Atmospheric Correction

In 2016 has been launched a scientific project supported by ESA in order to study the atmospheric correction for Sentinel-2 above coastal and inland waters. When successful, the overall idea is to made this algorithm available to the coastal community who is showing an increasing interest for Sentinel-2 through public tool (e.g. SNAP) and later on to allow systematic processing of L2A products dedicated to coastal zone in complement to land products. A consortium has been selected including ACRI, LOG, Informus and C-S. Each of the partners is bringing a particular expertise to each module of the whole processing chain: gaseous absorption correction, aerosols scattering, cloud/cloud shadow/land/water classifications, detection of topographic effects and, more specific to the water part and its interface to land; sun glint detection and correction, white caps and adjacency effects. After scientific evaluation of possible modules, first validation and a first round of implementation, the project is now entering into validation phase. Rationale for algorithm selections and validation plan and objectives will be presented.

From the Top Of Atmosphere (TOA) observations and in coastal areas, unmixing the high spatial-frequency water signal from the low spatial-frequency atmospheric signal is challenging as multiple sets of solution of vector {ρaer (λ),ρw(λ),Ttot(λ)} are possible for a single set of TOA(λ) observations. Aerosol spatial-scales are between one to tens of kilometres, while in the same time in coastal areas, water signal spatial-scales are between tens of meters to hundreds of meters. These differences are particularly interesting to be exploited for unmixing the water and the atmospheric signal in coastal areas, where the “black pixel hypothesis”= null contribution of the water, is not valid and the shape of the water and the aerosol signals in the infrared may be very similar, leading to inversion errors. The inversion scheme itself of the existing processors such as the OL2 OLCI processor addresses partially the reliability of the estimates. For example, aerosol optical thicknesses and types are estimated regardless the fact that it will result later in the processor in negative water reflectance estimates: the inversion has converged towards a local minimum. The final consequence for the users is an over-flagging of the products in coastal areas and for the spatial agencies an under-optimisation of the level 2. To limit the influence of this bad conditioned system, we use here the Maximum a posteriori (MAP) criterion to add to the minimization cost function a priori onto the variable to be estimated. MeetC2 is a Level 2 prototype processor for the Sentinels 2 based onto both a spatial-downscaling analysis and a probabilistic inversion scheme. The TOA signal is estimated using LUTs (interpolated on the fly) generated with the radiative transfer codes OSOAA and SMART-G. The spatial downscaling approach relies onto the estimation of the parameters using an ensemble of pixels, decreasing for each scale from 6 km to 600m resolution. The spatial continuity of the atmospheric components between scales is here modelled using 4 directional gradients onto the inversed atmospheric variables. The high resolution water signal is finally obtained by the subtraction of the high resolution S2 observations with the low resolution atmospheric signal. The validation results include 64 matchups on 6 Aeronet-OC sites and MOBY. We characterise also in this presentation the added-value of the spatial-regularisation terms in the inversion. The V1 of the MeetC2 processor, written in python, is parallelised and is able to process one S2 image at 60m resolution in 2-3 minutes (resp. 10m resolution in 15 minutes using 16 CPUs). Sub-region extraction is also possible to enhance the computation time. Figure 1: Functional scheme for the MeetC2 algorithm based onto i/ the Maximum a Posteriori (MAP) criterion for convergence, as much as possible, towards non-negative and realistic ρw and ii/ spatial gradient constraints onto the atmospheric parameter estimates. Sentinel 2 Image over the Laguna of Venice, 20160827, AAOT Aeronet-OC site.

The iCOR image correction, previously known as OPERA (Sterckx et al., 2015), is a scene and sensor generic atmospheric correction algorithm which can process images containing land and water pixels (coastal, transitional or inland waters). ICOR has the advantage that elevation, air-water interfaces and adjacency effects are considered and no assumptions are made on the turbidity of the water. Through the use of a single atmospheric correction implementation, discontinuities in the reflectance between land and the highly dynamic water areas are reduced. iCOR contains an image-based aerosol optical thickness retrieval module, based on the method developed by Guanter et al. (2007), and allows to perform a simplified adjacency correction over land using fixed ranges and the SIMilarity Environmental Correction (SIMEC) over water (Sterckx et al., 2014). This study illustrates the performance of iCOR for Sentinel-2 over coastal, transitional and inland waters by comparing the image retrieved water-leaving reflectanceswith in-situ optical data collected during field campaigns in European lakes and AERONET-OC measurements. Specific attention is given to the impact of applying an adjacency correction for inland waters . Keywords: iCOR; OPERA; adjacency effects; SIMEC; inland, coastal and transitional waters; AERONET-OC; in-situ optical data; water-leaving reflectance References Sterckx, S., Knaeps, E., Adriaensen, S., Reusen, I., Keukelaere, L. De, Hunter, P., 2015. Opera : an Atmospheric Correction for Land and Water. Proc. Sentin. Sci. Work. 3–6. Sterckx, S., Knaeps, S., Kratzer, S., Ruddick, K., 2014. SIMilarity Environment Correction (SIMEC) applied to MERIS data over inland and coastal waters. Remote Sens. Environ. doi:10.1016/j.rse.2014.06.017 Guanter, L., Del Carmen González‐Sanpedro, M., Moreno, J., 2007. A method for the atmospheric correction of ENVISAT/MERIS data over land targets. Int. J. Remote Sens. 28, 709–728. doi:10.1080/01431160600815525

With the advent of freely available data from the Landsat 8 (2013-...), Sentinel-2A (2015-...) and soon Sentinel-2B (2017-...) missions, uptake of high resolution water remote sensing has drastically increased in the coastal and inland water (hereafter "aquatic") community. No standard atmospheric correction and processor for aquatic applications was provided for any of these missions, so several algorithms were developed by different teams. In this presentation the ACOLITE atmospheric correction is discussed. ACOLITE is freely available and can process data from a number of sensors, including Landsat 5/7/8 and Sentinel-2. ACOLITE performs a two-step atmospheric correction: first, the Rayleigh reflectance is computed based on sun-sensor geometry and second, the aerosol reflectance over water is estimated using the SWIR bands and simply extrapolated to the visible and near-infrared bands in order to derive water-leaving reflectance. The performance of ACOLITE is evaluated using autonomous AERONET-OC measurements in Belgian coastal waters. Same-day imagery from L8 and S2 is compared in terms of water-leaving reflectance for the common band set. The impact of noise from the SWIR bands used in this approach is discussed. Processing challenges for inland waters are also highlighted, with proposed updates to the atmospheric correction algorithm. Novel aquatic applications are highlighted, such as monitoring of local sediment transport phenomena (turbid wakes) and events (dredging) using L8 and S2, and the detection of intense algal blooms with S2/MSI.

The Polymer algorithm is a robust and generic atmospheric correction method, designed to recover the ocean colour in presence of sun glint. It has been applied to multiple sensors, and is used within the Ocean Colour Climate Change Initiative project (OC-CCI). Initially developed for the open ocean, it has been extended to process case 2 waters, with the inherent advantage of not requiring negligible water reflectance in near infrared bands. The sun glint is a critical aspect of the Sentinel-2 MSI water observations: since the revisit of both S2A and S2B is done in the same viewing conditions, many areas are continuously affected by sun glint over a season, leading to long periods of no coverage if the sun glint is not recovered. The results of Sentinel-2 MSI processed with Polymer will be presented and evaluated.