1. Introduction

Particles suspended in seawater such as phytoplankton, detritus and mineral particles play an important role in the determination of optical properties of water [1,2]. Sunlight that enters the sea surface is either absorbed or scattered by suspended particles [3]. Light scattered in a backwards direction is of fundamental importance for many marine biogeochemical processes that involve optics [4,5]. In particular, it is essential for interpreting satellite ocean color data [6,7]. A detailed understanding of how particle assemblages affect backscattering properties is thus of importance for studies on water radiative transfer, remote sensing of ocean color and biogeochemical processes coupling optics based on in situ and satellite measurements [8].

According to Mie theory, assuming a spherical shape and material homogeneity for a given particle, the backscattering coefficient of particles (b_bp(λ)) at a given wavelength can be expressed as the sum of the product of the backscattering efficiency and the cross-sectional area of all particles [9]. Using the mean backscattering efficiency of all particles weighted by area (Q_bbe(λ)) and the total cross-sectional area concentration (AC), the calculation of Mie theory can be simplified as [10,11],

The results of several theoretical, experimental and field studies have shown that backscattering efficiency can significantly vary with changes in particle compositions and size distributions [7,8,11,19,21,35–37]. Based on theoretical principles, Stramski et al. [8] investigated the effects of particle compositions on backscattering efficiency and showed that when soft organic particles transform into mineral particles, backscattering efficiency levels increase by a factor of 30. Vaillancourt et al. [7] calculated backscattering efficiency levels at 620 nm for cultured 28 phytoplankton species of 11 marine classes and showed that backscattering efficiency levels vary from 0.001 to 0.068. In reference to the Bedford Basin, based on in situ observations made before and after a spring phytoplankton bloom, Flory et al. [36] found that the efficiency of backscattering at 589 nm decreased by a factor of nearly 4 when organic particles gradually became dominant and detrital particles began to aggregate with the starting and the progression of the blooming. More recently, along the west coast of Great Britain, Bowers et al. [19] observed that the backscattering efficiency level at 665 nm changes by an order of magnitude when the proportion of inorganic matter to TSM changes from 35% to 90%. Meanwhile, they found a strong positive relationship (R² = 0.62) between backscattering efficiency and the ratio of mineral particles to TSM, implying that particulate backscattering efficiency may be a good indicator for detecting particle compositions.

It is well known that particle assemblages often undergo dynamic responses to changes in hydrographic environments of water columns [38]. In the open ocean, suspended particles are mainly composed of phytoplankton cells and of their detrital products; in coastal regions, with the exception of phytoplankton and detritus, mineral particles introduced through river discharge, upwelling, mixing, etc., are also abundant [19]. Complex and dynamic particle compositions in coastal waters may consequently result in considerable variability in backscattering efficiency levels and thereby in b_bp(λ) and b_bp*(λ). Recently, based on field measurements collected during the late summer in two typical shallow and semi-enclosed seas, the Bohai Sea (BS) and Yellow Sea (YS), we found that the backscattering efficiency of suspended particles at 640 nm significantly vary in magnitude [21]. Meanwhile, backscattering efficiency levels were clearly clustered into two types. Type 1 samples containing relatively high proportions of organic or large particles were found to show low levels of backscattering efficiency and the backscattering efficiency of type 2 samples, which are mainly composed of relatively small mineral particles, was found to be high. Overall, as stated above, while some studies have examined the variability of particle backscattering properties with regards to biogeochemical variables, effects of backscattering efficiency on b_bp(λ), b_bp*(λ) and thus ocean color remote sensing and the extent to which such effects can be compared with respect to particle concentrations, density levels and size distributions still need to be completely examined and quantified.

In this paper, based on field observations made in the BS and YS over three seasons, i.e., late spring, late summer and late autumn, we first investigate backscattering efficiency variability levels and examine whether the clustering of backscattering efficiency as observed in the late summer in Wang et al. [21] is supported in other seasons. Subsequently, a simple scheme for particle classification based on backscattering efficiency is proposed. Finally, effects of backscattering efficiency on b_bp(λ) and b_bp*(λ) are quantified, and potential effects of particle types on ocean color remote sensing are discussed.

2. Materials and methods

2.1 Study area and sample collection

In this study, we report on field data collected from the BS and YS. The BS is a shallow sea with an average depth of 18 m [39]. The Yellow River, as the second largest sediment-load river in the world, flows into the BS with a total runoff of 890 × 10⁸ m³ each year [40]. The YS is also a relatively shallow sea with an average water depth of 44 m [39]. It is located between Mainland China and the Korean Peninsula and is connected to the BS to the north through the Bohai Strait. The topography of the YS in the central region is rather smooth but steep on both the western and eastern sides. These two relatively shallow seas are heavily influenced by monsoons in the winter and by freshwater discharge in summer. Interactions among winds, bottom topography, freshwater discharge, and tidal forcing cause significant regional and seasonal variability in water properties in the BS and YS [41]. Meanwhile, surrounding areas of the BS and YS are the most economically developed regions in northern China. The rapid proliferation of industries, agriculture, aquaculture and domestic sewage from these areas has seriously polluted the BS and YS over the past several decades [42].

Three cruises were conducted in the BS and YS during the late spring (May 2014), late summer (August 2015) and late autumn (November 2014). The locations of observation stations employed by the three cruises were generally similar [Fig. 1]. At each station, a profiling package with a HOBI Labs Hydroscat-6, Sequoia Scientific LISST-100X (type C) and Seabird SBE911P conductivity-temperature-depth (CTD) profiler was used to synchronously measure light backscattering, particle size distribution (PSD) and water column temperature and salinity values, respectively. The package was deployed in a surface layer (roughly 5 m) for several minutes to allow the equilibration of the sensor temperature and seawater temperature. The package was then lifted to the surface and slowly (approximately 0.2 m s⁻¹) lowered to the depth 2–3 m just above the bottom to measure a vertical profile of the water column. To avoid package perturbations to the water column, only downward looking measurements were used for our data analysis. These are denoted ‘downcast’ measurements in this study. Simultaneously, water samples were collected from three layers (the surface, the middle layer just below the mixing depth and the bottom layer) to measure TSM values. In addition, to determine remote sensing reflectance (R_rs(λ)) values, a Satlantic Hyper-Profiler II radiometer was performed whenever the observing condition was appropriate.

 

Fig. 1 Locations of sampling stations in the Bohai Sea (BS) and Yellow Sea (YS) during May 2014, November 2014 and August 2015.

Download Full Size | PDF

2.2 Particle mass concentration, backscattering and size distribution measurements

TSM concentrations were determined using a gravimetric technique. Seawater (0.5–2 L) collected using Niskin bottles was filtered onto pre-weighed 47-mm Whatman GF/F glass fiber filters under low vacuum pressure levels (< 0.01 MPa). To remove salt, the filters were rinsed three times using 50 ml MilliQ water after filtration and then were kept frozen at –20 °C until being dried at 105° for 4 h in a laboratory. These filters were then reweighed to determine TSM concentrations.

Total backscattering coefficients (b_b(λ)) at 410, 442, 488, 532, 550, and 640 nm were measured using a HOBI Labs Hydroscat-6, which was calibrated before the cruise to confirm its performance within factory specifications. This instrument records the total volume-scattering function at a fixed angle (approximately 140°) in a backwards direction [43]. To improve the accuracy of the backscattering measurements, path length attenuation effects were corrected using the sigma correction method. The absorption and scattering data used in sigma correction for the data in August 2015 were obtained from the simultaneous measurements of a WET labs AC-S, while those used in sigma correction for the data in May and November 2014 were estimated using the empirical models provided in Hydrosoft. The backscattering coefficient of particles b_bp(λ) was then calculated by subtracting the backscattering coefficient of pure water from the b_b(λ). Further information on the processing procedure can be found in the HOBI Labs User’s Manual.

An LISST-100X Type-C particle size analyzer (Sequoia Scientific Inc.) was used to determine the particle size distribution (PSD). The background scattering of LISST instrument was acquired before the cruise. In brief, this instrument measures the diffraction pattern produced by suspended particles in a volume of water. Based on Mie theory calculations, the diffraction pattern is used to determine the particle volume concentrations with a mean diameter of 32 sized bins that are logarithmically placed across a continuous size spectrum from 2.5 to 500 μm [44,45]. Volume concentrations and mean sizes of particles were processed using the LISST-SOP software program provided by the manufacturer (LISST-100X Particle Size Analyzer, 2013). Meanwhile, as noted in previous studies [11,46], LISST instrument measurements are not typically stable at the smallest and largest size ranges, likely due to the presence of particles that are smaller or larger than the measured size range. The instability of the smallest size ranges may also be associated with stray light [47]. Therefore, data from the smallest and largest size ranges were excluded in our analysis.

2.3 Particle backscattering efficiency determination

To determine the backscattering efficiency of particles, we first calculated the cross-sectional area concentration AC from LISST measurements of the volume concentration for a mean diameter. By assuming particles are spherical, the cross-sectional area concentration of particles of the ith size bin (AC_i) is calculated as,

Meanwhile, the mean apparent density ρ_a and mean diameter D_A of particles was calculated [10,11], to compare their effects on backscattering properties with those of Q_bbe(λ), as follows:

At this stage, it should be noted that there is a potential mismatch in the particle size ranges of the measurements of TSM, AC, and b_b(λ). The TSM represents the mass concentration of particles retained on the GF/F filter (nominal pore size of 0.7 μm), which have a size larger than approximately 0.4 μm [11]. The AC determined in this study accounts for the particles between 3.2 and 390 μm, due to the limited capability of the LISST measurement. Meanwhile, there is no size limitation for b_bp(λ) measured by the Hydroscat-6. Such particle size discrepancies of various measurements may bring uncertainties in analysis of bio-optical properties of this study, such as potential overestimation of b_bp*(λ) and Q_bbe(λ). Particularly, due to the mismatch in particle size ranges between b_bp(λ) and AC measurements, the key parameter Q_bbe(λ) obtained in this study may be not strictly the true particulate backscattering efficiency. Thus, it was referred to pseudo-backscattering efficiency (P_Q_bbe(λ)) hereafter. However, note that if the particles smaller than 3.2 μm and larger than 390 μm have consistent contribution proportion to AC for all of the samples, the P_Q_bbe(λ) may be proportional to the true backscattering efficiency. In this case, the clustering characteristics of P_Q_bbe(λ) obtained in this study should represent those of the true backscattering efficiency, though the clustering boundary may be not consistent with the natural condition. Despite these limitations, the LISST-derived particle size and AC (used to calculate P_Q_bbe(λ) in this study) are useful parameters, and have been commonly used to understand particulate backscattering, scattering and attenuation properties guided by the optical theory [10,11,28–34]. In addition, in the following analysis, we focuses on the P_Q_bbe(λ) variability at 640 nm (P_Q_bbe(640)) and consequent effects on optical properties to be consistent with the results of the previous study [21].

2.4 Remote sensing reflectance measurements

Remote sensing reflectance R_rs(λ) at daytime stations was derived from Satlantic Hyper-Profiler II measurements. This instrument includes a deck radiometer that measures above-water downwelling irradiance (E_d(λ, 0⁺)) values and two underwater radiometers that measure vertical profiles of downwelling irradiance (E_d(λ, z)) and upwelling radiance (L_u(λ, z)). The three radiometers were inter-calibrated by the manufacturer before the cruise. The measured spectra ranged from 349 to 804 nm with a mean bandwidth of approximately 3.3 nm.

While making our observations, a deck sensor was mounted on the deck to prevent shadows from the ship’s structure from having an effect. Underwater sensors were stabilized in the surface waters for several minutes using a hand-controlled cable. This allowed the instrument to drift further away from the ship to avoid contact with the ship’s shadow and to also equilibrate the sensor and seawater temperatures. Subsequently, the underwater sensors cluster was lowered from the surface in a vertically free-falling mode until they reached the euphotic layer (at a depth of 1% surface photosynthetically active radiation) [48]. We only used downcast radiometric measurements to perform calibration, data filtering, binning and interpolation based on the Prosoft 7.7.16 software program supplied by the manufacturer [49]. Data with tilt angles of > 5° or/and a downward velocity of > 0.5 m s⁻¹ were excluded during processing. The R_rs(λ) spectra were then calculated as

3. Results

3.1 Backscattering efficiency variability

Large variability in the pseudo-backscattering efficiency P_Q_bbe(640) was observed within and across different seasons in the BS and YS [Table 1]. During the late spring, for the samples combined in the surface, middle and bottom layers, the P_Q_bbe(640) results varied from 0.0022 to 0.1049 with a mean value of 0.0277 and a standard deviation of 0.0183 (coefficient of variation of 65.9%). During the late summer, the P_Q_bbe(640) data showed more significant variability, covering a range of 0.0022 to 0.1884 (mean ± standard deviation of 0.0319 ± 0.0268). The coefficient of variation correspondingly increased to 84.0%. In contrast, P_Q_bbe(640) variability in the late autumn decreased compared with that in the late spring and late summer. The coefficient of variation for the late autumn was the lowest at 41.7%, and the mean value of the P_Q_bbe(640) data was the highest at 0.0338 (standard deviation of 0.0141).

Table 1. Statistics on the P_Q_bbe(640) from different water layers in terms of maximum (Max), minimum (Min), standard deviation (SD) and coefficient of variation (CV).

View Table | View all tables in this article

To investigate P_Q_bbe(640) variation patterns, we analyzed the frequency distributions of the P_Q_bbe(640) data in log-space. The logarithmic normality tests of these distributions were examined using Kolmogorov–Smirnov test with the null hypothesis of log-normal distribution, and the p-values were provided for each test. As is shown in Fig. 2, clear differences were observed in distributions of the P_Q_bbe(640) results across different seasons. Similar to the P_Q_bbe(640) distribution for the late summer reported by Wang et al. [21], the P_Q_bbe(640) data of the late spring showed two clear clusters with low and high mean values, and each was approximately log-normally distributed. However, in the late autumn, with exception of low P_Q_bbe(640) values found for some samples, the P_Q_bbe(640) values of most samples were high. More importantly, we found that unlike two normal distribution patterns for the late spring and late summer, the P_Q_bbe(640) for the late autumn generally included one cluster, and the distribution was similar to that of the cluster with high P_Q_bbe(640) values observed for late spring and late summer.

 

Fig. 2 Frequency distribution of the pseudo-backscattering efficiency at 640 nm (P_Q_bbe(640)) for samples collected during the late spring (a), late summer (b) and late autumn (c). Red dashes denote the bound for P_Q_bbe(640) classification. Blue lines denote the log-normal fitting curves. The p-values indicate the significance level of the logarithmic normality tests.

Download Full Size | PDF

To further identify differences in P_Q_bbe(640) across different seasons, the P_Q_bbe(640) values of each sampling layer for the late spring, late summer and late autumn were compared [Table 1]. In surface waters, the mean P_Q_bbe(640) value was the lowest in the late summer at 0.0139 but was the highest in the late autumn at 0.0283. The lowest level of P_Q_bbe(640) variability was observed in the late autumn with a coefficient of variation of 41.8%. These observations were also true for samples drawn from middle layers. In bottom layers, the P_Q_bbe(640) values for the late autumn also showed the lowest level of variability (coefficient of variation of 24.9%). The mean P_Q_bbe(640) for the late summer was the highest at 0.0566 while it was the lowest in the late spring (a mean value of 0.0353).

Comparisons between P_Q_bbe(640) frequency distributions for the surface, middle and bottom layers of the late spring, late summer and late autumn are shown in Fig. 3. Clearly, for the bottom layer, although absolute values of the P_Q_bbe(640) showed some differences across the three periods, distribution patterns were generally similar, showing a single cluster with high P_Q_bbe(640) values. However, for the surface and middle layers, the P_Q_bbe(640) distribution for the late autumn differed from that of the late spring and late summer. Generally speaking, the P_Q_bbe(640) values in the late spring and late summer displayed approximately two clusters, while values for the late autumn generated roughly a single cluster. Overall, comparisons made between Figs. 2 and 3 show that P_Q_bbe(640) variability for the late autumn clearly differs from that of the late spring and late summer, and these differences are mainly attributable to the surface and middle water layers. Such differences in P_Q_bbe(640) distributions are likely attributable to particle types related to various water conditions as discussed in a later section.

 

Fig. 3 Frequency distribution of the pseudo-backscattering efficiency at 640 nm (P_Q_bbe(640)) for samples collected during the late spring from surface (a), middle (b) and bottom water layers (c); for samples collected during the late summer from surface (d), middle (e) and bottom water layers (f) and for samples collected during the late autumn from surface (g), middle (h) and bottom water layers (i). Red dashes denote the bound for P_Q_bbe(640) classification. Blue lines denote the log-normal fitting curves. The p-values indicate the significance level of the logarithmic normality tests.

Download Full Size | PDF

3.2 Particle classification scheme based on backscattering efficiency

The analysis presented in section 3.1 confirms that although the number of clusters in P_Q_bbe(640) showed some seasonal differences, the clustering of particles is generally stable across seasons. Backscattering efficiency is a key variable related to backscattering properties shown in Eqs. (7) and (8). Therefore, in this paper, we propose a simple scheme to classify particles based on distribution patterns of the P_Q_bbe(640) results, in order to improve our understanding of particle backscattering properties and in turn the remote sensing of ocean color.

As is shown in Fig. 4(a), when we combined all of the samples collected from three water layers of in the late spring, late summer and late autumn, two clusters of samples were clearly found. Based on the frequency distribution of the P_Q_bbe(640) data, a threshold of the P_Q_bbe(640) data with a value of 0.0134 was empirically determined for the classification. It should be stated that determination of the empirical threshold depends on the bin size of the histogram, while it was found that changes of bin size caused negligible influences on the particle classification for our data set. The P_Q_bbe(640) values of type 1 samples showed a mean value of 0.0063 with a standard deviation of 0.0027 while the mean P_Q_bbe(640) of the type 2 samples was much higher than that of the type 1 samples, generating a value of 0.0397 ± 0.0200. Upon applying this classification scheme to the data for different seasons and to seasonal surface data, the samples were reasonably classified [Figs. 2, 3 and 4(b)].

 

Fig. 4 Classification of particle types based on the frequency distribution of the pseudo-backscattering efficiency of particles at 640 nm (P_Q_bbe(640)) for all of the samples (a) and surface samples (b). Red dashes denote a P_Q_bbe(640) threshold of 0.0134 for the classification. Blue lines denote the log-normal fitting curves. The p-values indicate the significance level of the logarithmic normality tests.

Download Full Size | PDF

Using the proposed classification method, 47.4%, 36.6% and 1% of the samples respectively drawn from the surface, middle and bottom layers during the late spring were classified as type 1; and corresponding 52.6%, 63.4% and 99% of the samples drawn from the surface, middle and bottom layers were respectively classified as type 2. During the late summer, 65.4% of the surface samples, 34.7% of the middle layer samples and 100% of the bottom layer samples were classified as type 1 particles, and the remainder were classified as type 2 particles. Type 1 particles collected from the surface and middle layers during the late spring and late summer mainly derived from coastal regions (data not shown). During the late autumn, most samples were classified as type 2 particles, accounting for 80%, 88.4% and 100% of samples collected from the surface, middle and bottom layers, respectively.

Meanwhile, for the potential application of the P_Q_bbe(640) -based classification scheme to satellite observations, we propose a threshold for Q_bbe(550) to classify the particles, as the band at 550 nm is often used for most current ocean color sensors. We found that P_Q_bbe(550) was strongly related to P_Q_bbe(640) with an R² value of 0.991 in log-log space (data not shown). Consequently, a threshold of 0.0157 for P_Q_bbe(550) is designated for classifying particles from satellite measurements in the future.

3.3 Backscattering efficiency effects on backscattering properties

In theory, the particulate backscattering coefficient b_bp(λ) is regulated by the backscattering efficiency value together with the cross-sectional area or by combined effects of backscattering efficiency, TSM and product of apparent particle density and mean diameter (ρ_aD_A)⁻¹ [Eqs. (1) and (7)]. Similarly, the second-order variability of b_bp(λ) represented by the mass-specific backscattering coefficient b_bp*(λ) is controlled by backscattering efficiency, TSM and (ρ_aD_A)⁻¹ values [Eq. (8)]. Here we quantify the extent to which the P_Q_bbe(640) can affect the relationship between backscattering and biogeochemical properties by applying a linear regression in log-log space. We analyze these relationships in log-log space with respect to their log-normal distributions.

For b_bp*(640), it was found that P_Q_bbe(640) clearly separated relationships between b_bp*(640) and (ρ_aD_A)⁻¹ into two patterns [Fig. 5]. When we combined all of the samples, a coefficient of determination R² with a value of 0.363 was found between b_bp*(640) and (ρ_aD_A)⁻¹, and an R² value of 0.673 was obtained between b_bp*(640) and P_Q_bbe(640) [Fig. 5 and Table 2]. Note that P_Q_bbe(640) calculations are dependent on b_bp(640) while (ρ_aD_A)⁻¹ is independent of b_bp(640). These observations imply that for all of the samples, only 36.3% of the variability in b_bp*(640) can be explained by (ρ_aD_A)⁻¹, and the remaining 63.7% can likely be attributed to P_Q_bbe(640), which generally agrees with the R² value of 0.673 between b_bp*(640) and P_Q_bbe(640). After classification, the effects of P_Q_bbe(640) on b_bp*(640) significantly decreased, showing that b_bp*(640) variability for type 1 and 2 samples is mainly attributed to (ρ_aD_A)⁻¹ with contributions of 60.9% and 71.5%, respectively. For the type 2 samples, this observation was generally consistent with the relationship between b_bp*(640) and P_Q_bbe(640), showing an R² value of 0.320. However, we found that for the type 1 samples, the R² value for the relationship between b_bp*(640) and P_Q_bbe(640) was only 0.021. This observation implies that b_bp*(640) variability in type 1 samples cannot be fully explained by (ρ_aD_A)⁻¹ and P_Q_bbe(640). Other factors may also modulate b_bp*(640) variability as we discuss below.

 

Fig. 5 Scatter plots of the mass-specific backscattering coefficient of particles at 640 nm (b_bp*(640)) versus the inverse of the product of particle apparent density and mean diameter [(ρ_aD_A)⁻¹] (a) and the particulate pseudo-backscattering efficiency at 640 nm (P_Q_bbe(640)) (b). Black, green and red lines denote fitted functions for all of the samples, type 1 samples and type 2 samples, respectively.

Download Full Size | PDF

Table 2. The coefficient of determination (R²) of the regression for bio-optical variables.

View Table | View all tables in this article

The relationships of b_bp(640) with AC, P_Q_bbe(640), TSM and (ρ_aD_A)⁻¹ were analyzed [Fig. 6 and Table 2]. It was noted that P_Q_bbe(640) calculation depends on b_bp(640), while AC, TSM and (ρ_aD_A)⁻¹ are independent of b_bp(640). In regards to b_bp (640) variability in all of the samples, the effects of P_Q_bbe(640) decreased to 20.8%, and most b_bp (640) variability was attributed to AC, which showed high R² values of 0.792. The classification of particles shows that 92.0% of the variability in b_bp (640) of type 2 samples was explained by AC, and corresponding effects of P_Q_bbe(640) may decrease to 8%. For type 1 samples, AC was responsible for 61.2% of b_bp (640) variability and still more than 30% of the variability remained unexplained, echoing results of the b_bp*(640) analysis presented above. Regarding factors concerning TSM, (ρ_aD_A)⁻¹ and P_Q_bbe(640), we found that for all of the samples, the majority (65.7%) of the variability in b_bp (640) was attributed to TSM; and contributions of (ρ_aD_A)⁻¹ was 14.1%. In this sense, P_Q_bbe(640) effects likely accounted for approximately 20.2%. After classification, 71.0% and 26.8% of the variability in b_bp(640) of type 2 samples was attributable to TSM and (ρ_aD_A)⁻¹, respectively, and P_Q_bbe(640) may thus only explain 2.2% of the variability. For the type 1 samples, TSM and (ρ_aD_A)⁻¹ were respectively responsible for 38.9% and 1.1% of b_bp(640) variability, and the remaining percentage of variability remained unexplained. Overall, from above regression analysis of the bio-optical variables [Figs. 5 and 6, Table 2], it was generally found that P_Q_bbe(640) effects on b_bp(640) decreased compared to effects on b_bp*(640) for all of the samples. However, we note that the clustering of particles based on P_Q_bbe(640) generated clearly different patterns in the resulting relationships between b_bp(640) and various biogeochemical properties (i.e., AC, TSM and (ρ_aD_A)⁻¹).

 

Fig. 6 Scatter plots of the particulate backscattering coefficient at 640 nm (b_bp(640)) versus the concentration of total suspended matter (TSM) (a), the concentration of cross-sectional area particles (b), the inverse of the product of particle apparent density and mean diameter [(ρ_aD_A)⁻¹] (a) and the particulate pseudo-backscattering efficiency at 640 nm (P_Q_bbe(640)) (b). Black, green and red lines denote fitted functions for all of the samples, type 1 samples and type 2 samples, respectively.

Download Full Size | PDF

3.4 Effects of backscattering efficiency on remote sensing reflectance

TSM is commonly used to refer to the amount of particle, and remote sensing algorithms have been developed to derive TSM from remote sensing reflectance R_rs(λ) values. Therefore, we further examined effects of the clustering of particles generated by P_Q_bbe(640) on R_rs(λ) values based on 86 samples with concurrent measurements of R_rs(λ) and bio-optical properties. Figure 7 presents scatter plots for R_rs(640) and b_bp(640) and those for R_rs(640) and TSM. We found that the R_rs(640) of all of the samples strongly depended on b_bp(640) showed an R² value of 0.927 between them. Meanwhile, relationships between R_rs(640) and b_bp(640) for type 1 and type 2 samples were similar. However, as was expected, clear differences were observed in relationships between R_rs(640) and TSM in terms of R² values and in terms of covariation trends between type 1 and type 2 samples. Comparisons made between Figs. 6(c), 7(a) and 7(b) show that disparate covariation R_rs(640) trends with TSM for both types of samples were mainly related to different relationships between b_bp(640) and TSM caused by the clustering of P_Q_bbe(640).

 

Fig. 7 Scatter plots of the remote sensing reflectance at 640 nm (R_rs(640)) versus the particulate backscattering coefficient at 640 nm (b_bp(640)) (a) and the concentration of total suspended matter (TSM) (b). Black, green and red lines denote fitted functions for all of the samples, type 1 samples and type 2 samples, respectively.

Download Full Size | PDF

Correlations between R_rs(λ) and TSM were also further analyzed for type 1 and type 2 samples at between 400 – 700 nm to investigate the potential effects of particle clustering on the remote sensing of ocean color. As is shown in Fig. 8, clear differences in the spectra of these correlations were found between type 1 and type 2 samples. For the type 2 samples, R_rs(λ) showed similar high correlations with TSM from 400 to 700 nm in agreement with the optical properties of typical mineral-dominated particles [6]. However, for type 1 samples, correlations between R_rs(λ) and TSM were obvious spectral dependence. In particular, in the range of 400 – 500 nm, correlations showed negative values; within a range of 650 – 700 nm, correlations clearly decreased. These negative correlations between 400 and 500 nm and low correlations between 650 and 700 nm may be resulted from strong phytoplankton absorption processes, as the type 1 samples were likely dominated by organic matter as was shown in our previous work [21]. These results suggest that the clustering of particles based on backscattering efficiency levels would generate different variation patterns in the relationships between biogeochemical variables and optical properties for different types of particles, and discretion should be used in particle clustering in the development of algorithms for estimating TSM and potentially other biogeochemical variables from R_rs(λ).

 

Fig. 8 The spectral variability of the correlation coefficient (R) between remote sensing reflectance R_rs(λ) and the concentration of total suspended matter in log-log space.

Download Full Size | PDF

4. Discussion

4.1 Backscattering efficiency variability in relation to hydrological environments

Backscattering efficiency of suspended particles is a key factor that links their biogeochemical properties with backscattering properties. As was observed in this study, the pseudo-backscattering efficiency at 640 nm P_Q_bbe(640) varies considerably and shows clear seasonal differences [Table 1 and Fig. 2]. For all samples collected in the late spring, late summer and late autumn, the P_Q_bbe(640) varied from 0.0022 to 0.1884 by a factor of roughly 85. This degree of variability is greater than those reported in previous studies based on experimental data and in situ observations [7,11,19,35–37], while 90% of the P_Q_bbe(640) values found in this study range from 0.0040 to 0.0620, which were generally consistent with reported value ranges. For instance, the values of backscattering efficiency collected at 620 nm from 28 phytoplankton cultures varies between 0.0014 and 0.0681 as reported by Vaillancourt et al. [7]; and the backscattering efficiency values at 650 nm for mineral particles falls within a range of 0.0460 to 0.0620 [37].

Although the P_Q_bbe(640) values of all of the samples are highly varied, we found that P_Q_bbe(640) values were generally clustered into two types; within each type, P_Q_bbe(640) variability levels significantly decreased compared to that of the all of samples [Fig. 4]. This observation complements the findings of our previous study conducted in the late summer [21]. More importantly, we found that such P_Q_bbe(640) clustering holds well in different seasons [Fig. 2], although the clustering number showed seasonal differences. Based on measurements of particle compositions and size distributions for the late summer and based on comparisons made with the backscattering efficiency values dominated by organic and mineral particles presented in other studies, Wang et al. [21] inferred that type 1 samples with a mean value of 0.0054 ± 0.0018 are mainly composed of relatively high proportions of organic or large particles and that type 2 samples with a mean value of 0.045 ± 0.0305 mainly contain relatively small mineral particles. Consistent with this, although the present study considers more data for the late spring and late autumn, the mean P_Q_bbe(640) values of the type 1 and type 2 samples were generally similar at 0.0063 ± 0.0027 and 0.0397 ± 0.0200, respectively. These findings indicate that P_Q_bbe(640) clustering is a natural phenomenon in the BS and YS.

As noted above, although P_Q_bbe(640) clustering generally holds well in different seasons, the occurrence of type 1 and type 2 samples for all data showed some seasonal differences [Figs. 2 and 3]. For bottom water layers, Q_bbe(640) distributions were generally similar across seasons, showing that almost all of the samples can be classified as type 2 that are likely dominated by mineral particles [Fig. 2]. However, in the surface and middle layers, clear differences were found in P_Q_bbe(640) distributions between late autumn, late spring and late summer. During the late spring and late autumn, type 1 samples that were likely dominated by organic or large particles contributed comparable proportions while during the late autumn, samples collected from the surface and middle layers were still mainly composed of type 2 samples. These seasonal differences in P_Q_bbe(640) distributions in surface and middle layers may be related to the respective dynamic hydrographic environments found during different seasons.

During the late spring and summer, offshore waters in the BS and YS are heavily stratified, favoring the dominance of phytoplankton. Furthermore, in coastal regions, strong tides in shallow regions often induce vertical mixing. In the late autumn, driven by the strong NE monsoon, most of the water columns in the BS and YS are thoroughly mixed [41]. Heavy water column mixing may draw mineral particles from the bottom to middle and surface layers [50,51], and this may change particle compositions and consequently P_Q_bbe(640) variability levels. To confirm this inference, we analyzed the spatial distributions of P_Q_bbe(640) and the boundary of the clustering in surface water layers in the three seasons [Fig. 9], and examined the profiles of temperature and salinity at observation stations with type 1 and 2 samples [Fig. 10]. In generally, the type 2 samples were mainly from the coastal region and the type 1 samples were more located offshore. Meanwhile, the water columns with type 1 samples at the surface layers showed typical temperature and salinity profiles of stratified water columns while those with type 2 samples at surface layers were thoroughly mixed. In addition, with the exception of vertical mixing, river discharge, and especially the Yellow River which carries large loads of suspended sediment to the BS, may also affect the composition of particles in the estuary. Overall, the seasonal dynamics of hydrographic environments of these waters may change particle assemblages, which may further induce significantly different variability and distributions in P_Q_bbe(640) in different seasons; and caution should be used in regard to the associated optical properties.

 

Fig. 9 Distributions of pseudo-backscattering efficiency P_Q_bbe(640) in surface layers in late spring (a), late summer (b) and late autumn (c). Thick black lines denote the isograms with the value of 0.0134 which is defined as a threshold for classifying type 1 (< 0.0134) and type 2 samples (> 0.0134).

Download Full Size | PDF

 

Fig. 10 Profiles of temperature (a) and salinity (b) for the observation stations showing type 1 samples in the surface layer. Profiles of temperature (c) and salinity (d) for observation stations showing type 2 samples in the surface layer. Green, blue and red lines denote samples collected during the late spring, late summer and late autumn, respectively.

Download Full Size | PDF

4.2 Implications for remote sensing

Considerable P_Q_bbe(640) variability in the BS and YS consequently induced significant effects on backscattering properties [Figs. 5 and 6 and Table 2]. For all of the samples, the P_Q_bbe(640) value caused roughly 63.7% of b_bp*(640) variability. For b_bp(640), although variability was mainly attributed to the AC or the TSM and (ρ_aD_A)⁻¹, the P_Q_bbe(640) data was still able to explain roughly 20.8% of the variability. More importantly, P_Q_bbe(640) clustering clearly differentiated the covariation patterns both for b_bp*(640) with (ρ_aD_A)⁻¹ and for b_bp(640) with AC and TSM. After classifying samples using the P_Q_bbe(640)-based scheme, the effects of P_Q_bbe(640) on b_bp*(640) and b_bp(640) was reduced. Moreover, especially the trends in the co-variation of b_bp*(640) and b_bp(640) with biogeochemical properties could be more accurately described.

For type 2 samples, we found that the effects of P_Q_bbe(640) on the b_bp*(640) and b_bp(640) decreased to roughly 28.5% and 8.0%, respectively. For Type 1 samples, although 60.9% and 61.2% of the variability in b_bp*(640) and b_bp(640) was attributed to (ρ_aD_A)⁻¹ and AC, respectively, due to weak correlations between P_Q_bbe(640) and b_bp*(640) and b_bp(640), it is difficult to ascertain the effects of P_Q_bbe(640); and large proportions of b_bp*(640) and b_bp(640) variability still remained unexplained. These observations may be related to the presence of non-spherical shaped particles. According to Mie theory, particles are assumed to be spherical in shape and to have material homogeneity. However, this assumption may not hold true for some particles and especially for some algal species that are complex in shape (e.g., chains, irregular structures, etc.) [10,11]. Meanwhile, as noted in section 2.3, there is a mismatch in size ranges between backscattering and biogeochemical measurements. The LISST measurements only cover a size range of between 3.2 and 390 μm, while the size range of measurements of the Hydroscat-6 and TSM determined by water sample filtration on GF/F glass fiber filters (effectively a pore size of 0.4 μm) were larger. Such problems regarding particle shapes and measured size ranges may introduce uncertainties in P_Q_bbe(640) variability levels and in relationships between b_bp*(640) and b_bp(640).

As was expected, the remote sensing reflectance R_rs(640), which is strongly dependent on b_bp(640), was also affected by clustering in P_Q_bbe(640), showing two obvious patterns in relationships between R_rs(640) and TSM [Fig. 7]. After we classified the samples using the P_Q_bbe(640)-based scheme, co-variation trends of R_rs(640) with TSM were more accurately identifiable for the type 1 and type 2 samples, respectively, although considerable variability was still found in these relationships. Meanwhile, as is shown in Fig. 8, the TSM values of the two sample types showed clearly different correlations with R_rs(λ) between 400 and 700 nm. These observations imply that classification of particles or water types must be carefully considered when using or developing remote sensing algorithms for deriving TSM in the BS and YS. To more accurately understand water properties from satellite measurements, studies focusing on water classifications have been conducted in other regions [52,53]. Most classification methods are empirically based on the typical spectral characteristics of R_rs(λ) for different water types. By contrast, the classification scheme used in this study is proposed based on the different backscattering properties of particles guided by optical theory. Currently, b_bp(λ) can be derived from satellite measurements using an inherent optical algorithm [54,55]. Meanwhile, our recent work has shown that the cross-sectional area AC can be estimated from R_rs(λ) [56], and both evaluations based on in situ and satellite data indicate promising performances of the algorithm. Once satellite b_bp(λ) and AC are available, the classifications of particles based on the P_Q_bbe(λ) becomes feasible. Here, we need to point out that there are also some studies, such as Twardowski et al. [57] and Boss et al. [14], and the study of Chen et al. [20] in BS and YS, to suggest that particulate backscattering ratio (ratio of b_bp(λ) to scattering coefficient) can be used to study particle composition. However, it must be admitted that the classification of particle based on particulate backscattering ratio may be not applicable to current ocean color satellite measurements, since retrieval of particulate scattering coefficient is difficult from these measurements in theory. Although Roesler and Boss [58] proposed an algorithm for deriving attenuation from current satellite ocean color observations and Ibrahim et al. [59] also presented an inversion algorithm for attenuation from polarized radiometry measurements; the algorithm of Roesler and Boss [58] is actually based on the information of backscattering and absorption, and that of Ibrahim et al. [59] may only be applied in future ocean color missions.

We must note that the R_rs(λ) values is not only regulated by particulate backscattering properties but also by the absorption of particles and colored dissolved organic matter (CDOM) [60]. This study mainly examined the effects of P_Q_bbe(λ) at 640 nm on backscattering properties and thereby on R_rs(λ), at which the absorption of particles and CDOM is negligible. However, absorption at blue bands and about 675 nm may be strong due to the strong absorption of phytoplankton. Therefore, the effects of P_Q_bbe(λ) on R_rs(λ) in visible regions with respect to the absorption properties must be further clarified in future works to thoroughly understand the potential effects of particle types on R_rs(λ). Meanwhile, future studies will be conducted to investigate satellite applications of the P_Q_bbe(λ)-based clustering scheme and the potential influences of P_Q_bbe(λ) on ocean color remote sensing algorithms, and thereby to improve the algorithm for each particle type to achieve better estimations of water constitutes, such as TSM.

5. Conclusions

High levels of variability and clearly evident clustering were observed in pseudo-backscattering efficiency P_Q_bbe(640) in the Bohai Sea and Yellow Sea. After analyzing P_Q_bbe(640) variability levels during different seasons, we can confirm that the clustering of P_Q_bbe(640) is generally seasonally stable. Consequently, a simple scheme is proposed for the classification of particle types. Our analysis of optical effects of P_Q_bbe(640) shows that P_Q_bbe(640) can significantly affect the backscattering coefficient b_bp(640), mass-specific backscattering coefficient b_bp*(640) and remote sensing reflectance R_rs(640). In particular, as a result of P_Q_bbe(640) clustering, relationships between b_bp*(640), b_bp(640) and R_rs(640) and biogeochemical variables differ for different particle types, and discretion should be exercised when applying general bio-optical models that couple these relationships. After classifying using the scheme proposed in this study, the effects of P_Q_bbe(640) on b_bp*(640), b_bp(640) and R_rs(640) were reduced, showing that co-variation trends of b_bp*(640), b_bp(640) and R_rs(640) with biogeochemical variables can be more accurately demonstrated. The findings of this study may help clarify the biogeochemical processes that couple particulate backscattering properties and the remote sensing of ocean color in the Bohai Sea and Yellow Sea. Moreover, future studies will explore the satellite applications of the P_Q_bbe(640)-clustering scheme, and examine the effects of particle types on the whole spectral region of R_rs(λ) thereby improve remote sensing algorithms.