Skip to main content

Automatically Find New Buildings: A Pilot Project Using Sentinel-1

Point of contact
Anja Sundal
Norwegian Space Agency
Drammensveien 165
0277 Oslo
Phone: +4740885380


The aim of this action was to develop a service to find new buildings and detect building changes automatically, using Copernicus Sentinel-1 data. The now completed action (April 2022 – March 2023) was coordinated by the Norwegian Space Agency (NOSA) with the Norwegian Mapping Authority (NMA) as a partner and NORCE as a sub-contractor to the NMA.

Automated building-change map products would not only be cost beneficial for the NMA but could possibly ease the task of building construction mapping and administration for municipalities in Norway. The approach of using Sentinel-1 data for automatic change detection in extent and shape of building polygons was innovative, and a pilot project was needed to clarify whether a fully operational service could be established. Unfortunately, the ground truth data set applied for testing proved to be of insufficient quality for verifying the methods, and the resulting detectability was significantly lower than target. However, the algorithms tested and developed in the project using averaged Sentinel-1 backscatter data and study the use of machine learning, provide a valuable knowledge base and may be developed further by incorporating higher resolution, more accurate ground truth data.

Two neighborhoods in Kristiansand, a city in the southern part of Norway, were was selected as a pilot site, as there were potential synergies with another AI based mapping project there, using high resolution aerial images and LiDAR. The NMA developed two sets of ground truth datasets over the pilot area, and selected Sentinel-1 data from 2016-2021, utilizing a total of 1672 images. The first ground truth dataset was based on data from the local municipality and contained point data with information about the various buildings in the project area, which were converted to building polygons and rasterized to Sentinel-1 resolution. The initial data set was later modified to remove errors, by manual comparison with aerial images.

NORCE won an open ITT arranged by the NMA, for developing a methodology for change detection based on change detection in measurements of backscatter with different satellite geometries, in geometry-normalized measurements of backscatter, and in measurements of interferometric coherence, as well as combinations of these. Two methods were analyzed:

  • A traditional method based on thresholding changes between two averaged periods (annual averages)
  • A machine learning (ML) algorithm trained on quality assured training data. Because of the low quality of the ground truth data, however, ML proved unsuitable.

A solution for detecting building changes, utilizing data from Sentinel-1 was developed. The change detection algorithms return raster data sets, which are vectorized using another script, from which the output is a GML file (Shapefile or GeoJSON is also possible). The map projection of the output file (EPSG) can be set by the user. The scripts are openly available and run from terminal. The NMA have tested the method in the pilot area, and found detection rates of 18 – 26 % and index changes of 34 – 42 %, which was significantly lower than expected, due to inaccuracies in the ground truth. The method is likely to perform better for more accurate data sets and is highly likely transferable for other change detection problems.  

Outputs and Results:

  • Building detection algorithms using Copernicus Sentinel 1-data have been developed and tested. The tool is described and openly available for further testing here. Possible user groups include municipalities, mapping agencies, industry, SMEs and large private companies, insurance companies.
  • The project was presented at the: 13th Nordic NMA meeting on Aerial Photography & LiDAR Scanning, 29-30 November 2022, Gävle, Sweden. Acquired knowledge and workflows from the project distributed to individuals and organizations who can benefit from the technology, e.g., this webpage (see also Resources)