Mapping and map revision

We specialise in the research and development of algorithms specifically designed for mapping and map revision. Drawing on data collected from satellites, aircraft and drones, we are able to develop tailored methods that are applicable across various domains, including infrastructure, climate and even archaeology.

The figure shows an illustration of the entire African continent with areas marked in yellow and shades of green depending on vegetation in the region.
We use deep learning to map vegetation height across the African continent based on Sentinel-2 data. Illustration: NR.

A map is a visual representation that highlights relationships among spatial elements like objects, regions, terrains and themes. Maps simplify information about the world and play a central role in conveying data at different levels, from overarching information at national level, all the way down to structures, roads and other features at local level.

Maps are crucial to a country’s security, socioeconomic development and environmental sustainability. National maps include topographic maps, cadastral maps and various thematic maps like landcover, population density and climate. Urban landscapes are constantly evolving, and frequent map revision is needed for effective urban management. Updated maps are used to inform decisions on urban planning, building infrastructure, managing emergencies and overseeing real estate. They are also used to monitor green spaces and address environmental issues in densely populated areas.

In Norway, mapping and map revision is managed collaboratively between the Norwegian Mapping Authority (Kartverket) and local municipalities, with active participation from various public and private entities, including NR.

Extensive experience in mapping methodologies

We have been developing mapping methodologies for more than 40 years, and began our work with thematic mapping using Landsat and SPOT satellites in the 1980s. In the 1990s, we started using digitalised aerial images to achieve high-resolution forest mapping, and we used data from the European Space Agency (ESA) to test airborne hyperspectral sensors, also known as imaging spectrometers, and synthetic aperture radars (SAR). Today, digital aerial imagery is standard, while imaging spectrometers are used on a case-by-case basis.

For large-scale mapping, different satellite sensors are available, from very-high resolution sensors on commercial satellites, to high-resolution sensors like MSI aboard Sentinel-2, and sensors with moderate resolution based on Sentinel-3 and similar satellites, which are used for daily, global mapping.

A topographic map of an area on the Norwegian coast. Land is marked in green and yellow while the sea is dark blue.
An updated map based on Sentinel-2 image data shows a suggestion for map updates in an area of wilderness in Norway where wind turbines are being developed. Illustration: NR.

Automated methods for mapping and map revision

In several of our projects, NR is developing automated methods for mapping and map revision, including mapping vegetation height across the entire African continent, identifying new forest roads, detecting cultural heritage sites, classifying tree species, tracking building changes, and estimating birch pollen release.

The type of remote sensing data selected is determined by four factors:

  • The size or scale of objects or phenomena, which dictates what resolution or grid cell size is required
  • The area size
  • The visibility of the objects or phenomena in the image
  • The budget for data acquisition.

Typically, a compromise is made between these factors. For instance, when dealing with a vast area, using one image may result in a larger grid cell size. On the other hand, retrieving multiple images can be necessary in order to cover the area in sufficient detail.

Three images positioned horizontally showing the same area with different airborne sensors, ultimately revealing previously undetected grave mounds,
Three images positioned horizontally showing the same area with different airborne sensors, ultimately revealing previously undetected grave mounds, Cultural heritage mapping using airborne laser scanner data. Left: an aerial photo of a forested area in Larvik municipality. Middle: laser scanning data of the same area, including forest canopies and other vegetation. Right: laser scanning data with vegetation removed wherein several grave mounds are made visible. These grave mounds were discovered by NR in collaboration with local archaeologists using deep learning. Image: NR.

Automated mapping with deep learning

Automated mapping methods rely on artificial intelligence. This can be achieved by training deep neural networks with large amounts of labelled training data, or by utilising conventional methods if a well-established approach already exists. This can for example apply to assessments in land cover and land surface temperature.

Selected projects