Ocean color retrieval from MWI onboard the Tiangong-2 Space Lab: preliminary results

Xianqiang He; Yan Bai; Jun Wei; Jing Ding; Palanisamy Shanmugam; Difeng Wang; Qingjun Song; Xiaoxian Huang

doi:10.1364/OE.25.023955

1. Introduction

Satellite ocean color remote sensing plays an important role in global oceanic ecosystem monitoring [1]. In the past three decades, more than 20 satellite ocean color sensors were launched, and some of them including the Moderate Resolution Imaging Spectroradiometer (MODIS), Geostationary Ocean Color Imager (GOCI), Visible Infrared Imaging Radiometer Suite (VIIRS), and Ocean Land Colour Instrument (OLCI) are still on-orbit operation currently. On Sept. 15, 2016, China launched an ocean color sensor named the Moderate-resolution Wide-wavelengths Imager (MWI) onboard the Chinese Tiangong-2 Space Lab. The MWI is an experimental satellite sensor for the Chinese next generation ocean color satellites HY-1E and HY-1F, which are scheduled for launch around 2021. In contrast to the scanning imaging mode adopted for the Chinese first generation ocean color satellites HY-1A (2002-2004), HY-1B (2007-2016) [2], and the HY-1C/HY-1D (to be launched around 2018), MWI uses a push-broom imaging system similar to the Medium resolution Imaging Spectrometer (MERIS) onboard ENVISAT satellite and the OLCI onboard Sentinel-3A satellite that is capable of capturing the entire image frame at once in the flight track direction and high signal to noise ratio (SNR).

The MWI is a comprehensive sensor that has three imaging modules, including the visible light and near-infrared wavelength module (VIS/NIR), shortwave infrared wavelength module (SWIR), and thermal infrared module (TIR) (Fig. 1). The VIS/NIR module consists of 14 bands covering the wavelengths from 393nm to 1024nm for ocean color observations, and has the capability of on-orbit band programming. Table 1 presents the default setups of bands for the MWI observation, and the shortest band can be programed to 393nm according to requirements. During the whole testing phase, the default setups of bands were applied. The SWIR module has 2 bands (1240nm and 1640nm) for the atmospheric correction in turbid waters. Similarly, the TIR module has 2 bands (8.475µm and 9.125µm) for measuring the sea surface temperature. The spatial resolutions achieved for the VIS/NIR, SWIR and TIR modules are around 100m, 200m and 400m, respectively. To enlarge the observing swatch, the VIS/NIR module uses three aligned cameras, and the SWIR and TIR modules use two aligned cameras (Fig. 1). With the orbit height (~400km) of the Tiangong-2 Space Lab, the total swath of the MWI observation is about 300km.

Fig. 1 The layout of the opto-mechanical head assembly of the MWI.

Download Full Size | PDF

Table 1. The default bands configuration of the MWI sensor.

View Table

Having a spatial resolution of up to 100m and a good number of spectral bands (18 bands), the MWI is the highest spatial resolution sensor among other on-orbit ocean color sensors providing high quality spectral data for remote sensing applications over oceanic and inland waters. Moreover, the designed lifetime of the MWI is longer than 3 years. Thus, it is naturally interesting to examine the performance of the MWI for supporting applications in different regimes. In this study, we have developed data processing methods for retrieving the ocean color information in both turbid and clear waters and conducted preliminary validation using in situ data collected from turbid coastal oceanic waters.

2. Data and methods

2.1 In situ measurements

To validate the performance of the MWI sensor, a cruise was conducted in the Changjiang River estuary over Jan.1 to Jan. 12, 2017[Fig. 2(a)]. At the stations with favorable solar illumination conditions (usually during 9:00 AM to 3:00 PM), the spectral radiometric data were collected under using an ASD FieldSpec®3 full range (350-2500nm). Specifically, the upward radiances from the water surface ( $L_{t}$ ) and standard reflecting plate ( $L_{p}$ ) and downward sky radiance ( $L_{s k y}$ ) were measured with optimal geometries [3]. Then, the normalized water-leaving radiance ( $L_{w n}$ ) and remote sensing reflectance ( $R_{r s}$ ) were determined using

{\begin{cases} L_{w n} (λ) = F_{0} (λ) β_{p} (λ) [L_{t} (λ) - β_{s} (λ) L_{s k y} (λ)] / [π L_{p} (λ)] \\ R_{r s} (λ) = L_{w n} (λ) / F_{0} (λ) \end{cases}

where

F_{0}

is the mean extraterrestrial solar irradiance;

β_{p}

and

β_{s}

are the reflectances of the standard plate and sea surface, respectively;

λ

is the wavelength. In general,

β_{s}

varies from 0.022 to 0.05 depending on the sky condition, sea-surface roughness, viewing and illumination geometry, and also the wavelength [4]. Here, a constant value 0.028 was adopted for

β_{s}

. For this cruise, we got valid spectrums of

L_{w n}

or

R_{r s}

at 15 daytime stations, as shown in Fig. 2(b).

Fig. 2 Locations of the in situ measurements. (a) The sampling stations of the cruise measurement; (b)The spectrums of $R_{r s}$ at 15 daytime stations; (c)The two fixed stations (red points) at the Datong hydrologic station (west point) and Hangzhou Bay Bridge station (east point).

Download Full Size | PDF

Besides the cruise measurements, $L_{w n}$ data were also collected from two fixed stations in turbid waters – one at the center of the Hangzhou Bay Bridge and one at the Datong hydrologic station in the lower reaches of the Changjiang River [Fig. 2(c)]. At each fixed station, the above water optical system SAS (Surface Acquisition systems) was mounted above the water surface to measure the upward radiances from the water surface ( $L_{t}$ ), downward sky radiance ( $L_{s k y}$ ), and downward irradiance ( $E_{s}$ ) in the spectral range over 350-800nm. Then, the normalized water-leaving radiance ( $L_{w n}$ ) was determined using

L_{w n} (λ) = F_{0} (λ) [L_{t} (λ) - β_{s} (λ) L_{s k y} (λ)] / E_{s} (λ)

Since SAS measures the radiances continually with a time step around several minutes, data from this in situ platform can always match the MWI observation under cloud free conditions.

2.2 Satellite ocean color retrieval

The Level-1 MWI data were obtained from the National Ocean Satellite Application Center (NSOAS) of China. Two orbits MWI data were obtained over the Changjiang River and Hangzhou Bay on Dec.1, 2016 and Jan. 22, 2017, respectively. In addition, typical MWI data were obtained to compare with the GOCI, MODIS/Aqua and VIIRS data. For this purpose, the level-1 and Level-2 data of the GOCI and the Level-2 data of the MODIS/Aqua and VIIRS were obtained from the respective Korea Ocean Satellite Center (KOSC) and the NASA Goddard Space Flight Centre (oceancolor.gsfc.nasa.gov).

The atmospheric correction was performed on the Level-1 MWI data using the methods described in a later section. Based on the Lwn spectral data retrieved by the atmospheric correction methods, the ocean color products such as Chla and TSM were derived for further analysis. For the first time, this study focused on examining the capacity of the MWI in retrieving the ocean color information. In highly turbid waters, Chl retrieval is a challenging issue, and thus only the TSM product was generated and compared. In clear waters, since the TSM is quite low, only the Chla concentration was retrieved and compared.

(1) Chla retrieval

To compare with the MODIS/Aqua and VIIRS data products, the OC3M algorithm was commonly used to estimate Chla from the MWI data [5]. The general form of this algorithm is expressed as

{\begin{cases} C h l a = 10^{c_{0} + c_{1} r + c_{2} r^{2} + c_{3} r^{3} + c_{4} r^{4}} \\ r = \max [R_{r s} (443 n m), R_{r s} (490 n m)] / R_{r s} (555 n m) \end{cases}

with the constant coefficients C₀ = 0.283, C₁ = −2.753, C₂ = 1.457, C₃ = 0.659, and C₄ = −1.403. It should be noted that the standard OC3M algorithm for the MODIS/Aqua and VIIRS sensors uses the 555 nm band, however, MWI has 565nm band instead of the 555nm band. Here, we use the linear interpolation method to estimate the

R_{r s} (555 n m)

from the

R_{r s} (520 n m)

and

R_{r s} (565 n m)

.

(2) TSM retrieval

The accurate retrieval of the TSM is highly regional dependence. Many TSM algorithms were developed for different regions [6]. Usually, local TSM algorithm can achieve higher accuracy than the wide-regions applicable algorithms for a specific region. Here, an empirical TSM algorithm [7] developed for the Changjiang River Estuary and Hangzhou Bay was adopted to provide the TSM comparison results in these waters. The TSM algorithm takes the form as

{\begin{cases} T S M = 10^{1.0758 + 1.1230 \times r a t i o} \\ r a t i o = R_{r s} (750 n m) / R_{r s} (490 n m) \end{cases}

3. Atmospheric correction of the MWI data

Different atmospheric correction algorithms were applied to the Level-1 MWI data according to water turbidity. In clear waters, the NIR-AC algorithm was used. In turbid waters, the UV-AC, SWIR-AC and MUMM algorithms were used. These algorithms are briefly described in the following sections.

3.1 NIR-AC algorithm

The total radiance ( $L_{t}$ ) measured by the MWI sensor can be written as the sum of three main components of the ocean-atmospheric system

L_{t} (λ) = L_{r} (λ) + L_{a} (λ) + t_{v} (λ) L_{w} (λ)

where

L_{r}

is the Rayleigh scattering radiance contributed by air molecules in the absence of aerosol;

L_{a}

is the aerosol multiple-scattering reflectance (including the interactive scattering between molecules and aerosols);

t_{v}

is the atmospheric diffuse transmittance from the sea surface to satellite; and

L_{w}

is the desired water-leaving reflectance. Note that in Eq. (5), the sea surface reflectance from whitecaps and sun-glint is ignored. For convenience, the radiance can be converted to the reflectance

ρ (λ) = π L (λ) / [F_{0} (λ) \cos (θ_{0})]

Accordingly, Eq. (5) can be transformed into

ρ_{t} (λ) = ρ_{r} (λ) + ρ_{a} (λ) + t_{v} (λ) ρ_{w} (λ)

Using the vector radiative transfer model for the coupled ocean-atmosphere system (PCOART) [8, 9], a general Rayleigh-scattering look-up table was generated which can be applied to all ocean color sensors [10]. Based on the look-up table, $ρ_{r}$ can be calculated accurately with the inputs of the band-equivalent Rayleigh scattering optical thickness, extraterrestrial solar irradiance, and sea surface wind speed. With the prelaunch measured band response functions, the band-equivalent Rayleigh scattering optical thickness and extraterrestrial solar irradiance can be calculated [11]

{\begin{cases} τ_{R} (b a n d) = \sum_{λ_{i} = λ_{\min}}^{λ_{\max}} τ_{R} (λ_{i}) F_{0} (λ_{i}) B (λ_{i}) / \sum_{λ_{i} = λ_{\min}}^{λ_{\max}} F_{0} (λ_{i}) B (λ_{i}) \\ F_{0} (b a n d) = \sum_{λ_{i} = λ_{\min}}^{λ_{\max}} F_{0} (λ_{i}) B (λ_{i}) / \sum_{λ_{i} = λ_{\min}}^{λ_{\max}} B (λ_{i}) \end{cases}

where

τ_{R}

is the Rayleigh scattering optical thickness, and

B

is the band response functions. Then the Rayleigh scattering corrected reflectance can be obtained as

ρ_{r c} (λ) = ρ_{t} (λ) - ρ_{r} (λ) = ρ_{a} (λ) + t_{v} (λ) ρ_{w} (λ)

For clear waters, the water-leaving radiance at the NIR bands can be neglected which allows the approximation of aerosol scattering reflectances at two NIR bands (750nm and 865nm) from the Rayleigh-scattering corrected reflectances. Finally, the aerosol scattering reflectances at the shorter wavebands are extrapolated from the two NIR bands using the fairly accurate method [12]

{\begin{cases} c = \ln [ρ_{r c} (λ_{N I R 1}) / ρ_{r c} (λ_{N I R 2})] / (λ_{N I R 2} - λ_{N I R 1}) \\ ρ_{a} (λ_{b a n d}) = ρ_{r c} (λ_{N I R 2}) \exp [c (λ_{N I R 2} - λ_{b a n d})] \end{cases}

For non- and weakly-absorbing aerosols (maritime, coastal, and tropospheric aerosol models), the performance of this extrapolation method was truly remarkable with the errors usually within the permissible limit [13].

3.2 UV-AC algorithm

For highly turbid waters in the coastal and inland regions, a practical atmospheric correction algorithm was developed by which the aerosol scattering radiance is estimated using the reference ultraviolet wavelength (called “UV-AC” algorithm) [14, 15]. In turbid waters, the water-leaving radiances increase largely at longer VIS wavelengths and NIR due to strong particulate scattering, yet the strong combined absorption by detritus and CDOM contents cause a rapid decrease of the water-leaving radiance at the UV band. In extremely turbid waters, such as the Hangzhou Bay, the water-leaving radiance at the UV band is much lower than at the NIR band. In such turbid coastal and inland waters, the UV band is more suited than the NIR band for estimating the aerosol scattering radiance [7]. The performance of the UV-AC algorithm was previously verified in different estuarine waters, including the Changjiang River Estuary, Mississippi River Estuary, and Orinoco River Estuary [15].

In a recent study, the UV-AC algorithm was successfully applied to the hourly GOCI data to study the diurnal variations of the TSM in the Hangzhou Bay that provided the increased confidence level for the retrieval of water-leaving radiance in highly turbid waters [7]. Since the GOCI sensor has no such a UV band, the shortest waveband (412 nm) was used as the reference to estimate the aerosol scattering radiance. In case of the MWI, the UV-AC algorithm was applied with the 413nm band for estimating the aerosol radiance. Assuming that the water-leaving reflectance at 413 nm can be small in highly turbid waters, one can estimate the aerosol scattering reflectance at 413 nm [ $ρ_{a} (413 n m) = ρ_{r c} (413 n m)$ ]. Based on the extrapolation method as given in Eq. (10), one can estimate the aerosol scattering reflectance at 865 nm

{\begin{cases} ρ_{a} (865 n m) = ρ_{r c} (413 n m) \exp [c (413 - 865)] \\ c = \ln [ρ_{r c} (750 n m) / ρ_{r c} (865 n m)] / (865 - 750) \end{cases}

Note that because the UV-AC algorithm was applied to the whole selected region, some of the clear waters may also have been included in the region. To avoid the overestimation of the aerosol scattering reflectance in clear waters, we limit

ρ_{a} (865 n m)

to

ρ_{r c} (865 n m)

. Specially, when the estimated

ρ_{a} (865 n m)

is larger than

ρ_{r c} (865 n m)

,

ρ_{a} (865 n m)

is set as

ρ_{r c} (865 n m)

. Finally, we assume a “white” aerosol scattering reflectance spectrum [16], and the

ρ_{a}

at all bands are equal to

ρ_{a} (865 n m)

.

3.3 SWIR-AC algorithm

Wang et al. [17, 18] proposed an atmospheric correction algorithm for turbid waters, where the two SWIR bands are used to estimate the aerosol scattering radiances with the assumption that the water-leaving radiances at these SWIR bands are negligible due to strong pure seawater absorption. The aerosol scattering reflectance determined at two SWIR bands are then used to find suitable aerosol models. Based on the selected aerosol models, the aerosol scattering reflectance at the visible and NIR bands were extrapolated from the SWIR bands using the look-up tables generated by the radiative transfer model.

The MWI has two SWIR bands (1.242 µm and 1.642 µm) characterized by high signal-to-noise ratio values (Table 1), which can be used to estimate the aerosol scattering radiances in turbid waters. The SWIR-AC algorithm scheme is same as the NIR-AC algorithm scheme except the band selection. Here, we used the fairly accurate extrapolation method instead of the look-up tables to estimate and extrapolate the aerosol scattering reflectance at the VIS/NIR bands from the SWIR band.

3.4 MUMM algorithm

In turbid waters, the non-negligible water-leaving radiances at the NIR usually cause the failure of the NIR-AC algorithm. Ruddick et al. [19] proposed an analytical method (MUMM) to separate the aerosol scattering reflectance from the $ρ_{r c}$ at two NIR bands as

{\begin{cases} ρ_{a} (865 n m) = \frac{α γ ρ_{r c} (865 n m) - ρ_{r c} (750 n m)}{α γ - ε} \\ ρ_{a} (750 n m) = ε ρ_{a} (865 n m) \end{cases}

where

α

is the ratio of water-leaving reflectance at 750 nm and 865 nm normalized by the downward atmospheric diffuse transmittance;

γ

is the ratio of two-way atmospheric transmittances from Rayleigh and aerosol effects between 750 nm and 865 nm, which can be approximated as 1.0;

ε

is the ratio of aerosol scattering reflectances at 750 nm and 865 nm. Finally, the aerosol scattering reflectances at visible wavebands are extrapolated from the estimated

ρ_{a} (750 n m)

and

ρ_{a} (865 n m)

using the previously described method [Eq. (10)].

One of the key issues for the MUMM algorithm is to setup the values of $α$ and $ε$ . In the default of the MUMM algorithm, $α$ and $ε$ are assumed to be spatially homogeneous with fixed values, such as 1.945 [20] and 1.0 (SeaDAS), respectively. The assumption of spatial homogeneity of the $ε$ parameter has been adopted in many studies with the CZCS (Coastal Zone Color Scanner) [21, 22]. For the turbid coastal waters with clear shelf waters nearby, $ε$ can be estimated by the $ρ_{r c} (750 n m)$ and $ρ_{r c} (865 n m)$ at clear water pixels [19]. However, if the coastal and shelf regions are all covered by turbid waters, such as the Changjiang River estuary in winter season, it is difficult to estimate the $ε$ _, and the “white” aerosol approximation ( $ε$ = 1.0) is adopted. Actually, the “white” aerosol approximation is rationale for the coastal and maritime aerosol models, which are the dominating aerosol types in the coastal regions [23]. Moreover, the “white” aerosol has been adopted in the vicarious calibration of the CZCS for global data processing [16].

Compared to the relative uniformity of the $ε$ values, the variation of $α$ is more complex. With turbidity variations, the value of $α$ can vary from 1.0 to 2.0 [24]. In turbid waters where the inherent optical properties at the NIR are predominantly determined by TSM, $α$ can be estimated as a function of the TSM concentration [15]

α = \frac{a_{w} (865 n m) + a_{s}^{*} T S M}{a_{w} (750 n m) + a_{s}^{*} T S M}

where

a_{w}

is the absorption coefficient of pure seawater;

a_{s}^{*}

is the specific absorption coefficient of TSM. In a previous study, a value of

a_{s}^{*} = 0 .015

m²/g was adopted [15, 25]. Since the TSM concentration is unknown prior to the atmospheric correction, the MWI-retrieved TSM concentration based on the UV-AC algorithm was used. By this approach,

α

can be determined pixel-by-pixel according to Eq. (13).

4. Results and discussion

4.1 Comparison of Lwn retrieved by different atmospheric correction algorithms

Figure 3 shows the MWI-Lwn products for some typical ocean color bands retrieved by the UV-AC, SWIR-AC and MUMM algorithms in the Changjiang River estuary on Jan. 22, 2017. Overall, the spatial distributions of the retrieved Lwn are consistent well among the three atmospheric correction algorithms. Especially, the Lwn values retrieved by the UV-AC and MUMM algorithms are almost identical in both their spatial distribution and magnitude in the visible and NIR bands. The Lwn values generally increase from shelf waters to turbid Changjiang River estuary and its inland waters. Compared to the UV-AC and MUMM algorithms, the SWIR-AC algorithm produced slightly higher Lwn values in all wavebands, which could be attributed to the large spectral distance from the SWIR to the VIS/NIR.

Fig. 3 The MWI-retrieved Lwn products from the three atmospheric correction algorithms at 2:58 GMT on Jan. 22, 2017. (a) UV-AC algorithm; (b) SWIR-AC algorithm; (c) MUMM algorithm.

Download Full Size | PDF

The results from these three atmospheric correction algorithms demonstrate that the Lwn values reach higher than 10.0 mW/(cm²µmsr) in the Hangzhou Bay, Changjiang River mouth, river streams and lakes. In these extremely turbid waters, the Lwn values at 413 nm are much lower [usually less than 2.0 mW/(cm²µmsr)] than at longer wavelengths (VIS and NIR bands), which would indicate the rationale of using the UV bands (UV-AC) for estimating the aerosol scattering radiance. Notably, the Lwn values in the NIR bands go up to 4.0 mW/(cm²µmsr) which pose a major challenge for the NIR-based atmospheric correction algorithm to retrieve accurate Lwn values in these waters.

4.2 Validation of MWI-retrieved Lwn with in situ data

For the two orbits MWI data over the Changjiang River Estuary (Dec. 2, 2016 and Jan. 22, 2017), three matchups data between MWI observations and SAS measurements were collected at two fixed stations [i.e., Hangzhou Bay station and Datong hydrologic station shown in Fig. 2(c)]. On Jan. 22, 2017, both these stations allowed to find matchup data. On Dec. 2, 2016, the Hangzhou Bay station was affected by cloud cover, while the Datong hydrologic station was nearly cloud free and captured by the MWI sensor. Considering the high flow conditions (usually larger than 1.0 m/s) and relatively uniform water turbidity in the stream lines, the nearest river pixel adjacent to the Datong hydrologic station was found suitable for comparison with the in situ data. Since the SAS system measured the Lwn continuously with a time step of several minutes, the time difference between MWI observations and SAS measurements was quite small. Figure 4 shows the comparison between MWI-retrieved Lwn values from the three atmospheric correction algorithms and in situ Lwn values at the two fixed stations. Overall, the MWI-retrieved Lwn spectra from the UV-AC algorithm agree closely in shape and magnitude with the in situ Lwn spectra. The overall correlation coefficient for the matched data was better than 0.93 [Fig. 5(d)], whereas the mean relative errors computed on these data were 28.73% (413 nm), 12.64% (443 nm), 14.74% (490 nm), 15.19% (520 nm), 13.31% (565 nm), 11.77% (620 nm), 5.18% (665 nm), 5.94% (682.5 nm) and 41.47% (750 nm).

Fig. 4 Comparisons of the MWI-retrieved Lwn with the in situ data from the two fixed stations. (a) Lwn at the Hangzhou Bay Bridge station on Jan. 22, 2017; (b) Lwn at the Datong hydrologic station on Jan. 22, 2017; (c) Lwn at the Datong hydrologic station on Dec. 2, 2016; (d) comparison of the MWI-retrieved Lwn values from the UV-AC algorithm and the in situ data with combining all the three matchups and first MWI 9 bands together.

Download Full Size | PDF

Fig. 5 Comparisons of the MWI-retrieved Lwn with the cruise data. (a)-(j) are the comparisons of the Lwn spectrums at the matchup stations D4, E2, E3, E4, F3, F4, F5, H2, Q1 and Q7, respectively. (k) is the comparison between the UV-AC retrieved and in situ Lwn with combing all the first 11 bands of the MWI together.

Download Full Size | PDF

To increase the matchups, we also used the cruise-obtained Lwn to validate the MWI-retrieved Lwn on Jan. 22, 2017, though there was more than 10 days difference. Except the stations near the coastal and in the Hangzhou Bay, the Lwn in the shelf stations are expected to be stable within a short period. Figure 5 shows the comparisons results at the 10 matchup stations. Overall, the MWI-retrieved Lwn are consistent with the in situ Lwn in both the spectrum shapes and magnitudes. The UV-AC and MUMM results are quite similar, while the SWIR-AC results appear to be slight overestimated. The correlation coefficient between the MWI-retrieved Lwn with the UV-AC algorithm and the in situ Lwn was better than 0.93, which was similar as the validation results by the two fixed stations.

4.3 Comparisons of the MWI and GOCI derived Lwn products

Due to the limited number of matchup data between MWI observations and in situ measurements, it was difficult to evaluate the MWI-retrieved Lwn values in diverse water types. Thus, we compared the MWI-retrieved Lwn values with GOCI data at similar observation times. Since the GOCI sensor (geostationary ocean color satellite sensor) provides ocean color observations from 0:30am to 7:30am (GMT) with a time step of one hour, a GOCI image closely matching with the MWI data was selected for this comparative evaluation.

Figure 6(a) shows the GOCI-derived Lwn from the UV-AC algorithm on Jan. 22, 2017. For comparison, the standard Lwn products from the KOSC were presented in Fig. 6(b). Clearly, the standard product significantly underestimates the Lwn in the coastal turbid waters, especially at the longer VIS and NIR bands, which was evidenced by the previous validations in the Korean coastal waters [26]. Actually, the GOCI-derived Lwn values from the UV-AC algorithm in the Changjiang River estuary and Hangzhou Bay were evidenced to be fairly good agreement with in situ Lwn data, with the relative errors of 25.0% (412 nm), 11.8% (443 nm), 9.9% (490 nm), 6.6% (555 nm), 13.9% (660 nm), 6.8% (680 nm) and 29.1% (745 nm) [7]. These statistical results are similar to those obtained for the MWI-retrieved Lwn data as shown in Fig. 4(d).

Fig. 6 GOCI-retrieved Lwn at 2:30 GMT on Jan. 22, 2017. (a) produced from the UV-AC algorithm; (b) the standard Level-2 product from the KOSC.

Download Full Size | PDF

Figure 7 shows the comparison of MWI and GOCI retrieved Lwn values using the UV-AC algorithm on Jan. 22, 2017. Note that a slight difference between the wavelengths of MWI and GOCI sensors was ignored. For example, MWI has bands 413, 565, 665, 682.5 and 750 nm comparable to those of (412, 555, 660, 680 and 745 nm) of GOCI. Except the 413nm band, the correlation coefficients between MWI and GOCI retrieved Lwn values were 0.9069 (443 nm), 0.9292 (490 nm), 0.9422(565 nm), 0.9783 (665 nm), 0.9780(682.5 nm), 0.9694 (750 nm), and 0.9630 (865 nm). The correlation coefficient at 413 nm was slightly lower (0.7148). The mean relative errors were obtained in the range of 26.63% (413 nm), 4.72% (443 nm), 3.69% (490 nm), 7.15% (565 nm), 9.45% (665 nm), 8.11% (682.5 nm), 14.68% (750 nm) and 18.55% (865 nm), which indicate fairly good consistency between the MWI and GOCI Lwn data in highly turbid waters.

Fig. 7 Comparisons of the MWI and GOCI retrieved Lwn values with unit of mW/(cm²µmsr) on Jan. 22, 2017.

Download Full Size | PDF

4.4 Comparisons of the TSM retrieved by MWI and GOCI

Based on the satellite-derived Lwn values, the TSM concentration was estimated [Eq. (4)] for further analysis. Figure 8 shows the comparison of TSM maps provided by the MWI and GOCI sensors on Jan. 22, 2017. Clearly, the retrieved TSM patterns are spatially well consistent between the MWI and GOCI sensors. Generally, the MWI and GOCI products show maximum TSM concentrations in the Hangzhou Bay (>1000 mg/l) and minimum TSM concentrations in shelf waters (< 20 mg/l), which indicates high spatial variation in this region [7, 27]. Statistical analysis performed between the MWI and GOCI TSM retrievals yielded the correlation coefficient 0.9556 and mean relative error 18.59% (Fig. 9).

Fig. 8 Comparison of the TSM products derived from MWI and GOCI data on Jan. 22, 2017. (a) MWI-derived TSM at 2:58 GMT; (b) GOCI-derived TSM at 2:30 GMT.

Download Full Size | PDF

Fig. 9 Comparison of the MWI and GOCI retrieved TSM values on Jan. 22, 2017.

Download Full Size | PDF

With a high spatial resolution (100m), the MWI is capable of resolving fine structures of TSM in the coastal regions. For example, the patchiness of TSM structures in the Hangzhou Bay and the influences of Hangzhou Bay Bridge on the TSM dynamics are clearly seen in the MWI-derived TSM data (Fig. 10). The MWI’s capability is also high for monitoring the TSM variations in lakes and rivers, as one can see the TSM distribution in the Changjiang River sections [Fig. 8(a)]. Figure 11 shows another example of the MWI-retrieved TSM variation in the Amazon River. It is seen that the TSM concentration varies significantly along the stream course, in particular, the TSM concentrations increase from 50 mg/l to 300 mg/l in the lower reaches of the Amazon River and its major branch Madeira River. Interestingly, after the Madeira River feeding into the main stream, the TSM concentration decreases significantly due to the increased river width and decreased flow velocity and disturbance.

Fig. 10 The fine structures of the TSM in the Hangzhou Bay retrieved by the MWI at 2:58 GMT on Jan. 22, 2017.

Download Full Size | PDF

Fig. 11 The variation of the TSM along the Amazon River retrieved by the MWI at 13:40 GMT on Sept. 22, 2016.

Download Full Size | PDF

4.5 Comparisons of the Chla retrieved by MWI and other sensors

Space-borne ocean color observations in clear waters require a high radiance sensitivity or signal-to-noise ratio, since the water-leaving radiance signal is weak and significantly affected the atmosphere contributions before being received by a sensor at the top of atmosphere. Figure 12 provides a comparison of the satellite-derived Chla maps from the MWI, MODIS/Aqua and VIIRS sensors in the northeast part of the Gulf of Mexico on Oct. 13, 2016. Clearly, the MWI-retrieved Chla patterns are consistent with both the MODIS/Aqua and VIIRS Chla products. The MWI sensor with high spatial resolution (100m) and good S/N ratio resolves a finer structure within the sub-mesoscale eddies and filaments on the shelf. Although the MWI underestimates the Chla, the difference between MWI and MODIS/Aqua Chla products is small as indicated by the correlation coefficient of 0.9432 and relative difference of 25.67% [Fig. 13(a)]. The correlation coefficient between MWI and VIIRS Chla products is 0.9557 with the relative difference of 31.06% [Fig. 13(b)]. Similar results were obtained in the north part of the Bengal Bay (Fig. 14). In the oligotrophic basin of the Bengal Bay, MWI resolves fine structures in the weak bloom areas in contrast to the MODIS/Aqua and VIIRS sensors.

Fig. 12 Comparison of the Chla products derived different satellite sensors in the northeast part of the Gulf of Mexico on Oct. 13, 2016. (a) MWI Chla; (b) MODIS/Aqua Chla; (c) VIIRS Chla.

Download Full Size | PDF

Fig. 13 Comparison of the MWI-retrieved Chla with the MODIS/Aqua and VIIRS Chla values in the Gulf of Mexico on Oct. 13, 2016. (a) Comparison between MWI and MODIS/Aqua Chla data; (b) Comparison between MWI and VIIRS Chla data.

Download Full Size | PDF

Fig. 14 Comparison of the Chla products derived from different satellite sensors in the north part of the Bengal Bay on Jan. 22, 2017. (a) MWI Chla; (b) MODIS/Aqua Chla; (c) VIIRS Chla.

Download Full Size | PDF

5. Conclusion

For the first time, we demonstrated the capability of MWI sensor for retrieving ocean color information in both turbid and clear waters. Four atmospheric correction algorithms were tested on the MWI data, including the NIR-AC algorithm for clear waters, and the UV-AC, SWIR-AC and MUMM algorithms for turbid waters. The TSM and Chla products were further analyzed in turbid and clear waters. Using the in situ data collected from four different stations, the MWI-retrieved Lwn, TSM and Chla products were evaluated. In addition, the MWI-derived products were compared with the GOCI, MODIS/Aqua and VIIRS-derived products.

In turbid coastal and inland waters, the MWI-retrieved Lwn values from the UV-AC algorithm showed fairly good agreement with the in situ data from the two fixed stations, with the mean relative errors of 28.73% (413 nm), 12.64% (443 nm), 14.74% (490 nm), 15.19% (520 nm), 13.31% (565 nm), 11.77% (620 nm), 5.18% (665 nm), 5.94% (682.5 nm) and 41.47% (750 nm). Assessment of the MWI-derived Lwn products in comparison to those of the GOCI showed a high correlation coefficient and a low mean relative difference. The relative difference between the MWI and GOCI-derived TSM products was considerably low (18.59%), with the correlation coefficient close to unity. Examination of the Chla products from different sensors demonstrated that in clear open oceans, the spatial patterns of MWI-derived Chla were generally consistent with both the MODIS/Aqua and VIIRS-derived Chla products. The advantage of MWI is its capability with high spatial resolution to resolve finer structures in algal blooms and sediment plumes.

The MWI is truly a dedicated space-borne sensor that provides high spatial resolution data for a number of bands (18 bands) in the VIS/NIR/SWIR/TIR. These data are highly valuable for ocean color work in coastal and inland waters. The future generation Chinese ocean color satellites HY-1E and HY-1F will carry the MWI sensor for providing continuity in ocean color observations. In the future, a systematic vicarious calibration will be performed for the MWI sensor based on buoy measurement data and ocean color observations provided by other space-borne sensors. This will be accompanied by a more comprehensive validation with additional matchup data.

Funding

National Basic Research Programme (“973” Programme) of China (grant #2015CB954002); National High Technology and Development Program of China (grant #2014AA123301); National Natural Science Foundation of China (NSFC) (grants #41676170, #41676172, and #41621064); “Global Change and Air-Sea Interaction” project of China (grant #GASI-03-03-01-01).

Acknowledgments

We thank the NASA and the KIOST/KOSC for providing the MODIS/Aqua, VIIRS and GOCI data. We thank the “LORCE” cruise and all the crews for collecting the in situ data. We also thank the reviewer for giving us constructive comments.

References and links

1. C. R. McClain, “A decade of satellite ocean color observations,” Annu. Rev. Mar. Sci. 1(1), 19–42 (2009). [CrossRef] [PubMed]

2. X. He, P. Delu, Q. Zhu, Z. Hao, and F. Gong, “On-orbit assessment of the polarization response of COCTS onboard HY-1B satellite,” Proc. SPIE 7862, 78620W (2010). [CrossRef]

3. C. D. Mobley, “Estimation of the remote-sensing reflectance from above-surface measurements,” Appl. Opt. 38(36), 7442–7455 (1999). [CrossRef] [PubMed]

4. Z. Lee, K. L. Carder, R. G. Steward, T. G. Peacock, C. O. Davis, and J. L. Mueller, “Protocols for measurement of remote-sensing reflectance from clear to turbid waters,” Presented at SeaWiFS Workshop, Halifax (1996).

5. N. M. Komick, M. P. F. Costa, and J. Gower, “Bio-opticalalgorithm evaluation for MODIS for western Canada coast-al waters: an exploratory approach using in situ reflectance,” Remote Sens. Environ. 113(4), 794–804 (2009). [CrossRef]

6. B. Han, H. Loisel, V. Vantrepotte, X. Mériaux, P. Bryère, S. Ouillon, D. Dessailly, Q. Xing, and J. Zhu, “Development of a Semi-Analytical Algorithm for the Retrieval of Suspended Particulate Matter from Remote Sensing over Clear to Very Turbid Waters,” Remote Sens. 8(3), 211 (2016). [CrossRef]

7. X. He, Y. Bai, D. Pan, N. Huang, X. Dong, J. Chen, C.-T. A. Chen, and Q. Cui, “Using geostationary satellite ocean color data to map the diurnal dynamics of suspended particulate matter in coastal waters,” Remote Sens. Environ. 133, 225–239 (2013). [CrossRef]

8. X. He, D. Pan, Y. Bai, Q. Zhu, and F. Gong, “Vector radiative transfer numerical model of coupled ocean-atmosphere system using matrix-operator method,” Sci. China. Ser. D 50(3), 442–452 (2007). [CrossRef]

9. X. He, Y. Bai, Q. Zhu, and F. Gong, “A vector radiative transfer model of coupled ocean-atmosphere system using matrix-operator method for rough sea-surface,” J. Quant. Spectrosc. Radiat. Transf. 111(10), 1426–1448 (2010). [CrossRef]

10. X. He, D. Pan, Y. Bai, and F. Gong, “A general purpose exact Rayleigh scattering look-up table for ocean color remote sensing,” Acta Oceanol. Sin. 25(1), 48–56 (2006).

11. H. R. Gordon, “Remote sensing of ocean color: a methodology for dealing with broad spectral bands and significant out-of-band response,” Appl. Opt. 34(36), 8363–8374 (1995). [CrossRef] [PubMed]

12. M. Wang and H. R. Gordon, “A Simple,Moderately Accurate,Atmospheric Correction Algorithm for SeaWiFS,” Remote Sens. Environ. 50(3), 231–239 (1994). [CrossRef]

13. H. R. Gordon, “Atmospheric Correction of Ocean Color Imagery in the Earth Observing System Era,” J. Geophys. Res. 102(D14), 17081–17106 (1997). [CrossRef]

14. X. He, D. Pan, and Z. Mao, “Atmospheric correction of SeaWiFS imagery for turbid coastal and inland waters,” Acta Oceanol. Sin. 23(4), 609–615 (2004).

15. X. He, Y. Bai, D. Pan, J. Tang, and D. Wang, “Atmospheric correction of satellite ocean color imagery using the ultraviolet wavelength for highly turbid waters,” Opt. Express 20(18), 20754–20770 (2012). [CrossRef] [PubMed]

16. R. H. Evans and H. R. Gordon, “Coastal Zone Color Scanner system calibration: A retrospective examination,” J. Geophys. Res. 99(C4), 7293–7307 (1994). [CrossRef]

17. M. Wang and W. Shi, “Estimation of ocean contribution at the MODIS near infrared wavelengths along the east coast of the U.S.: two case studies,” Geophys. Res. Lett. 32(13), L13606 (2005). [CrossRef]

18. M. Wang, “Remote sensing of the ocean contributions from ultraviolet to near-infrared using the shortwave infrared bands: simulations,” Appl. Opt. 46(9), 1535–1547 (2007). [CrossRef] [PubMed]

19. K. G. Ruddick, F. Ovidio, and M. Rijkeboer, “Atmospheric correction of SeaWiFS imagery for turbid coastal and inland waters,” Appl. Opt. 39(6), 897–912 (2000). [CrossRef] [PubMed]

20. K. G. Ruddick, V. De Cauwer, Y.-J. Park, and G. Moore, “Seaborne measurements of near infrared water-leaving reflectance - the similarity spectrum for turbid waters,” Limnol. Oceanogr. 51(2), 1167–1179 (2006). [CrossRef]

21. M. Viollier, D. Tanré, and P. Y. Deschamps, “An algorithm for remote sensing of water color from space,” Boundary-Layer Meteorol. 18, 247–267 (1980). [CrossRef]

22. H. R. Gordon, O. B. Brown, R. H. Evans, J. W. Brown, R. C. Smith, K. S. Baker, and D. K. Clark, “A semianalytical radiance model of ocean color,” J. Geophys. Res. 93(D9), 10909–10924 (1988). [CrossRef]

23. H. R. Gordon and M. Wang, “Retrieval of water-leaving radiance and aerosol optical thickness over the oceans with SeaWiFS: a preliminary algorithm,” Appl. Opt. 33(3), 443–452 (1994). [CrossRef] [PubMed]

24. M. Doron, S. Bélanger, D. Doxaran, and M. Babin, “Spectral variations in the near-infrared ocean reflectance,” Remote Sens. Environ. 115(7), 1617–1631 (2011). [CrossRef]

25. G. F. Moore, J. Aiken, and S. J. Lavender, “The atmospheric correction of water colour and quantitative retrieval of suspended particulate matter in Case II waters: application to MERIS,” Int. J. Remote Sens. 20(9), 1713–1733 (1999). [CrossRef]

26. W. Kim, J. Moon, Y. Park, and J. Ishizaka, “Evalution of chlorophyll retrievals from Geostationary Ocean color Imager (GOCI) for the North-East Asian region,” Remote Sens. Environ. 184, 482–495 (2016). [CrossRef]

27. Y. Bai, X. He, D. Pan, Q. Zhu, H. Lei, B. Tao, and Z. Hao, “The extremely highconcentration of suspended particulate matter in Changjiang Estuary detected by MERIS data,” Proc. SPIE 7858, 78581D (2010). [CrossRef]

Bands	Center wavelength^a	Band width^a	Typical radiance^b	Signal-to-noise ratio^c
V1	413	403-423	7.86	390
V2	443	433-453	7.02	515
V3	490	480-500	5.31	557
V4	520	510-530	4.58	556
V5	565	555-575	3.39	555
V6	620	610-630	2.19	467
V7	665	655-675	1.60	399
V8	682.5	677.5-687.5	1.45	273
V9	750	740-760	0.93	302
V10	820	810-830	0.59	203
V11	865	845-885	0.62	255
V12	905	895-915	0.47	138
V13	940	930-950	0.36	100
V14	980	970-990	0.39	59
S1	1.242	1.232-1.252	0.54	510
S2	1.642	1.630-1.654	0.73	678
T1	8.475	8.125-8.825	330K	0.014K
T2	9.100	8.925-9.275	330K	0.026K

Ocean color retrieval from MWI onboard the Tiangong-2 Space Lab: preliminary results

Abstract

1. Introduction

2. Data and methods

2.1 In situ measurements

2.2 Satellite ocean color retrieval

(1) Chla retrieval

(2) TSM retrieval

3. Atmospheric correction of the MWI data

3.1 NIR-AC algorithm

3.2 UV-AC algorithm

3.3 SWIR-AC algorithm

3.4 MUMM algorithm

4. Results and discussion

4.1 Comparison of Lwn retrieved by different atmospheric correction algorithms

4.2 Validation of MWI-retrieved Lwn with in situ data

4.3 Comparisons of the MWI and GOCI derived Lwn products

4.4 Comparisons of the TSM retrieved by MWI and GOCI

4.5 Comparisons of the Chla retrieved by MWI and other sensors

5. Conclusion

Funding

Acknowledgments

References and links

Cited By

Figures (14)

Tables (1)

Equations (13)

Optics Express