1. Introduction

The Geostationary Ocean Color Imager (GOCI) [1] is the first geostationary ocean color satellite sensor which was launched on June 27, 2010. With its six visible bands centered at the wavelengths of 412, 443, 490, 555, 660, and 680 nm and two near-infrared (NIR) bands at wavelengths of 745 and 865 nm, GOCI can monitor the marine environment and provide a variety of ocean optical, biological, and biogeochemical property products for an area of about 2500 × 2500 km² around the Korean Peninsula. In GOCI measurement, the region is divided into 16 sections for separate slot-shots with a temporal frequency of eight times a day, i.e., from local times of 9:00 to 16:00. GOCI data can be used for various applications such as the short- and long-term regional ocean environment monitoring, disaster and ocean hazard monitoring and prevention, ocean ecosystem and water quality evaluation and analysis, as well as intelligence and national security applications [2, 3].

The western Pacific region (Fig. 1 ) which is covered by GOCI measurement, including the Bohai Sea (BS), Yellow Sea (YS), and East China Sea (ECS), has one of the most turbid waters in the world [4–6]. Total suspended matter (TSM) concentration in the region can reach up to the order of ~100 g m⁻³ [7]. Significant NIR ocean radiance contributions can be found along the coast of the Yellow Sea and the East China Sea [8, 9]. For example, the normalized water-leaving radiance nL_w(λ) [10–13] at the NIR wavelength of 748 nm (nL_w(748)) can reach ~3 mW cm⁻² μm⁻¹ sr⁻¹ in the Hangzhou Bay of China’s east coastal region [9]. The complexity of the water property in these regions suggests that some existing NIR-modeling schemes [14–16] may not work properly for GOCI ocean color data processing. Thus, satellite ocean color remote sensing in the western Pacific region is significantly limited in these highly turbid coastal regions. It has been shown that the ocean generally is still black for the shortwave infrared (SWIR) bands in the China’s east coastal regions [8], i.e., ocean water-leaving radiance contributions at the SWIR bands are negligible. Wang et al. (2007; 2011) [9, 17] also show that improved and reasonably accurate nL_w(λ) spectra data can be derived using the SWIR-based atmospheric correction algorithm [18] in the coastal regions of the Yellow Sea and East China Sea [9], as well as in the highly turbid inland Lake Taihu [17].

 

Fig. 1 Map of the GOCI coverage, which includes the Bohai Sea, Yellow Sea, East China Sea, Japan/East Sea, and part of the South China Sea. Locations of the in situ measurements for GOCI ocean color data evaluations are marked in solid red circles. The black box in the Bohai Sea is the region used to characterize the diurnal variability of the ocean environments measured by GOCI.

Download Full Size | PDF

However, because GOCI has no SWIR bands, atmospheric correction for the sensor has been a challenge in order to derive accurate ocean color products in these highly turbid ocean regions. In a recent study, a regional NIR-nL_w(λ) model has been proposed for atmospheric correction for ocean color data processing in the western Pacific region, including the Bohai Sea, Yellow Sea, and East China Sea [19]. The new algorithm that was developed based on long-term measurements (2002–2009) from the Moderate Resolution Imaging Spectroradiometer (MODIS) on the satellite Aqua using the SWIR atmospheric correction algorithm [19], provides an effective alternative method for GOCI ocean color data processing in these highly turbid ocean regions.

In a collaboration between NOAA Center for Satellite Applications and Research (STAR) and Korea Institute of Ocean Science and Technology (KIOST), we have been working on improving GOCI ocean color products from the standard GOCI data processing [20], e.g., nL_w(λ) spectra data, chlorophyll-a (Chl-a) concentration, diffuse attenuation coefficient at the wavelength of 490 nm (K_d(490)), etc. In fact, the current standard GOCI data processing [20] has been significantly improved resulted from our collaboration. In this paper, we report ocean color products derived from GOCI measurements using the new iterative NIR-corrected atmospheric correction algorithm [19], which is different from the GOCI standard algorithm. Our motivation in this work is to implement the new NIR-corrected atmospheric correction algorithm in the GOCI ocean color data processing for deriving accurate ocean color product data. The accuracy of nL_w(λ) spectra data derived from GOCI measurements is assessed and validated with in situ ocean optical and biological measurements, which were collected around Korean Peninsula. We also show example results of the diurnal variation of ocean optical and biogeochemical properties in the region. Furthermore, multi-month composites of ocean color products from GOCI observations are provided and discussed.

2. Data and method

2.1. GOCI ocean color data processing system

The ocean color data processing system at NOAA has been developed based on the NASA satellite ocean color data processing package, i.e., the Sea-Viewing Wide Field-of-View Sensor (SeaWiFS) Data Analysis System (SeaDAS) version 4.6. In addition, the NOAA Multi-Sensor Level-1 to Level-2 (NOAA-MSL12) data processing system has been modified and improved to include: (1) the SWIR and NIR-SWIR ocean color data processing options [18, 21, 22], (2) new aerosol lookup tables including polarization effects [23] and more accurate Rayleigh radiance computations [24], (3) the SWIR-based vicarious calibration approach for deriving consistent vicarious gain coefficients for ocean color data processing options of the NIR, SWIR, and NIR-SWIR approaches, (4) incorporated algorithms for detecting absorbing aerosols and turbid waters [25], (5) implementation of a SWIR-based cloud-masking scheme (particularly over turbid waters) [26], (6) implementation of an ice-detecting algorithm for global ocean color data processing [27], and some others, e.g., an approach to improve the performance of MODIS SWIR bands [28], a method of deriving regional sea ice optical property [29, 30], etc. In the last several years, the NOAA-MSL12 ocean (water) color data processing system has been used to generate improved satellite ocean color product data for global oceans [22], coastal highly turbid regions [9], as well as the inland fresh water lakes, e.g., China’s Lake Taihu [17] and Florida’s Lake Okeechobee [31].

For the GOCI ocean color data processing, various parameters and lookup tables (LUTs), e.g., Rayleigh radiance LUTs [32, 33], atmospheric diffuse transmittance LUTs [34, 35], spectral solar irradiance data and ozone absorption coefficients [36], etc., have been generated and modified specifically corresponding to the GOCI eight spectral bands. It is noted that these LUTs and various parameters were generated in the same way as those for MODIS-Aqua using the GOCI spectral response function data. In addition, to accommodate the requirement of the atmospheric correction for GOCI ocean color data processing, a regional NIR-nL_w(λ) model has been proposed for atmospheric correction for ocean color data processing in the western Pacific region [19]. Based on the regional empirical relationship between the NIR-nL_w(λ) and the diffuse attenuation coefficient K_d(490) [37], which is derived from long-term MODIS-Aqua measurements (2002–2009) using the SWIR-based ocean color data processing, an iterative scheme with the NIR-based atmospheric correction algorithm [38] has been developed. Specifically, the following empirical polynomial between the NIR nL_w(745) and the diffuse attenuation coefficient K_d(490) has been implemented in the GOCI ocean color data processing:

It is noted that in Eqs. (1) and (2) the same fitting coefficients that were derived from MODIS-Aqua measurements [19] were used for the GOCI ocean color data processing. The effect of NIR spectral wavelength differences between GOCI and MODIS-Aqua is generally small with small NIR wavelength differences between two sensors, i.e., differed by ~3–4 nm. In fact, the data noise errors in deriving these fitting coefficients [19] are likely much larger than errors due to spectral band differences between GOCI and MODIS-Aqua. In addition, MODIS-Aqua data from the region have been used to vicariously calibrate GOCI spectral bands (see next section), and in average the effect of the spectral band difference between two sensors has been somewhat accounted for.

2.2. Vicarious calibration for GOCI ocean color products

For the GOCI coverage, the waters in Japan/East Sea are typical of clear Case-1 water. In this study, we have used the MODIS-Aqua-measured nL_w(λ) and aerosol property data in the region within the box of 38.2°N–39.2°N and 132.5°E–133.5°E in the central Japan/East Sea as the reference to carry out the sensor on-orbit vicarious calibration [39–41] and derive the vicarious gain coefficients for GOCI eight spectral bands. In effect, GOCI vicarious gains are derived by forcing GOCI-derived nL_w(λ) and aerosol models in the region the same as those from MODIS-Aqua.

Specifically, both MODIS-Aqua and GOCI data were remapped to a region centered at 38.7°N and 133°E. MODIS-Aqua aerosol models (Ångström exponents) [42] and nL_w(λ) are derived in the region and used for vicariously calibrating GOCI spectral bands, i.e., GOCI spectral gain coefficients were derived by adjusting the GOCI gains such that the same aerosol Ångström exponent and nL_w(λ) values in average are obtained. This is an iterative data process to have final product mean values from GOCI and MODIS-Aqua matched for the calibration scene.

To understand and evaluate the vicarious calibration performance, three data processing methods were used for deriving MODIS-Aqua aerosol and ocean color parameters, i.e., atmospheric correction approaches of the NIR [38], SWIR [18], and Wang et al. (2012) NIR-model [19], respectively. As expected, the GOCI gains derived from the MODIS SWIR and NIR-model approaches are quite consistent, while there are some important differences using the MODIS NIR atmospheric correction method. For GOCI spectral bands of 412, 443, 490, 555, 660, 680, 745, and 865 nm, the MODIS-SWIR derived GOCI gains are: 0.9857, 0.9749, 0.9484, 0.9179, 0.9299, 0.9283, 0.9502, and 1.0, respectively, compared to the method of using Wang et al. (2012) NIR-model-derived GOCI gains of 0.9862, 0.9753, 0.9473, 0.9149, 0.9245, 0.9223, 0.9430, and 1.0, respectively. Results show that the two methods produced consistent GOCI gain coefficients, i.e., in the order of ~0.1%. With the MODIS NIR atmospheric correction method, however, the derived GOCI gains differed by ~1% in comparison with those from the other two approaches. This data analysis gives us confidence to use GOCI vicarious gains derived from the MODIS-Aqua SWIR data processing. Thus, the MODIS-SWIR-derived vicarious calibration gains are applied to the GOCI Level-1B data for ocean color data processing for generating nL_w(λ) spectra and other ocean biological and biogeochemical products, e.g., Chl-a concentration, K_d(490) data, etc.

2.3. In situ data

During the period between March and November of 2011, there were extensive field campaigns in ocean regions of the southwest coast of Korea (Yellow Sea), Japan/East Sea, and the East China Sea near Korean Peninsula for the purpose of collecting various in situ physical, optical, and biological ocean data in support of the GOCI calibration and validation, as well as algorithm development efforts [20, 43]. The in situ data collection and processing were carried out following the procedures outlined in the NASA ocean optics protocols [44].

In particular, in situ hyperspectral ocean water-leaving reflectance spectra data were acquired using the Analytical Spectral Devices, Inc. (ASD) FieldSpec and TriOS RAMSES hyperspectral radiometers [43]. Water-leaving reflectance spectra were derived from the above-surface radiances and sky radiances measured at the nadir angle of 40° and the relative azimuth angle of 90°. In addition, in situ optics data were obtained after applying data quality control procedures described in Moon et al. (2012) [43].

Water samples that were acquired using Niskin bottles for chlorophyll-a measurements were filtered through 47 mm GF/F filters with nominal pore size of 0.7 μm. From extracted pigments, which were obtained using 90% acetone 10 ml, a dual beam spectrophotometer was used to measure absorbance spectra of the acetone extracted pigment samples. In situ Chl-a data can then be obtained using the measured absorbance spectra following Jeffrey and Humphrey (1975) [45]. Details for in situ data collection, data processing, and data quality can be found in Moon et al. (2012) [43].

2.4. GOCI-measured ocean color products

Multi-month GOCI Level-1B data from March to December of 2011 were obtained from the Korea Ocean Satellite Center (KOSC). These data were processed into ocean color products with the new atmospheric correction algorithm [19]. We use the in situ data collected during the period of March to November of 2011 to quantify the data quality of GOCI-measured ocean color products and validate the performance of the new atmospheric correction algorithm for GOCI ocean color data processing. Furthermore, we examine the image-wise performance of GOCI products, e.g., ocean features, data spatial continuity and smoothness, data noise, etc., to qualitatively assess GOCI ocean color data product quality. Particularly, diurnal variation of the ocean optical, biological, and biogeochemical properties observed by GOCI are presented and discussed.

3. Results and discussions

3.1. GOCI ocean color products

The GOCI-derived products include nL_w(λ) spectra [13, 38], Chl-a concentration [46, 47], and diffuse attenuation coefficient at the wavelength of 490 nm (K_d(490)) [37, 48–50]. It should be noted that GOCI Chl-a data were derived using the empirical algorithm [46, 47] and GOCI K_d(490) data were derived using the Wang et al. (2009) algorithm [37].

Figure 2 provides examples of the GOCI-derived ocean color products in the GOCI coverage region, corresponding to GOCI data acquired at 12:00 (noon) local time on April 5, 2011. Figure 2(a) is the GOCI-measured true color image on April 5, 2011, while Figs. 2(b)–2(f) are nL_w(λ) images at GOCI wavelengths of 443, 490, 555, 660, and 865 nm, respectively. GOCI-derived Chl-a and K_d(490) images are shown in Figs. 2(g) and 2(h), respectively. It is noted that there are no retrievals over some extremely turbid regions, e.g., the Hangzhou Bay. This is because of failure in cloud masking, which uses the method from Wang and Shi (2006) [26] with the NIR spectral reflectance information. The NIR spectral reflectance method [26] sometimes fails to identify clear sky from extremely turbid waters, e.g., in the Hangzhou Bay in the case of April 5, 2011.

 

Fig. 2 GOCI-measured various ocean color product images which were acquired on April 5, 2011 at noon local time for (a) true color image, (b) nL_w(443), (c) nL_w(490), (d) nL_w(555), (e) nL_w(660), (f) nL_w(865), (g) chlorophyll-a concentration (Chl-a), and (h) diffuse attenuation coefficient at the wavelength of 490 nm (K_d(490)).

Download Full Size | PDF

Along the China’s east coastal region, GOCI-measured nL_w(λ) values rise with increase of the wavelength from the blue to the green band, and nL_w(λ) values for some regions actually peak at the red band. This is a typical characteristic of the sediment-dominated waters, which is consistent with the optical features observed by MODIS-Aqua [6]. Indeed, ocean properties derived from GOCI are comparable to those from MODIS-Aqua observations. It is noted that there are no obvious correlations between GOCI-derived nL_w(λ) and aerosol optical thickness (AOT) τ_a(865) [42] (results not shown), which implies that the iterative NIR-corrected atmospheric correction approach [19] performs well in removing the ocean radiance contribution at the NIR bands in order to carry out atmospheric correction properly for GOCI ocean color data processing.

3.2. GOCI ocean color products compared with in situ measurements

To compare the GOCI-derived and the in situ-measured nL_w(λ) data, GOCI-measured nL_w(λ) data were computed by averaging in 5 × 5 pixels surrounding the in situ measurement location. Specifically, GOCI-measured data at local times of 11:00, 12:00, and 13:00 are used for the comparison analysis. From these three GOCI measurements, the one with the closest measurement time matched with the in situ data was selected and used for the analysis. In fact, the time difference between the GOCI and in situ measurements is within a ~3-hour period, with the minimum and maximum time differences of 5 minutes and about 3 hours, respectively. To have more matchup data points, we used data with relatively large time differences between GOCI and in situ measurements (e.g., > one hour). It should be noted that we received almost all GOCI data at local times of 11:00, 12:00, and 13:00, and in some cases with daily eight images. Figure 1 shows the locations (red dots) of the in situ measurements corresponding to the matchup data set. For the GOCI versus in situ data matchup analyses, we have followed the procedure of Wang et al. (2009) [22].

Comparison results in Fig. 3 show that the new atmospheric correction algorithm performs reasonably well for the GOCI-derived nL_w(λ) data in these highly turbid ocean regions. Indeed, GOCI-derived nL_w(λ) spectra at 412, 443, 490, 555, 660, and 680 nm generally match well with the corresponding in situ nL_w(λ) measurements (Figs. 3(a)–3(c)). In Fig. 3, the fitting coefficients of slope and intercept (Int.) as well as the correlation coefficient (R) are provided in plots for matchup data of nL_w(λ) spectra (linear fitting) (Figs. 3(a)–3(c)) and Chl-a (in log-scale fitting) (Fig. 3(d)). The ratios in nL_w(λ) between the GOCI-derived and in situ data show that GOCI retrievals are slightly higher than the in situ measurements, and in particular, for low nL_w(λ) values at the blue (412 and 443 nm) bands (i.e., cases with turbid waters) (Fig. 3(a)). Specifically, mean ratios of nL_w(λ) at 412, 443, 490, 555, 660, and 680 nm between GOCI and in situ data are 1.166, 1.339, 1.238, 1.133, 1.229, and 1.370, respectively. In fact, the biased high values (high ratios) are mostly due to the cases with low nL_w(λ) values. The discrepancies between the GOCI-derived and the in situ-measured nL_w(λ) data can be attributed to the time difference between these two measurements (up to ~3 hours) (particularly for turbid waters, e.g., see results in next section), complex ocean environments, and the uncertainty (bias and noise errors) of the empirical optical model to estimate the NIR ocean radiance contributions [19], e.g., with temporal variation, as well as errors from the in situ measurements [43]. It is particularly noted that most of the in situ measurements were actually located in the coastal regions. Indeed, we found that, with restriction to the time difference within one-hour between GOCI and in situ measurements (but with significantly reduced matchup data amount), mean ratio values of nL_w(λ) at 412, 443, 490, 555, 660, and 680 nm can be improved to 1.175, 1.270, 1.178, 1.073, 1.098, and 1.254, respectively. The coastal dynamics such as tidal current, coastal current, and sediment re-suspension may lead to significant changes of the ocean’s optical property within the time difference (up to ~3 hours) between the GOCI and the in situ measurements.

 

Fig. 3 GOCI-derived ocean optical and biological properties compared with in situ measurements for (a) nL_w(412), nL_w(443), and nL_w(490), (b) nL_w(555), nL_w(660), and nL_w(680), (c) nL_w(745) and nL_w(865), and (d) Chl-a. In plots, the matchup fitting coefficients for slope, intercept (Int.), and correlation coefficient (R) are provided.

Download Full Size | PDF

The GOCI-derived two NIR nL_w(λ) data, nL_w(745) and nL_w(865) (Fig. 3(c)), are more or less consistent with the in situ measurements. As an example, the highest nL_w(745) from the in situ measurements is ~0.3 mW cm⁻² μm⁻¹ sr⁻¹, while the value derived from the GOCI is ~0.34–0.35 mW cm⁻² μm⁻¹ sr⁻¹. This shows that nL_w(λ) values at the NIR bands can be effectively estimated and removed from the GOCI-measured top-of-atmosphere (TOA) radiances using the Wang et al. (2012) model [19]. Thus, reasonably accurate atmospheric correction can be carried out to derive nL_w(λ) spectra data from GOCI measurements.

With GOCI-measured nL_w(λ) spectra data, other ocean biological and biogeochemical products such as Chl-a concentration and K_d(490) data can be derived. Figure 3(d) shows the comparison between GOCI-measured Chl-a data and the corresponding in situ measurements. Although Fig. 3(d) shows some overall consistence in GOCI-derived Chl-a data compared with in situ measurements, the mean ratio (GOCI versus in situ) of Chl-a in Fig. 3(d) is 1.435. This shows some overestimations of GOCI-measured Chl-a values. The GOCI-derived Chl-a uncertainty is primarily due to the limitation of the empirical Chl-a algorithm [46, 47], which is mostly applicable to clear open oceans and may not be valid over such highly turbid waters. Overall, however, these comparison results show that GOCI-derived ocean color products can be used to quantify and characterize ocean optical, biological, and biogeochemical properties and study their related ocean processes in the western Pacific region.

3.3. Diurnal variation of ocean property in the Bohai Sea

With possibly eight-time measurements daily, GOCI provides a unique capability to monitor the ocean environments in near real-time, and GOCI data can be used to address the diurnal variability in the ecosystem of the entire GOCI coverage region. Here, a case study of ocean diurnal changes from GOCI measurements is provided and discussed. The GOCI results demonstrate that GOCI can provide real-time monitoring of water optical, biological, and biogeochemical variability of the ocean ecosystem in the western Pacific region.

An example from GOCI measurements. Figure 4 shows an example of the diurnal change in K_d(490) in the Bohai Sea region on April 5, 2011. In general, the Bohai Sea is dominated with turbid waters with K_d(490) over ~1.0 m⁻¹ for most part of the region between local times 9:00 and 16:00. Even though variations in K_d(490) between two neighboring GOCI observations (one-hour apart) are not obviously notable, the progressive change in terms of K_d(490) patterns and magnitudes from local time 9:00 to 16:00 is clearly shown in Figs. 4(a)–4(h). At the local time of 9:00 on April 5, 2011, K_d(490) ranged between ~1.4–1.5 m⁻¹ for most part of the Bohai Sea (Fig. 4(a)). The coverage of waters with K_d(490) between ~1.4–1.5 m⁻¹ gradually decreased at local times of 10:00 (Fig. 4(b)), 11:00 (Fig. 4(c)), 12:00 (Fig. 4(d)), and 13:00 (Fig. 4(e)). At the local time of 13:00, coverage of waters with K_d(490) values of ~1.4–1.5 m⁻¹ is less than half of the Bohai Sea. The change of the ocean environments in terms of K_d(490) variation in a 4-hour period is quite notable. It is noted that the fingerlike features with south-north orientation actually mark the sea surface signatures of underwater sand ridges in the Bohai Sea [51].

 

Fig. 4 Maps of GOCI-derived diffuse attenuation coefficient at the wavelength of 490 nm (K_d(490)) in the Bohai Sea region on April 5, 2011 at the local time of (a) 9:00, (b) 10:00, (c) 11:00, (d) 12:00, (e) 13:00, (f) 14:00, (g) 15:00, and (h) 16:00. Region for further data analysis and quantification is marked with the black box in plot 4(a).

Download Full Size | PDF

At the local time 14:00, K_d(490) value in the region was slightly higher than that observed one hour earlier (Fig. 4(f)). Different from the K_d(490) trend between local times 9:00 and 13:00, K_d(490) in the Bohai Sea showed slightly increase after the local time 14:00 (Fig. 4(g)). This is reflected with slightly enhanced K_d(490) in the sand ridge regions near Liaodong Peninsula [51]. At a local time 16:00, most part of the Bohai Sea was covered with clouds (Fig. 4(h)). However, K_d(490) values still showed noticeable increase for the portion with K_d(490) retrievals in the northeastern Bohai Sea, suggesting that K_d(490) between local 15:00 and 16:00 also increased for a broad Bohai Sea region.

Some quantitative evaluations. Using a region of 50 × 50 km² centered at a location of [38.89°N, 120.17°E] in the central Bohai Sea as a marked box in Fig. 4(a), we further quantify the diurnal variability in nL_w(λ) and K_d(490) on April 5, 2011 (Fig. 5 ). For the central Bohai Sea region on April 5, 2011, nL_w(λ) values at wavelengths of 412, 443, and 490 nm increased from local times of 9:00 to 12:00 (noon), peaked at the local noon, and then reduced in afternoon. However, from the local time of 14:00 to 15:00, nL_w(443) and nL_w(490) slightly increased. On the other hand, nL_w(λ) values at wavelengths of 555, 660, and 680 nm decreased from the local morning (9:00) to the local afternoon at 13:00, and then slightly increased from local time of 13:00 to 15:00. Ocean biological changes can be inferred from nL_w(λ) spectra variations. The changes in nL_w(λ) at green and red bands in the region are consistent with the K_d(490) variations. Indeed, K_d(490) in the central Bohai Sea showed significant changes between local times of 9:00 and 15:00. From a period of 9:00 to 13:00, K_d(490) in the central Bohai Sea trended lower from ~1.5 m⁻¹ to ~1.05 m⁻¹. After local time 13:00, K_d(490) inched higher to ~1.15 m⁻¹. Diurnal variations of other ocean properties such as TSM concentration [7] in the Bohai Sea are also assessed and show similar trends (results not shown).

 

Fig. 5 GOCI-measured diurnal variability on April 5, 2011 for (a) nL_w(λ) at GOCI bands of 412, 443, 490, 555, 660, and 680 nm and (b) K_d(490) in the location of central Bohai Sea as marked in Fig. 4(a).

Download Full Size | PDF

Discussions. Results in Figs. 4 and 5 show that the diurnal variability of ocean environments can be effectively monitored with the geostationary ocean color satellite as suggested by Neukermans et al. (2012) [52]. Diurnal variability in nL_w(λ) is also remarkable as shown in Fig. 5(a). GOCI-measured nL_w(λ) values at the blue bands have highs at local noon, while nL_w(λ) values at green and red bands have lows at early afternoon (at local time 13:00). Different spectral shapes in nL_w(λ) suggest that diurnal variability as revealed by GOCI in the Bohai Sea is not the artifact caused by the inaccurate atmospheric correction due to different geometries for the 8-GOCI measurements (as also shown from nL_w(λ) validation results in the previous section). In addition, the optical spectra for all the 7 observations as shown in Fig. 5 are all reasonable and similar to the typical optical spectra of the turbid waters in the Bohai Sea [6]. Further examination of AOT data (results not shown here) for the seven GOCI measurements shows that the diurnal changes of AOT is relatively small in the Bohai Sea region between local times of 9:00 and 15:00. There is no regional-wide AOT increase or decrease between different measurements. All these provide evidences that atmospheric correction for GOCI ocean color data processing is robust for different GOCI observations on the same day, and the variability of K_d(490) and nL_w(λ) reflects the real diurnal changes of ocean environment in the Bohai Sea.

For the ocean environment, diurnal variability can generally be caused by a variety of ocean’s physical and biological processes such as tides, biological cycles, diurnal winds, etc. For the case of April 5, 2011 (Figs. 4 and 5), the wind speed was quite low and stable. The mean wind speeds in the study region in Fig. 4 varied from ~3.5 to ~3.7 m/s during the GOCI eight measurements on that day. Because the results of notable diurnal variability (Fig. 5) are from the region with water depth over 30 m, wind-driven wave is not the driving force that could re-suspend the sediment from the ocean bottom and consequently was observed by GOCI. It requires much shallow water and large winds to show the effect of the wind-driven wave on the sediment re-suspension from the ocean bottom.

Even though the diurnal biological variability can occur (such as showing in Chl-a and inherent optical property (IOP) variations) [53, 54], the ocean biological diurnal variation is generally small and the maximum of Chl-a (as well as K_d(490)) is normally observed in the early afternoon as the result of phytoplankton growth, grazing, and physiological responses [55]. In the GOCI observations of this study, K_d(490) actually reached minimum in the early afternoon (Fig. 5). This indicates that the diurnal variability as shown in Figs. 4 and 5 is not primarily from a result of the ocean diurnal biological change.

In the Bohai Sea, Yellow Sea, and East China Sea, however, a semidiurnal and diurnal tide plays a significant role on the dynamics of the ocean environments [56–58]. In fact, variations of the satellite ocean color observations within a spring-neap tidal cycle are in the same order as the seasonal change as observed by MODIS-Aqua in this region [59]. Some studies have shown that TSM in the water column is strongly correlated to tidal currents in a semidiurnal/diurnal cycle [60, 61]. Thus, of all the possible ocean processes that can drive the diurnal change of ocean environments in the region, diurnal variability as observed in this study is mainly driven by the change of the tidal currents in the Bohai Sea. The tidal dynamics (tidal current) is the most important factor influencing the variation of the TSM (thus K_d(490)) in the region.

3.4. GOCI-measured composite ocean color products

Using GOCI measurements at local 12:00 (noon) from March to December of 2011 (10 months data), GOCI composite ocean color images are generated for nL_w(412) (Fig. 6(a) ), nL_w(443) (Fig. 6(b)), nL_w(490) (Fig. 6(c)), nL_w(555) (Fig. 6(d)), nL_w(660) (Fig. 6(e)), nL_w(680) (Fig. 6(f)), nL_w(865) (Fig. 6(g)), Chl-a (Fig. 6(h)), and K_d(490) (Fig. 6(i)). The enhancements of nL_w(λ) at blue (443 nm), green (555 nm), and red (660 nm) are consistent with the elevated K_d(490) in coastal regions. In the Japan/East Sea, K_d(490) values are lower than those in the Bohai Sea, Yellow Sea, and East China Sea. This is also reflected with a flat nL_w(660) (red band) distribution and near zero nL_w(865) in the NIR wavelength. In the southeastern portion of GOCI coverage region in east of the East China Sea and South China Sea, waters are featured with low Chl-a, low K_d(490), and high nL_w(443) (blue band). The boundary and ocean front between the open ocean oligotrophic and coastal waters can be clearly identified in the nL_w(412) and nL_w(443) (Figs. 6(a) and 6(b)), Chl-a (Fig. 6(h)), and K_d(490) (Fig. 6(i)), and match the classic path of the Kuroshio Current.

 

Fig. 6 Composite images of GOCI-measured ocean color products at the local noon derived from GOCI measurements from March to December of 2011 (10 months) for (a) nL_w(412), (b) nL_w(443), (c) nL_w(490), (d) nL_w(555), (e) nL_w(660), (f) nL_w(680), (g) nL_w(865), (h) Chl-a (logscale), and (i) K_d(490) (logscale).

Download Full Size | PDF

Significantly enhanced nL_w(555) and nL_w(660) are located in the Yellow River estuary in the Bohai Sea, the Subei Shoal in the Yellow Sea, and the Yangtze River estuary and the Hangzhou Bay in the East China Sea. Values of nL_w(555) and nL_w(660) are over ~5 mW cm⁻² μm⁻¹ sr⁻¹ for these highly turbid ocean regions. For the modestly turbid regions, such as plumes in the central East China Sea [5], nL_w(λ) is more enhanced in the green band than that in the red band. Compared to the climatology of the ocean color products derived with 8-year observations from MODIS-Aqua [6], GOCI-measured nL_w(λ) data in this study are quantitatively consistent with the climatology of the nL_w(λ) in the region. This provides further evidence that GOCI-derived ocean color products from the proposed atmospheric correction approach are reasonably accurate and can be used to quantify and characterize both short- and long-term variability of the ocean ecosystem in the highly turbid ocean regions.

It should be noted that, since the complete image for GOCI coverage is generated with 16 slot-shots as a mosaic, the boundary effect of different GOCI observation slots is shown in the composite image maps in Fig. 6, as well as in other GOCI Level-2 images (e.g., Fig. 2). The artificial boundary caused by GOCI’s 2D frame image capture mode is still an ongoing issue, and the issue is being resolved in an effort at KIOST.

5. Conclusion

We have derived the GOCI normalized water-leaving radiance spectra nL_w(λ) for the GOCI coverage region using an iterative NIR-corrected atmospheric correction algorithm [19]. GOCI-derived ocean color products are compared with the in situ measurements. The validation results from this effort show a reasonably good agreement between GOCI-derived values and in situ measurements. Multi-month composites of GOCI ocean color products in this region are also quantitatively consistent with corresponding composites of MODIS-Aqua ocean color products that were derived using the SWIR-based atmospheric correction algorithm. This demonstrates that, using the new atmospheric correction algorithm in processing GOCI data, ocean color products can be used to characterize and quantify the ocean environments in the western Pacific region.

This study also shows that GOCI observations can be used to characterize and quantify the diurnal variability of the marine ecosystem for the western Pacific region. Our results show that there are significant diurnal variations in ocean optical, biological, and biogeochemical properties in the region. In particular, as an example for the Bohai Sea in April, the diurnal variability in nL_w(λ) has different spectral variation. GOCI-measured nL_w(λ) values at the blue bands have highs at local noon, while nL_w(λ) values at green and red bands have lows at early afternoon (13:00 at local time). Such large scale (spatial) and high temporal resolution measurements can only be achieved by geostationary satellite sensor [62, 63]. Furthermore, this unique capability from geostationary satellite sensor can complement the ocean color observations from other polar-orbiting satellite sensors such as MODIS and the Visible Infrared Imaging Radiometer Suite (VIIRS), which have a global coverage, but lack the temporal resolution to monitor the dynamics of marine environments on an hourly basis.