1. Introduction

Water molecules, together with colored dissolved organic matter (CDOM) and particulate matter, are the optically important components in natural waters. The absorption and scattering properties of these components, known as the inherent optical properties (IOPs), determine the propagation of light within the water body [1]. In Case 1 waters, defined by Morel [2], pure seawater, phytoplankton pigments and co-varying associates are the dominant absorbing components. Therefore, the spectral absorption coefficient of pure seawater (a_w(λ), in m⁻¹) may play a vital role in ocean color remote sensing studies in Case 1 waters. This is especially true for the subtropical ocean gyres, which account for approximately 40% of the Earth’s surface area [3] and are reportedly expanding [4]. In addition, the fidelity of a_w(λ) is also essential for better understanding and prediction of in-water penetration of radiation and water transparency in oceanic waters [5,6].

There is a long history in the determination of a_w(λ) [7], and values of a_w(λ) were determined conventionally by laboratory measurements of purified water or estimated from the “clearest” natural water. For brevity, the same term, a_w(λ), is used herein to represent the “pure water” absorption coefficient obtained from purified water or “clearest” freshwater or seawater. For the ocean color community, the a_w(λ) values from Smith and Baker [8] were generally accepted and used in the 1980’s and 1990’s, which were the upper bound of pure water (seawater) absorption coefficients estimated from the freshwater of Crater Lake, Oregon and the seawater of the Sargasso Sea. Since the 2000’s, the community has adopted the a_w(λ) values of Pope and Fry [9], referred to as a_w_PF97, as the “standard”, which were determined from carefully prepared “pure water” using an integrating cavity. Recently, new spectral values of a_w(λ) were obtained from remote sensing reflectance (R_rs(λ), in sr⁻¹) [10], referred to as a_w_Lee15, for “pure seawater” based on optical closure in the “clearest” natural waters in the South Pacific Gyre. Moreover, Mason et al. [11] improved the measurements of Pope and Fry [9] using an updated integrating cavity, and their measurements hereafter are referred to as a_w_Mason16. It is noted that the two recently obtained a_w_Lee15 and a_w_Mason16 are lower than a_w_PF97 in the blue-green domain (410–550 nm), especially in the blue domain where the differences are about a factor of 2. Given a_w(λ) is a critical input to semi-analytical algorithms for IOPs retrieval, it is logical and necessary to answer the question: what is the impact on the derived IOPs using a_w_Lee15 instead of using a_w_PF97 that has been adopted for decades?

In this study, a high-quality in-situ dataset is first employed to evaluate the validity of a_w_PF97 and a_w_Lee15 for IOPs retrievals in oligotrophic waters. The impacts of a_w(λ) on the derived IOPs in global oceans are further demonstrated using the Visible Infrared Imaging Radiometer Suite (VIIRS) data, followed by detailed discussions.

2. Data and methods

2.1 The absorption coefficients of “pure water” and “pure seawater”

Two sets of a_w(λ) data, a_w_PF97 and a_w_Lee15, are primarily evaluated in this study to demonstrate the impacts of a_w(λ) on the derived IOPs. Values of a_w_PF97 were determined from purified water by measuring its absorption and attenuation coefficients using an integrating cavity from 380 to 720 nm, and has been adopted by the ocean color community as a “standard”. Values of a_w_Lee15, on the other hand, were inverted from R_rs(λ) measured in the South Pacific Gyre for the spectral range between 350 and 550 nm. Specifically, R_rs(λ) is a function of the bulk absorption and backscattering coefficients (a(λ) and b_b(λ), in m⁻¹) [12,13]. With concurrently measured component IOPs and R_rs(λ), a_w_Lee15 was derived analytically from measured R_rs(λ) and the absorption coefficients of phytoplankton, suspended particulate matter and colored dissolved organic matter [10]. Note that the values of a_w_Lee15 beyond 550 nm were adopted from a_w_PF97, as a_w(λ) beyond 550 nm are found very consistent among various determinations [11]. The spectral values of a_w_PF97 and a_w_Lee15 are illustrated in Fig. 1 to present their differences in the 350–550 nm domain. In addition, a_w_Mason16, acknowledged as scattering-independent a_w(λ) measurements of laboratory-grade “pure water” prepared from a sophisticated purification system, is also included in Fig. 1 for comparison. Note that all the three a_w(λ) data were linearly interpolated to 1 nm interval to calculate the differences among two a_w(λ) data, and to obtain a_w(λ) at the VIIRS bands.

 

Fig. 1. Comparison of three spectral a_w(λ) reported in the literature: a_w_PF97 (black line, [9]), a_w_Lee15 (red dots, [10]), and a_w_Mason16 (magenta squares, [11]). The blue dashed and dotted lines, aligned to the right y-axis, represent the differences between a_w_PF97 and that of a_w_Lee15 and a_w_Mason16, respectively.

Download Full Size | PDF

2.2 In-situ measurements and satellite data

A globally compiled bio-optical in-situ dataset [14] was adopted in this study to evaluate the impacts of a_w_PF97 and a_w_Lee15. Specifically, the distributed ± 2 nm binned data at VIIRS bands, centered at 410, 443, 486, 551, and 671 nm, respectively, were employed after the following criteria applied: R_rs(λ) at the five VIIRS bands and concurrent phytoplankton absorption coefficient (a_ph(λ), in m⁻¹) at 443 nm are available and positive. The spectral quality of R_rs(λ) was later checked by the quality assurance score (QA) [15], which ranges between 0 and 1 with 0 stands for questionable data and 1 for perfect quality. In this study, only high-quality R_rs(λ) data (QA ≥ 0.8) were retained for the subsequent analysis. As a result, 736 out of 1004 paired measurements of R_rs(λ) and IOPs were used and are hereafter referred to as the Valente_16 dataset.

Given that the relative contribution of a_w(λ) to a(λ) in the blue-green domain is negligible in coastal waters where the contributions from CDOM and suspended particulates dominate, it is preferred to evaluate the a_w(λ) impacts on the derived IOPs in oligotrophic waters. For oligotrophic waters, such as ocean gyres, a(λ) at 443 nm is typically less than 0.02 m⁻¹, as demonstrated in the global IOPs products distributed by NOAA CoastWatch (figure not shown here). However, the Valente_16 dataset has no data falling within the range of a(443) < 0.02 m⁻¹. As an alternative, we selected the in-situ data with a(443) < 0.05 m⁻¹ to represent the oligotrophic water samples in the Valente_16 dataset and eventually obtained 74 paired measurements in total.

In addition to the in-situ dataset, satellite data were also employed to provide a synoptic view of a_w(λ) impacts on the derived IOPs in global oceans. The monthly-binned science-quality VIIRS product of normalized water-leaving radiance (nL_w(λ), in W/m²/µm/sr), distributed by NOAA CoastWatch, is selected for the demonstration. Specifically, one scene of the monthly-binned product for May 2018 with a spatial resolution of 4 km (‘V2018121_2018151_D31_WW00.nc’) was downloaded in this effort. The NOAA VIIRS ocean color Level-2 data were processed with the Multi-Sensor Level-1 to Level-2 (MSL12) ocean color data processing system at the NOAA Center for Satellite Applications and Research (STAR). The procedure for monthly-binned Level-3 data is explained in the documents for NOAA ocean color products (Algorithm Theoretical Basis Document (ATBD), version 1.0, June 2017).

2.3 IOPs retrieval algorithms

Two well-known semi-analytical algorithms were employed in this study to derive IOPs from R_rs(λ). The first one is the quasi-analytical algorithm (QAA) [16], which is a stepwise algorithm using algebraic equations to solve the absorption and backscattering coefficients of different water components. The latest update (version 6) of QAA, referred to as QAAv6 herein, can be found at http://www.ioccg.org/groups/Software_OCA/QAA_v6_2014209.pdf. The other algorithm is the generalized IOPs algorithm (GIOP) [17], which derives three eigenvalues that parameterize the spectral absorption coefficients of phytoplankton (a_ph(λ)), colored dissolved organic matter and non-algae particles (a_dg(λ)), and the spectral backscattering coefficients of suspended particles (b_bp(λ)) using spectral optimization. For simplicity, GIOP in this effort employed all default settings, as described in Werdell et al. [17], except for a_w(λ). The backscattering coefficients of pure seawater (b_bw(λ)) in both QAAv6 and GIOP were adopted from Morel [18].

Note that the inelastic scattering processes, such as the Raman scattering, also contribute photons to R_rs(λ), especially in the longer wavelengths for oligotrophic waters [19]. Therefore, prior to the implementation of QAAv6 and GIOP, a simple correction on R_rs(λ) for the Raman-scattering effect was carried out for the Valente_16 dataset and the VIIRS data according to Eqs. (11) and (13) of Lee et al. [20].

2.4 Statistical measures

The slope and the coefficient of determination (R²), determined from the Model-II regression [21], and the median absolute percentage difference (MAPD), were calculated to quantitatively evaluate the agreement between the derived IOPs and the measured IOPs. For demonstration with satellite data, the absolute percentage difference (APD) was also introduced to assess the difference in the derived IOPs using the two different a_w(λ), with the derived IOPs using a_w_PF97 as the reference. The MAPD and APD are defined as,

3. Results

3.1 Validations with Valente_16 dataset

For the oligotrophic water samples in the Valente_16 dataset, validations of the derived a_ph(λ) and a_dg(λ) from both GIOP and QAAv6 using a_w_Lee15 and a_w_PF97, respectively, are shown in Fig. 2. Here examples are presented for the derived IOPs at 410 nm, which is because the relative difference between a_w_PF97 and a_w_Lee15 is the largest at 410 nm for the VIIRS bands (see Fig. 1). A validation of the derived b_bp(λ) using the two a_w(λ) could not be done here because no b_bp(λ) measurements can be found from the oligotrophic subset of the Valente_16 dataset.

 

Fig. 2. Validation of R_rs(λ)-derived a_ph(410) and a_dg(410) from GIOP and QAAv6 and using a_w_Lee15 and a_w_PF97, respectively, for the oligotrophic subset of the Valente_16 dataset. The upper and lower panels present the results of the derived a_ph(410) and a_dg(410), respectively.

Download Full Size | PDF

As shown in Fig. 2, the derived a_ph(410) data using a_w_Lee15 have shown improved agreement with the in-situ data compared to the retrievals using a_w_PF97 for both GIOP and QAAv6. Statistically, MAPDs of the derived a_ph(410) using a_w_Lee15 are reduced by 7% and 10% compared to the retrievals using a_w_PF97 for GIOP and QAAv6, respectively. Also, the derived a_ph(410) using a_w_PF97 are systematically underestimated, as shown in Figs. 2(b) and 2(d). In contrast, the derived a_ph(410) using a_w_Lee15 are greater than the retrievals using a_w_PF97 and agree better with the measured a_ph(410) (see Figs. 2(a) and 2(c)), indicating that a_w_Lee15 could be more appropriate to represent the “pure seawater” absorption coefficient and for the retrieval of a_ph(λ) in oligotrophic waters. As shown in Figs. 2(e)–2(h), the changes in a_w(λ) only have very small impacts on the derived a_dg(410) for both GIOP and QAAv6, with slightly improved retrievals using a_w_Lee15 than that of using a_w_PF97.

The a_w(λ) impacts on the derived IOPs at other VIIRS blue bands are further presented in Fig. 3 for the oligotrophic subset of the Valente_16 dataset. Here only results of the derived a_ph(λ) and a_dg(λ) at 410, 443, and 486 nm are presented, as the measured R_rs(λ) and IOPs at 551 and 671 nm have more uncertainties due to very small values. The a_w(λ) impacts on the derived b_bp(λ) will be discussed in Section 3.2 and Section 4.

 

Fig. 3. Evaluation of the derived IOPs at three VIIRS bands by GIOP and QAA6 using a_w_Lee15 and a_w_PF97, respectively. The left two boxes in each subfigure represent the results of GIOP and the right two boxes for QAAv6. The upper and lower panels present the results of the derived a_ph(λ) and a_dg(λ), respectively. Calculations of the APDs were based on the same oligotrophic subset used in Fig. 2.

Download Full Size | PDF

On each box in Fig. 3, the central red mark indicates the MAPD, and the bottom and top edges of the box represent the 25^th and 75^th percentiles, respectively. The whiskers extend to the most extreme data points not considered as outliers, and the outliers are plotted individually using the red cross symbol. As shown in Fig. 3, MAPD of the derived a_ph(λ) are overall smaller for the retrievals using a_w_Lee15 for both GIOP and QAAv6 (see Figs. 3(a)–3(c)), suggesting that more accurate retrievals of a_ph(λ) are obtained when a_w_Lee15 is employed. Further, as shown in Figs. 3(d)–3(f), it appears that the selection of different a_w(λ) data has minimal impacts on the derived a_dg(λ) for both GIOP and QAAv6, as MAPD of the derived a_dg(λ) using the two a_w(λ) are quite comparable. The minimal impacts of a_w(λ) selection on the derived a_dg(λ) can be well explained by the spectral feature of the difference between a_w_Lee15 and a_w_PF97 in the blue domain.

For QAAv6, a_w(λ) are primarily employed in the steps to decompose the algebraically solved a(λ) into a_dg(λ) and a_ph(λ). Because the differences between a_w_PF97 and a_w_Lee15 in the ∼410–470 nm domain have a spectral pattern that matches the spectral feature of a_ph(λ) (see the blue dashed line in Fig. 1), the use of a_w_Lee15 rather than a_w_PF97 will have more impacts on the derived a_ph(λ), instead of a_dg(λ). In other words, the difference in using a_w_Lee15 rather than using a_w_PF97 is well compensated by a change of a_ph(λ). Moreover, QAAv6-derived b_bp(λ) are only affected by the difference between a_w_PF97 and a_w_Lee15 at 550 nm (see Table 2 of Lee et al. [16]), which is, however, only about 0.53%. Therefore, using a_w_PF97 or a_w_Lee15 will have limited impacts on the derived b_bp(λ).

For GIOP, however, changes in a_w(λ) will generally be propagated to all three eigenvalues due to the use of spectral optimization with all available R_rs(λ) in the visible domain as the input. Again, because the differences between a_w_PF97 and a_w_Lee15 in the blue domain have a similar spectral pattern to a_ph(λ), and that a_ph(λ) and a_dg(λ) do not contribute much to the total absorption in the green band for oligotrophic waters, the differences between a_w_PF97 and a_w_Lee15 are generally compensated by the changes of a_ph(λ), rather than the changes of a_dg(λ) and b_bp(λ).

3.2 Demonstrations with VIIRS data

Validation with the in-situ dataset shows that a_w_Lee15 could be more appropriate for IOPs retrievals in oligotrophic waters. Here, NOAA VIIRS monthly-binned data are employed to compare the retrieved IOPs of global waters using the two spectral a_w(λ) and to demonstrate the impacts of a_w(λ). Figure 4 shows the comparisons of derived a_ph(443) using a_w_Lee15 and a_w_PF97 for the two inversion algorithms.

 

Fig. 4. Global distribution of derived a_ph(443) from VIIRS data using a_w_Lee15 and a_w_PF97 and the absolute percentage difference between the two derived a_ph(443). The left panel presents the results from GIOP while the right panel for QAAv6. The red box superimposed on panels (e) and (f) highlights the oligotrophic core zone in the North Atlantic Gyre, defined by Morel et al. [22], for statistical analysis in Table 1.

Download Full Size | PDF

Table 1. The range (minimum - maximum) of derived IOPs and APD for the selected oligotrophic core zone in the North Atlantic Gyre, as highlighted in Fig. 4(e). The number in the brackets implies the median value.

View Table

As shown in Figs. 4(a)–4(d), the primary differences between the derived a_ph(443) using the two a_w(λ) are observed in oligotrophic waters, such as the ocean gyres, where the derived a_ph(443) using a_w_Lee15 are much higher than the retrievals using a_w_PF97. As explicitly shown in Figs. 4(e) and 4(f), APD of the derived a_ph(443) using the two sets of a_w(λ) is found over 30% in the ocean gyres and increases rapidly towards the gyres’ core. For instance, in the oligotrophic core zone of the North Atlantic Gyre (hereafter referred to as cNAG), defined according to Morel et al. [22] and highlighted in Figs. 4(e) and 4(f), the derived a_ph(443) using a_w_Lee15 are up to 146% and 238% larger than that of using a_w_PF97 for GIOP and QAAv6, respectively. Detailed statistical results regarding the derived component IOPs within cNAG are tabulated in Table 1. Also, consistent with the validation results with the Valente_16 dataset, a_w(λ) impacts on the derived a_ph(443) are less significant for GIOP, where a smaller median APD of 59.2% is found compared to the median APD of 68.2% for QAAv6. Moreover, as shown in Figs. 4(e) and 4(f), for optically complex waters, such as most coastal waters, a_w(λ) impacts on the derived a_ph(443) for both GIOP and QAAv6 can be neglected given that the APD is generally less than 2%.

Comparisons of the derived a_ph(λ) at other VIIRS bands are not shown here for brevity, but a_w(λ) impacts on the derived a_ph(λ) at other wavelengths are deducible. For QAAv6, a_w(λ) impacts on the derived a_ph(λ) are wavelength-dependent, which are determined by both the difference between a_w_Lee15 and a_w_PF97 and the magnitude of the derived a_ph(λ) at the specific wavelength. The impacts of a_w(λ) on the derived a_ph(λ) beyond 550 nm would be negligible. For GIOP, on the other hand, a_w(λ) impacts on the derived a_ph(λ) would be wavelength-independent, as it is a scaling of the pre-determined spectral shape of a_ph(λ).

Demonstrations for QAAv6-derived global a_dg(λ) and b_bp(λ) are not presented here, as the impacts of a_w(λ) (between a_w_Lee15 and a_w_PF97) on QAAv6-derived a_dg(λ) and b_bp(λ) are expected to be relatively small (see Fig. 2 and Fig. 3). As tabulated in Table 1, the median APD of QAAv6-derived a_dg(443) and b_bp(443) within cNAG are only 1.5% and 0.9%, respectively. Moreover, because a_dg(λ) and b_bp(λ) are based on spectral models in QAAv6, the impacts of a_w(λ) on QAAv6-derived a_dg(λ) and b_bp(λ) would be wavelength-independent.

The impacts of a_w(λ) on GIOP-derived a_dg(λ) and b_bp(λ) are further evaluated, and demonstrations for the derived a_dg(λ) and b_bp(λ) at 443 nm are presented in Fig. 5 as an example. As expected, the differences between a_w_Lee15 and a_w_PF97 also result in discrepancies in the derived a_dg(λ) and b_bp(λ), but the impacts are much less significant than that of the derived a_ph(λ), with APDs of a_dg(443) and b_bp(443) are generally less than 10% for the ocean gyres (see Figs. 5(e) and 5(f)). Specifically, for the cNAG, the median APDs for GIOP-derived a_dg(443) and b_bp(443) are 6.6% and 7.3%, respectively (see Table 1 for details). In the non-oligotrophic waters, a_w(λ) impacts on GIOP-derived a_dg(λ) and b_bp (λ) are negligible with APD generally less than 1% (see Figs. 5(e) and 5(f)).

 

Fig. 5. Same as Fig. 4, but for GIOP-derived a_dg(443) and b_bp(443). The left panel presents the results for a_dg(443) and the right panel for b_bp(443).

Download Full Size | PDF

4. Discussion and conclusions

Lee et al. [10] argued that a_w_ PF97 could be less reliable than a_w_Lee15 in oligotrophic waters from indirect validation results, as using a_w_PF97 resulted in negative retrievals of a_ph(488) and a wider range of a_ph(488)/a_ph(443) that was inconsistent with field measurements. The present validation results with the Valente_16 dataset confirm that a_w_Lee15 is more appropriate for IOPs retrievals in oligotrophic waters. This could be explained by the uncertainties of a_w_PF97 in the ultraviolet-blue domain due to several limitations in the measurement, such as difficulties in preparing laboratory-grade “pure water”, inadequate instrument sensitivity for low absorption measurements, and insufficient correction of the scattering effects. In addition, the integrating cavity used in Pope and Fry [9] was made of the diffuse reflecting material Spectralon, which could absorb light between 380 and 440 nm [11].

The questionable quality of a_w_PF97 within the blue-green domain is also confirmed when comparing it with a_w_Mason16 (see Fig. 1), which is considered to be of higher accuracy as the aforementioned issues in the a_w_PF97 measurement have been largely addressed with more advanced instrumentation. Demonstration for the impacts of using a_w_Mason16 on the derived IOPs at the global scale is not presented in this study, but the derived IOPs using a_w_Mason16 could be deduced. The a_w(λ) impacts on the derived IOPs in oligotrophic waters depend mainly on the spectral pattern of the difference between the two a_w(λ) data. As shown in Fig. 1, the spectral pattern of the difference between a_w_PF97 and a_w_Mason16 (see the blue dotted line in Fig. 1) is similar to that of a_dg(λ). Therefore, the use of a_w_ Mason16 will likely have more impacts on the derived a_dg(λ), rather than a_ph(λ). Here we compared the impacts of using a_w_Mason16 and a_w_Lee15 on the derived IOPs, with retrievals using a_w_PF97 as the reference, and presented the results in Fig. 6 with the derived IOPs at 410 nm as the example. As shown in Fig. 6, compared to the use of a_w_PF97, a use of a_w_ Mason16 will result in a significant difference in the derived a_dg(410), while at the same time, quite small differences in the derived a_ph(λ) and b_bp(λ). This pattern of impact is opposite to the use of a_w_Lee15. In other words, a use of a_w_Mason16 will not improve much on the retrieval of a_ph(λ) in oligotrophic waters; on the other hand, a use of a_w_Mason16 will negatively impact the retrieval of a_dg(λ) in oligotrophic waters (MAPD increased from 19.2% to 21.8% for QAAv6-derived a_dg(λ) for the oligotrophic subset of the Valente_16 dataset, figure not shown here).

 

Fig. 6. The impacts of a_w_Mason16 on the derived IOPs at 410 nm compared to that of using a_w_Lee15. Note that the derived IOPs using a_w_PF97 were used as the reference to calculate the APD, and the same oligotrophic subset used in Fig. 2 was employed here.

Download Full Size | PDF

Again, there is not a significant impact on the retrieval of b_bp(λ) (see Fig. 6(c)) when a_w_Mason16 is used, which is because b_bp is primarily determined from R_rs in the longer wavelengths where the absorption coefficient of pure seawater plays a major role [23]. Therefore, the retrieved b_bp(λ) by both QAAv6 and GIOP will be less sensitive to the changes of a_w(λ) in the blue.

Further, it is necessary to point out that a_w_Lee15 represents the absorption coefficient of “pure seawater”, while a_w_Mason16 represents the absorption coefficient of “pure water”. The significant divergences between a_w_Lee15 and a_w_Mason16 for λ < ∼410 nm are likely due to the impact of sea salts [11]. Thus, a_w_Lee15 might be more appropriate for applications of oceanic waters, especially the use of a_w_Mason16 will not improve the retrieval of a_ph(λ) in oligotrophic waters (see Fig. 6(a)). Also, note that there will be more measurements or sensors that are capable of measuring R_rs(λ) in the 350–410 nm domain, such as the Ocean and Land Colour Instrument (OLCI) and the forthcoming Plankton, Aerosol, Cloud, ocean Ecosystem (PACE) mission, thus a validation of a_w_Lee15 in the 350–410 nm domain demands more field measurements in the ocean gyres.

Assuming the derived IOPs using a_w_Lee15 are closer to the “true” IOPs in natural oceanic waters, the demonstrations here with the VIIRS data indicate larger uncertainties of the derived IOPs using a_w_PF97 in oligotrophic waters, especially for the derived a_ph(λ). Note that a_w_PF97 has been accepted as the “standard” a_w(λ) in the past two decades, the uncertainty of a_w_PF97 will be propagated not only to the derived IOPs but also to the IOPs derivative products. Due to the comprehensive observation by ocean color satellites, remotely sensed IOPs had been widely used in interdisciplinary studies to investigate ocean biological processes and global biogeochemical cycles [24,25]. For instance, the values of a_ph(λ) were adopted in estimating the primary production [26], phytoplankton pigment concentrations [27], and phytoplankton size spectrum and size classes [28,29]. The significant underestimation of the derived a_ph(λ) using a_w_PF97 would introduce uncertainties to these derivative products, especially for applications in oligotrophic waters. Take the estimation of primary production for example, when it uses a_ph(443) to scale up [26], the estimated primary production in cNAG could be up to 248% lower if a_w_PF97 is employed to derive a_ph(443) instead of a_w_Lee15.

More importantly, although the a_w(λ) impacts on the derived IOPs are only evaluated for QAA and GIOP in this study, it is reasonable to assume that a_w(λ) may have similar impacts on the derived component IOPs of oligotrophic waters for other semi-analytical inversion algorithms [30–32]. This is mainly because R_rs(λ) is a function of bulk absorption and backscattering coefficients, the changes in a_w(λ) will be inevitably compensated by the derived IOPs of other components when a semi-analytical algorithm is applied.

To conclude, this study shows that the commonly accepted a_w_PF97 by the ocean color community could introduce large uncertainties in remotely sensed IOPs in oligotrophic waters, especially for a_ph(λ). Therefore, it is important for the community to adopt a more appropriate spectral a_w(λ) in the practices of IOPs retrievals at global scales, and a_w_Lee15 is recommended before more convincing a_w(λ) data become available for “pure seawater”.