Next Article in Journal
Towards Circumpolar Mapping of Arctic Settlements and Infrastructure Based on Sentinel-1 and Sentinel-2
Previous Article in Journal
PlanetScope Radiometric Normalization and Sentinel-2 Super-Resolution (2.5 m): A Straightforward Spectral-Spatial Fusion of Multi-Satellite Multi-Sensor Images Using Residual Convolutional Neural Networks
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Evaluation of Ocean Color Remote Sensing Algorithms for Diffuse Attenuation Coefficients and Optical Depths with Data Collected on BGC-Argo Floats

1
State Key Laboratory of Satellite Ocean Environment Dynamics, Second Institute of Oceanography, Ministry of Natural Resources, Hangzhou 310012, China
2
Pilot National Laboratory for Marine Science and Technology (Qingdao), Qingdao 266000, China
3
School of Marine Sciences, University of Maine, Orono, ME 04469, USA
4
Laboratoire d’Océanographie de Villefranche, Sorbonne Université, 06238 Villefranche-sur-Mer, France
*
Author to whom correspondence should be addressed.
Remote Sens. 2020, 12(15), 2367; https://doi.org/10.3390/rs12152367
Submission received: 22 June 2020 / Revised: 20 July 2020 / Accepted: 21 July 2020 / Published: 23 July 2020
(This article belongs to the Section Ocean Remote Sensing)

Abstract

:
The vertical distribution of irradiance in the ocean is a key input to quantify processes spanning from radiative warming, photosynthesis to photo-oxidation. Here we use a novel dataset of thousands local-noon downwelling irradiance at 490 nm (Ed(490)) and photosynthetically available radiation (PAR) profiles captured by 103 BGC-Argo floats spanning three years (from October 2012 to January 2016) in the world’s ocean, to evaluate several published algorithms and satellite products related to diffuse attenuation coefficient (Kd). Our results show: (1) MODIS-Aqua Kd(490) products derived from a blue-to-green algorithm and two semi-analytical algorithms show good consistency with the float-observed values, but the Chla-based one has overestimation in oligotrophic waters; (2) The Kd(PAR) model based on the Inherent Optical Properties (IOPs) performs well not only at sea-surface but also at depth, except for the oligotrophic waters where Kd(PAR) is underestimated below two penetration depth (2zpd), due to the model’s assumption of a homogeneous distribution of IOPs in the water column which is not true in most oligotrophic waters with deep chlorophyll-a maxima; (3) In addition, published algorithms for the 1% euphotic-layer depth and the depth of 0.415 mol photons m−2 d−1 isolume are evaluated. Algorithms based on Chla generally work well while IOPs-based ones exhibit an overestimation issue in stratified and oligotrophic waters, due to the underestimation of Kd(PAR) at depth.

Graphical Abstract

1. Introduction

Light from the sun fuels oceanic primary production, heats the upper ocean, and oxidizes chemical compounds such as organic molecules. In order to accurately model these processes, the subsurface light distribution is needed. As light attenuates near-exponentially in water, to describe the subsurface light field, the exponent describing this attenuation needs to be known (referred to as the diffuse attenuation coefficient [1]). This exponent is a function of in-water components themselves and, to a lesser degree, a function of the illumination conditions [2]. Additionally, in order to constrain the layer in which certain processes take place, specific light horizons are of interest. For example, for photosynthesis, a euphotic or isolume depth is defined based on a relative light level (e.g., 1% [3]) or an absolute intensity (e.g., 0.415 mol photons m−2 d−1 [4]) below which photosynthesis is assumed as zero. Moreover, accurately quantifying the attenuation of sunlight within the upper ocean is essential for physical and biogeochemical models, affecting the modeled upper-ocean temperature (e.g., [5,6]) and ecosystem dynamics (e.g., [7]). Sea-surface sunlight is globally available from space agencies ([8]), and a variety of models to describe its attenuation through the water have been devised (e.g., [9,10]). However, lack of data has limited their validation on global scales.
In the past decade, with the rapid progress of the Biogeochemical-Argo (BGC-Argo) float technology, the BGC-Argo dataset has become the largest data source for optical and biogeochemical observations in the global ocean [11,12]. Downwelling radiometry is one of the six core BGC variables in the International Argo program [13,14], and has been used to study global bio-optical relationships and anomalies [15,16], used to determine the depths of the euphotic zone and a specific isolume [17,18], to improve the parameterization scheme of sunlight attenuation in a biological model [19], as well as to correct and derive other bio-optical measurements performed on the same float [20,21,22]. An additional important application of BGC-Argo is to validate ocean color remote sensing products (e.g., [23,24,25]) over the whole year and on the global scale.
In this study, based on a recently compiled global BGC-Argo dataset (described in [15,26,27]), first, we evaluate several ocean color remote sensing algorithms for the diffuse attenuation coefficient at 490 nm (Kd(490)), including two empirical [28,29] and two semi-analytical algorithms [9,30]; then, we evaluate the performance of the algorithm for the diffuse attenuation coefficient (Kd(PAR)) of photosynthetically available radiation (PAR) [10], at different optical depths; finally, we evaluate several empirical [29,31] and semi-analytical algorithms [32] for the euphotic depth (zeu) and the 0.415 mol photons m−2 d−1 isolume depth (z0.415). Given our dataset, we focus only on open ocean algorithms. The need for validation stems from the fact that these algorithms have not been validated globally and over the whole year.

2. Data and Methods

2.1. BGC-Argo Data

The globally distributed BGC-Argo dataset used here was published in the SEA scieNtific Open data Edition (SEANOE, www.seanoe.org/data/00383/49388/) and is comprised of more than 5000 local-noon profiles of downwelling irradiance (different numbers for different wavelengths due to quality control filtering [26]) obtained using 103 Satlantic OCR504 radiometers deployed on Argo floats, spanning 3 years from 20 October 2012 to 26 January 2016. Each radiometry data depth profile includes downward irradiance (Ed) at three wavelengths (380, 412, 490 nm, [μW cm−2 nm−1]), and the instantaneous photosynthetically available radiation (iPAR, [μmol photons m−2 s−1]), which is the downwelling photon flux integrated over a wavelength range spanning from 400 to 700 nm. All radiometry data has been quality controlled [26], removing the points and profiles that are significantly affected by clouds and wave focusing. The ocean basin and trophic environment of each profile have been identified in the dataset (www.seanoe.org/data/00383/49388/data/49825.pdf), however, in this study, we reorganized them into 10 regions (Table 1 and Figure 1). All symbols used in this study are listed in Table 2. We note that no correction for dark offset has been applied in this dataset beyond that obtained from the manufacturer. Small (O(0.03 μW cm−2 nm−1)) nonzero dark values are observed in data from profiling floats but these values are variable in sign and hence will not introduce a bias in our current work.

2.1.1. Layer-Averaged Diffuse Attenuation Coefficients, Penetration, and Euphotic Depths

The attenuation of Ed(490) and of PAR with depth is approximated as an exponential decrease, and the layer-averaged diffuse attenuation coefficients, Kd(490)z and Kd(PAR)z, from surface (0) to any given depth, z, are defined as:
K d ( 490 ) z = 1 z ln ( E d ( 490 , 0 ) E d ( 490 , z ) )
K d ( PAR ) z = 1 z ln ( iPAR ( 0 ) iPAR ( z ) )
The BGC-Argo radiometry data processing flow chart is shown in Figure 2, where the surface Ed(490,0) and iPAR(0) are determined first. Since BGC-Argo has no radiometry exactly at z = 0, they are computed by extrapolating a linear regression for ln(Ed(490,z)) and a second-degree polynomial regression for ln(iPAR(z)) with depth, z, respectively, at the top 10 m of each profile. To ensure the accuracy of extrapolation, we only process the profiles with at least 5 valid values within the top 10 m (which include 4882 Ed(490) and 2548 iPAR profiles, as shown in Table 1). This extrapolation method is validated with in-situ cruise data in Appendix A.
We then determine the penetration depth at 490 nm (zpd) and euphotic depth (zeu), as the depth where Ed(490,z) reaches Ed(490,0) × e−1 ~ 0.37 × Ed(490,0) [36] and the depth (denoted by zeu) where iPAR(z) reaches iPAR(0) × 1% [3], respectively. zpd and zeu, respectively, are determined through an interpolation of Ed(490) and iPAR profiles to depths where Ed(490) and iPAR reach the calculated Ed(490,zpd) = Ed(490, 0)/e and iPAR(zeu) = 0.01 iPAR(0) (Figure 2). Note that, the definition of penetration depth is spectrally dependent (i.e., zpd varies with wavelength), in this study, the wavelength is omitted for simplicity, and all zpd without designated wavelength represents zpd(490). By definition, the remotely-sensed diffuse attenuation coefficient Kd(490) is the layer-averaged one from surface to zpd, and it is determined in situ as:
K d ( 490 ) zpd = 1 z pd
Note that Organelli et al. [15] used an approximation to estimate the penetration depth, to derive Kd(490)zpd, namely zpd = zeu/4.6. In Appendix B we show that this approximation results in a significant bias in Kd(490)zpd in oligotrophic regions.

2.1.2. The 0.415 mol Photons m−2 d−1 Isolume Depth (z0.415)

Since the work of Letelier et al. [4], the isolume depth z0.415, defined as the depth where the daily PAR reaches 0.415 mol photons m−2 d−1, has been used as a threshold depth below which light is insufficient to support photosynthesis, and has been regarded as a useful estimate of the euphotic depth [31,37,38,39]. Phytoplankton living at depth are expected to react to absolute light level rather than a relative level [40], and thus, the isolume is more applicable for phytoplankton growth than the depth of 1% of sea surface light intensity (zeu). Note that recently, Behrenfeld and Boss [41] argued for a lower threshold irradiance value.
Generally, it is difficult to directly measure z0.415, as well as any other daily-isolume depth, as it needs the radiometry at both continuous vertical scale (at least a resolution of meters) and continuous temporal scale (at least hourly). It can be estimated using a single observation during a day (on Argo and ships), with the assumption that the layer-averaged Kd(PAR) at z0.415 is constant during the day (not a constant Kd(PAR) profile), i.e.,:
iPAR ( z 0.415 ) iPAR ( 0 ) = 0.415 PAR sat × α
Equation (4) allows us to determine iPAR(z0.415) based on a single profile of radiometry and PARsat, the surface daily photosynthetically available radiation product (see Section 2.2.1). α is the transmission of sun light through the air-sea interface, dependent on the latitude and day of year [33]. z0.415 is determined as the depth where iPAR(z) reaches iPAR(z0.415) which is calculated from Equation (4). Using a similar procedure, one can derive the depth of any other isolume.

2.2. Satellite Data

2.2.1. Auxiliary Satellite Data Products

We use several satellite-derived products (Level-3, daily, 4 km-resolution) provided by the Moderate Resolution Imaging Spectroradiometer (MODIS) on the satellite “EOS PM (Aqua)”: (1) Daily photosynthetically available radiation (PARsat [mol photons m−2 d−1]) (https://oceandata.sci.gsfc.nasa.gov/MODIS-Aqua/Mapped/Daily/4km/par/) for calculating z0.415. This product is derived from the radiative transfer model [8] and the atmospheric properties inferred from MODIS-Aqua; (2) Surface chlorophyll-a concentration (Chlasat) (https://oceandata.sci.gsfc.nasa.gov/MODIS-Aqua/Mapped/Daily/4km/chlor_a/). This product combines two algorithms, band-ratio OCx algorithm [35], and the Hu’s color index (CI) algorithm [42]; (3) Remote sensing reflectance (Rrs [sr−1]) at four wavelengths (443, 488, 555, and 667 nm) (https://oceandata.sci.gsfc.nasa.gov/MODIS-Aqua/Mapped/Daily/4km/Rrs_443/) are used as input to the quasi-analytical algorithm (QAA) [34] to retrieve the absorption and backscattering coefficients at 490 nm (a(490) and bb(490)), and in turn, to calculate Kd(490) following [9] and [30], and to calculate Kd(PAR) following [10]. Note that the Raman effect on Rrs is corrected following [30].

2.2.2. Satellite-Based Kd(490)

We download three available satellite Kd(490) products and derive three other literature-based Kd(490) for evaluation using the floats’ data. They encompass different approaches (empirical vs. semi-analytical) and different sources (MODIS-Aqua vs. GlobColour):
(i) Kd(490)M-KD2M is an operational MODIS product derived from an empirical algorithm “KD2M” that uses the blue-green reflectance ratio (“B/G”) of satellite-measured remote sensing reflectance [28] and is available from NASA (https://oceandata.sci.gsfc.nasa.gov/MODIS-Aqua/Mapped/Daily/4km/Kd_490/);
(ii) Kd(490)GC-M07 is an operational product retrieved from an empirical relationship between Kd(490) and Chla (Equation (5), [29]), provided by the GlobColour project [43], using products merged between MODIS and a visible infrared imaging radiometer (VIIRS) for our data period (ftp://ftp.hermes.acri.fr/GLOB/merged/day/);
K d ( 490 ) = 0.0166 + 0.077298 × [ Chla ] 0.67155
(iii) Kd(490)M-M07 is derived from MODIS-retrieved Chlasat and Equation (5) (the same empirical algorithm as Kd(490)GC-M07);
(iv) Kd(490)GC-L05a is an operational product retrieved from a semi-analytical IOPs-Kd(490) algorithm [9] available from GlobColour (ftp://ftp.hermes.acri.fr/GLOB/merged/day/);
(v) Kd(490)M-L05a is derived from MODIS-Aqua Rrs, based on the same algorithm as Kd(490)GC-L05a; and
(vi) Kd(490)M-L13 is derived from MODIS-Aqua Rrs, based on a similar but updated semi-analytical algorithm [30]. Compared to L05a, the L13 algorithm added a new input: the backscattering coefficient of pure sea water (bbw(490)). Here we take it as a constant, 0.001387 m−1, which corresponds to the sea-surface water with temperature at 20 °C and salinity at 35 psu, based on [44]. We have tried to use the Argo-observed temperature and salinity to compute a more accurate bbw(490) following [44], but found no obvious improvement for Kd(490) retrieval, when compared to using a constant value.

2.2.3. Satellite-Based Kd(PAR)z

Based on radiative-transfer computations with Hydrolight [2], Lee et al. [10] proposed an IOPs-Kd(PAR) algorithm, which can be used to estimate the layer-averaged Kd(PAR) at any depth, with the inputs of remotely-sensed surface a(490) and bb(490). In this study, we evaluate Kd(PAR)M-L05b at six optical depths (from zpd to 6zpd).

2.2.4. Satellite-Based zeu and z0.415

Two euphotic layer depth algorithms are evaluated here:
(i) zeu-L07, is an operational MODIS product provided by NASA (https://oceandata.sci.gsfc.nasa.gov/MODIS-Aqua/Mapped/Daily/4km/Zeu_lee/), based on [32], which is an implementation of the IOPs-Kd(PAR) model [10];
(ii) zeu-M07, is derived from MODIS-retrieved Chlasat, based on an empirical Chla-zeu relationship proposed by Morel et al. [29]:
log 10 ( z eu - M 07 ) = 1.524 0.436 X 0.0145 X 2 + 0.0186 X 3
Here, X = log10(Chlasat).
For satellite-based z0.415 products, two products are computed:
(i) z0.415-B10, is derived following [31], based on PARsat and Morel’s euphotic layer depth (zeu-M07):
z 0.415 - B 10 = ln ( 0.415 / PAR sat / α ) × ( z eu - M 07 / ln ( 0.01 ) )
While α = 0.98 in [31], in this study we use the look-up table provided by Mobley and Boss [33] which takes into account the sun angle and a wind ruffled surface.
(ii) z0.415-L07, is calculated following [32], whose method for zeu, can also be used for z0.415.

2.2.5. Satellite-Float Matchup Criteria

For each float profile, the median value of the corresponding satellite data within a 3 × 3 pixel box centered on the profile’s surface position was used if at least 5 values in the box were available and that the profile was within 3 hrs of satellite overpass (Figure 3), consistent with the criteria of Bailey and Werdell [45].

2.3. Statistical Metrics

Three statistical metrics are used to evaluate the differences between products derived from the float database and satellite products: (1) Mean absolute difference (MAD), which represents the absolute differences between the measured (observation) values, and satellite-derived or model-estimated values; (2) Mean absolute percentage difference (MAPD), which represents the relative differences; (3) and Mean percentage difference (MPD), which represents the relative system bias. They are defined as:
MAD = 1 n i = 1 n | E i M i |
MAPD = 100 n i = 1 n | E i M i | M i   ( % )
MPD = 100 n i = 1 n ( E i M i ) M i   ( % )
Here, Mi is the in-situ measured value, Ei represents the satellite-retrieved or model-estimated value, and n is the number of observations.

3. Results and Discussion

3.1. Distribution of Kd(490)zpd, zeu, and z0.415 for the BGC-Argo Dataset

In the BGC-Argo dataset, Kd(490)zpd varies by over an order of magnitude, from 0.02 m−1 to 0.3 m−1 (Figure 4a). Regionally, the median Kd(490)zpd is observed to be lowest in subtropical gyres (0.024 m−1) and highest in the Black Sea (0.112 m−1), with the highest dynamic ranges in the Western Mediterranean (Med.) Sea, Southern Ocean, and subpolar regions (Figure 4b). zeu spans about an order of magnitude in this dataset, from ~ 25 m in the (North Atlantic) subpolar gyre to ~ 250 m in subtropical gyres (Figure 4c), with a median value being shallowest (34.0 m) in the Black Sea (Figure 4d) where high CDOM concentration contributes significantly to attenuation [46]. z0.415 has a similar dynamic range and regional distribution as zeu (Figure 4e,f) with median values varying from 23.5 m in the Black Sea to 118.5 m in the subtropical gyres. It exhibits a few very shallow values (<10 m, see Figure 4e) corresponding to very cloudy days (daily PAR above sea surface are very low, e.g., <2 mol photons m−2 d−1). In fact, z0.415 is often shallower than zeu, because the highest daily PAR at sea surface is about 70–80 mol photons m−2 d−1 for a clear summer day [33], and then the corresponding PAR(zeu) is about 0.7–0.8 mol photons m−2 m−1, slightly higher than 0.415 mol photons m−2 m−1 (as light intensity in water decreases with depth exponentially, about 5.2 m (6.6 m) are needed to decrease PAR from 0.7 (0.8) mol photons m−2 m−1 to 0.415 mol photons m−2 m−1 for a moderate Kd(PAR) as 0.1 m−1). Thus, for a clear summer day, z0.415 may be deeper than zeu by about 5 m; but when clouds cause highly depressed daily PAR values, or in other seasons, zeu is deeper than z0.415. In the dataset analyzed here, the median zeu is always deeper than the median z0.415 in all regions (Figure 4d,f), by 3.64 m in subtropical gyres and by 14.7 m in the Southern Ocean.

3.2. Assessment of Satellite Algorithms for Kd(490)

The performance of classical B/G algorithm (Kd(490)M-KD2M) and semi-analytical algorithm (Kd(490)M-L13) outperform the others with this dataset (Figure 5a,b, Table 3). The former has the lowest relative differences (MAPD = 14.1%) and no significant mean bias (MPD = −0.3%), but it has a few points overestimated in the high-value range (>0.2 m−1); the latter has the lowest absolute differences (MAD = 0.009 m−1) and it does not exhibit the overestimation issue at high-values as the B/G algorithm. Note, however, that both algorithms underestimate Kd(490) in the Black Sea. It is likely due to high colored dissolved organic matter (CDOM) concentration in this region [16,46], which affects the performance of both band-ratio and semi-analytical algorithms that have been devised using open-ocean data. Kd(490)M-L05a (Figure 4c) also performs well in these highly attenuating waters with similar but slightly worst results than Kd(490)M-L13. Chla-based Kd(490)M-M07 has the highest MAPD (19.1%) among all 4 MODIS-Aqua products, with an overestimation bias (MPD = 8.7%) mainly in oligotrophic waters (Figure 4d). As for the GlobColour products, Kd(490)GC-L05a exhibits an obvious overestimation in all waters (Figure 4e), having high relative and absolute differences (MAPD = 52.4%, MPD = 51.5%). Comparing MODIS-Aqua and GlobColour with the same Chla-based empirical algorithm (M07), we find little difference (Table 3); however, for the semi-analytical algorithm (L05a), Kd(490)M-L05a, is much closer to float-observed Kd(490)zpd, than Kd(490)GC-L05a, suggesting that the overestimation of Kd(490)GC-L05a is associated with GlobColour, rather than the model of Lee et al. itself.
In addition, we find that the relationship of float-observed Kd(490)zpd and Chlasat (Figure 5f) is different than that in [29], having a larger scale factor and a higher exponent (Figure 5f and Equation (11)):
K d ( 490 ) zpd = 0.0166 + ( 0.1056 ± 0.002 ) × Chla sat ( 0.886 ± 0.010 )   ( R 2 = 0.94 )
We applied this relationship to retrieve Kd(490)zpd from Chlasat and found it to have similar statistical differences to Kd(490)M-KD2M (MAD = 0.013 m−1, MAPD = 15.1%, MPD = 2.1%). Although our results show that the updated Chla-Kd(490) equation has limited improvement on the Kd(490) retrieval from satellite, it suggests the need to re-consider the bio-optical relationship with more in-situ data.
In summary, the GlobColour-provided semi-analytical Kd(490) product seems to have a retrieval problem while the classical B/G-based KD2M algorithm performs best except in highly turbid waters. The semi-analytical algorithm [30] also performs well in the open ocean, and even better in the high-value range as it is designed for solving both clear and turbid waters.

3.3. Assessment of the Satellite Algorithm for Kd(PAR)

The IOPs-Kd(PAR) algorithm [10] was the only one to date to estimate the layer-averaged Kd(PAR) at any depth. It was developed on the basis of Hydrolight simulation to provide a layer-averaged Kd(PAR) at any depth from remote sensing. However, the Hydrolight simulation conducted by Lee et al. [10] assumed a homogeneous distribution of IOPs in the water column. Since the surface a(490) and bb(490) only represent the IOPs within the upper mixed layer, changes of IOPs below the mixed layer depth (MLD) are not accounted for in the IOPs-Kd(PAR) algorithm. Satellite-retrieved Kd(PAR)z-L05b is evaluated with the BGC-Argo dataset from zpd to 6zpd (Figure 6 and Table 4). It performs well at all six optical depths in the high latitude (e.g., Southern Ocean, Subpolar Gyre, and Arctic Ocean), where most water-columns are well-mixed. However, in the stratified waters (including almost all profiles in subtropical gyres, most in the Eastern Mediterranean Sea, and many in the Western Mediterranean Sea), the algorithm increasingly and gradually underestimates Kd(PAR) from 2zpd to 4zpd (from 5.7% to 0.8% of relative light intensity for PAR). In subtropical gyres, MPD decreases from 4.4% at zpd to −14.6% at 4zpd (Table 4). Since Chla increases with depth from MLD to the deep chlorophyll maximum (DCM) in oligotrophic waters [47], both a(490) and bb(490) are expected to increase with depth as well. As a consequence, the IOPs-Kd(PAR) algorithm inevitably underestimates Kd(PAR), even if the QAA algorithm could retrieve sea-surface a(490) and bb(490) correctly. zeu corresponds roughly to 3–4 times zpd (see Appendix B), and is near the DCM in oligotrophic waters [4,47]. It follows that the largest underestimation of Kd(PAR) is at 4zpd. At 5zpd and 6zpd (correspondingly 0.43% and 0.28% of relative light intensity for PAR), there are fewer valid BGC-Argo observations and the underestimation issue is not clear. In addition, given that below the DCM Chla, a(490) and bb(490) are observed to decrease with depth, the underestimation of Kd(PAR) at such depths will weaken.

3.4. Assessment of Satellite Algorithms for zeu and z0.415

Next, we evaluate the satellite products of two important light-level horizons, the euphotic layer depth zeu, and isolume depth z0.415, which represent a relative and an absolute light horizon, respectively. Both zeu and z0.415 have optical and biological significance, and have been used to estimate primary production [48], retrieve the vertical distributions of Chla [47], and understand the mechanisms on formation and variability of the DCM [17,18]. Based on the BGC-Argo dataset, the Chla-based zeu algorithm (zeu-M07) performs better, with the lower absolute and percentage differences of 8.0 m and 12.5%, respectively (Figure 7a), than the IOPs-based one (Figure 7b). Regionally, zeu-M07 exhibits a slight underestimation in the subtropical gyres (MPD = −8.4%) and the eastern Mediterranean Sea, and an overestimation in the Black Sea where CDOM deviates significantly from its global relationship to Chla [16,49]. The CDOM index (relative CDOM concentration to the global mean CDOM-Chla relationship) is extremely high in the Black Sea and low in oligotrophic gyres [50]. Therefore, the Black Sea (Subtropics) is expected to have a shallower (deeper) zeu than the empirical Chla-zeu relationship would provide (High CDOM shoals zeu and low CDOM deepens zeu). The overall performance of zeu-L07 is still satisfactory (Figure 7b), with MAD of 14.0 m, but with a bias (MPD = 8.4%) which appears in the clearest waters (zeu > 80 m). In subtropical gyres, the overestimation is most remarkable, with MPD reaching 12.0% (Table 4). This issue is likely due to the constant IOPs with depth assumed in this model design which does not include a DCM, consistent with the underestimation of Kd(PAR)-L05b at 3zpd and 4zpd.
We assess two satellite z0.415 algorithms (Figure 7c,d), which to our knowledge has never been done, although the equation (Equation 7) proposed by Boss and Behrenfeld [31] has been used in several subsequent studies (e.g., [51,52,53,54]). Overall, the patterns of two scatter plots of z0.415 (Figure 7c,d) are similar to those for zeu (Figure 7a,b), with similar statistics (Table 4). z0.415-B10 displays a slight but obvious underestimation in Subtropical Gyres and Eastern Mediterranean Sea as it is based on zeu-M07 (Figure 7c) and z0.415-L07 exhibits the same overestimation in subtropical gyres (Figure 7d) as zeu-L07 (Figure 7b).
In summary, the evaluation of satellite-retrieved zeu and z0.415 suggests that the Chla-based algorithms outperform the IOPs-based ones, with lower scatter and lower relative system bias. Both IOPs-based zeu and z0.415 exhibit the overestimation in subtropical gyres due to the underestimation of Kd(PAR)z-L05b at depth.

4. Final Remarks and Conclusions

The synergy and joint use of BGC-Argo and satellite remote sensing data contribute to studies using both observing assets [55]. BGC-Argo floats provide the largest dataset for validation and evaluation of satellite products in the global ocean, extend the satellite ocean color observations from surface to depth, and fill missing data in satellite coverage due to low sun angle, high latitude winters and clouds; remote sensing is helpful in guiding the deployment of BGC-Argo floats (e.g., in the subtropical gyres or seasonal bloom regions), identifying the spatial scale of float-observed phenomena (basin-, meso- or submeso-scale), extending the BGC-Argo observation from discrete locations to continuous temporal and spatial distributions, and even as a method for calibration of chlorophyll fluorometers deployed on floats (e.g., [23]).
In this study, we present the use of BGC-Argo data to assess existing satellite products. First, for the MODIS-Aqua Kd(490) products, both the B/G algorithm and the semi-analytical algorithm [30] perform well, in the open ocean. Moreover, our dataset shows, the B/G algorithm, while having the lowest bias in the open ocean, exhibits an overestimation at large-values region (Kd(490) > 0.2 m−1), not exhibited by the semi-analytical one. All Kd(490) algorithms underestimate in the Black Sea due to extremely high CDOM. The GlobColour Kd(490)GC-L05a product has an abnormal overestimation for all the data, which is likely not related to Lee et al.’s model embedded in it. The agreement between float and satellite platforms can be regarded as a “consistency check”: on one hand, it provides the validation of existing products, and on the other hand, it suggests that the quality-controlled float-observed Ed(490) [26] is of high quality.
Kd(PAR) is a critical variable for the retrieval of zeu, and z0.415, and is also useful for marine ecosystem modeling (e.g., [56]) and the estimation of biological heat effect on the upper-layer oceans [5]. However, Kd(PAR) is affected by differences in diffuse attenuation coefficient at different wavelengths. Near the surface, the decrease of PAR is mainly dominated in the open ocean by the losses at the red band due to water absorption [57]. With the depth increasing, as red and near-UV light nearly disappear, the green and blue bands make increasing contributions to Kd(PAR). Finally, Kd(PAR) at deep waters will be close to Kd of the blue/green band (440–510 nm), which is most penetrative in the clear waters [5]. Owing to its spectral sensitivity, the layer-averaged Kd(PAR)z at a certain depth (z) is not a linear average of Kd(λ)z spectra, but a weighted average, and the weight varies as function of both downwelling irradiance spectra (Ed(z)) and attenuation spectra Kd(λ)z [57]. This characteristic makes devising an algorithm for it challenging close to the sea surface [58]. While, in the deeper waters, the vertical change of IOPs becomes the main error sources of retrieval algorithm of Kd(PAR)z. Subsurface structure of IOPs can vary for the same surface conditions, therefore, it is difficult to predict the whole PAR profile accurately, when only relying on the sea-surface IOPs information. Even so, overall, the IOPs-based model [10] performs well near the sea surface until ~2zpd, and also estimates the vertical change of Kd(PAR) well in well-mixed waters, but underestimate remarkably below 2zpd in stratified waters (including almost all profiles in Subtropical Gyres, most in the Eastern Mediterranean Sea, and quite a few in the Western Mediterranean Sea), due to the presence of a DCM not accounted for in its design. In turn, affected by the underestimation of Kd(PAR) below 2zpd, both IOPs-derived zeu and z0.415 exhibit a bias in the oligotrophic waters. Chla-based algorithms to estimate zeu and z0.415 [29,31] perform better although both exhibit some underestimation in subtropical gyres and eastern Mediterranean Sea.
Besides the validation of remote sensing products performed here (and the identification of the significant bias associated with some regions), the statistics associated with our validation can be used to assign errors to the remote sensing products, for example, to propagate them when used in models. For example, many primary-production algorithms parametrize light attenuation using the euphotic depth or include the diffuse attenuation of light attenuation directly (e.g., [48,59]). Simple error propagation (e.g., [60]) can provide the uncertainty in primary production resulting from their uncertainties.

Author Contributions

Conceptualization, X.X.; Methodology, X.X., E.B.; Investigation, X.X., E.B.; Data Curation, X.X., J.Z.; Writing—Original Draft Preparation, X.X.; Writing—Review & Editing, E.B.; Visualization, J.Z.; Funding Acquisition, F.C., E.B. All authors have read and agreed to the published version of the manuscript.

Funding

This This research was funded by the Scientific Research Fund of Pilot National Laboratory for Marine Science and Technology (Qingdao) [grant number QNLM2016ORP0103], the Scientific Research Fund of the Second Institute of Oceanography, MNR, China [grant number QNYC1702, 14283], the Marine S&T Fund of Shandong Province for Pilot National Laboratory for Marine Science and Technology (Qingdao) [grant number 2018SDKJ0206], and NASA’s Ocean Biology and Biogeochemistry program.

Acknowledgments

The authors are grateful to Hervé Claustre (Laboratoire d’Océanographie de Villefranche, Sorbonne Université, Villefranche-sur-Mer, France) for providing the radiometry data of BIOSOPE cruise, all the BGC-Argo data providers, and the principal investigators of related BGC-Argo float missions and projects.

Conflicts of Interest

The authors declare no conflict of interest.

Appendix A. Independent Evaluation of Processing Methods of BGC-Argo Data

In Figure A1, the processing methods of BGC-Argo radiometry data are validated with traditional ship-borne measurements in the Biogeochemistry and Optics South Pacific Experiment (BIOSOPE) cruise [61], which was conducted in the Southeast Pacific during October-December, 2004. Its radiometry dataset had 39 spectral-irradiance profiles recorded by a Satlantic profiler and corresponding surface irradiance measurements recorded by the Satlantic TSRB (Tethered Spectral Radiometer Buoy), spanning various trophic environments, from eutrophic (west of Marquesas Island, and the upwelling conditions off Chile) to ultra-oligotrophic (center of South Pacific subtropical gyre). We first conducted the same quality-control procedure as Organelli et al. [26] on all BIOSOPE radiometry data, removing the noisy profiles and points, then the measured irradiance values above the sea surface (Es) are converted to the ones just below sea surface Ed(0), by multiplying the transmission coefficient α [33], and using a similar procedure to obtain iPAR(0). Following the same steps as Figure 2, zpd, zeu, Kd(490), and Kd(PAR) are calculated. All these values are denoted as “measured”, as they are obtained based on measured Es. We follow the same extrapolation methods mentioned above to estimate Ed(490,0) and iPAR(0) from below water measurements. Then, extrapolation-based derived zpd, zeu, Kd(490), and Kd(PAR) are denoted as “estimated”.
After quality control procedures similar to those used for radiometry measured with floats, there remain 23 Ed(490) and 21 iPAR profiles from BIOSOPE for validation (due to the requirement of at least 5 samples within the top 10 m). We find no obvious differences between the linear and non-linear extrapolation methods for Ed(490), although both have a slight overestimation (MPD = 6.1% and 4.9%) (Figure A1a). Since the attenuation coefficient at 490 nm is mainly determined by IOPs, Kd(490) is expected to vary little within the top 10 m (it will vary some due to the adjustment in the mean-cosine at that wavelength and Raman scattering). Thus, ln(Ed(490)) is expected to be a nearly linear function of depth. As for iPAR, a second-degree polynomial extrapolation is necessary due to the strong attenuation of red wavelengths especially in the top 10 m just below the sea surface due to strong attenuation of red wavelengths [57], causing the linear extrapolation to yield a significant underestimation (MPD = −15.6%, Figure A1b). Derived zpd, zeu, Kd(490)zpd, and Kd(PAR)zpd are computed to evaluate the error due to propagation of estimated sea surface Ed(490,0) and iPAR(0) (Figure A1c–f). The influence of extrapolation is very limited, although Kd(490)zpd and Kd(PAR)zpd exhibit a slight bias.
We conclude from this validation exercise that our extrapolation methods to obtain iPAR(0) and Ed(490,0) and computing zpd, zeu, Kd(490)zpd, and Kd(PAR)zpd from radiometers on profiling floats to have well constrained and small uncertainties.
Figure A1. Evaluation of BGC-Argo data processing methods suggested here using the BIOSOPE data. Scatter plot of Measured and Estimated values, of Ed(490,0) (a) iPAR(0), (b) zpd, (c) zeu, (d) Kd(490)zpd, (e) and Kd(PAR)zpd, (f) respectively. “Measured (meas.)” means determination of Ed(490,0) and iPAR(0) based on measured radiometry above sea surface; “Estimated (Est.)” means determination of Ed(490,0) and iPAR(0) is based on a linear (blue) or second-degree polynomial extrapolation (red) within the upper10 m of the ocean. The black solid lines represent the 1:1 lines.
Figure A1. Evaluation of BGC-Argo data processing methods suggested here using the BIOSOPE data. Scatter plot of Measured and Estimated values, of Ed(490,0) (a) iPAR(0), (b) zpd, (c) zeu, (d) Kd(490)zpd, (e) and Kd(PAR)zpd, (f) respectively. “Measured (meas.)” means determination of Ed(490,0) and iPAR(0) based on measured radiometry above sea surface; “Estimated (Est.)” means determination of Ed(490,0) and iPAR(0) is based on a linear (blue) or second-degree polynomial extrapolation (red) within the upper10 m of the ocean. The black solid lines represent the 1:1 lines.
Remotesensing 12 02367 g0a1

Appendix B. Should We Use zpd = zeu/4.6?

Organelli et al. [15] proposed a method to derived Kd(490)zpd from BGC-Argo data by first determining zeu from iPAR profile as we do here (Section 2.1), then determining zpd based on the proposed equation: zpd = zeu/4.6 (derived based on the fact that -ln(0.01) ≈ 4.6). Finally, they applied a linear regression between ln(Ed(490)) and depth to retrieve Kd(490)zpd. In contrast, here we use Equation (3) to compute zpd.
When we follow their method to derive Kd(490)zpd-O17, it displays a slight underestimation in oligotrophic waters in comparison to ours (Figure A2a). The relationship zpd = zeu/4.6 is likely biased in most oligotrophic waters. On one hand, the diffuse attenuation of visible light is lowest in the blue band for such waters [62], and hence zeu (the 1% light depth for PAR) must be shallower than z1%490 (the 1% light depth for Ed(490), i.e., zeu < z1%490 in most waters [57], as shown in Figure A2b). On the other hand, from Equation (1):
K d ( 490 ) z 1 % 490 × z 1 % 490 = ln ( 0.01 ) × K d ( 490 ) zpd × z pd
which implies that the ratio of z1%490 to zpd could reach –ln(0.01) (approximately 4.6) only when Kd(490) is uniform from surface to z1%490 (i.e., Kd(490)z1%490 = Kd(490)zpd). However, due to the deep chlorophyll-a maximum (DCM) observed in highly clear and stratified waters [47], one would expect Kd(490) to increase with depth to the DCM depth, which means z1%490/4.6 < zpd (Figure A2c). It follows that zeu/4.6 < zpd in highly stratified (oligotrophic) waters, as shown in Figure A2d, consistent with the observed bias in [15] obtained in such waters (Figure A2a). Statistically, for all valid 2272 samples in our BGC-Argo dataset (Figure A2d), zeu = (3.53 ± 0.83) × zpd. Regionally, the ratio of (zeu/zpd) reaches its highest values in the North Atlantic subpolar gyre (4.05 ± 0.83), and its lowest values in the subtropical gyres (3.05 ± 0.46) and New Caledonia (2.98 ± 0.60). We note that Lee et al. [30] also showed that, z1% of blue light (the arithmetic average of z1%412, z1%443, z1%488, and z1%531) was generally 30%–40% deeper than zeu in open oceans.
Figure A2. Comparison of different methodology of Kd calculation. (a) float-observed Kd(490)zpd-O17 following the methodology of [15] vs. our determined Kd(490)zpd; (b) zeu vs. z1%490; (c) z1%490/4.6 vs. zpd; (d) zeu/4.6 vs. zpd.
Figure A2. Comparison of different methodology of Kd calculation. (a) float-observed Kd(490)zpd-O17 following the methodology of [15] vs. our determined Kd(490)zpd; (b) zeu vs. z1%490; (c) z1%490/4.6 vs. zpd; (d) zeu/4.6 vs. zpd.
Remotesensing 12 02367 g0a2

References

  1. Kirk, J.T.O. A theoretical analysis of the contribution of algal cells to the attenuation of light within natural waters. III. Cylindrical and spheroidal cells. New Phytol. 1976, 77, 341–358. [Google Scholar] [CrossRef]
  2. Mobley, C.D. Light and Water: Radiative Transfer in Natural Waters; Elsevier: New York, NY, USA, 1994. [Google Scholar]
  3. Ryther, J.H. Photosynthesis in the ocean as a function of light intensity. Limnol. Oceanogr. 1956, 1, 61–70. [Google Scholar] [CrossRef]
  4. Letelier, R.M.; Karl, D.M.; Abbott, M.R.; Bidigare, R.R. Light driven seasonal patterns of chlorophyll and nitrate in the lower euphotic zone of the North Pacific Subtropical Gyre. Limnol. Oceanogr. 2004, 49, 508–519. [Google Scholar] [CrossRef] [Green Version]
  5. Chen, J.; Zhang, X.; Xing, X.; Ishizaka, J.; Yu, Z. A spectrally selective attenuation mechanism-based Kpar algorithm for biomass heating effect simulation in the open ocean. J. Geophys. Res. Oceans 2017, 122, 9370–9386. [Google Scholar] [CrossRef] [Green Version]
  6. Pimentel, S.; Tse, W.-H.; Xu, H.; Denaxa, D.; Jansen, E.; Korres, G.; Mirouze, I.; Storto, A. Modeling the near-surface diurnal cycle of sea surface temperature in the Mediterranean Sea. J. Geophys. Res. Oceans 2019, 124, 171–183. [Google Scholar] [CrossRef] [Green Version]
  7. Skákala, J.; Bruggeman, J.; Brewin, R.J.W.; Ford, D.A.; Ciavatta, S. Improved representation of underwater light field and its impact on ecosystem dynamics: A study in the North Sea. J. Geophys. Res. Oceans 2020, 125, e2020jc016122. [Google Scholar] [CrossRef]
  8. Frouin, R.; McPherson, J.; Ueyoshi, K.; Franz, B.A. A time series of photosynthetically available radiation at the ocean surface from SeaWiFS and MODIS data. In Proceedings of the SPIE Asia-Pacific Remote Sensing; SPIE-Intl Soc Optical Eng, San Jose, CA, USA, 12–16 February 2012; p. 85251. [Google Scholar] [CrossRef]
  9. Lee, Z.; Du, K.; Arnone, R. A model for the diffuse attenuation coefficient of downwelling irradiance. J. Geophys. Res. Oceans 2005, 110, C02016. [Google Scholar] [CrossRef]
  10. Lee, Z.; Du, K.; Arnone, R.; Liew, S.C.; Penta, B. Penetration of solar radiation in the upper ocean—A numerical model for oceanic and coastal waters. J. Geophys. Res. Oceans 2005, 110, C09019. [Google Scholar] [CrossRef]
  11. Johnson, K.S.; Claustre, H. Bringing Biogeochemistry into the Argo Age. Eos 2016, 97, 11–15. [Google Scholar] [CrossRef]
  12. Claustre, H.; Johnson, K.S.; Takeshita, Y. Observing the Global Ocean with Biogeochemical-Argo. Annu. Rev. Mar. Sci. 2020, 12, 23–48. [Google Scholar] [CrossRef]
  13. Biogeochemical-Argo Planning Group. The Scientific Rationale, Design, and Implementation Plan for a Biogeochemical-Argo Float Array; Johnson, K., Claustre, H., Eds.; Ifremer: Issy-les-Moulineaux, France, 2016. [Google Scholar] [CrossRef]
  14. Roemmich, D.; Alford, M.H.; Claustre, H.; Johnson, K.; King, B.; Moum, J.; Oke, P.; Owens, W.B.; Pouliquen, S.; Purkey, S.; et al. On the Future of Argo: A Global, Full-Depth, Multi-Disciplinary Array. Front. Mar. Sci. 2019, 6, 439. [Google Scholar] [CrossRef] [Green Version]
  15. Organelli, E.; Barbieux, M.; Claustre, H.; Schmechtig, C.; Poteau, A.; Bricaud, A.; Boss, E.; Briggs, N.; Nencioli, F.; D’Ortenzio, F.; et al. Two databases derived from BGC-Argo float measurements for marine biogeochemical and bio-optical applications. Earth Syst. Sci. Data 2017, 9, 861–880. [Google Scholar] [CrossRef] [Green Version]
  16. Organelli, E.; Claustre, H.; Bricaud, A.; Barbieux, M.; Uitz, J.; D’Ortenzio, F.; Dall’Olmo, G. Bio-optical anomalies in the world’s oceans: An investigation on the diffuse attenuation coefficients for downward irradiance derived from Biogeochemical Argo float measurements. J. Geophys. Res. Oceans 2017, 122, 3543–3564. [Google Scholar] [CrossRef] [Green Version]
  17. Mignot, A.; Claustre, H.; Uitz, J.; Poteau, A.; D’Ortenzio, F.; Xing, X. Understanding the seasonal dynamics of phytoplankton biomass and the deep chlorophyll maximum in oligotrophic environments: A Bio-Argo float investigation. Glob. Biogeochem. Cycles 2014, 28, 856–876. [Google Scholar] [CrossRef]
  18. Barbieux, M.; Uitz, J.; Gentili, B.; De Fommervault, O.P.; Mignot, A.; Poteau, A.; Schmechtig, C.; Taillandier, V.; Leymarie, E.; Penkerch, C.; et al. Bio-optical characterization of subsurface chlorophyll maxima in the Mediterranean Sea from a Biogeochemical-Argo float database. Biogeosciences 2019, 16, 1321–1342. [Google Scholar] [CrossRef] [Green Version]
  19. Terzić, E.; Lazzari, P.; Organelli, E.; Solidoro, C.; Salon, S.; D’Ortenzio, F.; Conan, P. Merging bio-optical data from Biogeochemical-Argo floats and models in marine biogeochemistry. Biogeosciences 2019, 16, 2527–2542. [Google Scholar] [CrossRef] [Green Version]
  20. Xing, X.; Morel, A.; Claustre, H.; Antoine, D.; D’Ortenzio, F.; Poteau, A.; Mignot, A. Combined processing and mutual interpretation of radiometry and fluorimetry from autonomous profiling Bio-Argo floats: Chlorophyll a retrieval. J. Geophys. Res. Oceans 2011, 116, C06020. [Google Scholar] [CrossRef]
  21. Xing, X.; Morel, A.; Claustre, H.; D’Ortenzio, F.; Poteau, A. Combined processing and mutual interpretation of radiometry and fluorometry from autonomous profiling Bio-Argo floats: 2. Colored dissolved organic matter absorption retrieval. J. Geophys. Res. Oceans 2012, 117, C04022. [Google Scholar] [CrossRef] [Green Version]
  22. Xing, X.; Briggs, N.; Boss, E.; Claustre, H. Improved correction for non-photochemical quenching of in situ chlorophyll fluorescence based on the synchronous irradiance profile. Opt. Express 2018, 26, 24734–24751. [Google Scholar] [CrossRef]
  23. Boss, E.; Swift, D.; Taylor, L.; Brickley, P.; Zaneveld, R.; Riser, S.; Perry, M.J.; Strutton, P.G. Observations of pigment and particle distributions in the western North Atlantic from an autonomous float and ocean color satellite. Limnol. Oceanogr. 2008, 53, 2112–2122. [Google Scholar] [CrossRef] [Green Version]
  24. Johnson, K.S.; Plant, J.N.; Dunne, J.P.; Talley, L.D.; Sarmiento, J.L. Annual nitrate drawdown observed by SOCCOM profiling floats and the relationship to annual net community production. J. Geophys. Res. Oceans 2017, 122, 6668–6683. [Google Scholar] [CrossRef] [Green Version]
  25. Haëntjens, N.; Boss, E.; Talley, L.D. Revisiting Ocean Color algorithms for chlorophyll a and particulate organic carbon in the Southern Ocean using biogeochemical floats. J. Geophys. Res. Oceans 2017, 122, 6583–6593. [Google Scholar] [CrossRef] [Green Version]
  26. Organelli, E.; Claustre, H.; Bricaud, A.; Schmechtig, C.; Poteau, A.; Xing, X.G.; Prieur, L.; D’Ortenzio, F.; Dall’Olmo, G.; Vellucci, V. A Novel Near-Real-Time Quality-Control Procedure for Radiometric Profiles Measured by Bio-Argo Floats: Protocols and Performances. J. Atmos. Oceans Technol. 2016, 33, 937–951. [Google Scholar] [CrossRef]
  27. Barbieux, M.; Organelli, E.; Claustre, H.; Schmechtig, C.; Poteau, A.; Boss, E.; Bricaud, A.; Briggs, N.; Dall’Olmo, G.; D’Ortenzio, F.; et al. A global database of vertical profiles derived from Biogeochemical Argo float measurements for biogeochemical and bio-optical applications. SEANOE 2017. [Google Scholar] [CrossRef]
  28. NASA (2020). Available online: https://oceancolor.gsfc.nasa.gov/atbd/kd_490/ (accessed on 22 July 2020).
  29. Morel, A.; Huot, Y.; Gentili, B.; Werdell, P.J.; Hooker, S.B.; Franz, B.A. Examining the consistency of products derived from various ocean color sensors in open ocean (Case 1) waters in the perspective of a multi-sensor approach. Remote. Sens. Environ. 2007, 111, 69–88. [Google Scholar] [CrossRef]
  30. Lee, Z.; Hu, C.; Shang, S.; Du, K.; Lewis, M.; Arnone, R.; Brewin, R. Penetration of UV-visible solar radiation in the global oceans: Insights from ocean color remote sensing. J. Geophys. Res. Oceans 2013, 118, 4241–4255. [Google Scholar] [CrossRef] [Green Version]
  31. Boss, E.; Behrenfeld, M. In situ evaluation of the initiation of the North Atlantic phytoplankton bloom. Geophys. Res. Lett. 2010, 37, 18603. [Google Scholar] [CrossRef] [Green Version]
  32. Lee, Z.; Weidemann, A.; Kindle, J.; Arnone, R.; Carder, K.L.; Davis, C. Euphotic zone depth: Its derivation and implication to ocean-color remote sensing. J. Geophys. Res. Oceans 2007, 112, C03009. [Google Scholar] [CrossRef] [Green Version]
  33. Mobley, C.D.; Boss, E.S. Improved irradiances for use in ocean heating, primary production, and photo-oxidation calculations. Appl. Opt. 2012, 51, 6549–6560. [Google Scholar] [CrossRef]
  34. Lee, Z.; Carder, K.L.; Arnone, R. Deriving inherent optical properties from water color: A multi-band quasi-analytical algorithm for optically deep waters. Appl. Opt. 2002, 41, 5755–5772. [Google Scholar] [CrossRef]
  35. NASA (2020). Available online: https://oceancolor.gsfc.nasa.gov/atbd/chlor_a/ (accessed on 22 July 2020).
  36. Gordon, H.R.; McCluney, W.R. Estimation of the depth of sunlight penetration in the sea for remote sensing. Appl. Opt. 1975, 14, 413–416. [Google Scholar] [CrossRef]
  37. Cetinić, I.; Perry, M.J.; D’Asaro, E.; Briggs, N.; Poulton, N.; Sieracki, M.E.; Lee, C.M. A simple optical index shows spatial and temporal heterogeneity in phytoplankton community composition during the 2008 North Atlantic Bloom Experiment. Biogeosciences 2015, 12, 2179–2194. [Google Scholar] [CrossRef] [Green Version]
  38. Schulien, J.A.; Behrenfeld, M.J.; Hair, J.W.; Hostetler, C.A.; Twardowski, M.S. Vertically- resolved phytoplankton carbon and net primary production from a high spectral resolution lidar. Opt. Express 2017, 25, 13577. [Google Scholar] [CrossRef]
  39. Thushara, V.; Vinayachandran, P.N.M.; Matthews, A.J.; Webber, B.G.M.; Queste, B.Y. Vertical distribution of chlorophyll in dynamically distinct regions of the southern Bay of Bengal. Biogeosciences 2019, 16, 1447–1468. [Google Scholar] [CrossRef] [Green Version]
  40. Banse, K. Should we continue to use the 1% light depth for estimating the compensation depth of phytoplankton for another 70 years? Limnol. Oceanogr. 2004, 13, 49–52. [Google Scholar] [CrossRef]
  41. Behrenfeld, M.J.; Boss, E.S. Student’s tutorial on bloom hypotheses in the context of phytoplankton annual cycles. Glob. Chang. Biol. 2018, 24, 1–23. [Google Scholar] [CrossRef]
  42. Hu, C.; Lee, Z.; Franz, B. Chlorophyll a algorithms for oligotrophic oceans: A novel approach based on three-band reflectance difference. J. Geophys. Res. Oceans 2012, 117, C01011. [Google Scholar] [CrossRef] [Green Version]
  43. Fanton d’Andon, O.; Mangin, A.; Lavender, S.; Antoine, D.; Maritorena, S.; Morel, A.; Barrot, G.; Demaria, J.; Pinnock, S. GlobColour—The European Service for Ocean Colour. In Proceedings of the 2009 IEEE International Geoscience & Remote Sensing Symposium, Cape Town, South Africa, 12–17 July 2009. [Google Scholar]
  44. Zhang, X.; Hu, L.; He, M.-X. Scattering by pure seawater: Effect of salinity. Opt. Express 2009, 17, 5698–5710. [Google Scholar] [CrossRef] [Green Version]
  45. Bailey, S.W.; Werdell, P.J. A multi-sensor approach for the on-orbit validation of ocean color satellite data products. Remote Sens. Environ. 2006, 102, 12–23. [Google Scholar] [CrossRef]
  46. Churilova, T.; Suslin, V.; Krivenko, O.; Efimova, T.; Moiseeva, N.; Mukhanov, V.; Smirnova, L. Light absorption by phytoplankton in the upper mixed layer of the Black Sea: Seasonality and Parametrization. Front. Mar. Sci. 2017, 4, 90. [Google Scholar] [CrossRef] [Green Version]
  47. Uitz, J.; Claustre, H.; Morel, A.; Hooker, S.B. Vertical distribution of phytoplankton communities in open ocean: An assessment based on surface chlorophyll. J. Geophys. Res. Oceans 2006, 111, C08005. [Google Scholar] [CrossRef]
  48. Behrenfeld, M.J.; Falkowski, P.G. A consumer’s guide to phytoplankton primary productivity models. Limnol. Oceanogr. 1997, 42, 1479–1491. [Google Scholar] [CrossRef] [Green Version]
  49. Morel, A.; Claustre, H.; Gentili, B. The most oligotrophic subtropical zones of the global ocean: Similarities and differences in terms of chlorophyll and yellow substance. Biogeosciences 2010, 7, 3139–3151. [Google Scholar] [CrossRef] [Green Version]
  50. Morel, A.; Gentili, B. A simple band ratio technique to quantify the colored dissolved and detrital organic material from ocean color remotely sensed data. Remote Sens. Environ. 2009, 113, 998–1011. [Google Scholar] [CrossRef]
  51. Brody, S.R.; Lozier, M.S. Characterizing upper-ocean mixing and its effect on the spring phytoplankton bloom with in situ data. Ices J. Mar. Sci. 2015, 72, 1961–1970. [Google Scholar] [CrossRef] [Green Version]
  52. Itoh, S.; Yasuda, I.; Saito, H.; Tsuda, A.; Komatsu, K. Mixed layer depth and chlorophyll a: Profiling float observations in the Kuroshio-Oyashio Extension region. J. Marine Systems 2015, 151, 1–14. [Google Scholar] [CrossRef] [Green Version]
  53. Mayot, N.; D’Ortenzio, F.; Taillandier, V.; Prieur, L.; de Fommervault, O.P.; Claustre, H.; Bosse, A.; Testor, P.; Conan, P. Physical and Biogeochemical Controls of the Phytoplankton Blooms in North Western Mediterranean Sea: A Multiplatform Approach Over a Complete Annual Cycle (2012–2013 DEWEX Experiment). J. Geophys. Res. Oceans 2017, 122, 9999–10019. [Google Scholar] [CrossRef]
  54. Balaguru, K.; Doney, S.C.; Bianucci, L.; Rasch, P.J.; Leung, L.R.; Yoon, J.-H.; Lima, I.D. Linking deep convection and phytoplankton blooms in the northern Labrador Sea in a changing climate. PLoS ONE 2018, 13, e0191509. [Google Scholar] [CrossRef] [Green Version]
  55. Chai, F.; Johnson, K.S.; Claustre, H.; Xing, X.; Wang, Y.; Boss, E.; Riser, S.; Fennel, K.; Schofield, O.; Sutton, A. Monitoring ocean biogeochemistry with autonomous platforms. Nat. Rev. Earth Environ. 2020, 1, 315–326. [Google Scholar] [CrossRef]
  56. Xiu, P.; Chai, F. Connections between physical, optical and biogeochemical processes in the Pacific Ocean. Progr. Oceanogr. 2014, 122, 30–53. [Google Scholar] [CrossRef]
  57. Morel, A. Optical modeling of the upper ocean in relation to its biogenous matter content. J. Geophys. Res. 1988, 93, 10749–10768. [Google Scholar] [CrossRef] [Green Version]
  58. Lee, Z. KPAR: An optical property associated with ambiguous values. J. Lake Sci. 2009, 21, 159–164. [Google Scholar] [CrossRef] [Green Version]
  59. Westberry, T.; Behrenfeld, M.J.; Siegel, D.A.; Boss, E. Carbon-based primary productivity modeling with vertically resolved photoacclimation. Glob. Biogeochem. Cycles 2008, 22, GB2024. [Google Scholar] [CrossRef] [Green Version]
  60. Joint Committee for Guides in Metrology. Evaluation of Measurement Data—Guide to the Expression of Uncertainty in Measurement. Available online: https://ncc.nesdis.noaa.gov/documents/documentation/JCGM_100_2008_E.pdf (accessed on 22 July 2020).
  61. Claustre, H.; Sciandra, A.; Vaulot, D. Introduction to the special section: Bio-optical and biogeochemical conditions in the South East Pacific in late 2004—The BIOSOPE program. Biogeosciences 2008, 5, 679–691. [Google Scholar] [CrossRef] [Green Version]
  62. Morel, A.; Maritorena, S. Bio-optical properties of oceanic waters: A reappraisal. J. Geophys. Res. Oceans 2001, 106, 7163–7180. [Google Scholar] [CrossRef] [Green Version]
Figure 1. BGC-Argo profile data distribution maps used in this study, for all points (a) with valid surface Ed(490,0) (N = 4882), (b) with valid iPAR(0) (N = 2548), (c) with matchup with all satellite Kd(490) products (N = 946), and (d) with matchup with all satellite Kd(PAR) products (N = 374, same for zeu and z0.415 products), respectively.
Figure 1. BGC-Argo profile data distribution maps used in this study, for all points (a) with valid surface Ed(490,0) (N = 4882), (b) with valid iPAR(0) (N = 2548), (c) with matchup with all satellite Kd(490) products (N = 946), and (d) with matchup with all satellite Kd(PAR) products (N = 374, same for zeu and z0.415 products), respectively.
Remotesensing 12 02367 g001
Figure 2. Flow chart of BGC-Argo radiometry data processing. Note that all derived Kd values are layer-averaged ones (the layer from sea surface to depth z, thus, z as the subscript in the symbols, e.g., Kd(490)z), rather than Kd at a specific depth. The same procedure could be used to derive the layer-averaged attenuation coefficient to any depth of interest.
Figure 2. Flow chart of BGC-Argo radiometry data processing. Note that all derived Kd values are layer-averaged ones (the layer from sea surface to depth z, thus, z as the subscript in the symbols, e.g., Kd(490)z), rather than Kd at a specific depth. The same procedure could be used to derive the layer-averaged attenuation coefficient to any depth of interest.
Remotesensing 12 02367 g002
Figure 3. Diagram of a good match-up between satellite and float data. The 9 squares represent 9 pixels closest to the location of float surfacing, with 5 valid pixels (blue) and 4 invalid ones (white). The spatial resolution is 4 km. In such a condition or with more than 5 valid-value pixels out of 9, a pair of valid satellite-float matchup data is obtained.
Figure 3. Diagram of a good match-up between satellite and float data. The 9 squares represent 9 pixels closest to the location of float surfacing, with 5 valid pixels (blue) and 4 invalid ones (white). The spatial resolution is 4 km. In such a condition or with more than 5 valid-value pixels out of 9, a pair of valid satellite-float matchup data is obtained.
Remotesensing 12 02367 g003
Figure 4. Histogram and boxplot of the distribution of the (a,b) near surface diffuse attenuation coefficient at 490 nm (Kd(490)zpd), (c,d) euphotic layer depth (zeu), (e,f) and isolume depth (z0.415), for each of the 10 regions (SO: Southern Ocean; AS: Arctic Sea; SPG: Subpolar Gyre; BS: Black Sea; TZ: Transition Zone; WMS: West Med. Sea; RS: Red Sea; EMS: East Med. Sea; NC: New Caledonia; STG: Subtropical Gyre). In the histogram, black and red columns (as well as numbers) represent the full dataset and the satellite-matched ones, respectively. In the boxplot, red points beyond the end of the whiskers represent outliers beyond the 1.5 × IQR (IQR = interquartile range) threshold.
Figure 4. Histogram and boxplot of the distribution of the (a,b) near surface diffuse attenuation coefficient at 490 nm (Kd(490)zpd), (c,d) euphotic layer depth (zeu), (e,f) and isolume depth (z0.415), for each of the 10 regions (SO: Southern Ocean; AS: Arctic Sea; SPG: Subpolar Gyre; BS: Black Sea; TZ: Transition Zone; WMS: West Med. Sea; RS: Red Sea; EMS: East Med. Sea; NC: New Caledonia; STG: Subtropical Gyre). In the histogram, black and red columns (as well as numbers) represent the full dataset and the satellite-matched ones, respectively. In the boxplot, red points beyond the end of the whiskers represent outliers beyond the 1.5 × IQR (IQR = interquartile range) threshold.
Remotesensing 12 02367 g004
Figure 5. Assessment of satellite Kd(490) products based on BGC-Argo dataset. Scatter plot of float-observed Kd(490)zpd v.s. MODIS-Aqua B/G-based empirical Kd(490)M-KD2M (a) semi-analytical Kd(490)M-L13, (b) semi-analytical Kd(490)M-L05a, (c) Chla-based empirical Kd(490)M-M07, (d) GlobColour semi-analytical Kd(490)GC-L05a, (e) Scatter plot of float-observed Kd(490)zpd vs. MODIS-Aqua-retrieved Chla, (f) black dashed line represents the empirical equation (Equation (5)) used in Kd(490)GC-M07 and Kd(490)M-M07, and the red solid line represents the relationship obtained in this study (Equation (11)). Black solid lines are the 1:1 lines. Note that in Panel (a)–(e) we use only data for which all algorithms could be evaluated.
Figure 5. Assessment of satellite Kd(490) products based on BGC-Argo dataset. Scatter plot of float-observed Kd(490)zpd v.s. MODIS-Aqua B/G-based empirical Kd(490)M-KD2M (a) semi-analytical Kd(490)M-L13, (b) semi-analytical Kd(490)M-L05a, (c) Chla-based empirical Kd(490)M-M07, (d) GlobColour semi-analytical Kd(490)GC-L05a, (e) Scatter plot of float-observed Kd(490)zpd vs. MODIS-Aqua-retrieved Chla, (f) black dashed line represents the empirical equation (Equation (5)) used in Kd(490)GC-M07 and Kd(490)M-M07, and the red solid line represents the relationship obtained in this study (Equation (11)). Black solid lines are the 1:1 lines. Note that in Panel (a)–(e) we use only data for which all algorithms could be evaluated.
Remotesensing 12 02367 g005
Figure 6. Assessment of satellite Kd(PAR) products based on BGC-Argo dataset. Scatter plots of float-observed Kd(PAR) and satellite-derived Kd(PAR)z-L05b [10] at (a) zpd, (b) 2zpd, (c) 3zpd, (d) 4zpd, (e) 5zpd, and (f) 6zpd, respectively. All black solid lines represent 1:1 lines.
Figure 6. Assessment of satellite Kd(PAR) products based on BGC-Argo dataset. Scatter plots of float-observed Kd(PAR) and satellite-derived Kd(PAR)z-L05b [10] at (a) zpd, (b) 2zpd, (c) 3zpd, (d) 4zpd, (e) 5zpd, and (f) 6zpd, respectively. All black solid lines represent 1:1 lines.
Remotesensing 12 02367 g006
Figure 7. Assessment of satellite zeu and z0.415 algorithms through comparison with the BGC-Argo dataset. zeu retrieved from a Chla-based algorithm (zeu-M07; [29]) (a) and from the IOPs-based algorithm (zeu-L07; [32]); (b) z0.415 retrieved from a Chla-based algorithm (z0.415-B10; [31]); (c) and from IOPs-based algorithm (z0.415-L0 [32]); (d) Black solid lines are the 1:1 lines.
Figure 7. Assessment of satellite zeu and z0.415 algorithms through comparison with the BGC-Argo dataset. zeu retrieved from a Chla-based algorithm (zeu-M07; [29]) (a) and from the IOPs-based algorithm (zeu-L07; [32]); (b) z0.415 retrieved from a Chla-based algorithm (z0.415-B10; [31]); (c) and from IOPs-based algorithm (z0.415-L0 [32]); (d) Black solid lines are the 1:1 lines.
Remotesensing 12 02367 g007
Table 1. Regions and Basins Classified in This Dtudy and Corresponding Profile Numbers of Ed(490) and iPAR.
Table 1. Regions and Basins Classified in This Dtudy and Corresponding Profile Numbers of Ed(490) and iPAR.
Region Name in This StudyBasin Name Classified in the SEANOE-BGC-Argo DatasetEd(490) Num *iPAR
Num *
Southern OceanATL (Southern Ocean Atlantic sector)
ATOI (Southern Ocean Atlantic to Indian sector)
IND (Southern Ocean Indian sector)
974 (894)1097 (471)
Subpolar GyreLAS (Labrador Sea)
IRM (Irminger Sea)
ICB (Iceland Basin)
SLAS (South Labrador Sea)
1618 (1465)1610 (885)
Transition ZoneNASTZ (North Atlantic Transition Zone)
EQNASTZ (North Atlantic South Transition Zone)
SASTZ (South Atlantic Transition Zone)
197 (137)205 (83)
Red SeaRED (Red Sea)62 (39)56 (8)
Black SeaBLACK (Black Sea)128 (118)134 (84)
West Mediterranean (Med.) SeaNW (Northwestern)
SW (Southwestern)
TYR (Tyrrhenian Sea)
1121 (939)1141 (472)
East Med. SeaION (Ionian Sea)
LEV (Levantine Sea)
886 (683)801 (206)
Subtropical GyreNASTG (North Atlantic Subtropical Gyre)
ENASTG (Eastern North Atlantic Subtropical Gyre)
WNASTG (Western North Atlantic Subtropical Gyre)
SASTG (South Atlantic Subtropical Gyre)
SSASTG (South Atlantic South Subtropical Gyre)
SPSTG (South Pacific Subtropical Gyre)
549 (505)630 (282)
Arctic SeaNOR (Norwegian Sea)61 (57)83 (39)
New CaledoniaNC (New Caledonia)52 (45)58 (18)
TOTAL 5648 (4882)5815 (2548)
* The numbers in the parentheses represent the numbers of profiles used in this study, with valid surface Ed(490,0) and iPAR(0).
Table 2. Symbols used in this study.
Table 2. Symbols used in this study.
SymbolDefinitionAlgorithmReference
αAir-sea interface transmission factorModeled[33]
a(490)Absorption coefficient at 490 nmQuasi-analytical[34]
bb(490)Backscattering coefficient at 490 nmQuasi-analytical[34]
ChlasatDownloaded MODIS-Aqua chlorophyll-a concentrationEmpirical[35]
Ed(490)Downwelling irradiance at 490 nmObserved[26]
iPARInstantaneous photosynthetically available radiationObserved[26]
Kd(490)M-L05aDerived MODIS-Aqua Kd(490) productSemi-analytical[9]
Kd(490)M-L13Derived MODIS-Aqua Kd(490) productSemi-analytical[30]
Kd(490)M-KD2MDownloaded MODIS-Aqua Kd(490) productEmpirical[28]
Kd(490)M-M07Derived MODIS-Aqua Kd(490) productEquation (5)[29]
Kd(490)GC-L05aDownloaded GlobColour Kd(490) productSemi-analytical[9]
Kd(490)GC-M07Downloaded GlobColour Kd(490) productEquation (5)[29]
Kd(PAR)zLayer-averaged Kd(490) from surface to any depthEquation (1)/
Kd(490)zpdFloat-observed near surface Kd(490)Equation (3)[36]
Kd(490)zpd-O17Float-observed near surface Kd(490)Appendix A[15]
Kd(PAR)z-L05bDerived MODIS-Aqua layer-averaged Kd(PAR) from surface to any depthIOPs-based[10]
Kd(PAR)zLayer-averaged Kd(PAR) from surface to any depthEquation (2)/
PARsatDownloaded MODIS-Aqua daily PAR/[8]
z0.415Isolume depth at 0.415 mol photons m−2 d−1Equation (4)[4]
z0.415-B10Derived MODIS-Aqua z0.415 productEquations (6) and (7)[31]
z0.415-L07Derived MODIS-Aqua z0.415 productIOPs-basedModified from [32]
z1%490Float-observed 1% light depth for Ed(490)Equation (1)/
zeuFloat-observed 1% light depth for iPAREquation (2)[3]
zeu-L07Downloaded MODIS-Aqua zeu productIOPs-based[32]
zeu-M07Derived MODIS-Aqua zeu productEquation (6)[29]
zpdPenetration depth at 490 nmEquation (3)[36]
Table 3. Statistical results of evaluation on the satellite-based Kd(490) products *.
Table 3. Statistical results of evaluation on the satellite-based Kd(490) products *.
ProductPlatformSourceAlgorithmMADMAPDMPDFigure
Kd(490)M-KD2MMODIS-AquaDownloadedB/G-based0.010 m−114.1%−0.3%Figure 5a
Kd(490)M-L13MODIS-AquaCalculatedIOPs-based0.009 m−114.4%2.2%Figure 5b
Kd(490)M-L05aMODIS-AquaCalculatedIOPs-based0.009 m−115.2%4.6%Figure 5c
Kd(490)GC-L05aGlobColourDownloadedIOPs-based0.025 m−152.4%51.5%Figure 5e
Kd(490)M-M07MODIS-AquaCalculatedChla-based0.010 m−119.1%8.7%Figure 5d
Kd(490)GC-M07GlobColourDownloadedChla-based0.011 m−119.5%11.7%/
* We compared 946 matchup values for which all products were available.
Table 4. Statistical results of evaluation on the satellite-based Kd(PAR) and related products (all products compared here are calculated for the MODIS-Aqua platform).
Table 4. Statistical results of evaluation on the satellite-based Kd(PAR) and related products (all products compared here are calculated for the MODIS-Aqua platform).
ProductAlgorithmNumber #MAD#MAPD #MPD #Figure
Kd(PAR)zpd-L05bIOPs-based374(49)0.013 m−1 (0.005 m−1)12.8%(11.1%)8.3%(4.4%)Figure 6a
Kd(PAR)2zpd-L05bIOPs-based374(49)0.011 m−1 (0.004 m−1)12.5%(10.9%)−0.4%(−1.3%)Figure 6b
Kd(PAR)3zpd-L05bIOPs-based374(49)0.012 m−1 (0.006 m−1)14.7%(14.7%)−4%(−10.6%)Figure 6c
Kd(PAR)4zpd-L05bIOPs-based364(40)0.012 m−1 (0.007 m−1)15.2%(17.1%)−2.5%(−14.3%)Figure 6d
Kd(PAR)5zpd-L05bIOPs-based315(14)0.012 m−1 (0.005 m−1)15.1%(12.5%)4.2%(−2.9%)Figure 6e
Kd(PAR)6zpd-L05bIOPs-based264(5)0.014 m−1 (0.006 m−1)17.9%(15.5%)12.6%(1.6%)Figure 6f
zeu-M07Chla-based374(49)8.0 m (12.6 m)12.5%(9.9%)−1.7%(−8.4%)Figure 7a
zeu-L07IOPs-based 374(49)14.0 m (20.5 m)19.7%(16.7%)8.4%(12.0%)Figure 7b
z0.415-B10Chla-based374(49)7.7 m (12.7 m)12.2%(9.8%)−1.1%(−8.2%)Figure 7c
z0.415-L07IOPs-based374(49)16.4 m (32.7 m)21.3%(24.7%)10.2%(20.6%)Figure 7d
# The numbers in the parentheses represent the statistics in subtropical gyres.

Share and Cite

MDPI and ACS Style

Xing, X.; Boss, E.; Zhang, J.; Chai, F. Evaluation of Ocean Color Remote Sensing Algorithms for Diffuse Attenuation Coefficients and Optical Depths with Data Collected on BGC-Argo Floats. Remote Sens. 2020, 12, 2367. https://doi.org/10.3390/rs12152367

AMA Style

Xing X, Boss E, Zhang J, Chai F. Evaluation of Ocean Color Remote Sensing Algorithms for Diffuse Attenuation Coefficients and Optical Depths with Data Collected on BGC-Argo Floats. Remote Sensing. 2020; 12(15):2367. https://doi.org/10.3390/rs12152367

Chicago/Turabian Style

Xing, Xiaogang, Emmanuel Boss, Jie Zhang, and Fei Chai. 2020. "Evaluation of Ocean Color Remote Sensing Algorithms for Diffuse Attenuation Coefficients and Optical Depths with Data Collected on BGC-Argo Floats" Remote Sensing 12, no. 15: 2367. https://doi.org/10.3390/rs12152367

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop