Space | 22. March 2021 | posted by Manfred Gottwald

Using principles from painting – creating 3D-effect satellite images in true colour

Die physische Karte des Nördlinger Ries, abgeleitet aus einem farbkodierten, schräg beleuchteten digitalen Höhenmodell von TanDEM%2dX
Credit: © DLR
A physical map of the Nördlinger Ries, derived from a colour-coded, diagonally illuminated digital elevation model created using data acquired by TanDEM-X

How can an impression of three-dimensionality be created using a two-dimensional medium? In art, this question arose centuries ago. Certain painting techniques have since evolved to simulate effects of light and shadow, creating a 3D effect for the viewer. Such effects are referred to as trompe-l'œil – they 'deceive the eye'.

Earth observation satellites with multispectral sensors provide images in natural colours when their red, green and blue (RGB) channels are combined. However, they tend to appear somewhat 'flat'. To turn them into attractive, three-dimensional representations, they must first be 'transformed' into three dimensions. This can be achieved if we use proven methods from art as a guide in science, but this requires elevation information.##markend##

The TanDEM-X elevation model of Earth is a dense grid that gives a height value for points on Earth's surface that are approximately 12 metres apart. It does not convey a three-dimensional impression. This only happens when the scene is virtually illuminated – that is, when the position of the Sun is defined, and the corresponding light and shadow cast is simulated. If the elevation model is also colour-coded, the physical maps familiar from school atlases can be created, which illustrate the topography of an area.

Topography is important, especially when researching terrestrial impact structures that were created when asteroids collided with Earth. Simple or complex craters on Earth's surface stand out from their surroundings due to their unique morphology. Such structures are clearly evident in maps created using TanDEM-X elevation model data. In contrast, they appear less obvious in the visual representations of data from multispectral sensors such as the Sentinel-2 satellites of the European Copernicus Earth observation missions. Only by introducing an effect like a trompe-l'œil can satellite images appear three-dimensional. To achieve this, the multispectral data must be combined with the TanDEM-X elevation model.

The Sentinel-2 satellites use multispectral imaging instruments to acquire data in their RGB channels with a spatial resolution of 10 metres. This brings them very close to the values produced by TanDEM-X, at 12 metres. The Sentinel-2 images provide the natural colours. The elevation, and thus the light and shadow information, come from TanDEM-X.

Credit: Image processing using Copernicus Sentinel-2 data / ESA

The summery Nördlinger Ries on 6 July 2017 in a Sentinel-2 RGB image (left), characterised by the pattern of agricultural areas, and the same scene again (right), but now as a combination of TanDEM-X elevation model and Sentinel-2 RGB images.

Firstly, the section of the TanDEM-X elevation model to be examined is 'illuminated' with a virtual Sun from the direction (azimuth) that corresponds to the position of the Sun during the RGB data acquisition by Sentinel-2. The virtual height of the Sun (elevation) should be chosen so that the shadows are not too long or too short. A lighting angle of 45 degrees gives reasonably good results. In the grey-scale, diagonally illuminated TanDEM-X scene, shadows appear deep black and the surfaces facing the virtual Sun appear very bright. Water surfaces give rise to high altitude noise and must be masked out.

Overlaying RGB and elevation data – adopting a painter's approach to natural contours and colours

If both representations are now superimposed, the slightly different scales adjusted and a colour correction carried out, one can obtain a representation of the crater with clear light and shadow effects. These give the viewer a natural-looking 3D impression that differs only slightly from the original spatial resolution of Sentinel-2.

Not all Sentinel-2 data are suitable for this technique. The imaged area must have been largely free of clouds during the RGB recording. For large impact structures requiring more than one Sentinel-2 tile, all acquisitions must be from the same day. Especially in regions with vegetation or large bodies of water, a time difference of just a few days can make it extremely difficult to combine the image tiles.

Case study – the Nördlinger Ries as a 'digital trompe-l'œil'

The Nördlinger Ries is located approximately 120 kilometres northwest of Munich. It is the result of an impact that took place about 15 million years ago, when an asteroid measuring 1.5 kilometres across and an accompanying projectile measuring around 150 metres hit the Swabian Jura. The larger projectile produced the Ries Crater, which measures 26 kilometres across, while the smaller one created the Steinheim Basin, 40 kilometres to the southwest.

Credit: Image processing using Copernicus Sentinel-2 data / ESA
The snow-covered Nördlinger Ries on 21 February 2018 in a Sentinel-2 RGB image (left), the obliquely illuminated TanDEM-X elevation model of the same scene, illuminated with a virtual Sun from the same direction as in the Sentinel-2 RGB image (centre), and the two images combined (right).

The Nördlinger Ries example shows that impact structures can be represented as digital trompe-l'œil images using data acquired by TanDEM-X and Sentinel-2. The data used come from two completely different seasons. One shows a completely snow-covered scene in winter, which is something of a rarity, while the other shows the Ries crater in midsummer, when the vegetation is at its thickest. While the crater structure in the Sentinel-2 RGB images can only be distinguished indirectly via patterns in the forest, the merged-data representations convey both the visual, correct-colour impression and the morphology of the crater through the simulated 3D effect.


About the author

Manfred Gottwald was affiliated with DLR since 1991. In September 2018 he retired but still pursues remote sensing work, particularly by using data from the TanDEM-X mission for studying impact craters. to authorpage