Published on 27 March 2025
Because optical sensors onboard earth observation satellites are passive devices (https://eo.belspo.be/en/ii22-main-types-instruments) they are only able to capture the light (re)emitted by Earth. That means also that this signal is affected by the atmosphere state and composition with as consequence a difference in the radiation emitted by the Earth and the one captured by the satellite.. To make the satellite images as close as possible to the ground truth they must be corrected and this is where the processing tool developed by the RADCOR project is a game changer.
Improving Satellite Data for Clearer Earth Observations
When a satellite images the earth, it is orbiting above the atmosphere at an altitude of several hundred km! These satellite images hence represent a combination of the surface and atmosphere signals. The atmosphere modifies the surface signal by scattering or absorbing part of the reflected light, and to get accurate information, scientists must correct for this atmospheric interference. In the simplified case, only the effects of light absorption (i.e. atmospheric gases are absorbing part of the light) and scattering amount (i.e. particles and atmospheric gases change the direction of light) need to be corrected for — this is the case when observing large, uniform areas. However, in regions with high contrast (e.g. a dark lake next to bright land, or vegetation next to desert sand or ice), the atmospheric blurring between the different surface types also needs to be taken into account. The blurring causes brighter areas to appear darker, and vice versa. This effect results in a reduction of image contrast, a loss of sharp edges, mixing of optical signals, and makes small objects appear less distinct.
To address this issue, the STEREO RAdCor project has developed an advanced image processing algorithm to address this atmospheric blurring effect, which is now freely available in the open-source ACOLITE software. A more detailed description of the processor is available in the press release.
RAdCor for small heterogeneous targets
For surfaces that are homogeneous over large spatial scales, only the magnitude of atmospheric absorption and scattering needs to be considered to correct satellite imagery. Over areas with high surface contrast on the other hand, the blurring effect of the atmosphere also needs to be corrected. The smaller the target, the higher the contrast with its surroundings, and the more turbid the atmosphere, the higher the impact of this blurring will be.
The new processor developed in the RAdCor project tackles this issue with computationally efficient algorithms and is made available in the free and open source software ACOLITE (from version 20250114.0 onward). A more detailed description of the processor is available in the press release.
Figure 1. An overview of the RAdCor processing approach
RAdCor for water applications
Accounting for the atmospheric blurring effect is essential for satellite remote sensing of inland and coastal waters, which are typically much darker than the surrounding land cover in the near-infrared (NIR, wavelengths around 700-1000 nm), but can also be darker in the visible wavelength range (wavelengths between 400-700 nm), for example in lakes with a high concentrations of humic substances, or lakes surrounded by ice and snow (More info about spectral bands can be found here: https://eo.belspo.be/en/remote-sensing-images). The visible-NIR range is the most useful for water quality remote sensing, but is also a range with significant atmospheric scattering. One extreme example is a clear marine water target surrounded by ice, as demonstrated in Figure 2.
Figure 2: In this image of water (red arrows) surrounded by sea ice (black arrows), the top row shows the RGB composite of the imagery not corrected for blurring on the left, and corrected for blurring on the right. The bottom row shows the difference (left) and ratio (right) of the different corrections for band B02 of Sentinel-2.
The atmospheric blurring effect can be an issue for water quality parameter retrieval for small or narrow inland waters. Monitoring of these waters is required for e.g. the Water Framework Directive, and as in situ sampling efforts are limited, the monitoring is frequently supplemented with remote sensing data. For example, to determine the chlorophyll a concentration, a proxy for phytoplankton biomass in the water, a band combination is typically used that is impacted by land and vegetation adjacency. If the adjacency effect is not corrected for, this will lead in general to an overestimation of the chlorophyll a concentration (Figure 3). The adjacency pattern may follow the seasonality of the land cover and land use, and may not represent the actual state of the water system.
A more reliable band ratio, and hence chlorophyll a concentration estimate for small inland waters can be retrieved using RAdCor processing.
Figure 3: This figure shows the RedEdge/Red waveband ratio used in retrieval algorithms for a river and two lakes in Belgium. The Standard Processing provides too high a ratio for the lakes in the centre of the scene, whereas RAdCor gives lower and more realistic values, as it corrects for the adjacency effect.
RAdCor for land applications
The compensation for blurring is also useful for land applications, as different land cover types can be either darker or brighter than their surroundings (e.g. dark pivot farms in desert areas. Figure 4). For more information about the analysis on Figure 4 and the application for land, see here.
Figure 4: RGB composites and NDVI (Normalised Difference Vegetation Index, a measure of vegetation health) maps of centre pivot farms in Egypt as observed with Sentinel-2. The image on the left has not been corrected for the atmospheric blurring, and the image on the right has been corrected.
Project Team
RBINS: Quinten VANHELLEMONT
Ghent University: Alexandre CASTAGNA - Koen SABBE