Ocean Colour Standard Atmospheric Correction module

The Clear Water Atmospheric Correction (CWAC) is a key component of Ocean Colour data processors and has been used operationally by agencies, for more than forty years, to provide Ocean Colour Radiometry (i.e. marine reflectance at sea level) and downstream water bio-optical products. In essence, the role of CWAC is to detect the amount and type of aerosols, variable in space and time, with minimum assumptions on the marine optical properties. With this aim, CWAC relies on radiometric measurements in the near infrared (NIR) and possibly short-wave infrared (SWIR), where water absorption is high and the top-of-atmosphere (TOA) signal is assumed to result from the atmosphere only, and not from the water. To deal with geophysical perturbations of the pure atmospheric signal by sea surface, CWAC is preceded by sunglint correction, whitecaps correction and Bright Pixel Correction (BPC), all together forming the so-called Ocean Colour Standard Atmospheric Correction (OC-SAC). OC-SAC differs from alternative atmospheric corrections (OC-AAC), such as spectral matching methods (e.g. EUMETSAT SACSO study) or artificial neural networks whose ocean model over the full spectrum implies further assumptions, and possibly limitations, on the inversion.

The OLCI operational processor CWAC algorithm, up to EUMETSAT Collection 3, dated 2021, comes directly from the MERIS processor developed in the 1990s (Antoine and Morel, 1999). It is known to create spatial noise (uncertainty amplification, aerosol discontinuity), it has limitations in terms of aerosol optical thickness and Angstrom (e.g. Zibordi et al, 2022), air mass or viewing and solar dependence and its outputs need to be flagged data in many situations (e.g. negative reflectance, sensitivity to perturbations, failure for absorbing aerosols). A complete review and upgrade of this algorithm was required for EUMETSAT, in order to deliver Copernicus Ocean Colour products of higher quality in the near future (Collection 4).

Objectives

The objective of this study was to provide a robust and versatile OC-SAC module, based on state-of-the-art aerosol modelling and inversion, giving advanced possibilities for future Copernicus Ocean Colour data processing for Sentinel-3 OLCI and, potentially, other contributing polar and geostationary missions (FCI, 3MI, MetImage). The module has integrated the Bright Pixel Correction (BPC) recently developed for OLCI Collection 3, with refinements in terms of convergence, and, above all, a totally revised CWAC algorithm with new aerosol modelling, detection and new capabilities regarding aerosol layer height (ALH) retrieval. The module has been integrated into both the OLCI operational processor and the SACSO prototype processor.

Overview

Considering the many options of the CWAC components (aerosol modelling, radiative transfer, aerosol detection, uncertainty propagation, etc.), developing a successful CWAC module in the operational processor was a challenging objective.

The overall strategy was:

1. At science level, narrow down the options to approaches successfully demonstrated in the OC community. Scientific investigations were then conducted within this circumscribed frame.
2. At code level, develop a unique OC-SAC module (in C/C++), interfacing with both the existing operational processor and the SACSO prototype processor.

The project relied on five main tasks:

Task 1: Scientific review and requirements consolidation — A detailed review of existing CWAC algorithms, together with small-case prototyping, supported the design of the CWAC module in terms of science (aerosol modelling, numerical strategy for aerosol selection) and implementation.

More specifically, the development of the CWAC component was guided through an objective criterion on aerosol reflectance (accuracy of 2*10^(-4) in the blue bands) based on an uncertainty analysis to reach the OC requirements over a wide range of oceanic and atmospheric conditions. Key features of the new module are:

• State-of-the-art radiative transfer (RT), with two vector RT codes: a full Spherical Shell Atmosphere (SSA) code for the pure Rayleigh contribution (SMART-G, Ramon et al., 2019) and a Plane Parallel Atmosphere (PPA) code for the aerosol-Rayleigh signal (ARTDECO, Dubuisson et al., 2016).
• Full review of the OLCI weakly-absorbing aerosol models: the algorithm relies on the microphysics of Ahmad et al. (2010) based on a mixture of fine and coarse modes, whose size and real refractive index are driven by relative humidity (RH); the storage in Look-up tables (LUT) follows an innovative grid-optimisation procedure to meet the required accuracy.
• Introduction of a new family of OLCI strongly-absorbing models: the approach of Ahmad et al. (2010) is extended with increased spectral refractive indices for fine and coarse modes, keeping size distribution and its RH dependence.
• New Aerosol Layer Height Assessment (ALH) module: ALH is a new dimension of the aerosol LUT because the aerosol vertical profile is a key factor when multiple scattering occurs and, in particular, in the presence of strongly absorbing models. An assessment is done before the AC itself, for all models, thanks to the ratio of the O2 absorbing channel to non-absorbing channel (following Dubuisson et al., 2009).
• Robust aerosol identification: in terms of modelling, the proposed aerosol models present a large range of spectral variation (aerosol epsilon) with a continuous and non-ambiguous transition as a function of fine mode fraction. Numerically, the aerosol optical thickness (AOT), bracketing fine mode fractions and mixing ratio are computed through a spectral matching algorithm using the six OLCI bands in the NIR-SWIR range, following the method already developed for the BPC (Mazeran et al., 2021).
• Detection of strongly absorbing aerosols: the current approach of Nobileau and Antoine (2005) is extended over all water types thanks to a new climatology of remote-sensing reflectance (ESA Ocean Colour CCI, Sathyendranath et al., 2021) and a chi-square metrics using various bands in the visible.

Task 2: OC-SAC algorithm development — This comprised the LUT generation and code implementation. The OC-SAC module was developed and consolidated in a new OC-SAC branch of the SACSO prototype. Support and training were given to EUMETSAT in parallel to transfer and use the same module in the operational processor.

The core module, CWAC, comprises schematically four steps, starting from a family of standard aerosol models (Figure 2):

1. A first guess estimate of AOT at 865nm for each aerosol model, based on the observation at 865nm (green box).
2. An assessment of ALH thanks to OLCI O2-A bands, for each aerosol model (purple box).
3. An aerosol identification, based on the spectral matching algorithm, to retrieve the best fine mode fractions, aerosol mixing and optical thickness that match the observation, followed by a correction for this aerosol in the visible band (red boxes).
4. A detection of strongly absorbing aerosols in the visible bands, based on climatological values of marine reflectance (blue boxes). If such situation occurs, the three previous steps are run one more time in a 2nd pass, with the second family of predefined strongly absorbing models (blue arrows). The difference in step 3 is that bands in the VIS are used, with weighing associated to the climatological uncertainty.

The algorithm stops either after the 1st pass with the standard aerosols, or after the 2nd pass.

Task 3: Product generation — This task first covered the computation of dedicated SVC gains with the EUMETSAT Ocean Colour system vicarious calibration tool (Figure 3) and then the generation of Level-2 products over the Diagnostic Dataset of the SACSO study (see examples on Figure 4), with the OC-SAC-SACSO prototype).

Task 4: Evaluation and validation — This comprised the assessment of the new marine reflectance reflectance in comparison with existing processor (OLCI Collection 3) over individual scenes (Figure 4), global maps, time-series over selected sites and validation and atmospheric by-products against in-situ radiometric data (AERONET-OC, MOBY, Figure 5).

OC-SAC-SACSO prototype results analysed below include a different cloud screening scheme than the existing Collection 3 processor, which may be not as effective. Nevertheless, the main conclusions of this task were that that the new OC-SAC module presents, in comparison to the existing processor:

• significantly reduced residual error and mean absolute deviation, but a slightly higher bias that has to be investigated with more robust statistics and data screening;
• less noise, more homogenous retrievals, less anomalous and flagged data (typically less negative reflectance);
• systematically higher Angstrom and lower AOT, while Collection 3 is known to underestimate Angstrom and overestimate AOT;
• a negative shift with respect to airmass, which could also compensate the known bias of Collection 3.

Task 5: Project outreach — This comprised the promotion of the study to the Sentinel-3 Validation Team, scientific discussions with NASA’s Ocean Biology Processing Group and provision of a detailed roadmap for extended validation and refinements targeting operationalisation of the new OLCI OC-SAC in 2023.