Atmospheric correction algorithm aims at removing the impact of the atmosphere (mainly the aerosols) from the top-of-atmosphere signal. While over open ocean waters this process can use the hypothesis of a total absorbant ocean in the near infra-red, it is not the case for the inland and coastal waters. This means that this task is more difficult. But it is the most critical one as the output of the atmospheric correction is the remote-sensing reflectance which is the most important parameter that allows us to estimate the optical inherent properties of the seawater (absorption and scattering) and biogeochemical parameters critical to evaluate the water quality (chlorophyll-a concentration, suspended particulate matter, harmful algae, ...).

We investigate here the accuracy of several atmospheric correction algorithms for the MSI and OLCI sensors on-board of Sentinel-2 and Sentinel-3 respectively, using radiometric measurements over contrasted coastal areas.

Several atmospheric correction algorithms for MSI and OLCI have been developed and proposed. While some of them are only dedicated to correcting atmospheric effects over land for MSI, others have been specifically designed for ocean activities. For MSI, six algorithms are benchmarked, among them: MAJA (Hagolle et al., 2010, 2015), Sen2Cor (Wilm et al., 2016), Polymer (Steinmetz et al., 2011), NASA (Pahvelan et al., 2017) and ICOR (Sterckx et al., 2014, 2015). The LAC algorithm has been specifically consolidated in the frame of this project and aims at taking advantage of having a scene with a large variability of radiometry as observing both land and water and which blends specificities of these two types of observations

Validation of the atmospheric correction schemes will be presented for selected MSI images of French Guiana and the eastern English Channel for which in situ validation measurements have been collected in 2016 and 2017 using TriOS and ASD FieldSpec 4 radiometers. Further images over European coastal and inland waters are being inspected, to indicate the respective performances of cloud masks, cloud shadow masks and glint algorithms. The use of AERONET-OC data is also an option that will be shown.

Similar work will be presented for the OLCI sensor on-board of Sentinel-3 with the same in-situ dataset for four atmospheric correction algorithm (baseline, C2R (Doerffer et al., 2007) , NASA (Bailey et al., 2010), Polymer).

The availability of Sentinel-2 data boosted spaceborne, multispectral analyses of aquatic ecosystems. Various studies showed that Sentinel-2 enables water constituent derivation and the delineation of submersed aquatic vegetation and water depths in near-shore, shallow water. These parameters can be derived using analytical bio-optical inversion algorithms; their performance relies on both the intensity and the shape of the water bodies’ generally low reflectance signal. Thus, atmospheric correction algorithms have a high importance in aquatic remote sensing.
In the past years, various atmospheric correction algorithms have been developed for Sentinel-2, some of them with a specific emphasise on the characteristics of water surfaces (e.g. strong adjacency effects, sun glint and sky glint). Comparisons among atmospheric correction algorithms demonstrated the variety of resulting surface reflectance spectra by applying different algorithms. Due to the low reflectance signal, small absolute variation among algorithms may result in large relative differences. The objective of this study therefore is to assess the impact of different atmospheric correction approaches on the retrieval of shallow water parameters such as bottom coverage and water depths. To this end, we selected a coastal area ‘Heidkate’ (54°26’ N, 10°18’ E) in the turbid waters of the Baltic Sea/ Northern Germany. Predominating sandy sediment and seagrass (Zostera marina) covered patches characterise the shallow water bottom coverage. Water depths slope gently interrupted by elevated sandbanks. In 2017 and 2018, we conducted in situ measurements comprising water constituent concentration measurements, radiometric measurements of the water body and bottom coverages (e.g. sand, gravel, seagrass, other brown macroalgae, blue mussels), multibeam sonar for obtaining a bathymetric map with high spatial detail. Information on bottom coverage with seagrass or bare sediment originates from diver videos and UAV RGB imagery. We further applied different atmospheric correction algorithms, i.e. Sen2Cor, ATCOR-2, Modular Inversion and Processing System, ACOLITE, iCOR and C2RCC, on cloud-free Sentinel-2 A and B scenes. Comparisons based on homogenous patches of bare ground or densely covered seagrass as well as radiometric ground measurements helped to identify the variability among atmospheric correction algorithms. In a next step, we used the analytical bio-optical modelling software WASI-2D for deriving water depths and bottom coverage. WASI-2D inversely models the reflectance spectrum of atmospherically corrected image data using a set of constant and variable fit parameters (i.e. water depths and bottom coverage). The model parameterisation was the same for each of the differently atmospherically corrected images to ensure comparability. For assessing the influence of atmospheric correction algorithms on bio-optical modelling results, we focused on the retrieved water depths as primary and bottom coverage as secondary parameter. The variation of results underpins the need for a careful selection of atmospheric correction algorithms in aquatic remote sensing.

The Sentinels 2A&B missions offer today opportunities to propose innovative atmospheric correction schemes for waters in coastal areas i.e. enhanced Level 2 products. The estimated water reflectances, i.e. the level 2 products, are then used in the estimation of the water constituents such as the chlorophyll-a (chl-a) and the suspended particulate matter concentrations. These parameters are themselves used in some European countries to assess the quality of the coastal waters in the frame of the Marine Strategy Framework Directive (MSFD). For the MSFD, quality assessment is a function of the deviation from the reference conditions, following the definitions of the Directive. ‘Good status’ means ‘slight’ deviation, ‘moderate status’ means ‘moderate’ deviation and so on. Several criterions are taken into account to define the ecological status of a water body. Amongst them, the eutrophication risk is assessed through the chlorophyll-a concentration, which is a proxy of phytoplankton biomass. The evaluation of the eutrophication risk of a water body is established through the percentile 90, P90, of the chlorophyll-a concentration integrated on its whole area. P90 represents the chlorophyll-a level such that 90% of the observations are equal to or less than this value.
In the first part of the presentation we discuss about the advantages and the limitations of the use of the Sentinels 2 in coastal areas for the MSFD compared with the in-situ data, including the uncertainties on the both estimates. We show that the spatial and temporal coverage of the Sentinel 2 has the potential to offer much more reliable estimates of the eutrophication status than the in-situ data. Nevertheless we show how the atmospheric corrections issues in coastal area influence the P90 estimates, i.e. the evaluation of the eutrophication risk.
In the second part, we compare the chl-a P90 and derived MSFD eutrophication status, using in-situ data and two satellite processors: the Sen2cor and the Meet_C2 algorithms. The Sen2cor algorithm is the official ESA level 2 processor for the Sentinels 2 and the Meet_C2 algorithm is an innovative processor developed by ACRI-ST dedicated for oceans and inland waters. The case study of the MSFD and the eutrophication for coastal waters is addressed using the ESA Coastal Thematic Plateform (C-TEP). The C-TEP platform allows performing on-demand processing on a dedicated cluster, and benefits from a large global archive of Sentinel-2 images on coastal areas. It offers also a web interface to launch the processing and visualize the results on-line, and compare them with in-situ data. We show how the WPS API of the C-TEP can be used to process the Sentinels 2 data and derive the descriptor 5 for the MSFD. We discuss about the proposed improvements brought by our atmospheric correction scheme regarding the MSFD thematic.
MSFD: http://ec.europa.eu/environment/marine/good-environmental-status/index_en.htm
ESA Coastal–TEP: www.coastal-tel.eu/geobrowser
References:
Gohin, F., Saulquin, B., Oger-Jeanneret, H., Lozac'h, L., Lampert, L., Lefebvre, A., ... & Bruchon, F. (2008). Towards a better assessment of the ecological status of coastal waters using satellite-derived chlorophyll-a concentrations. Remote Sensing of Environment, 112(8), 3329-3340.
Saulquin, B., Fablet, R., Bourg L, Mercier, G., and Fanton d'Andon, O. (Remote Sensing of Environment (2015), http://dx.doi.org/10.1016/j.rse.2015.10.035): MEETC2: Ocean Color Atmospheric corrections in coastal complex waters using a Bayesian latent class model and potential for the incoming Sentinel 3-OLCI mission.
Saulquin Bertrand, Martin-Lauzer François-Régis, Bajhat Alhammoud: Level 2 validation of the Sentinels 2 A&B over waters: state of the art and opportunities

Coastal waters are vital resources for aquaculture, fisheries, wind and tidal energy, cooling waters, and shipping and tourism. However, these waters are often the most susceptible to pollution, eutrophication, habitat loss and harmful algal blooms (HABs). Earth observation (EO) provides the ability to monitor environmental change in coastal waters at broad spatial and fine temporal scales as compared to in situ monitoring. However, monitoring coastal waters from EO data is often challenging because of the broad range of optical water types resulting from complex physical and biogeochemical gradients that characterise these environment. The H2020 CoastObs project is developing a range of data products for coastal waters from Sentinel-2 (MSI) and Sentinel-3 (OLCI) data, including seagrass and macroalgae mapping, primary production (PP), phytoplankton size classes (PSCs) and the detection of HABs. The study sites include coastal waters in Italy (Venice Lagoon and the Adriatic), Spain (Ria de Vigo) and the Netherlands (Wadden Sea and Eastern Scheldt).

Here we present the preliminary results from the validation of PP, PSC and HAB algorithms. In the first instance, a range of atmospheric corrections were tested for MSI and OLCI over coastal waters using TriOS and WISP-3 in situ reflectances for validation. The resulting atmospherically corrected data were then used for the development of products. Depth-integrated PP (mg C m-² day-¹) values were modelled from Envisat MERIS and AATSR data using a modified version of the Vertically Generalised Production Model (mVGPM) (Behrenfeld and Falkowski, 1997). Modelled PP was validated with historic ¹⁴C PP measurements from Venice Lagoon and the Adriatic Sea, indicating good agreement with PP modelled from both in situ (RMSE_log < 0.50) and satellite parameters (RMSE_log < 0.45). The resulting mVGPM will be tested with Sentinel-3 OLCI and SLSTR data over the coastal waters of Italy and Spain following upcoming field campaigns in summer 2019. PSCs were retrieved from MSI and OLCI data using the abundance based approach of Brewin et al. (2010) and validated with in situ measurements of fractionated chlorophyll-a (Chl-a). Size classes of 0.2-2 μm (picoplankton), 2-20 μm (nanoplankton) and >20 μm (microplankton) were retrieved. Spectral based methods exploiting variations in the optical properties of pigments to characterise PSCs were also explored. The detection of HABs in coastal waters from OLCI data was based on previous work on the retrieval of Chl-a and HAB species (e.g. Pseudo-nitzschia sp.) from MERIS (Gonzalez et al, 2011; Spyrakos et al., 2011) using fuzzy c-means logic and neural network techniques.

Coastal zones comprise a wide variety of important habitats and ecosystems that attracted population and human activities. It is essential to make use of long term management tools to protect coastal and marine resources and increase the efficiency of their use.

Identification of the most favorable areas for fishing and aquaculture activities, based on modern techniques, is essential for a sustainable development. Nevertheless, this task has become a challenging one due to fish stocks decreasing trend and offshore general migration of the remaining ones. This situation can trigger longer search times and, therefore, increased costs. Earth Observation data (both satellite acquired and in-situ) can be successfully used to raise the efficiency of fishing and aquaculture efforts.
The SkyFISH platform, developed as a CMEMS downstream service, is a dedicated web-based service that can be used by the end-users as a decision support tool for finding the most favorable fishing zones or analyze the suitability of specific areas for aquaculture activities and monitoring of the existing ones. The area of interest covered by the service is the Romanian coastal zone, but can be scaled up to other regions. SkyFISH takes advantage of the wealth of data delivered under the Copernicus program (CMEMS) and other data sources. It makes use of advanced Earth Observation data analysis and web-mapping technologies in order to offer access to a wide variety of physical ocean parameters. Users can retrieve, in a simple and user-friendly way, information regarding both essential water related variables (i.e. temperature, salinity, turbidity etc) and other contextual information (i.e. lighthouses location, bathymetry, meteorological conditions etc). A suitability index for fishing and aquaculture is also being developed. The service directly addresses multiple categories of stakeholders, from individual fishermen, to mid-size and industrial fishing companies.

During Phase 1, an initial SkyFISH prototype service is currently developed under the Space Technology and Advanced Research (STAR) financial scheme of ESA. The second phase of the project is developed under the User uptake activities of CMEMS Development and promotion of demonstration of CMEMS downstream service – DEM4, Lot 9 (Black Sea). The main objective of the second phase is to make the service operational.

The poor state of the Baltic Sea again became apparent during summer 2018 in form of massive and long-lasting cyanobacteria blooms. Warm and sunny weather, combined with good availability of nutrients led to the worst algae situation in a decade. Climate change is expected to cause further warming in this region making these events more and more common in the future.

The decades of dumping untreated waste water into the Baltic Sea and the use of fertilizers in agriculture have resulted in strong internal loading. While the water treatment situation has improved and fertilizers are being used more responsibly, the flux of carbon and nutrients from land to sea is still great and in many areas largely unknown. The Sentinel satellites of the Copernicus programme offer an excellent opportunity for characterizing and monitoring the fluxes and processes occurring in coastal zones. This in turn will lead to improved process understanding.

In addition to frequent cyanobacteria blooms, the high absorption by colored dissolved organic matter (CDOM) causes problems to the utilization of EO for monitoring the state of the Baltic Sea. The available processors for Sentinel-3 (S3) OLCI and Sentinel-2 (S2) MSI often provide overestimated values for Chlorophyll a and underestimate CDOM. The main source of these problems is the failure of the atmospheric correction to provide reasonable marine reflectances.

The overall goal of the ESA funded project Baltic+ SeaLaBio (Sea-Land Biogeochemical linkages) running from 2018 to 2020 is to develop methods for assessing carbon dynamics and eutrophication in the Baltic Sea through integrated use of EO, models, and ground-based data.

We will present the first results on the following topics:

- Improvements in atmospheric correction of satellite data in the complex absorbing waters of the Baltic Sea. Here we combine the methodological approaches of the C2RCC neural network for the in-water modelling (well adapted to represent the natural variability in the Baltic Sea) and of spectral matching algorithms for the atmospheric part (to gain in robustness).
- Improvements in the in-water bio-optical inversion, in particular through a regional training of the neural network.
- Synergistic use of S2 and S3 for Level-2 products in order to improve the coverage in coastal regions. S3 OLCI has a better band combination for water quality estimation than S2 but its spatial resolution limits its use in river estuaries and archipelagos common in the coastal areas of the Baltic Sea. Hence, combining these two data sources can lead to high resolution water quality products without compromising the thematic quality of the data.
- Comparisons of EO, model and in situ data in order to investigate how well carbon dynamics can be inferred from satellite data. In order to facilitate the comparison we modify the biogeochemical model of IOW (Ecological ReGional Ocean Model, ERGOM) by improving its spatial resolution in test areas.