1. Introduction

Satellite ocean color retrievals such as normalized water-leaving radiance spectra (nL_w(λ)) [1,2], chlorophyll-a concentration (Chl-a) [3], and water diffuse attenuation coefficient at the wavelength of 490 nm (K_d(490)) [4–6] are computed after removing atmospheric radiance contributions and ocean surface effects from the top-of-atmosphere (TOA) radiance L_t(λ) measured by satellite ocean color sensors such as the Sea-viewing Wide Field-of-view Sensor (SeaWiFS), the Moderate-resolution Imaging Spectroradiometer (MODIS), and the Medium Resolution Imaging Spectrometer (MERIS). For the blue wavelengths, the atmosphere radiance contribution, which includes atmosphere molecules (Rayleigh) radiance L_r(λ) [7–10], atmospheric aerosol radiance L_A(λ) (including Rayleigh-aerosol interactions) [1, 11, 12], can account for up to ~90% of the TOA radiance measurements [2]. Thus, accurate estimation of the atmospheric contributions in the TOA radiance is crucial in order to compute the normalized water-leaving radiance spectra nL_w(λ), and consequently derive ocean optical, biological, and biogeochemical properties for studying, understanding, and monitoring the global and regional ocean variability and changes.

Particularly, the satellite-derived water diffuse attenuation coefficient K_d(490) can be related to light penetration and availability in aquatic system. The K_d(490) is an important parameter to understand not only physical processes such as the heat transfer in the ocean upper layer [13, 14], but also biological processes such as phytoplankton photosynthesis in the ocean euphotic zone [15]. For global open oceans, K_d(490) can be accurately derived from satellite ocean color sensors (e.g., SeaWiFS, MODIS) using empirical models either through the blue-green nL_w(λ) ratio [4] or using the relationship of K_d(490) and chlorophyll-a concentration [5]. Semi-analytical approaches based on radiative transfer models have also been used [16]. In effect, these models all use the blue-green nL_w(λ) data for deriving satellite K_d(490) product. To provide improved K_d(490) data in coastal turbid waters, a K_d(490) model using nL_w(λ) at blue, green, and red bands has been proposed and shown significantly improved K_d(490) results in turbid waters [6]. It has been shown that for turbid waters K_d(490) has strong correlation to nL_w(λ) at the red band [6]. In fact, for highly turbid waters, K_d(490) can be well correlated to nL_w(λ) at the near-infrared (NIR) bands.

For the current and future global ocean color satellite sensors such as SeaWiFS, MODIS, and Visible Infrared Imager Radiometer Suite (VIIRS), the atmospheric correction for producing the global ocean color products is based on the Gordon and Wang (1994) [1] algorithm. Specifically, the algorithm uses two NIR bands to determine the aerosol type and compute the atmospheric effects in the visible by extrapolating the aerosol effect from the NIR into visible bands using aerosol models [1, 11, 17]. One assumption for the Gordon and Wang (1994) [1] algorithm is that the NIR ocean contribution is negligible in order to determine the aerosol type. For global open ocean waters, both SeaWIFS and MODIS have been producing high-quality ocean color products [18, 19].

In the coastal turbid regions and highly productive waters, the NIR ocean contributions are not negligible [20–24]. In order to remove the NIR ocean contributions in the turbid or productive waters so that the “black pixel assumption” in the NIR wavelengths can be valid for employing the Gordon and Wang (1994) [1] type atmospheric correction algorithm, various approaches are investigated. Siegel et al. (2000) [22] used the chlorophyll-a estimation to determine the NIR nL_w(λ); Lavender et al. (2005) [21] used the sediment estimation to compute NIR nL_w(λ); Ruddick et al. (2000) [23] fixed the aerosol and backscattering type and then solved for both the NIR nL_w(λ) and NIR aerosol reflectance simultaneously; and Stump et al. (2003) [24] used a bio-optical model for absorptions in the red band, and then used that and nL_w(λ) in the red band to compute NIR nL_w(λ). In a more recent effort, Bailey et al. (2010) [25] further tuned and improved the Stump et al. (2003) [24] approach by applying optical modeling to compute the nL_w(λ) from the absorption coefficient and nL_w(λ) in the red band.

In recent years, an atmospheric correction algorithm using the shortwave infrared (SWIR) bands has been proposed [26, 27] and demonstrated to derive improved ocean (water) color products in turbid coastal ocean regions [28, 29] and inland lake waters [30]. Waters are more strongly absorbent at the SWIR bands (i.e., 1240, 1640, and 2130 nm for MODIS) than at the NIR wavelengths [31], i.e., water absorptions at the three MODIS SWIR bands are at least one order higher than those in the NIR bands. Significantly higher water absorptions at the SWIR bands indicate that for the highly turbid regions or productive waters, ocean can still be black for the SWIR bands even though there are significant nL_w(λ) contributions in the NIR wavelengths [32]. Thus, water-leaving radiance in the visible and NIR wavelengths can be derived from the SWIR atmospheric correction algorithm for the highly turbid waters.

The western Pacific region (Fig. 1 ) covers some of the most turbid waters in the world, i.e., the Bohai Sea (BS), Yellow Sea (YS), and East China Sea (ECS) [33, 34]. Total suspended matter (TSM) concentration in the region can reach up to ~100 g m⁻³ [35]. Significant NIR ocean radiance contributions can be found along the coast of the Yellow Sea and the East China Sea [29, 33]. For example, the normalized water-leaving radiance at the NIR 748 nm (nL_w(748)) can reach over ~3 mW cm⁻² μm⁻¹ sr⁻¹ in the Hangzhou Bay of China’s east coastal region [29]. The complexity of the water property in these regions suggests that the NIR-modeling scheme such as that proposed by Stump et al. (2003) [24] or Bailey et al. (2010) [25] cannot work properly; thus satellite ocean color remote sensing in the western Pacific is significantly limited in these highly turbid coastal zones. On the other hand, Wang et al. (2007; 2011) [29, 30] demonstrated that the SWIR-based atmospheric correction algorithm can be used to derive reasonably accurate nL_w(λ) spectra data and other ocean color products in the Bohai Sea, Yellow Sea, and East China Sea, as well as in a turbid inland Lake Taihu. These studies indeed quantify and characterize the ocean optical features such as the nL_w(λ) from deep blue (412 nm) to NIR (869 nm) wavelengths as well as the other optical, biological, and biogeochemical parameters such as K_d(490) [6] and TSM concentration [35] in these turbid waters. Thus, the SWIR-based atmospheric correction algorithm provides opportunity to quantitatively estimate the NIR nL_w(λ) in the region in order to carry out the NIR-based satellite ocean color data processing without the SWIR radiance measurements.

 

Fig. 1 Map of the western Pacific region, which includes the Bohai Sea, Yellow Sea, and East China Sea with isobaths of 20, 50, and 100 m. The highly turbid waters in this study are outlined in the box with marked pseudo-stations l (33.24°N, 121.87°E) and 2 (33.07°N, 121.43°E) for detailed data analyses.

Download Full Size | PDF

Our motivation for this work is to derive ocean color product data in the highly turbid western Pacific region using the Geostationary Ocean Color Imager (GOCI) onboard South Korean Communication, Ocean, and Meteorological Satellite (COMS), which was launched in June of 2010. GOCI has local area coverage of the western Pacific region (Fig. 1) and eight spectral bands covering from the blue to the NIR wavelengths (412–865 nm), but no SWIR bands. For such often highly turbid ocean regions, previously proposed NIR radiance models for atmospheric correction usually do not work properly. Thus, a new regional NIR radiance model needs to be developed. In this study, we use the long-term MODIS-Aqua-measured nL_w(λ) in the NIR bands that have been derived using the SWIR-based atmospheric correction algorithm and water diffuse attenuation coefficient at 490 nm (K_d(490)) [6] to build an empirical relationship between the NIR nL_w(λ) and K_d(490). It is noted that K_d(490) data are derived from MODIS-Aqua measurements using Wang et al. (2009) K_d(490) algorithm [6] that provides improved K_d(490) data in turbid coastal waters [6]. An iterative scheme with the NIR-based atmospheric correction algorithm has been developed in order to derive valid nL_w(λ) in highly turbid coastal regions. Normalized water-leaving radiance spectra and other ocean color products derived from MODIS-Aqua using this new NIR-corrected atmospheric correction approach are further compared, evaluated, and validated quantitatively with the same ocean color parameters derived from the SWIR-based ocean color data processing.

2. Empirical relationship of the NIR nL_w(748) and nL_w(869) vs. K_d(490)

As discussed previously, in order to carry out the NIR-based atmospheric correction algorithm [1] for satellite ocean color data processing in coastal turbid waters, it is crucial that the NIR ocean radiance contributions be accurately accounted for [21–24]. Therefore, accurately and effectively estimating the NIR nL_w(λ) contributions in coastal turbid water regions, e.g., the western Pacific region, for carrying out the NIR-based atmospheric correction (without using the SWIR bands) for satellite ocean color data processing is the main focus of this section.

In a recent study, Shi and Wang (2009) [32] show that nL_w(λ) in the red band (nL_w(645)) and the NIR band (nL_w(859)) has a quasi-linear relationship when nL_w(859) is less than ~1 mW cm⁻² μm⁻¹ sr⁻¹ in the turbid waters of the Yellow Sea and East China Sea. When nL_w(859) is larger than ~1 mW cm⁻² μm⁻¹ sr¹, the relationship becomes nonlinear until nL_w(645) reaches its maxima with no nL_w(645) changes corresponding to the increase of nL_w(859) as waters get extremely turbid. This phenomenon is attributed to the dominance of b_b(λ) for extremely turbid waters in a function of b_b(λ)/(a(λ) + b_b(λ))—a(λ) and b_b(λ) are total absorption coefficient and backscattering coefficient, respectively—of which the normalized water-leaving radiance (or normalized water-leaving reflectance) can be expressed as a function [36]. On the other hand, Wang et al. (2009) [6] show that water diffuse attenuation coefficient at the wavelength of 490 nm K_d(490) can be estimated using nL_w(488) and nL_w(645) (or nL_w(667)) in the turbid waters. In addition, Shi and Wang (2010) [33] show that the NIR normalized water-leaving radiance nL_w(748), nL_w(859), and nL_w(869) from MODIS-Aqua are highly correlated to ocean turbidity, which can be quantified and characterized with K_d(490) in the western Pacific’s highly turbid waters, such as those of the Yellow Sea and East China Sea.

Using the normalized water-leaving radiance at the NIR wavelengths nL_w(748) and nL_w(869), and the corresponding K_d(490) derived using the SWIR atmospheric correction algorithm from MODIS-Aqua measurements in the turbid waters of the western Pacific between 2002 and 2009, the statistical relationships between nL_w(748) and nL_w(869) vs. K_d(490) were derived. Figure 2(a) shows the scatter plots of nL_w(748) vs. K_d(490), as well as an optimal polynomial fitting function between these two parameters. In fact, the empirical polynomial between nL_w(748) and K_d(490) can be expressed as:

 

Fig. 2 Scatter plots and empirical polynomial fitting functions for (a) nL_w(748) vs. K_d(490) and (b) nL_w(869) vs. nL_w(748). Note that nL_w(748), nL_w(869), and K_d(490) were derived from MODIS-Aqua measurements (2002 to 2009) using the SWIR atmospheric correction algorithm in this region.

Download Full Size | PDF

3. Implementation of the NIR-corrected atmospheric correction algorithm

Figure 3 shows the flow chart of the MODIS-Aqua NIR-corrected atmospheric correction and ocean color data processing. An iterative approach is conducted to estimate nL_w(748) and nL_w(869) using Eqs. (1) and (2) with inputs of K_d(490). Three criteria are predefined for the iteration procedure to make sure that nL_w(748) and nL_w(869) are accurately computed and removed in the MODIS-Aqua ocean color data processing:

 

Fig. 3 Schematic flow chart of the new NIR-corrected atmospheric correction algorithm for MODIS-Aqua ocean color data processing in the western Pacific region.

Download Full Size | PDF

Using stations 1 and 2 as marked in Fig. 1 in the Yellow Sea, the convergence performance of this iterative atmospheric correction scheme has been tested and evaluated for the turbid waters at station 1 (Fig. 4(a) ) and highly turbid waters at station 2 (Fig. 4(b)). At station 1, the value of K_d(490) derived from the SWIR atmospheric correction is ~1.9–2.0 m⁻¹. With the iterative NIR atmospheric correction approach, K_d⁽¹⁾(490) was initially computed with the assumption of no NIR ocean radiance contributions at 748 and 869 nm for the 1st iteration (N = 1). With the derived K_d⁽¹⁾(490), nL_w⁽²⁾(748) and nL_w⁽²⁾(869) are estimated in the

 

Fig. 4 Changes of nL_w(748) and nL_w(869) estimations as a function of the number of iterations using the NIR-corrected atmospheric correction algorithm for the case of (a) at station 1 and (b) at station 2 for MODIS-Aqua measurements acquired on October 19, 2003. The final K_d(490) values for plots (a) and (b) are 1.96 and 3.49 m⁻¹, respectively.

Download Full Size | PDF

2^nd iteration and K_d⁽²⁾(490) is recomputed in order to estimate nL_w⁽³⁾(748) and nL_w⁽³⁾(869) in the 3rd iteration. At station 1, nL_w(748) and nL_w(869) pairs for the 2nd, 3rd, 4th, and 5th iterations are (0.259, 0.099), (0.304, 0.117), (0.331, 0.127), and (0.347, 0.134) mW cm⁻² μm⁻¹ sr⁻¹, respectively. nL_w(748) and nL_w(869) values eventually converged to (0.374, 0.144) mW cm⁻² μm⁻¹ sr⁻¹ at N = 5.

For the highly turbid waters like those at station 2 (Fig. 4(b)), the large NIR ocean radiance contributions can lead to negative nL_w(λ) in the visible bands at the 1st iteration, thus with no derived K_d⁽¹⁾(490). In this case, K_d(490) is setup to be 5 m⁻¹ in order to start the iteration process to estimate nL_w(748) and nL_w(869). The estimations of nL_w(748) and nL_w(869) in the following iterations quickly converge as shown in Fig. 4(b) with the final estimations of the nL_w(748) and nL_w(869) pair to be (1.607, 0.698) mW cm⁻² μm⁻¹ sr⁻¹. In general, the criteria of (2) can be met, and nL_w(748) and nL_w(869) can be determined with less than ~5–6 iterations for both turbid waters like those at station 1 and highly turbid waters like those at station 2 in the turbid western Pacific region.

Figure 5 shows nL_w(λ) retrievals in the visible and NIR wavelengths, K_d(490), and aerosol optical thickness (AOT) at 869 nm (τ_a(869)) [37] after carrying out the proposed NIR-corrected atmospheric correction approach as described above for the scene acquired by MODIS-Aqua on October 19, 2003. To further evaluate and validate this NIR-corrected atmospheric correction algorithm, the same corresponding ocean color parameters are also produced using the SWIR atmospheric correction algorithm. In general, the normalized water-leaving radiance maps of nL_w(443) (Fig. 5(b)), nL_w(551) (Fig. 5(c)), nL_w(645) (Fig. 5(d)), nL_w(748) (Fig. 5(e)), and nL_w(869) (Fig. 5(f)) derived from this NIR-corrected atmospheric correction are comparable to the corresponding nL_w(λ) maps (Figs. 5(i), 5(j), 5(k), 5(l), and 5(m)) derived using the SWIR atmospheric correction algorithm. Some slight differences in nL_w(λ) reflect the uncertainty of the atmospheric correction procedure as well as the difference between nL_w(748) (or nL_w(869)) estimation from the empirical model and the actual nL_w(λ) in the NIR wavelengths as shown in Fig. 2. For the K_d(490) (Figs. 5(a) and 5(h)) with this scene, the values derived from these two atmospheric correction approaches are almost identical. In the AOT τ_a(869) map derived from the NIR-corrected atmospheric correction approach (Fig. 5(g)), the turbid regions, which show enhanced nL_w(λ) in the green (551 nm), red (645 nm), and NIR (748 and 869 nm) wavelengths in the coastal regions of Yellow Sea and East China Sea, do not have any signatures (Fig. 5(g)). It is consistent with the τ_a(869) map derived from the SWIR atmospheric correction approach (Fig. 5(n)). This indicates that the ocean radiance contributions in the NIR bands are effectively removed from the MODIS-Aqua TOA radiance measurements using the proposed iterative approach in carrying out the NIR-based atmospheric correction for ocean color data processing.

 

Fig. 5 Comparisons of ocean color retrievals on October 19, 2003 derived from the SWIR atmospheric correction (lower panel) and from the NIR-corrected atmospheric correction (upper panel) for K_d(490) (5(a) and 5(h)), nL_w(443) (5(b) and 5(i)), nL_w(551) (5(c) and 5(j)), nL_w(645) (5(d) and 5(k)), nL_w(748) (5(e) and 5(l)), nL_w(869) (5(f) and 5(m)), and τ_a(869) (5(g) and 5(n)).

Download Full Size | PDF

To further evaluate nL_w(λ) quantitatively using the NIR-corrected atmospheric correction approach, three sets of nL_w(λ) spectra derived from the NIR-corrected, SWIR, and standard-NIR (i.e., no NIR nL_w(λ) correction) atmospheric correction algorithms in the turbid waters at station 1 and in highly turbid waters at station 2 are compared and shown in Fig. 6 . At station 1, nL_w(λ) spectra from the NIR-corrected and the SWIR atmospheric correction approaches are similar with some small differences in the blue and green bands. This is consistent with the nL_w(λ) maps derived from these two atmospheric correction algorithms as shown in Fig. 5. Specifically, nL_w(748) and nL_w(869) values derived from these two approaches match very well. In comparison, significant errors of nL_w(λ) can be found if the NIR nL_w(λ) contributions are not removed when the standard-NIR atmospheric correction was carried out. nL_w(412) values are less than ~0.5 mW cm⁻² μm⁻¹ sr⁻¹ from the standard-NIR atmospheric correction approach, while values of nL_w(412) are all over ~2.0 mW cm⁻² μm⁻¹ sr⁻¹ from the SWIR and the new NIR-corrected atmospheric correction methods.

 

Fig. 6 nL_w(λ) spectra derived from the new NIR-corrected atmospheric correction and the SWIR atmospheric correction algorithm for case of (a) at station 1 and (b) at station 2 for MODIS-Aqua measurements acquired on October 19, 2003. For comparison, nL_w(λ) spectra derived from standard-NIR atmospheric correction algorithm are also shown in plots (a) and (b).

Download Full Size | PDF

Similar to station 1, nL_w(λ) spectra derived from the new NIR-corrected and SWIR atmospheric correction approaches are almost identical from the blue to NIR wavelengths at station 2. It is noted that the NIR ocean radiance contributions are very significantly higher at this station than those at station 1. Specifically, nL_w(748) and nL_w(869) values for station 1 (Fig. 6(a)) are 0.36 and 0.15 mW cm⁻² μm⁻¹ sr⁻¹, respectively, compared with those of 1.60 and 0.69 mW cm⁻² μm⁻¹ sr⁻¹ at station 2 (Fig. 6(b)). This further demonstrates that the NIR-corrected atmospheric correction approach can be applied to both turbid waters and highly turbid waters. On the other hand, Fig. 6(b) shows that erroneous nL_w(λ) can be resulted if the NIR ocean radiance contributions are not accounted for in the atmospheric correction process. At the wavelength of 412 nm, nL_w(412) is less than –3 mW cm⁻² μm⁻¹ sr⁻¹ from the standard-NIR atmospheric correction approach at station 2. In fact, actual values of nL_w(412) derived from the SWIR and the new NIR-corrected atmospheric correction algorithms are ~1.5 mW cm⁻² μm⁻¹ sr⁻¹_.

4. Evaluation of the NIR-corrected atmospheric correction in the western Pacific region

With a scene MODIS-Aqua acquired on October 19, 2003, we show that the new NIR-corrected atmospheric correction algorithm can be applied to western Pacific turbid ocean regions to derive reasonably accurate nL_w(λ) spectra that are consistent with those from the SWIR atmospheric correction algorithm. Using the MODIS-Aqua measurements between 2002 and 2010 at stations 1 and 2, the long-term performance of the NIR-corrected atmospheric correction approach has been further evaluated (Fig. 7 ). Figure 7(a) shows that nL_w(748) and nL_w(869) derived from the SWIR atmospheric correction (indicated as nL_w^(SWIR)(λ)) match well with those from the NIR-corrected atmospheric correction method (indicated as nL_w^(NIR-Corr)(λ)) at station 1. The mean values of nL_w^(NIR-Corr)(748)/nL_w^(SWIR)(748) and the nL_w^(NIR-Corr)(869)/nL_w^(SWIR)(869) are 1.02 and 1.00, respectively, with STD values of 0.26 and 0.32. At station 2, Fig. 7(b) shows that nL_w^(NIR-Corr)(748) and nL_w^(NIR-Corr)(869) are consistent with nL_w^(SWIR)(748) and nL_w^(SWIR)(869). A slight low bias for both nL_w^(NIR-Corr)(748) and nL_w^(NIR-Corr)(869) in the high NIR nL_w(λ) range reflects the fact that K_d(490) becomes nonlinear and less-sensitive to the increase of nL_w(λ) in the NIR wavelengths in the highly turbid water [33]. It is also noted that the number of matchups at station 2 is significantly less than that at station 1. At station 2, MODIS-Aqua sensor for the NIR bands at 748 and 869 nm are more often saturated due to higher ocean turbidity and large nL_w(λ) in these two NIR bands.

 

Fig. 7 Matchups of the MODIS-Aqua-measured nL_w(λ) spectra (from 2002 to 2010) derived from the new NIR-corrected and SWIR atmospheric correction algorithms for (a) nL_w(748) and nL_w(869) at station 1, (b) nL_w(748) and nL_w(869) at station 2, (c) all nL_w(λ) in the visible and NIR bands at station 1, and (d) all nL_w(λ) in the visible and NIR bands at station 2.

Download Full Size | PDF

Figures 7(c) and 7(d) further characterize the performance of the NIR-corrected atmospheric correction algorithm with nL_w(λ) retrievals in the MODIS visible bands at stations 1 and 2. At station 1, the mean ratio of nL_w^(NIR-Corr)(λ)/nL_w^(SWIR)(λ) for wavelengths of 412, 443, 488, 531, 551, and 645 nm are 1.00, 1.02, 1.01, 1.00, 0.99, and 1.00, respectively. For station 2, nL_w^(NIR-Corr)(λ) values also appear to be consistent with the corresponding nL_w^(SWIR)(λ) values. Unlike the slight bias of nL_w^(NIR-Corr)(748) and nL_w^(NIR-Corr)(869) in the highly turbid waters, no obvious bias of nL_w^(NIR-Corr)(λ) in the visible bands can be identified. In fact, for station 2 (Fig. 7(d)), the nL_w(λ) ratios for bands 412, 443, 488, 531, 551, and 645 nm are 1.03, 1.02, 1.01, 1.03, 0.99, and 1.05, respectively. The consistence of nL_w^(NIR-Corr)(λ) and nL_w^(SWIR)(λ) spectra at these two stations suggests that ocean optical, biological, and biogeochemical properties in the highly turbid western Pacific region can also be computed consistently. However, there are obvious noise errors in the derived products.

Figure 8 shows the comparisons of nL_w^(NIR-Corr)(λ) and nL_w^(SWIR)(λ) composite images for wavelengths of 443, 488, 551, 645, 748, and 869 nm from MODIS-Aqua measurements in 2010. Indeed, the nL_w^(NIR-Corr)(λ) composite images for each wavelength shows notable similarity to the corresponding nL_w^(SWIR)(λ) images in terms of both spatial patterns and their magnitudes. In the turbid regions of the Yellow Sea and East China Sea, enhanced nL_w(λ) spectra from the blue to NIR wavelengths are observed. The seasonal sediment plume [34] in the central East China Sea can be identified clearly with nL_w(488) (Figs. 8(b) and 8(h)) and nL_w(551) (Figs. 8(c) and 8(i)). In the Hangzhou Bay, Yangtze River estuary, and coastal region of the Yellow Sea, significant elevations of nL_w(551), nL_w(645), nL_w(748), and nL_w(869) were derived from both the new NIR-corrected atmospheric correction algorithm (Figs. 8(c)-8(f)) and the SWIR algorithm (Figs. 8(i)-8(l)). The nL_w(551), nL_w(645), nL_w(748), and nL_w(869) images from these two atmospheric correction approaches are visually identical. It is also worth noting that nL_w(748) and nL_w(869) reach over ~2 and ~1 mW cm⁻² μm⁻¹ sr⁻¹ in the Yangtze River Estuary and coastal regions of Yellow Sea. This provides further evidences that the new NIR-corrected atmospheric correction approach cannot only be carried out in moderately turbid regions such as in the central East China Sea plume, but also be applied in the highly turbid coastal zones of the Yangtze River estuary to derive the ocean properties with reasonable accuracy.

 

Fig. 8 Comparison of composites of MODIS-Aqua ocean color retrievals in 2010 derived from the new NIR-corrected and the SWIR atmospheric correction algorithms for nL_w(443) (8(a) and 8(g)), nL_w(488) (8(b) and 8(h)), nL_w(551) (8(c) and 8(i)), nL_w(645) (8(d) and 8(j)), nL_w(748) (8(e) and 8(k)), and nL_w(869) (8(f) and 8(l)).

Download Full Size | PDF

5. Discussions and summary

In this study, we developed an iterative approach to carry out a new NIR-corrected atmospheric correction in order to derive MODIS-Aqua ocean color products in the turbid coastal regions of the western Pacific. This approach is for ocean color sensors that do not have the SWIR bands. The NIR nL_w(λ) contributions at wavelengths of 748 and 869 nm are estimated from empirical relationships between nL_w(748) (or nL_w(869)) and K_d(490), which were derived from the MODIS-Aqua long-term (2002–2009) ocean color retrievals using the SWIR atmospheric correction algorithm. Comparisons between ocean color products with the NIR-corrected atmospheric correction approach and those from the SWIR atmospheric correction algorithm show that the derived nL_w(λ) spectra from this new NIR-corrected atmospheric correction algorithm in the western Pacific region match well with the corresponding nL_w(λ) from the SWIR atmospheric correction algorithm. Shi and Wang (2009) [32] demonstrated that the ocean, in general, is still black for the SWIR bands in China’s east coastal regions. Wang et al. (2007; 2011) [29, 30] also show that improved and reasonably accurate nL_w(λ) spectral data can be achieved with the SWIR-based atmospheric correction in the coastal regions of the Yellow Sea and East China Sea [29], as well as in the highly turbid Lake Taihu [30, 38]. The ocean color products derived from the NIR-SWIR combined atmospheric correction approach have been used to study the ecosystems of the turbid waters, such as the sea surface signature of a sand ridge in the Bohai Sea [39], ocean responses to the Korean dump site of the Yellow Sea [40], the environmental response to a land reclamation project in Korea [41], and the spring-neap tidal effects in this region [42]. The consistence of the ocean color products derived from these two atmospheric correction approaches suggests that the ocean color products from the proposed NIR-corrected atmospheric correction algorithm in the western Pacific region can be used to provide quantitative evaluations of the optical features in the turbid zones.

In the proposed NIR-corrected atmospheric correction approach, K_d(490) plays a key role in the process to iteratively compute the nL_w(748) and nL_w(869) values. Since K_d(490) for the turbid waters is based on semi-analytical optical modeling [6], the NIR-corrected atmospheric correction approach thus implicitly related to the ocean’s inherent optical property (IOP) [16]. Even though the relationship between nL_w(645) and nL_w(748) (or nL_w(869)) can also be derived from the SWIR-measured nL_w(λ) data and applied to estimate nL_w(748) (or nL_w(869)), extensive tests (not shown here) show that using the nL_w(748)-K_d(490) (or nL_w(869)-K_d(490)) relationship can make the nL_w(748) (or nL_w(869)) estimation less sensitive to the noise of nL_w(λ) values. It should be further emphasized that the NIR-corrected atmospheric correction algorithm is based on a regional nL_w(748)-K_d(490) (or nL_w(869)-K_d(490)) model for the Bohai Sea, Yellow Sea, and East China Sea.

There are some limitations for this K_d(490)-based, NIR-corrected atmospheric correction approach. First, relationships between nL_w(748) (or nL_w(869)) and K_d(490) are developed from long-term MODIS-Aqua observations. Since nL_w(748) (or nL_w(869)) and K_d(490) are both dependent on the IOPs in the water, the IOP variability caused by the sediment, phytoplankton, and colored dissolved organic matter (CDOM) can make the nL_w(748)-K_d(490) (or the nL_w(869)-K_d(490)) relationship not exactly the same in the short term as in the long term. On the other hand, the regional dependence of IOPs also indicates that the empirical relationship between K_d(490) and nL_w(748) (or nL_w(869)) developed in this study might not be suitable for other highly turbid regions such as the Amazon River estuary and La Plata River estuary, etc. Second, it has been shown [32, 33] that nL_w(λ) at the red band (such as nL_w(645)) is no longer sensitive to the changes of nL_w(748) (or nL_w(869)) in extremely turbid waters. At a certain level, nL_w(645) becomes a constant and does not change with the increase of nL_w(748) (or nL_w(869)). This suggests that K_d(490) derived from nL_w(645) for extremely turbid waters (K_d(490) > ~5.0 m⁻¹) cannot be used to estimate nL_w(748) (or nL_w(869)) for the NIR-corrected atmospheric correction algorithm.

Due to the limitations of the K_d(490)-based, NIR-corrected atmospheric correction approach, it is necessary to use the SWIR-based algorithm to carry out global ocean color data processing with a uniform atmospheric correction procedure for the turbid regions. On the other hand, the SWIR-based atmospheric correction can also be used to derive accurate ocean color products in extremely turbid waters, where the proposed NIR-corrected atmospheric correction can fail. For the global satellite ocean color observations, some of the sensors such as SeaWiFS and MERIS do not have the SWIR bands. Thus, this study is useful to address the atmospheric correction issues in the highly turbid regions for these sensors. Particularly, GOCI has a local coverage of highly turbid waters in the western Pacific region. Because GOCI has no SWIR bands, atmospheric correction has been a challenge in order to derive reasonably accurate ocean color products in these turbid ocean regions. Methodology and results demonstrated in this study provide an effective alternative for conducting atmospheric correction for GOCI (and other satellite sensors without SWIR bands) ocean color data processing in the turbid regions of the Bohai Sea, Yellow Sea, and East China Sea.