Open Access
Add to BibSonomy
Abstract
In this study, six algorithms (both empirical and semi-analytical) developed for the estimation of Kd in the ultraviolet (UV) domain (specifically 360, 380, and 400 nm) were evaluated from a dataset of 316 stations covering oligotrophic ocean and coastal waters. In particular, the semi-analytical algorithm (Lee et al. 2013) used remote sensing reflectance in these near-blue UV bands estimated from a recently developed deep learning system as the input. For Kd(380) in a range of 0.018 - 2.34 m−1, it is found that the semi-analytical algorithm has the best performance, where the mean absolute relative difference (MARD) is 0.19, and the coefficient of determination (R2) is 0.94. For the empirical algorithms, the MARD values are 0.23–0.90, with R2 as 0.70–0.92, for this evaluation dataset. For a VIIRS and in situ matchup dataset (N = 62), the MARD of Kd(380) is 0.21 (R2 as 0.94) by the semi-analytical algorithm. These results indicate that a combination of deep learning system and semi-analytical algorithms can provide reliable Kd(UV) for past and present satellite ocean color missions that have no spectral bands in the UV, where global Kd(UV) products are required for comprehensive studies of UV radiation on marine primary productivity and biogeochemical processes in the ocean.
© 2022 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Ultraviolet (UV) radiation has a complex impact on aquatic ecosystems [1]. It could be an energy source for phytoplankton [2], may also prohibit phytoplankton growth, can contribute to the photochemical transformation of dissolved organic matter (DOM) [3,4], and can affect the vertical movement of zooplankton [5]. In addition, ultraviolet radiation modulates the release of methane and other climate-related gases by phytoplankton cells, which directly contributes to global climate change [6]. Thus, quantifying UV radiation in ocean is an important task for a wide range of oceanographic studies. The amount of UV radiation reaching the ocean surface is primarily determined by the layer of ozone in the atmosphere, which is further modulated by water’s attenuation when UV radiation propagates from surface to deeper depths [1,7]. This attenuation parameter is commonly termed as the diffuse attenuation coefficient of downwelling irradiance (Kd; m−1). While UV radiation at surface varies in space and time, Kd, an important apparent optical property [8], also varies with water constituents. Therefore, to evaluate and understand UV related processes in the ocean, it is necessary to accurately map the spatial variation of Kd in the UV domain year around [9,10], a task that could only be achieved from satellite measurements.
The traditional satellites aimed at ocean’s bio-optical properties, such as the Coastal Zone Color Scanner (CZCS), Sea-viewing Wide Field-of-view (SeaWiFS), Moderate Resolution Imaging Spectroradiometer (MODIS), Medium Resolution Imaging Spectrometer (MERIS), Visible Infrared Imaging Radiometer Suite (VIIRS), etc., however, don’t have spectral bands shorter than 410 nm, thus it is a challenge to obtain global Kd properties in the UV from these ocean color satellites. While modern ocean color satellites start to include bands in the UV (e.g., Plankton, Aerosol, Cloud and ocean Ecosystem (PACE), Second generation Global Imager (SGLI), Ocean and Land Colour Instrument (OLCI), etc.), sophisticated algorithms must be developed in order to generate products in the UV for the earlier satellites, so that UV properties in the ocean could be extended back 20 + years. For this goal, as described in more details in Section 2, various algorithms have been developed in the past decades for the estimation of Kd in the UV (termed as Kd(UV) in the following, with the shortest wavelength limited to 360 nm here), either for one specific wavelength or many wavelengths [10–12]. But the applicability of these algorithms for global applications has not been evaluated. More importantly, most of these algorithms are empirical [9,13–15], where their applicability is usually limited to the data or region used during the phase of algorithm development. Recently it is found that remote sensing reflectance (Rrs, sr−1) in the near-blue UV bands (360–400 nm) can be well estimated from Rrs in the visible bands using deep-learning approaches [16,17], thus it is possible to employ a semi-analytical algorithm to estimate Kd(UV) for data obtained by traditional ocean color satellites.
Thus in this study, based on measurements from oligotrophic ocean to coastal waters, we evaluated six algorithms developed for Kd(UV), as well as a performance with VIIRS measurements. This paper is such organized that Section 2 describes the six algorithms (empirical and semi-analytical); Section 3 provides information of both field measured data and satellite data used for algorithm evaluations; Section 4 presents results and discussion about these algorithms; and Section 5 summarizes our findings and presents some future prospects.
2. Brief description of algorithms for Kd(UV)
A wide range of algorithms have been developed for the estimation of Kd(UV) in the past decades [9–12,14,15,18,19]. These can be grouped into empirical and semi-analytical algorithms, with empirical algorithms including one-step and multi-step schemes. A brief description of the essence of such representative algorithms is provided below.
2.1 One-step empirical algorithm (J2003)
Based on field-measured data (53 stations, Kd(380) in a range of 0.034 - 4.4 m−1) in the Mid-Atlantic Gulf and the Bering Sea from 1996 to 1998, following the scheme for Kd(490) [20,21], Johannessen et al. [11] developed a simple empirical relationship between the ratio of Rrs and Kd(380),
Here Rrs(sr−1) is the remote sensing reflectance of a water body, which is defined as the ratio of water-leaving radiance to downwelling irradiance just above the surface. Rrs is a standard product for many satellite ocean color missions.
2.2 Two-step empirical algorithm (M2001)
Centered on the “Case 1” concept and system, Morel and Maritorena [12] developed empirical relationships between Kd and chlorophyll concentration ([Chl], mg/m3) based on nearly 30 years of measurements (1969–1996, total 255 data stations). The relationship is written as:
Here χ and e are empirical coefficients developed from the dataset, while aw and bbw are the absorption and backscattering coefficients of pure seawater. In the development of this scheme [12], values of aw(λ) were taken from the literature [22,23], with values of bbw(λ) from Morel [24]. Although both aw and bbw values were revised in the recent decade [25,26], but we used the aw and bbw values originally adopted for this scheme, as χ and e were estimated using those aw and bbw values. For the estimation of Kd, [Chl] is a standard product from ocean color satellites, which is generally produced using empirical algorithms [27],
Here λ1∼3 are wavelengths in the range between 440 and 510 nm, and λ4 is around 550 nm. α0∼4 are empirical coefficients derived by pooling global measurements.
2.3 Empirical algorithm based on principal component analysis (C2014)
Fichot et al. [9] and Cao et al. [14] used principal component analysis (PCA) to decompose an Rrs dataset into a few principal components (PCs). For a given Rrs spectrum, the scores of the first four PCs are used as predictors of Kd(λ) (the algorithm is termed as SeaUV and SeaUVc), written as
2.4 Multi-step empirical algorithm (V2001)
Two decades ago, Vasilkov et al. [10] developed a Kd(UV) algorithm based on a relationship between Kd and the inherent optical properties (IOPs) [8]. From data obtained via Monte Carlo simulations, Gordon [28] modeled subsurface Kd as a function of water’s absorption (a) and backscattering (bb) coefficients as:
where f is the fraction of direct sunlight in the incident irradiance transmitted through the sea surface under clear sky, θ0w is the solar zenith angle in water. The value of f can be calculated based on solar zenith angle and wavelength from atmospheric models. It then requires values of a and bb for the calculation of Kd. Vasilkov et al. [10] used [Chl] and empirical averages of phytoplankton-specific absorption coefficients for the estimations. Specifically, a and bb are modeled as [29–31]: with subscripts “p, ph, DOM” representing particles, phytoplankton pigments, and dissolved organic matter, respectively. The values of aw and bbw were adopted from Smith and Baker [22].The absorption coefficient of phytoplankton pigments (aph(λ)) is expressed as
with $a_{\rm{ph}}^\ast$(λ) (the chlorophyll-specific absorption coefficient) taken from Vernet et al. [32].Spectra of bbp and aDOM are modeled as:
where the spectral slope (S) of aDOM was taken as 0.017 nm−1, and the backscattering exponent (η) was assumed as 1.0 [33]. Further, following Morel [34], aDOM(440) was assumed as 20% of the sum of (aw(440) + aph(440)). After aDOM(440) and S values are known, spectrum of aDOM(λ) can then be calculated following Eq. 8(a). The value of bbp(490) was estimated based on the standard SeaWiFS product Kd(490) [20,35] through the combination of Eqs. 5–8, which can solve for bbp(490).Although this complex scheme (V2001) started with the Kd - IOPs model developed by Gordon [28], it relies on [Chl] and Kd(490) as the inputs, where both are empirically estimated from the band ratio of Rrs, it is thus essentially a multi-step empirical algorithm for the estimation of Kd(UV).
2.5 Absorption-based empirical method (S2011)
Based on measured data (2007∼2009, total 106 data stations), Smyth [15] developed an empirical relationship between Kd(380) and a(443) as:
with A, B and C the fitting coefficients for targeted wavelength (380 nm here in particular) [15]. Kd(305, 325, 340) was further empirically linked with Kd(380) [15]. In this system, the required a(443) was derived from a look-up-table algorithm developed by Smyth et al. [36], with data obtained from Hydrolight [37] simulations. Since we don’t have the detailed look-up-table of this algorithm, while Smyth [15] showed that this method can obtain similar total absorption as the quasi-analytical algorithm (QAA) [38], here we applied a(443) obtained from QAA to Eq. 9, and focused on Kd(380). We did not evaluate the proposed relationships for Kd(305, 325, 340) [15] as the shortest wavelength from field measurements is 350 nm. Also, note that this system modeled Kd(305, 325, 340) as functions of Kd(380), respectively, which suggests Kd(UV) of different wavelengths co-vary for the global ocean, a relationship not generally supported from measurements [16].2.6 Semi-analytical algorithm (L2013)
Through derivations of the radiative transfer equation, Lee et al. [18] found that Kd can be described as a function of a and bb through
Based on Hydrolight [37] simulations, the two parameters m and ν were further described as functions of sun angle and IOPs, and Kd is then expressed as [19]:
The required inputs for the calculation of Kd are θs, a and bb, with the IOPs (a&bb) derived following the quasi-analytical algorithm [38]. Its recent version (QAA_v6, https://ioccg.org/wp-content/uploads/2020/11/qaa_v6_202011.pdf) was employed for this study. The contribution of Raman scattering to Rrs cannot be ignored for oceanic waters [39], which were removed following the scheme developed in Lee et al. [19].
3. Data
3.1 Field-measured data
Two data sources, which cover waters from the ultra-oligotrophic ocean to turbid coastal waters, are employed in this effort. The measurements of ultra-oligotrophic ocean were around the center of the South Pacific Gyre (SPG, see Fig. 1(a)) from the BIOSOPE cruise, which were carried out from 26 October to 11 December 2004 (a total of 25 stations). For these measurements, the concentration of chlorophyll is as low as ∼0.01 mg/m3, while Kd(380) is as low as 0.018 m−1 (depth of 1% surface light is deeper than 200 m). There have been many publications about this important and valuable field campaign [40–42], we thus omitted the details here.
Fig. 1. Locations of field measurements for the evaluation of Kd(UV) algorithms. (a) The data came from the BIOSOPE cruise from 26 October to 11 December 2004. (b) The data came from the NOAA VIIRS Cal/Val cruise off the East Coast of the United States and the Gulf of Mexico from 2014 to 2019. The white stars are the measurement locations, the blue stars represent stations for satellite and field measurements matched up, with the red star for abnormal data.
Since we are not simply focusing on oceanic waters, measurements of many coastal waters are also included in this effort. Specifically, data from the VIIRS Cal/Val campaigns were utilized, which generally cover waters in the East Coast of the United States [43–47]. The measurements were taken from 2014 to 2019 (a total of 294 stations), covered both open ocean and relatively turbid coastal waters (see Fig. 1(b)), where Kd(380) is in a range of ∼0.026 - 2.34 m−1.
For these VIIRS Cal/Val field measurements, profiles of Lu(λ,z) (spectral upwelling radiance, μW cm−2 nm−1 sr−1) and Ed(λ,z) (spectral downwelling irradiance, μW cm−2 nm−1) were measured with HyperOCR radiometers manufactured by Satlantic, Inc. The instruments were always deployed in the sunny side of the operation ship and at least 20 m away, with the maximum depth for some of these profiling measurements ∼50 m. To minimize uncertainties associated with an individual profiling measurement, following recommendations of the in-water optical measurement protocol [48,49], usually 4 to 5 profiles were taken for each station, which resulted in 100’s of measurements per meter, thus allowing an accurate representation of the Lu(λ,z) and Ed(λ,z) profiles. Further, spectral surface irradiance (Es(λ), μW cm−2 nm−1) was simultaneously measured on the ship deck with another HyperOCR radiometer for irradiance. This data was used in the Satlantic Prosoft Version 8.1.6 to remove short-term variations in the incident radiation at sea surface. Note that the spectral range of these HyperOCR is ∼350–800 nm, with a spectral bandwidth of 10 nm sampled every 3.3 nm.
From the Ed(λ,z) profiles, after excluding data with sensor tilt lager than 5 deg, following Zibordi et al. [50], Kd of each band was calculated as the slope of the linear regression between z and ln[Ed(λ, z)]. The depth range for regression is dependent on the wavelength and was set using the optical depth (OD). The shallowest depth used in the regression is 0.5 m from surface, which is the minimum depth a HyperPro can take measurements. The maximum depth used in the regression varies by wavelength and is defined by the maximum optical depth used in the data processing or 20 m whichever is shallower. This maximum optical depth, which varied from 0.5 to 2 for the coastal waters in this study, is determined based on that all the regressions are linear for all wavelengths.
Rrs(λ) is defined as
where Lw(λ) is the spectral water-leaving radiance (μW cm−2 nm−1 sr−1). For profiling measurements, Lw(λ) was derived from the relationship: where t is the Fresnel transmittance of radiance from water to air, approximating 0.975. n is the refractive index of water, with a common value as 1.34. Lu(0−, λ) is the upward radiance just beneath the sea surface, which was calculated by extrapolating the Lu(z, λ) profile [48] using the Satlantic Prosoft Version 8.1.6. We did not correct the self-shading effect in the Lu(z, λ) measurements, as the uncertainty from this factor is just a few percent [49], which is much less than the uncertainties introduced by an algorithm for the inversion of Kd.For the 294 total stations of clear and cloudy days having HyperPro measurements, it resulted in 291 valid Kd and Rrs spectra. The other 3 stations showing abnormal Kd values (approaching 0 at ∼500 nm) were excluded in this report, although an inclusion of these stations has negligible impact on the statistics. Fig. 2 shows examples of Rrs(λ) and Kd(λ) spectra from these measurements. As an indicator of the wide coverage of waters included in this effort, the range of Rrs(440)/Rrs(550) is 0.33 to 13.23.
Fig. 2. Examples of field measurements used in this study. (a) measured Rrs spectra; (b) measured Kd spectra.
3.2 Satellite data
To gain insights on applications to satellite ocean color remote sensing, taking the Visible Infrared Imaging Radiometer Suite (VIIRS) as an example, we used satellite-field matchup data to evaluate algorithm estimated Kd(UV). VIIRS Level-2 normalized water-leaving radiance (nLw; mW cm−2 sr−1 μm−1) in the visible band were downloaded from the official website of NOAA CoastWatch (https://coastwatch.noaa.gov). These nLw(λ) were converted to Rrs through
where F0 is the extra-terrestrial solar radiation, which is constant and available in the literature [51].For a field station, the median value of 3 × 3 VIIRS pixels centered on this station was used to represent satellite measurement [52]. Further, the time window between in situ and satellite data was set as ± 5 hours [53]. Also, the quality of the spectral data was judged according to a quality-assurance (QA) score [54] and Level-2 Processing Flags (l2_flags), where only data with QA score > 0.6 were kept, and excluded VIIRS data containing these l2_flags (atmospheric correction failure, land pixel, probable cloud or ice contamination, strong sun glint contamination and cloud straylight or shadow contamination). We subsequently obtained 62 matchups between VIIRS and field measurements.
3.3 Rrs(UV)
Since the shortest band of VIIRS (and SeaWiFS, MODIS, etc.) is ∼410 nm, in this study Rrs at 360, 380, and 400 nm were generated following a deep-learning scheme (UVISRdl) recently developed by Wang et al. [16]. UVISRdl has 4 hidden layers along with various neurons; it uses Rrs(visible) as the input to generate Rrs(UV), with the estimated Rrs(UV) found matching field-measured value within ∼10%. The details of UVISRdl can be found in Wang et al. [16].
To be consistent with satellite applications, for the evaluation of L2013 using field measurements that requires Rrs(UV) for the analytical calculation of Kd(UV), the field dataset also employed UVISRdl to obtain Rrs(UV), although there are Rrs(UV) data available from the profiling measurements. In short, for the evaluation of the six algorithms, it is the same visible Rrs spectrum applied to all algorithms.
3.4 Accuracy assessment
To quantify the accuracy of the estimated Kd(UV), the following statistical indicators are used for quantitative evaluation: root mean square difference (RMSD), mean absolute relative difference (MARD), mean absolute unbiased relative difference (MAURD), bias and coefficient of determination (R2). They are defined as follows (N is the number of samples):
4. Results and discussion
4.1 Evaluation of Kd(UV) from field-measured reflectance
The above-described six empirical and semi-analytical algorithms for the estimation of Kd(UV) were evaluated with the dataset described in Section 3, which covers waters from the oligotrophic ocean to coastal areas. Figs. 3∼4 and Table 1 summarize the performance of these algorithms, where some of these algorithms [9,11,14,15] focused on Kd(380) and/or other specific wavelengths (Fig. 3), while the other three algorithms also cover 360 nm and 400 nm (Fig. 4). Outcomes of each algorithm are described below.
Fig. 3. Comparison between measured and algorithm-estimated Kd(380) (the blue dots represent data from the BIOSOPE cruise, the green dots represent data from the NOAA Cal/Val cruise). (a) J2003, (b) S2011, (c) C2014, (d) M2001, (e) V2001, (f) L2013.
Fig. 4. As Fig. 3, but for Kd(360) ((a) M2001, (b) V2001, (c) L2013) and Kd(400) ((d) M2001, (e) V2001, (f) L2013).
Table 1. Kd(360), Kd(380) and Kd(400) algorithm inversion results statistical analysis.
4.1.1 J2003
The empirical algorithm (J2003) for Kd(380) developed by Johannessen et al. (2003) is one of the earlier attempts for the estimation of Kd(UV) from ocean color measurements. For this evaluation dataset where Kd(380) is in a range of 0.018–2.34 m−1, J2003 performed very well (Fig. 3(a)). The values of RMSD, MARD, and R2 are just slightly worse than that of L2013 (Table 1). This is likely due to that this empirical algorithm was developed also based on data measured in the East Coast of the United States (Kd(380) was in a range of 0.033–4.39 m−1), where there are similar data characteristics between the data for algorithm development and data for algorithm evaluation. For the lower end of Kd(380) (<0.1 m−1) mostly obtained from the BIOSOPE cruise, types of waters not covered by Johannessen et al. [11] in the phase of algorithm development, Kd(380) estimated by J2003 appeared generally underestimated (by ∼47%) compared to field-measured Kd(380). This degraded performance is expected when the application of an empirical algorithm is extended beyond its range. Note that, if the true Kd(380) is 0.02 m−1, a 47% underestimation will result in ∼1.5 times more estimation of UV radiation at 100 m, thus significantly over-emphasize the impact of UV.
4.1.2 S2011
S2011 is an empirical algorithm centered on the total absorption coefficient for Kd(380), which performed reasonably well for this evaluation dataset (see Fig. 3(b)). However, the performance is not uniform across the values of Kd(380), where S2011 underestimated (by 39%) Kd(380) for values ∼0.1 - 1.0 m−1. On the other hand, for Kd(380) lower than about 0.04 m−1, S2011 overestimated Kd(380) by about 58%, which is likely due to that the range of Kd(380) is higher than the lower limit of ∼0.04 m−1 when S2011 was developed. This overestimation of Kd(380) will significantly under-represent UV radiation at deeper depths (by a factor about 3 at 100 m). In addition to the data dependency of any empirical algorithms, another reason for the deviations of S2011 is likely in the model itself (Eq. 9), where Kd(380) is simply modeled as a second-order polynomial function of a(443). However, radiative transfer indicates that Kd is jointly determined by both a and bb, thus uneven performance will be resulted when there are different levels of contributions from backscattering.
4.1.3 C2014
C2014 uses PCA to reduce the dimension of Rrs, and then relates Kd(380) with the scores of the pre-established principal components through a multiple-component regression. This algorithm does not perform well for this evaluation dataset (see Fig. 3(c), Table 1). For Kd(380) < ∼0.05 m−1, C2014 overestimated Kd(380) by more than 100%, which, similarly as the results of S2011, is likely due to that the lower limit of Kd(380) is ∼0.05 m−1 when SeaUVc was developed (Fichot et al. [9]). The meaning of the principal components derived from a group of Rrs spectra is vague, thus the physics of a PCA-based algorithm for an optical or geophysical property is not straightforward. Further, since the principal components are data dependent, the performance of such kind of algorithms may vary for different datasets.
4.1.4 M2001
The two-steps scheme (M2001) based on Morel and Maritorena [12] performed better at the lower end (Kd(380) < ∼0.06 m−1) for the three wavelengths (360, 380, and 400 nm) evaluated here, but generally underestimated (by ∼57%) Kd(UV) for Kd(380) > 0.1 m−1 (see Figs. 3(d), 4(a), and 4(d)). The Morel and Maritorena [12] system was developed based on a large dataset mostly collected in oceanic waters (Kd(420) is in a range of 0.007 - 0.94 m−1), but most of the measurements of this evaluation dataset were from coastal waters, which might explain the very good performance at the lower end (including the ultra-oligotrophic ocean), but unsatisfactory results for coastal waters.
4.1.5 V2001
Similar as M2001, V2001 estimates Kd(UV) based on the “Case 1” concept, but taking a much more complex data processing scheme. For this evaluation dataset that covers both oceanic and coastal waters. Interestingly, similarly as M2001, V2001 also underestimated Kd(UV) for Kd(380) > 0.1 m−1 (see Figs. 3(e), 4(b), and 4(e)), but this underestimation (∼44%) is not as severe as that of M2001. But for the lower end (Kd(380) < ∼0.03 m−1), V2001 significantly (89%) overestimated Kd(UV) (see Figs. 3(e), 4(b), and 4(e)), likely due to an extension of this algorithm beyond its limit (the lower end was 0.03 m−1 when the algorithm was developed). It is found that the estimated Kd(400) agreed better with field measurements compared to Kd(360) (see Table 1), which in part is due to the selection of the S value in the system (Eq. 8a), as a large S value (0.017 nm−1) suggests higher estimation of aDOM(360) due to the exponential function, which further propagates to the estimated Kd(360). For oceanic waters, Bricaud et al. [55] suggested that S is around 0.014 nm−1.
4.1.6 L2013
L2013 is an algorithm centered on the IOPs (absorption and backscattering coefficients in particular), with the relationship established from the radiative transfer equation. More importantly, contrary to empirical algorithms, the algorithm coefficients of L2013 were not derived from any datasets obtained from field measurements, thus theoretically its application to an Rrs spectrum is not subjective to datasets. As a result, it is not surprising to see that L2013 has the best performance among the six algorithms for this evaluation dataset that covers wide range of waters (see Figs. 3(f), 4(c), 4(f), and Table 1). Especially, the performance is uniform for Kd(380) in a range of ∼0.02–2.0 m−1, rather showing any obvious deviations for any specific ranges or values of Kd(UV). This further highlights the power of systems based on radiative transfer, where the performance is the same for all waters and for all wavelengths, instead of varying performances with water types or wavelengths showing by the empirical algorithms (e.g., J2003, S2011). In addition, these results further support the deep-learning system (Wang et al. 2021) for the estimation of Rrs(UV) as well as QAA for the algebraic derivation of IOPs from an Rrs spectrum, for waters from ultra-oligotrophic ocean to coastal waters.
4.2 Evaluation of VIIRS Kd(UV) with in situ measurements
Since the ultimate goal of an algorithm is to apply it to ocean color satellites to obtain global distribution of Kd(UV), here we further evaluated Kd(UV) obtained by VIIRS measurements, where the locations of the satellite-insitu matchup stations are shown in Fig. 1(b) with blue stars. As the other five algorithms showed limited accuracy to some waters, the evaluation is focused on Kd(UV) from L2013 only. Also included for this comparison is Kd(410, 443, 486) obtained from L2013. Fig. 5 (a - f) presents scatterplots between inversion and measured Kd (Kd(360) is in a range of 0.046–0.84 m−1), while statistical measures are presented in Table 2.
Table 2. Statistical analysis between matchup VIIRS and in situ Kd(λ). N for the number of matchup stations.
Fig. 5. Comparison between VIIRS and measured Kd. (a) Kd(360), (b) Kd(380), (c) Kd(400), (d) Kd(410), (e) Kd(443), and (f) Kd(486). The red triangles represent the matching data of the red-star location in Fig. 1(b).
Overall, for these Kd(UV), the R2 values are greater than 0.92, with MARD between 0.19 (at 360 nm) and 0.28 (at 443 nm), and bias in a range of ∼ -0.031 to -0.015 m−1. These evaluation indicators are slightly worse than that obtained from the field-measured Rrs, a result not surprising for ocean color satellite remote sensing. Reasons contributing to this lowered performance include: 1) No exact “matchup” between satellite and field measurements due to gaps in time and space [56,57], which is one of the major sources of data deviation [53]. 2) Imperfect atmosphere correction in both oceanic and coastal areas [58–61], consequently the errors in the nLw (Rrs) product will be propagated to those further estimated from Rrs. Fig. 5 (d - f) compares VIIRS Kd(410, 443, 486) with in situ Kd(410, 443, 486), where the MARD values range are ∼ 0.27 - 0.28 and RMSD ∼ 0.050 m−1, which are just slightly better than those of Kd(UV), further indicating the deep-learning estimated Rrs(UV) has similar quality as those of Rrs in the visible (Wang et al. 2021).
There are a few stations (the red triangles in Fig. 5) where VIIRS Kd(360 - 486) are much lower than in situ Kd (Kd(360) is around 1.0 m−1). These four stations were from a nearshore area (Fig. 1(b), the red star) and collected on Oct. 14, 2016. Based on the VIIRS Level-2 flags processed by NOAA MSL12, there are stray-light contaminations in this VIIRS image, which may be the reason for the larger deviation of these stations. In addition, there was heavy stratification in these coastal waters with a high-sediment fresh-water layer at the surface from runoff due to Hurricane Matthew. This complicates satellite measures to the corresponding profiles [62].
4.3 Kd(UV) of global ocean from VIIRS
As the above shows very good estimation of Kd(UV) from VIIRS, examples of Kd(360, 380, 400 and 486 (with the Sun at zenith) of the global ocean are produced from VIIRS for climatology Rrs (see Fig. 6). Compared to the partial Kd(360) map obtained by Lee et al. [19], the results extend the estimation of Kd(UV) to the global ocean, which then help providing solar radiation distribution in the upper column for the evaluation of photochemical and photobiological processes in the ocean.
Fig. 6. Global distribution of VIIRS Kd(λ) climatology data. (a) Kd(360), (b) Kd(380), (c) Kd(400), (d) Kd(486).
Spatially, while Kd(380) is as low as ∼0.02 m−1 at the South Pacific Gyre, but due to the influence of terrestrial sources [15], Kd(380) can be as high as ∼3.0 m−1 in coastal regions such as the Yangtze Estuary. If the depth of 10% surface light can be a measure of the penetration of solar radiation, the penetration depth of solar radiation at 380 nm is ∼100 m at South Pacific Gyre. Such a strong penetration is expected to have comprehensive impacts on biogeochemical processes in the oceans [63,64].
5. Conclusions
From a large dataset covering super clear ocean gyre waters and coastal turbid waters, six empirical and semi-analytical algorithms developed for the estimation of Kd(UV) from the measurement of ocean color were evaluated. The five empirical algorithms showed reasonable results for some Kd(UV) values or waters, but indicated limitations for other waters, suggesting challenges to apply such empirical algorithms for the generation of Kd(UV) for the global oceans from satellite ocean color missions. The semi-analytical algorithm centered on waters’ inherent optical properties, however, not only obtained the best estimates compared to field measurements, also show uniform performance from ultra-oligotrophic ocean to coastal waters, as well as for the three wavelengths in the UV evaluated. These results further highlight the power of algorithms based on the radiative transfer, which is not limited to a single wavelength (such as the empirical algorithms for Kd(490) or Kd(380), for instance), rather applicable for an accurate estimation of a Kd spectrum, as long as the spectra of the total absorption and backscattering can be accurately produced from an Rrs spectrum. Also, the excellent agreement between Kd estimated from VIIRS ocean color data (with the semi-analytical algorithm) and field measurements indicates that VIIRS normalized water-leaving radiance (equivalent to Rrs) has very high quality. On the other hand, it is necessary to keep in mind that this dataset does not include waters with phytoplankton blooms containing mycosporine-like amino acids (MAAs). Due to the strong absorption of MAAs in the UV domain [65,66], the remote sensing of Kd(UV) for such kind of waters deserve further studies.
To summarize, with the deep learning system developed for the estimation of Rrs(UV) in the near-blue domain from Rrs in the visible, it is now feasible to produce highly reliable Kd(UV) from ocean color satellites dating back to the SeaWiFS, MODIS era. A generation of global data product in the past 20 + years will then lay the foundation for the study of long-term interactions between UV radiation and biogeochemical processes in the global ocean.
Funding
Ministry of Science and Technology of the People's Republic of China (2018YFC1407506); National Natural Science Foundation of China (#41830102, #41890803, #41941008); NOAA ocean color calibration and validation; University of Massachusetts Boston.
Acknowledgments
We are grateful to Dr. Marlon Lewis for sharing profiling measurements obtained during the BIOSOPE cruise and NOAA STAR for providing VIIRS Ocean Color measurements. Comments and suggestions from four anonymous reviewers are greatly appreciated. The views, opinions, and findings contained in this paper are those of the authors and should not be construed as an official NOAA or U.S. Government position, policy, or decision.
Disclosures
The authors declare no conflicts of interest.
Data availability
Data used in this study are available in either NASA SeaBASS or NOAA data portal, can also be obtained from the authors upon request.
References
1. M. Tedetti and R. Sempéré, “Penetration of ultraviolet radiation in the marine environment. A review,” Photochem Photobiol 82(2), 389–397 (2006). [CrossRef]
2. K. Gao, E. W. Helbling, D.-P. Häder, and D. A. Hutchins, “Responses of marine primary producers to interactions between ocean acidification, solar radiation, and warming,” Mar. Ecol. Prog. Ser. 470, 167–189 (2012). [CrossRef]
3. R. Zepp, D. Erickson Iii, N. Paul, and B. Sulzberger, “Interactive effects of solar UV radiation and climate change on biogeochemical cycling,” Photochem. Photobiol. Sci. 6(3), 286–300 (2007). [CrossRef]
4. C. Piccini, D. Conde, J. Pernthaler, and R. Sommaruga, “Alteration of chromophoric dissolved organic matter by solar UV radiation causes rapid changes in bacterial community composition,” Photochem. Photobiol. Sci. 8(9), 1321–1328 (2009). [CrossRef]
5. K. C. Rose, C. E. Williamson, J. M. Fischer, S. J. Connelly, M. Olson, A. J. Tucker, and D. A. Noe, “The role of ultraviolet radiation and fish in regulating the vertical distribution of Daphnia,” Photochem. Photobiol. Sci. 57(6), 1867–1876 (2012). [CrossRef]
6. A. McLeod, T. Brand, C. Campbell, K. Davidson, and A. Hatton, “Ultraviolet radiation drives emission of climate-relevant gases from marine phytoplankton,” Journal of Geophysical Research: Biogeosciences, e2021JG006345 (2021).
7. C. Booth and J. Morrow, “The penetration of UV into natural waters,” Photochem Photobiol 65, 254–257 (1997). [CrossRef]
8. R. W. Preisendorfer, “Secchi disk science: Visual optics of natural waters,” Photochem. Photobiol. Sci. 31(5), 909–926 (1986). [CrossRef]
9. C. G. Fichot, S. Sathyendranath, and W. L. Miller, “SeaUV and SeaUVC: Algorithms for the retrieval of UV/Visible diffuse attenuation coefficients from ocean color,” Remote Sensing of Environment 112, 1584–1602 (2008). [CrossRef]
10. A. Vasilkov, N. Krotkov, J. Herman, C. McClain, K. Arrigo, and W. Robinson, “Global mapping of underwater UV irradiances and DNA-weighted exposures using Total Ozone Mapping Spectrometer and Sea-viewing Wide Field-of-view Sensor data products,” J. Geophys. Res. 106(C11), 27205–27219 (2001). [CrossRef]
11. S. C. Johannessen, W. L. Miller, and J. J. Cullen, “Calculation of UV attenuation and colored dissolved organic matter absorption spectra from measurements of ocean color,” J. Geophys. Res. 108(C9), 3301 (2003). [CrossRef]
12. A. Morel and S. Maritorena, “Bio-optical properties of oceanic waters: A reappraisal,” J. Geophys. Res. 106(C4), 7163–7180 (2001). [CrossRef]
13. K. Sun, T. Zhang, S. Chen, C. Xue, B. Zou, and L. Shi, “Retrieval of Ultraviolet Diffuse Attenuation Coefficients From Ocean Color Using the Kernel Principal Components Analysis Over Ocean,” IEEE Transactions on Geoscience and Remote Sensing 59(6), 4579–4589 (2021). [CrossRef]
14. F. Cao, C. G. Fichot, S. B. Hooker, and W. L. Miller, “Improved algorithms for accurate retrieval of UV/visible diffuse attenuation coefficients in optically complex, inshore waters,” Remote sensing of environment 144, 11–27 (2014). [CrossRef]
15. T. J. Smyth, “Penetration of UV irradiance into the global ocean,” J. Geophys. Res. 116(C11), 2011JC007183 (2011). [CrossRef]
16. Y. Wang, Z. Lee, J. Wei, S. Shang, M. Wang, and W. Lai, “Extending satellite ocean color remote sensing to the near-blue ultraviolet bands,” Remote Sensing of Environment 253, 112228 (2021). [CrossRef]
17. H. Liu, X. He, Q. Li, S. Kratzer, J. Wang, T. Shi, Z. Hu, C. Yang, S. Hu, and Q. Zhou, “Estimating ultraviolet reflectance from visible bands in ocean colour remote sensing,” Remote Sensing of Environment 258, 112404 (2021). [CrossRef]
18. Z. P. Lee, K. P. Du, and R. Arnone, “A model for the diffuse attenuation coefficient of downwelling irradiance,” J. Geophys. Res. 110(C2), C02016 (2005). [CrossRef]
19. Z. Lee, C. Hu, S. Shang, K. Du, M. Lewis, R. Arnone, and R. Brewin, “Penetration of UV-visible solar radiation in the global oceans: Insights from ocean color remote sensing,” J. Geophys. Res. 118(9), 4241–4255 (2013). [CrossRef]
20. J. L. Mueller, “SeaWiFS algorithm for the diffuse attenuation coefficient, K(490), using water-leaving radiances at 490 and 555 nm,” in SeaWiFS Postlaunch Calibration and Validation Analyses, Part 3, S. B. Hooker, ed. (NASA, 2000), 24–27.
21. R. W. Austin and T. J. Petzold, “The determination of the diffuse attenuation coefficient of sea water using the coastal zone color scanner,” in Oceanography from Space, J. F. R. Gower, ed. (Plenum Press, 1981), pp. 239-256.
22. R. C. Smith and K. S. Baker, “Optical properties of the clearest natural waters,” Appl. Opt. 20(2), 177–184 (1981). [CrossRef]
23. R. Pope and E. Fry, “Absorption spectrum (380 - 700 nm) of pure waters: II. Integrating cavity measurements,” Appl. Opt. 36(33), 8710–8723 (1997). [CrossRef]
24. A. Morel, “Optical properties of pure water and pure sea water,” in Optical Aspects of Oceanography, N. G. Jerlov and E. S. Nielsen, eds. (Academic, 1974), pp. 1-24.
25. Z. Lee, J. Wei, K. Voss, M. Lewis, A. Bricaud, and Y. Huot, “Hyperspectral absorption coefficient of “pure” seawater in the range of 350–550 nm inverted from remote sensing reflectance,” Appl. Opt. 54(3), 546–558 (2015). [CrossRef]
26. X. Zhang and L. Hu, “Estimating scattering of pure water from density fluctuation of the refractive index,” Opt. Express 17(3), 1671–1678 (2009). [CrossRef]
27. J. E. O’Reilly, S. Maritorena, B. G. Mitchell, D. A. Siegel, K. L. Carder, S. A. Garver, M. Kahru, and C. McClain, “Ocean color chlorophyll algorithms for SeaWiFS,” J. Geophys. Res. 103(C11), 24937–24953 (1998). [CrossRef]
28. H. R. Gordon, “Dependence of the diffuse reflectance of natural waters on the sun angle,” Photochem. Photobiol. Sci. 34(8), 1484–1489 (1989). [CrossRef]
29. K. L. Carder, S. K. Hawes, K. A. Baker, R. C. Smith, R. G. Steward, and B. G. Mitchell, “Reflectance model for quantifying chlorophyll a in the presence of productivity degradation products,” J. Geophys. Res. 96(C11), 20599–20611 (1991). [CrossRef]
30. L. Prieur and S. Sathyendranath, “An optical classification of coastal and oceanic waters based on the specific spectral absorption curves of phytoplankton pigments, dissolved organic matter, and other particulate materials 1,” Photochem. Photobiol. Sci. 26(4), 671–689 (1981). [CrossRef]
31. R. P. Bukata, J. H. Jerome, K. Y. Kondratyev, and D. V. Pozdnyakov, Optical Properties and Remote Sensing of Inland and Coastal Waters (CRC Press, 1995).
32. M. Vernet, E. A. Brody, O. Holm-Hansen, and B. G. Mitchell, “The response of Antarctic phytoplankton to ultraviolet radiation: absorption, photosynthesis, and taxonomic composition,” Ultraviolet radiation in Antarctica: measurements and biological effects 62, 143–158 (1994). [CrossRef]
33. H. R. Gordon and A. Morel, Remote assessment of ocean color for interpretation of satellite visible imagery: A review (Springer-Verlag, 1983).
34. A. Morel, “Optical modeling of the upper ocean in relation to its biogenous matter content (case I waters),” J. Geophys. Res. 93(C9), 10749–10768 (1988). [CrossRef]
35. C. R. McClain, W. E. Esaias, W. Barnes, B. Guenther, D. Endres, S. B. Hooker, G. Mitchell, and R. Barnes, “Calibration and validation plan for SeaWiFS., “ in NASA Tech. Memo., 104566, Vol. 3 S. B. Hooker and E. R. Firestone, eds. (NASA Goddard Space Flight Center, 1992), p. 41.
36. T. J. Smyth, G. F. Moore, T. Hirata, and J. Aiken, “Semianalytical model for the derivation of ocean color inherent optical properties: description, implementation, and performance assessment,” Appl. Opt. 45(31), 8116–8131 (2006). [CrossRef]
37. C. D. Mobley, “Hydrolight 3.0 Users’ Guide,” (SRI International, 1995).
38. Z. P. Lee, K. L. Carder, and R. Arnone, “Deriving inherent optical properties from water color: A multi-band quasi-analytical algorithm for optically deep waters,” Appl. Opt. 41(27), 5755–5772 (2002). [CrossRef]
39. T. K. Westberry, E. Boss, and Z. Lee, “The Influence of Raman Scattering on Ocean Color Inversion Models,” Appl. Opt., under revision (2013).
40. M. Tedetti, R. Sempéré, A. Vasilkov, B. Charrière, D. Nérini, W. L. Miller, K. Kawamura, and P. Raimbault, “High penetration of ultraviolet radiation in the south east Pacific waters,” Geophysical Research Letters, 34 (2007).
41. A. Morel, B. Gentili, H. Claustre, A. Babin, A. Bricaud, J. Ras, and F. Tieche, “Optical properties of the “clearest” natural waters,” Limnol. Oceanogr. 52(1), 217–229 (2007). [CrossRef]
42. M. Twardowski, H. Claustre, S. Freeman, D. Stramski, and Y. Huot, ““Optical backscattering properties of the” clearest” natural waters,” Biogeosciences 4(6), 1041–1058 (2007). [CrossRef]
43. M. Ondrusek, V. P. Lance, M. Wang, E. Stengel, C. Kovach, S. Freeman, A. Mannino, A. Gilerson, C. Carrizo, and P. Grötsch, “Report for Dedicated JPSS VIIRS Ocean Color Calibration/Validation Cruise September 2019,” (2021).
44. M. Ondrusek, V. P. Lance, M. Wang, E. Stengel, C. Kovach, R. A. Arnone, S. Ladner, W. Goode, A. Gilerson, and A. El-Habashi, “Report for Dedicated JPSS VIIRS Ocean Color Calibration/Validation Cruise May 2018,” (2019).
45. M. Ondrusek, V. P. Lance, M. Wang, E. Stengel, C. Kovach, R. A. Arnone, S. Ladner, W. Goode, A. Gilerson, and S. A. Ahmed, “Report for dedicated JPSS VIIRS Ocean Color Calibration/Validation Cruise, October 2016,” (2017).
46. M. Ondrusek, V. P. Lance, M. Wang, R. A. Arnone, S. Ladner, W. Goode, R. Vandermeulen, S. Freeman, J. E. Chaves, and A. Mannino, “Report for dedicated JPSS VIIRS ocean color December 2015 calibration/validation cruise,” (2016).
47. M. Ondrusek and V. P. Lance, “Report for dedicated JPSS VIIRS ocean color calibration/validation cruise,” (2015).
48. J. L. Mueller, A. Morel, R. Frouin, C. Davis, R. Arnone, K. Carder, Z. Lee, R. Steward, S. Hooker, and C. Mobley, “Ocean Optics Protocols For Satellite Ocean Color Sensor Validation, Revision 4. Volume III: Radiometric Measurements and Data Analysis Protocols,” (2003).
49. G. Zibordi, K. J. Voss, B. C. Johnson, and J. Mueller, “IOCCG Ocean optics and biogeochemistry protocols for satellite ocean colour sensor validation,” IOCCG Protocols Series 3 (2019).
50. G. Zibordi, K. Ruddick, I. Ansko, G. Moore, S. Kratzer, J. Icely, and A. Reinart, “In situ determination of the remote sensing reflectance: an inter-comparison,” Ocean Science 8(4), 567–586 (2012). [CrossRef]
51. G. Thuillier, M. Hersé, T. Foujols, W. Peetermans, D. Gillotay, P. Simon, and H. Mandel, “The solar spectral irradiance from 200 to 2400 nm as measured by the SOLSPEC spectrometer from the ATLAS and EURECA missions,” Solar Physics 214(1), 1–22 (2003). [CrossRef]
52. S. W. Bailey and P. J. Werdell, “A multi-sensor approach for the on-orbit validation of ocean color satellite data products,” Remote Sens. Environ. 102(1-2), 12–23 (2006). [CrossRef]
53. P. J. Werdell and S. W. Bailey, “An improved in-situ bio-optical data set for ocean color algorithm development and satellite data product validation,” Remote sensing of environment 98(1), 122–140 (2005). [CrossRef]
54. J. Wei, Z. Lee, and S. Shang, “A system to measure the data quality of spectral remote-sensing reflectance of aquatic environments,” J. Geophys. Res. 121, 8189–8207 (2016).
55. A. Bricaud, A. Morel, and L. Prieur, “Absorption by Dissolved Organic Matter of the Sea (Yellow Substance) in the UV and Visible Domains,” Photochem. Photobiol. Sci. 26(1), 43–53 (1981). [CrossRef]
56. D. Antoine, F. d’Ortenzio, S. B. Hooker, G. Bécu, B. Gentili, D. Tailliez, and A. J. Scott, “Assessment of uncertainty in the ocean reflectance determined by three satellite ocean color sensors (MERIS, SeaWiFS and MODIS-A) at an offshore site in the Mediterranean Sea (BOUSSOLE project),” J. Geophys. Res. 113(C7), C07013 (2008). [CrossRef]
57. G. Zibordi, J.-F. Berthon, F. Mélin, D. D’Alimonte, and S. Kaitala, “Validation of satellite ocean color primary products at optically complex coastal sites: Northern Adriatic Sea, Northern Baltic Proper and Gulf of Finland,” Remote Sensing of Environment 113(12), 2574–2591 (2009). [CrossRef]
58. J. Wei, Z. Lee, R. Garcia, L. Zoffoli, R. A. Armstrong, Z. Shang, P. Sheldon, and R. F. Chen, “An assessment of Landsat-8 atmospheric correction schemes and remote sensing reflectance products in coral reefs and coastal turbid waters,” Remote sensing of environment 215, 18–32 (2018). [CrossRef]
59. IOCCG, “Atmospheric Correction for Remotely-Sensed Ocean-Colour Products,” in Reports of the International Ocean-Colour Coordinating Group, M. Wang, ed. (IOCCG, 2010), p. 83.
60. M. Wang, “Remote sensing of the ocean contributions from ultraviolet to near-infrared using the shortwave infrared bands: simulations,” Appl. Opt. 46(9), 1535–1547 (2007). [CrossRef]
61. J. Wang, Z. Lee, J. Wei, and K. Du, “Atmospheric correction in coastal region using same-day observations of different sun-sensor geometries with a revised POLYMER model,” Opt. Express 28(18), 26953–26976 (2020). [CrossRef]
62. Z. Lee, S. Shang, Y. Wang, J. Wei, and J. Ishizaka, “Nature of optical products inverted semianalytically from remote sensing reflectance of stratified waters,” Photochem. Photobiol. Sci. (2019).
63. J. J. Cullen and P. J. Neale, “Ultraviolet radiation, ozone depletion, and marine photosynthesis,” Photosynthesis Research 39(3), 303–320 (1994). [CrossRef]
64. R. C. Smith, B. Prezelin, K. e. a. Baker, R. Bidigare, N. Boucher, T. Coley, D. Karentz, S. MacIntyre, H. Matlick, and D. Menzies, “Ozone depletion: ultraviolet radiation and phytoplankton biology in Antarctic waters,” Science 255(5047), 952–959 (1992). [CrossRef]
65. M. Kahru, C. Anderson, A. D. Barton, M. L. Carter, D. Catlett, U. Send, H. M. Sosik, E. L. Weiss, and B. G. Mitchell, “Satellite detection of dinoflagellate blooms off California by UV reflectance ratios,” Elem Sci Anth 9(1), 00157 (2021). [CrossRef]
66. M. Kahru and B. G. Mitchell, “Spectral reflectance and absorption of a massive red tide off southern California,” J. Geophys. Res. 103(C10), 21601–21609 (1998). [CrossRef]
