography from individual radar images
If you want to derive 3D data from images, you usually need two or more images – as with three-dimensional vision with our eyes. We have developed an AI-based method that allows the use of a single radar image as a basis: images can be generated from radar signals (SAR: synthetic aperture radar), which look like black and white photographs but offer the advantage that they can also be taken from the satellite at night or in the event of cloud cover. This offers exciting possibilities, for example in situations where images are needed at short notice under bad weather conditions. For example, the height of buildings can also be represented by the 3D reconstruction of urban topography from individual radar images.

AI-based derivation of vegetation parameters from radar images
While known optical images capture the chemical properties of the surface depicted and thus can easily be used to draw conclusions about materials or plant health etc., radar images mainly capture physical characteristics, namely roughness and humidity of surfaces. We develop AI procedures that make it possible to also use radar images to draw conclusions on the health of vegetation (e.g. crops, forestry). 

Algorithms for the geolocation of thermal satellite imagery
For the detection of forest fires, satellites with thermal cameras are used to record the temperature of the earth's surface. Since the images of these cameras are relatively poorly resolved and their location on the earth's surface is not given with sufficient precision, the location of a detected fire cannot be determined precisely. We are developing procedures that allow the most precise localization of such images and thus of the fires identified in them.