Advances in the Ocean Color Component of the Aerosol Robotic Network (AERONET-OC)

a. Measurement and data reduction methods

The values of ρ(θ, φ, θ₀, W) applied in AERONET-OC data processing are those determined by Mobley (1999) at λ = 550 nm through HydroLight simulations (Mobley 1994; Mobley and Sundman 2000) using the Cox and Munk (1954) slope distribution of surface waves. It is recalled that the theoretical sky radiance distribution, neglecting polarization effects but accounting for multiple scattering and aerosol contributions, was constructed benefitting of an irradiance model (Gregg and Carder 1990) and experimental sky radiance patterns (Harrison and Coombes 1988).

AERONET-OC, in agreement with AERONET requirements, relies on CE-318 and the more recent CE-318T radiometers conceived to perform multispectral measurements in the ultraviolet, visible and near-infrared spectral regions with a full-angle field of view of 1.2°. Specifically, CE-318 and CE-318T radiometer systems configured for ocean color applications and called the Sea-Viewing Wide Field-of-View Sensor (SeaWiFS) Photometer Revision for Incident Surface Measurements (SeaPRISM), autonomously acquire on a channel-by-channel basis (i.e., spectrally asynchronously) the following:

1. the direct solar irradiance E(θ₀, φ₀, λ) as a function of θ₀, solar azimuth angle φ₀, and λ to determine the aerosol spectral optical depth τ_a(λ) required to compute t_d(λ);

2. N_T sea-radiance measurements for determining L_T(θ, φ, λ);

3. N_i sky-radiance measurements for determining L_i(θ′, φ, λ).

The sky and sea measurements for determining L_i(θ′, φ, λ) and L_T(θ, φ, λ) are performed with N_i = 3 and N_T = 11, respectively. The larger number of N_T measurements with respect to N_i, is suggested by the need to statistically address the impact of wave perturbations. Measurements are not performed in the presence of clouds affecting E(θ₀, φ₀, λ).

For each measurement sequence, L_i(θ′, φ, λ) is determined by simply averaging the N_i sky-radiance data. Conversely, L_T(θ, φ, λ) is determined from the average of a fixed percent of the N_T sea-radiance measurements exhibiting the lowest radiance levels (i.e., 2 out of 11). This approach has been suggested by studies (Zibordi et al. 2002; Hooker et al. 2002) highlighting the need for aggressive filtering of above-water measurements to minimize the effects of wave perturbations in L_T(θ, φ, λ).

Equivalent to AERONET atmospheric data products (Holben et al. 2001), ocean color data are also archived at three incremental quality levels. Level 1.0 only includes L_WN(λ) data for which (i) L_i(θ′, φ, λ) and L_T(θ, φ, λ) are determined from measurement sequences not exhibiting missing data; (ii) the dark values are below a given threshold; (iii) the value of φ₀ is within site-dependent limits to minimize superstructure perturbations in L_T(θ, φ, λ); (iv) τ_a(λ) has been determined; and (v) W is below the maximum threshold of 15 m s⁻¹.

Level 1.5 includes L_WN(λ) derived from level 1.0 data for which (i) cloud screened AERONET τ_a(λ) exist at level 1.5 in the AERONET database (Smirnov et al. 2000; Giles et al. 2019); (ii) a series of empirical thresholds are satisfied (e.g., L_WN(λ) > −0.01 mW cm⁻² μm⁻¹ sr⁻¹ indicating absence of exceedingly negative values at any λ); L_WN(412) < L_WN(443) for coastal sites; L_WN(1020) < 0.1 mW cm⁻² μm⁻¹ sr⁻¹, except for sites exhibiting very turbid waters, to exclude measurements appreciably affected by the presence of obstacles in the sight of the sea-viewing sensor; and (iii) the N_T sea-radiance measurements and N_i sky-radiance measurements exhibit low variance indicating low wave perturbations and negligible cloud contamination, respectively. Specifically, the test on the variance of the N_T sea-radiance data removes measurement sequences affected by random high radiance values due to glint or foam reflectance. The test on the variance of the N_i sky-radiance data, in combination with the AERONET cloud screening for τ_a(λ), aims at removing those measurement sequences likely perturbed by clouds that may hinder the determination of L_i(θ′, φ, λ) and consequently decrease the accuracy of L_WN(λ). Still, additional perturbations weakening the assumption of clear sky, may result from sparse clouds not affecting the sun disk and the portion of sky in the direction (θ′, φ). Nevertheless, it is expected that uncertainties resulting from those unfavorable measurement conditions are (at least partially) accounted for by the uncertainties assigned to t_d(λ) and to environmental perturbations (see Zibordi et al. 2009b).

Level 2.0 data include L_WN(λ) determined from level 1.5 products for which (i) the level 2.0 AERONET τ_a(λ) exist; (ii) the N_T sea-radiance measurements and N_i sky-radiance measurements satisfy lower variance thresholds with respect to those applied for level 1.5 (details on these thresholds are provided in section 3); (iii) the differences between pre- and postdeployment calibration coefficients for the AERONET-OC radiometer exhibit values smaller than 5% (it is mentioned that generally these differences do not exceed 1% per year); and (iv) L_WN(λ) do not show questionable values during the final spectrum-by-spectrum assessment performed by a practiced analyst. This final step has been often supported by an automatic process (D’Alimonte and Zibordi 2006) aiming at rejecting L_WN(λ) spectra exhibiting (i) low statistical representativeness within the dataset itself (self-consistency test) and (ii) anomalous features with respect to a reference set of quality-checked data (relative-consistency test). The automatic procedure has been effectively applied to data from a variety of sites (see Zibordi et al. 2009b). However, the analysis of results has shown limits in addressing unique measurement conditions not represented in both the reference dataset or the AERONET-OC data going to be screened. Because of this, the supervised spectrum-by-spectrum quality check is still the fundamental step toward level 2.0 data quality.

Statistically, approximately 44% of the level 1.0 L_WN(λ) are raised to level 1.5 and only 28% of the level 1.0 L_WN(λ) are qualified for level 2.0, as determined with data from 2002 until mid-2019 for which L_WN(λ) values have been available at all quality levels for the major sites.

b. Calibration

Calibrations of each AERONET-OC radiometer system comprise independent actions performed at the Goddard Space Flight Center of the National Aeronautics Space Administration (NASA GSFC) for direct solar irradiance and radiance measurements.

Pre- and postdeployment calibrations for direct solar irradiance are obtained indirectly by relying on a reference instrument in turn calibrated through the Langley method (Holben et al. 1998). Pre- and postdeployment calibrations for radiance measurements, which are relevant to above-water radiometry, are performed using an integrating sphere (IOCCG 2019). These derived pre- and postdeployment calibration coefficients are interpolated over time to account for any sensitivity change of the radiometers during field operation. All calibration measurements and retrieved coefficients are permanently stored in the AERONET database.

Fully independent radiance calibrations performed at the Joint Research Centre (JRC) on a number of AERONET-OC radiometers (tentatively ⅓ of those annually deployed), add a further check to the overall quality process. These independent radiance calibrations rely on the use of National Metrology Institutes (NMI) traceable 1000 W FEL lamps and reference 99% reflectance plaques (IOCCG 2019).

Verification of the accuracy of both NASA GSFC and JRC absolute radiance calibrations were performed by the National Institute of Standards and Technology (NIST) using a CE-318 radiometer. Results indicate uncertainties within approximately 1% in the visible spectral region of major interest for ocean color applications (Johnson et al. 2021).

Also, since 2010 the temperature sensitivity was characterized for each instrument at all spectral bands (Giles et al. 2019). This process has mostly led to an increase of the accuracy for τ_a determined from the near-infrared spectral bands (i.e., 870 and 1020 nm). Still, it does not have an appreciable impact on the determination of τ_a and L_WN in the visible spectral region.

c. Uncertainties

Uncertainties are a fundamental aspect of any measurement and are essential for a confident application of data products. In the case of AERONET-OC, a quantification of L_WN(λ) uncertainties was proposed in Zibordi et al. (2009b). These uncertainties were provided in relative terms (i.e., in percent) making explicit reference to the AAOT site representative of a variety of measurement conditions. The analysis embraced uncertainty contributions from absolute radiance calibrations, sensitivity decay during deployments, corrections for bidirectional effects, water surface reflectance, atmospheric diffuse transmittance, environmental perturbations (sky radiance variability and wave effects). Assuming each uncertainty independent from the others, their combined spectral values resulting from the quadrature sum of the various contributions, lead to values approaching 5% in the blue–green spectral regions and 8% in the red.

The former results provide a term of reference for typical relative uncertainties affecting L_WN(λ) AERONET-OC data. Still, these uncertainty values may not equally apply to measurements performed with observation conditions significantly different from those characterizing the AAOT site. The issue has been addressed following the Guide to the Expression of Uncertainty in Measurement (GUM) by quantifying uncertainties affecting data products from a number of AERONET-OC sites. Results from Gergely and Zibordi (2014) indicate the following:

1. The relative uncertainties determined for AAOT L_WN(λ) through the application of GUM accounting for correlations among the various uncertainty contributions, confirm within 0.5% the values earlier provided in Zibordi et al. (2009b). The study also shows a reduction of uncertainties when excluding the highest values of solar zenith angle and of aerosol optical depth.

2. The relative uncertainties may amply vary from site to site as a function of the water type, which implies different spectral shapes and ranges of L_WN(λ). These relative uncertainties may largely exceed the 5% threshold commonly considered for validation activities (Zibordi and Voss 2014).

3. The absolute uncertainties affecting L_WN(λ) also appreciably vary from site to site, but they do not necessarily mirror the values of relative uncertainties. In particular, sites exhibiting relative uncertainties well above those determined for the AAOT site, may exhibit much lower absolute uncertainties.

Table 1.

Relative (u_r; %) and absolute (u_a; mW cm⁻² μm⁻¹ sr⁻¹) uncertainties determined for L_WN(λ) from the AAOT and HLT sites at nominal center wavelengths from 412 to 667 nm [after Gergely and Zibordi (2014)]. The symbol m indicates the median of the L_WN(λ) spectra ( mW cm⁻² μm⁻¹ sr⁻¹) included in the uncertainty analysis.

View Table

Aside from suggesting the need to compute uncertainties for each specific AERONET-OC site, the above results confirm the importance of determining both relative and absolute uncertainties, whose values are expected to be representative for the range of considered radiances.

d. Products assessment

AERONET-OC L_WN(λ) rely on the basic measurement equation defined by Eq. (1) and a relatively small number of data collected with a narrow field of view (i.e., 1.2° full angle). The major drawback, however, arises from the spectrally asynchronous measurement capability of CE-318 and CE-318T radiometers: measurements are performed at different time for each spectral band. While this technical aspect is not relevant for sky or sun measurements benefitting of sources (i.e., sun and sky) relatively stable over a short time during clear sky conditions, it definitively affects sea measurements as a result of wave perturbations implicitly leading to differences in surface perturbations across the various bands.

The measurement capabilities offered by CE-318 and CE-318T systems are thus far away from being ideal for ocean color field applications (IOCCG 2019). Still, strict quality assurance of the measurement sequences can lead to data products meeting requirements for validation activities. As stated in section 1, an early step of the quality assurance process, which includes a number of quality control checks, is the removal of any measurement sequence exhibiting variance above given thresholds in the total radiance from the sea in at least one spectral band (Zibordi et al. 2009b). This means that quality assured measurement sequences raised to level 1.5 and level 2.0 are not affected by large surface perturbations. As a result of the filtering applied, AERONET-OC L_WN(λ) are only determined for low-to-mild sea state conditions bound by wind speeds typically not exceeding 5 m s⁻¹. Direct benefit of such quality assurance process is thus the low variability of the water surface reflectance factor ρ as a function of wind speed across a wide range of solar zenith angles (Zibordi 2016).

The measurement sequences passing the former quality assurance step might still be affected by excessive glint resulting from the reflection of sky radiance from brighter portions of the sky. Because of this, as already anticipated, L_T(θ, φ, λ) is determined from each quality assured measurement sequence by averaging a percent of the N_T data exhibiting the lowest radiance values in each band. The use of the average of the lowest radiance values instead of the mean of all values, still supported by experimental evidence (Zibordi 2012, 2016; Pitarch et al. 2020), might be responsible for an underestimate of L_WN(λ) more pronounced at low solar zenith angles θ₀ (i.e., below 20°) and increasing with the wind speed. Table 2 summarizes comparison results between L_WN(λ) from AERONET-OC and in-water measurements. These results show an excellent agreement between the independent determinations of L_WN(λ), nevertheless limited to cases with θ₀ > 20° and benefitting of glint filtering (i.e., not including measurement sequences heavily affected by surface perturbations and consequently mostly restricted to relatively low wind speeds). Still, recognizing the lack of a fundamental element justifying the use of the mean of relative minima alternative to the mean of all values, the agreement of the data shown in Table 2 is largely explained by the combined uncertainties of the compared measurements and corroborate the applied AERONET-OC processing scheme.

Table 2.

Comparison results between L_WN(λ) from AERONET-OC and independent in-water measurements accounting for 185 matchups at various center wavelengths in the 412–667-nm spectral interval [after Zibordi (2016)]. The symbol ψ indicates the mean of signed percent differences (i.e., an index for the bias affecting the compared data) of AERONET-OC-derived L_WN(λ) with respect to in-water data products. The symbol |ψ| indicates the mean of absolute percent differences (i.e., an index for the dispersion of data).

View Table

e. Polarization effects

A number of recent studies highlighted the dependence of above-water radiometry on polarization effects (Harmel et al. 2012; Mobley 2015; Hieronymi 2016; D’Alimonte and Kajiyama 2016; Foster and Gilerson 2016; Zhang et al. 2017; Gilerson et al. 2018). Recognizing that an operational network such as AERONET-OC can only rely on consolidated findings, an effort was made to evaluate the impact of polarization through tabulated values of ρ determined accounting (Mobley 2015) and not accounting (Mobley 1999) for polarization effects. The analysis was performed by comparing AERONET-OC L_W(λ) data determined using the different ρ values with L_W(λ) from independent in-water measurements (Zibordi 2016). Results, restricted to the measurement conditions for which AERONET-OC data can be produced (i.e., wind speeds generally lower than 5 m s⁻¹) indicated a slightly better performance of L_W(λ) determined without accounting for polarization effects and justifies the current application of the ρ values from Mobley (1999). Still, alternative ρ values, hopefully accounting for their spectral dependence, can be applied as soon as available and supported by community consensus.

f. Bidirectional effects

The nonisotropy of the in-water light field implies the correction of L_W(θ, φ, λ) data for bidirectional effects due to the solar zenith angle and the nonnadir view of the sensor. The correction approach commonly applied for bidirectional effects is that proposed for Case-1 (i.e., chlorophyll-a dominated) waters by Morel et al. (2002) hereinafter called Chla based. Considering that most of the AERONET-OC sites are located in optically complex coastal waters, the corrections relying on the Chla-based approach are likely affected by large uncertainties. Because of this, alternative corrections applicable to both Case-1 and optically complex waters have been evaluated with a view to their applicability to AERONET-OC data.

Among the various approaches proposed in literature (e.g., Park and Ruddick 2005), the one considered for dedicated investigations was that by Lee et al. (2011). This correction, hereinafter called IOP-based, provides the major advantage of relying on the retrieval of water IOPs from L_W(θ, φ, λ) itself. An experimental study focused on the comparison of Chla- and IOP-based approaches, still restricted to the bidirectional effects due to nonnadir view, shows average relative differences of corrections varying between 10% and 40% in the blue and green spectral regions with standard deviation of 10%–25% (Talone et al. 2018). By using in situ reference data to quantify the uncertainties affecting the correction factors from the two approaches, the study indicates relative spectrally and water dependent uncertainties varying between 20% and 60% for the Chla-based approach, with the highest values affecting the blue and red regions. Conversely, the IOP-based approach shows relative uncertainties within 20%–35% and lower dependence on wavelength and water type. These results have motivated the implementation of the IOP-based correction approach in the AERONET-OC processing in addition to the Chla-based one.

g. Perturbations by deployment structures

AERONET-OC radiometers are deployed on fixed platforms. The location of the instrument on each platform needs to ensure the minimization of structure perturbations. This implies identifying locations ideally free from obstacles above the radiometer, and allowing the sight of view of the sensor to point at the water surface well away from the main structure during the central hours of the day when satellite overpasses occur (at least for deployments underpinning ocean color validation activities). The former requirement may face challenges due to the size and shape of structures, deployment restrictions due to the presence of obstacles or of highly reflective objects. An early investigation carried out at the AAOT (Hooker and Zibordi 2005) indicated that observations performed pointing at the water surface at a distance at least as large as the height of the deployment structure would minimize perturbations due to the structure itself. This finding, also supported by field measurements performed on a research vessel (Hooker and Morel 2003), was proposed as a practical rule. However, its widespread applicability has not been demonstrated. In particular, the spectral impact of superstructures in L_WN(λ) remained an open issue. This latter point has been tackled in a recent investigation performed at the AAOT by collecting above-water radiometry data at various distances from the structure adopting the AERONET-OC viewing geometry (i.e., θ = 40° and φ = 90°) with the deployment platform exhibiting its regular appearance, or alternatively covered by white sheets (Talone and Zibordi 2019).

Results from this investigation indicate that for the low reflectance case (i.e., the AAOT without white cover), the structure perturbations affect L_WN(λ) by less than 1% in the visible spectral region when the sea-radiance measurements are performed at distances from the structure larger than the height of the structure itself. However, they may exceed 2% beyond 800 nm. For identical measurement geometries, but for the high reflectance case (i.e., the AAOT with white cover), results show that the impact of the superstructure approaches 1% in the blue–green and may exceed 2% beyond approximately 600 nm. Recognizing that the strict applicability of these results is confined to a specific deployment platform, still, these findings confirm previous deployment recommendations and additionally provide new elements for the quantification of spectral uncertainties affecting data products.

h. Adjacency effects

AERONET-OC sites are almost totally located in coastal regions. This particular condition, generally depending on the larger availability of fixed structures in costal rather than open sea, may impact the exploitation of data products in validation processes. In fact, close regions exhibiting different surface reflectance such as sea and land, naturally lead to adjacency effects in top-of-the-atmosphere data (Bulgarelli and Zibordi 2018a). A number of targeted studies (e.g., Bulgarelli and Zibordi 2018b) showed that adjacency effects may lead to unwanted perturbations in satellite data products up to more than 10 n mi (1 n mi = 1.852 km) away from the coast. These perturbations, which spectrally vary with the reflectance of land surfaces, water type, illumination and viewing geometries, coastline and the specific atmospheric correction method applied, may become responsible for seasonal trends in satellite data (Bulgarelli et al. 2018). Recognizing that adjacency effects only concern satellite data products, still, the application of AERONET-OC data from sites located within a very few nautical miles from the coast may be challenged in validation activities not supported by schemes allowing to remove adjacency perturbations in remote sensing data. This stresses the need to establish AERONET-OC sites at locations well away from the coast when prioritizing support to ocean color validation activities. Nevertheless, time series of AERONET-OC data from regions located within a few nautical miles from the coast are expected to gather increased value when operational atmospheric correction codes will include schemes for the minimization of adjacency effects.

a. Instrument and band setting

Excluding some early CE-318 analog radiometer systems, most AERONET-OC deployments rely on CE-318 9-channel digital systems and on the recent CE-318T 12-channel ones.

Table 3 shows the reference center wavelengths for the 9-channel and 12-channel systems. The CE-318 9-channel radiometer system largely exhibits spectral bands corresponding to those of the Moderate Resolution Imaging Spectroradiometer (MODIS) and the Visible Infrared Imaging Radiometer Suite (VIIRS) satellite sensor. Conversely, the CE-318T 12-channel system exhibits bands matching most of the Ocean Land Color Instrument (OLCI) and it is definitively expected to overlap the major ocean color ones of the forthcoming Plankton, Aerosol, Cloud, and Ocean Ecosystem (PACE) mission.

Table 3.

Corresponding nominal center wavelengths λ (nm) of CE-318 9-channel and CE-318T 12-channel radiometer systems (all having 10-nm bandwidth), in comparison with those of recent satellite ocean color sensors (OLCI, VIIRS, and MODIS). The symbols s and i indicate the marine and inland water spectral band configurations, respectively.

View Table

Also relevant, considering the growing interest in remote sensing applications focusing on lake waters (e.g., Moore et al. 2019), is that the 12-channel standard system has been conceived with two different spectral band settings: one for marine applications and the other more suitable for inland waters.

Briefly, the CE-318T radiometer system for marine applications, when compared with the CE-318 version, features additional spectral bands centered at 400, 510, 620, and 779 nm. While the supplemental bands in the visible allow for a more comprehensive spectral determination of L_WN(λ), the band centered at 779 nm combined with that at 865 nm allow for investigating τ_a(λ) at center wavelengths equivalent to those commonly supporting the atmospheric correction of satellite ocean color data.

The CE-318T radiometer system for inland water applications, when compared to the marine one, has spectral bands centered at 681 and 709 nm replacing those at 400 and 779 nm. The spectral setting comprising the 667, 681 and 709 nm center wavelengths aims at supporting the detection and quantification of phytoplankton blooms in the near-surface water layer.

While the foreoptics, optics and detector components of CE-318 and CE-318T radiometers are substantially the same for the relevant visible and near-infrared spectral bands, the CE-318T units exhibit an improved analog-to-digital resolution (21 versus 16 effective bits of the CE-318 units). Additionally, CE-318T systems can transmit relatively large volumes of data through General Packet Radio Service (GPRS) [opposite to the limited capability of satellite data collection platforms (DCP)]. Further relevant feature of the new CE-318T systems is the possibility to store large volumes of data (up to 32 gigabytes) on Secure Digital (SD) cards, definitively relevant for remote sites not always within GPRS coverage and constrained by the low transmission rate of DCP (Zibordi et al. 2009b).

CE-318T systems also exhibit some measurement protocol improvement with respect to the former CE-318 version. First, the sign of the relative azimuth φ can be varied during the day or across the year (e.g., from −90° to +90°). This feature is relevant for structures with deployment locations that may not allow respecting the basic geometric requirements during the full day or during part of the year. It is mentioned that the parameters determining the measurement geometry (i.e., θ and φ, in addition to the exact center wavelengths of the instrument) are accessible in each data record at any quality level.

An additional major feature of the CE-318T systems is the capability of performing an increased number of above-water marine or lake measurement sequences. By recalling that for CE-318 systems the number of above-water measurements is limited to a single sequence every 30 min, the marine or lake sequences of relevant measurements performed with CE-318T systems can now be increased by up to a factor of 3, with each sequence completed in 5 min. This may allow producing measurement replicates at a finer temporal scale, relevant to enhance the quality assurance of data and better quantify the environmental perturbations contributing to L_WN(λ) uncertainties.

b. Sites

Figure 2 displays the location of the 31 AERONET-OC sites to date. Notable, while the Northern Hemisphere exhibits a substantial midlatitude distribution of sites across continents, the Southern Hemisphere shows a very few active ones. The marine sites represent a number of water types embracing chlorophyll-a-, sediment-, and CDOM-dominated waters. When considering the lake sites, they embrace a variety of inland waters characterized by different seasonal conditions including the occurrence of harmful algal blooms. Finally, note the number of decommissioned marine sites. This is mostly an indication of the difficulty to maintain access to offshore locations often implying complex logistics and extraordinary safety measures governed by agreements with nonresearch bodies owning the structures.

Location of current AERONET-OC sites contributing to the version-3 database (https://aeronet.gsfc.nasa.gov/ocean_color.html). The blue, green, and gray filled circles indicate the active marine water, inland water, and decommissioned sites, respectively.

Citation: Journal of Atmospheric and Oceanic Technology 38, 4; 10.1175/JTECH-D-20-0085.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Location of current AERONET-OC sites contributing to the version-3 database (https://aeronet.gsfc.nasa.gov/ocean_color.html). The blue, green, and gray filled circles indicate the active marine water, inland water, and decommissioned sites, respectively.

Citation: Journal of Atmospheric and Oceanic Technology 38, 4; 10.1175/JTECH-D-20-0085.1

• Download Figure
• Download figure as PowerPoint slide

Location of current AERONET-OC sites contributing to the version-3 database (https://aeronet.gsfc.nasa.gov/ocean_color.html). The blue, green, and gray filled circles indicate the active marine water, inland water, and decommissioned sites, respectively.

Citation: Journal of Atmospheric and Oceanic Technology 38, 4; 10.1175/JTECH-D-20-0085.1

• Download Figure
• Download figure as PowerPoint slide

The temporal expansion of the AERONET-OC sites is illustrated in Fig. 3 together with the satellite ocean color missions that mostly benefit from the network data. Table 4 provides the comprehensive list of AERONET-OC sites (to date) together with details on location, type of structure and responsible institution.

(bottom) Temporal expansion of the AERONET-OC sites from 2002 until 2020 in comparison with (top) the major satellite ocean color missions supported by the network. The blue, green, and gray colors indicate the active marine water, inland water, and decommissioned AERONET-OC sites, respectively.

Citation: Journal of Atmospheric and Oceanic Technology 38, 4; 10.1175/JTECH-D-20-0085.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

(bottom) Temporal expansion of the AERONET-OC sites from 2002 until 2020 in comparison with (top) the major satellite ocean color missions supported by the network. The blue, green, and gray colors indicate the active marine water, inland water, and decommissioned AERONET-OC sites, respectively.

Citation: Journal of Atmospheric and Oceanic Technology 38, 4; 10.1175/JTECH-D-20-0085.1

• Download Figure
• Download figure as PowerPoint slide

(bottom) Temporal expansion of the AERONET-OC sites from 2002 until 2020 in comparison with (top) the major satellite ocean color missions supported by the network. The blue, green, and gray colors indicate the active marine water, inland water, and decommissioned AERONET-OC sites, respectively.

Citation: Journal of Atmospheric and Oceanic Technology 38, 4; 10.1175/JTECH-D-20-0085.1

• Download Figure
• Download figure as PowerPoint slide

Table 4.

AERONET-OC sites contributing to the version-3 database (to date). Here, EU indicates the European Union, US indicates the United States, and UK indicates the United Kingdom.

View Table

c. Processing

A number of changes have been implemented in the AERONET-OC processing supporting the version-3 database. A primary feature of the new processing release, when compared to the previous one supporting the version-2 database, is the capability to handle data from the CE-318T 12-channel instruments in addition to the CE-318 9-channel ones. This does not simply imply the need for handling a higher number of spectral bands, but also the necessity to address the impact of a larger number of spectral values on the quality assurance process or, to deal with an increased number of spectrally resolved parameters such as the correction factors for bidirectional effects.

The L_WN(λ) data in the version-2 database relied on the application of spectral thresholds to the standard deviations of the N_T sea-radiance measurements and N_i sky-radiance measurements, identical for both level 1.5 and level 2.0. These spectral thresholds constrained in the range of 0.1–0.2 mW cm⁻² μm⁻¹ sr⁻¹ for the N_T sea-radiance measurements and in the range of 0.02–0.03 mW cm⁻² μm⁻¹ sr⁻¹ for the N_i sky-radiance measurements, were determined from the analysis of data collected at the AAOT over a variety of water types during ideal observation conditions: clear sky and low-to-mild sea state. It is mentioned that the standard deviation of CE-318 and CE-318T radiance measurements determined with a stable laboratory source is generally lower than 0.002 and 0.001 mW cm⁻² μm⁻¹ sr⁻¹, respectively, in the spectral range of interest.

As already stated, the application of thresholds to the standard deviation of successive acquisitions aims at removing any data record likely (i) affected by significant wave perturbations or foam making difficult the application of an accurate surface reflectance factor ρ for the determination of the sky glint contribution or (ii) perturbed by clouds in the direction (θ′, φ). Obviously, an increase in the number of spectral bands naturally enhances the probability of excluding measurement sequences just because data at a single spectral band do not meet requirements. Aiming at preserving the same rigor characterizing level 2.0 L_WN(λ) in the version-2 database (i.e., level 2.0 is the quality level suggested for the validation of satellite data products), level 2.0 L_WN(λ) data in the version-3 database have been obtained applying the same thresholds declared above. Conversely, level 1.5 L_WN(λ) data in the version-3 database have been obtained by relaxing (i.e., doubling) the thresholds. This solution provides the advantage of increasing by some percent (depending on the site) the number of L_WN(λ) data at level 1.5 in the version-3 database at the expense of an increased uncertainty due to environmental perturbations. This revision of the processing scheme is solely supported by the assumption that an increase of the number of data at level 1.5, despite a potentially lower accuracy, may provide benefit to a variety of real-time applications spanning from early assessment of satellite data products to water quality monitoring.

The AERONET-OC version-2 database relied on the Chla-based approach specifically proposed to minimize bidirectional effects in Case-1 waters (Morel et al. 2002). The related lookup tables to compute the correction factors applied to CE-318 data, however, exhibit spectral limitations. In fact the f/Q table by Morel et al. (2002) hereinafter identified as f/Q-2002, are restricted to key ocean color center wavelengths (i.e., 412, 443, 490, 510, 560, 620, and 660 nm), and consequently do not confidently support corrections for bidirectional effects at significantly different center wavelengths. Because of this, new lookup tables (hereinafter identified as f/Q-2013) produced by the Observatoire Océanologique de Villefranche-sur-Mer applying the same processing solutions as in Morel et al. (2002), but with a spectral resolution of 5 nm and extending over the 350–700 nm spectral range (B. Gentili, unpublished data), have now been incorporated in the current AERONET-OC processing. The f/Q-2013 lookup tables with higher spectral resolution and spectral range relative to f/Q-2002, provide the immediate advantage to allow determining the correction factors C_fr for the Chla-based approach minimizing uncertainties due to spectral extrapolations. Further advantage offered by the f/Q-2013 with respect to the f/Q-2002 tables is the possibility of indexing the clear sky illumination conditions as a function of τ_a with values of 0.05, 0.10, 0.2, 0.4, and 1.0.

As already anticipated, an additional implementation incorporated in the current AERONET-OC processing supporting the version-3 database is the IOP-based correction approach to generate L_WN(λ) likely free from the water type restrictions intrinsic of the Chla-based one.

A brief introduction to both the Chla- and IOP-based correction approaches is provided in the following subsections, while an evaluation of their impact on AERONET-OC data representative of different water types is addressed in the discussion section.

1) Chla-based correction approach

The Chla-based approach minimizes the dependence of L_W(θ, φ, λ) on the viewing geometry and solar zenith following Eq. (2) with C_fr(θ, φ, θ₀, λ, τ_a, IOP, W) given by

$C_{\mathrm{fr}}^{\mathrm{Chla}}\left(\theta ,\phi ,{\theta }_0,\lambda ,{\tau }_a,\mathrm{IOP},W\right)={\displaystyle \frac{\Re \left(0,W\right)}{\Re \left(\theta ,W\right)}}{\displaystyle \frac{Q\left(\theta ,\phi ,{\theta }_0,\lambda ,{\tau }_a,\mathrm{Chla}\right)}{f\left({\theta }_0,\lambda ,{\tau }_a,\mathrm{Chla}\right)}}\times {\displaystyle \frac{f\left(0,\lambda ,{\tau }_a,\mathrm{Chla}\right)}{Q\left(0,0,0,\lambda ,{\tau }_a,\mathrm{Chla}\right)}},$(3)

where ℜ accounts for combined reflection/refraction effects on the downward irradiance and upwelling radiance propagating through the water surface, Q is the Q factor indicating the ratio of upward irradiance to upwelling radiance just below the surface, and f relates the irradiance reflectance to the water IOPs. The aerosol optical depth τ_a expresses dependence on the atmospheric optical properties and the chlorophyll-a concentration Chla indicates dependence on IOPs. It is recalled that the values of Chla applied for the determination of $C_{\mathrm{fr}}^{\mathrm{Chla}}$ and accessible in both the version-2 and version-3 databases, are computed through band-ratio algorithms (Zibordi et al. 2009b). When available, as for the Adriatic, Baltic and Black Sea sites (see Zibordi et al. 2009a, 2015), regional algorithms are applied. Conversely, the OC2v4 algorithm (O’Reilly et al. 2000) developed relying on a global dataset representative of both oligotrophic and eutrophic waters, is used. Because of this, AERONET-OC Chla values need to be considered best estimates with uncertainties that may largely vary across sites.

With reference to Morel et al. (2002), by assuming Case-1 waters, simulated values of ℜ and f/Q included in the f/Q-2002 and f/Q-2013 tables were determined as a function of the relevant variables. Taking into account the different f/Q values, Fig. 4 shows a basic comparison of the correction factors $C_{\mathrm{fr}}^{\mathrm{Chla}}$ extracted from the f/Q-2002 and f/Q-2013 lookup tables. The analysis is restricted to the AERONET-OC viewing geometry (i.e., θ = 40°, φ = 90°) with W = 4 m s⁻¹, Chla = 1 mg m⁻³, τ_a = 0.2 and various solar zenith angles (i.e., θ₀ = [20°, 40°, 60°]). Results indicate a close agreement between $C_{\mathrm{fr}}^{\mathrm{Chla}}$ from the f/Q-2002 and f/Q-2013 table values. Still, differences exhibit an impact that may slightly exceed ±1% in L_WN(λ) for the specific case considered.

(a) Correction factors $C_{\mathrm{fr}}^{\mathrm{Chla}}$ determined from the f/Q-2002 and f/Q-2013 lookup tables with parameters θ = 40°, φ = 90°, W = 4 m s⁻¹, Chla = 1 mg m⁻³, τ_a = 0.2, and θ₀ = [20°, 40°, 60°], and (b) the related percent differences ε between $C_{\mathrm{fr}}^{\mathrm{Chla}}$ determined from the f/Q-2013 and f/Q-2002 values.

Citation: Journal of Atmospheric and Oceanic Technology 38, 4; 10.1175/JTECH-D-20-0085.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

(a) Correction factors $C_{\mathrm{fr}}^{\mathrm{Chla}}$ determined from the f/Q-2002 and f/Q-2013 lookup tables with parameters θ = 40°, φ = 90°, W = 4 m s⁻¹, Chla = 1 mg m⁻³, τ_a = 0.2, and θ₀ = [20°, 40°, 60°], and (b) the related percent differences ε between $C_{\mathrm{fr}}^{\mathrm{Chla}}$ determined from the f/Q-2013 and f/Q-2002 values.

Citation: Journal of Atmospheric and Oceanic Technology 38, 4; 10.1175/JTECH-D-20-0085.1

• Download Figure
• Download figure as PowerPoint slide

(a) Correction factors $C_{\mathrm{fr}}^{\mathrm{Chla}}$ determined from the f/Q-2002 and f/Q-2013 lookup tables with parameters θ = 40°, φ = 90°, W = 4 m s⁻¹, Chla = 1 mg m⁻³, τ_a = 0.2, and θ₀ = [20°, 40°, 60°], and (b) the related percent differences ε between $C_{\mathrm{fr}}^{\mathrm{Chla}}$ determined from the f/Q-2013 and f/Q-2002 values.

Citation: Journal of Atmospheric and Oceanic Technology 38, 4; 10.1175/JTECH-D-20-0085.1

• Download Figure
• Download figure as PowerPoint slide