RAdCor is here! adjacency effects atmospheric correction….

RAdCor, a new atmospheric correction module that accounts for adjacency is now available in ACOLITE

Ghent, Belgium, January 21, 2025
Content provided by the RAdCor consortium.

The latest release of the free and open-source ACOLITE software, 20250114.0, includes a new atmospheric correction processor that accounts for atmospheric blurring, i.e. the adjacency effect. The new processor, RAdCor, was developed by Alexandre Castagna and Quinten Vanhellemont in a two year STEREO project funded by the Belgian Science Policy Office (BELSPO), and is described in an upcoming paper. This new processor further expands the capabilities of ACOLITE, which includes several processors for optical and thermal satellite imagery.

*Figure 1 – Overview of the new processor available in ACOLITE.*

RAdCor is a generalisation of the standard ACOLITE processor for optical imagery, and is applicable for heterogeneous surfaces under turbid atmospheres, conditions where atmospheric blurring can become problematic. The standard processor estimates the aerosol properties with the Dark Spectrum Fitting (DSF) method, and calculates the surface reflectance with the measurement equation for the sensor signal over homogeneous surfaces that was first described by Van de Hulst in the 1940s. The DSF generalises the NIR or SWIR dark pixel approach to estimate aerosol properties, and requires no assumption on the spectral signature of surface targets nor the presence of specific sensor wavebands.

Figure 2 – Example application of TSDSF+RAdCor to a MSI/Sentinel-2B image over the Spuikom lagoon in Ostend, Belgium. The RGB colour composite shows a 5 km x 5 km image centred on an in situ reference station (red cross). The spectral plot shows the in situ measurement, the reflectance at the sensor (ρt), and the reflectances at the surface (ρs) estimated with DSF and with TSDSF+RAdCor. More information about this example can be found *here*.

Over heterogeneous surfaces overlaid by a turbid atmosphere, the Van de Hulst equation is not an appropriate descriptor of the observed signal. It cannot account for the blurring caused by (1) the downward diffuse reflection, or the spherical albedo of the atmosphere, and (2) the upward diffuse transmittance of the atmosphere. These processes introduce a signal from surrounding areas to the signal observed over a given location. RAdCor restates the measurement equation for the observed signal over heterogeneous surfaces, providing a mathematically flexible and computationally efficient solution for the surface reflectance, regardless of atmospheric blurring.

Figure 3 – The cumulative distribution function (CDF) of the (nadir) atmospheric diffuse Point Spread Function (PSF) and Spatially-resolved Spherical Albedo (SAF) as a function of radial distance from a pixel position. The burring arises from the fractional contribution from surrounding areas described by these functions.

As a physics-based approach, RAdCor must be provided with the appropriate transmittance and reflectance quantities that ultimately depend on the aerosol properties. Despite the flexibility of DSF, it was developed with the homogeneous surface framework and its core assumptions are not valid under significant atmospheric blurring: (1) the darkest pixel at the top of the atmosphere for a given waveband might not be the darkest pixel at the surface; and (2) a part of the signal over the darkest pixel arises from diffuse transmittance from surrounding areas, not only from the atmospheric path reflectance. We generalised the DSF, as the TOA-surface Dark Spectrum Fitting (TSDSF) method, keeping its sensor-generic properties, but removing these two limitations that occur under atmospheric blurring. This estimation requires no image iteration, resulting in a computational efficient module.

Figure 4 – Example application of TSDSF+RAdCor to a OLI/Landsat 8 image over the Donkmeer lake in Berlare, Belgium. The RGB composite shows a 5 km x 5 km image centred on an in situ reference station (red cross). The spectral plot shows the in situ measurement, the reflectance at the sensor (ρt), and the reflectances at the surface (ρs) estimated with DSF and with TSDSF+RAdCor. Note that the panchromatic waveband (centred at around 592 nm) is included in ACOLITE processing.

RAdCor can be run using aerosol properties estimates from TSDSF, directly derived from the imagery, or from external sources. RAdCor can also be used in optimisation mode, if a reference spectrum for a target in the scene is available. In this case, an image iteration is performed to find the aerosol properties that minimise the difference between the sensor estimate and the reference spectrum. While this approach is less computationally efficient than TSDSF+RAdCor, it provides a robust correction option, with optimal aerosol properties.

Figure 5 – Example application of TSDSF+RAdCor to MSI/Sentinel-2 imagery over an area with banana plantations in Costa Rica. Both images show the ratio of the surface reflectances retrieved in the homogeneous (hom) and heterogeneous (het) surface framework. The ratios indicate the over or underestimation when using the homogenous framework. The left image shows the ratio for the red waveband (665 nm) and the right image shows the ratio for a NIR waveband (864 nm). More information about this example can be found *here*.

The current version supports multispectral nadir-viewing sensors from the Landsat, Sentinel-2, and PlanetScope constellations. Check out the ACOLITE user manual for installation and use instructions, and the ACOLITE user forum for support, questions and feedback. The RAdCor website contains further demonstrations and analysis in blog posts.