Open Access
Add to BibSonomy
Abstract
An empirical algorithm is proposed to estimate suspended particulate matter (SPM) ranging from 0.675 to 25.7 mg L−1 in the turbid Pearl River estuary (PRE). Comparisons between model predicted and in situ measured SPM resulted in R2s of 0.97 and 0.88 and mean absolute percentage errors (MAPEs) of 23.96% and 29.69% by using the calibration and validation data sets, respectively. The developed algorithm demonstrated the highest accuracy when compared with existing ones for turbid coastal waters. The diurnal dynamics of SPM was revealed by applying the proposed algorithm to reflectance data collected by a moored buoy in the PRE. The established algorithm was implemented to Hyperspectral Imager for the Coastal Ocean (HICO) data and the distribution pattern of SPM in the PRE was elucidated. Validation of HICO-derived reflectance data by using concurrent MODIS/Aqua data as a benchmark indicated their reliability. Factors influencing variability of SPM in the PRE were analyzed, which implicated the combined effects of wind, tide, rainfall, and circulation as the cause.
© 2018 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
Pearl River (PR) is the second largest river in terms of discharge volume in China. It has a catchment area of 453,690 km2. It is composed of a drainage basin including the West, North, and East rivers. Its annual average discharge rate is about 10,000 m3 s−1. It debouches into the South China Sea through eight outlets. It transports an estimated 75.29 Mt of sediment to the ocean between 1955 and 2005, of which 89% is from the West River [1]. The region is characterized by a complicated hydrodynamic system regulated by a number of forcing mechanisms containing bottom topography, freshwater discharge, wind, tide, and coastal current, all of which operate in concert to control the local circulation and water properties [2].
Sediment transport in rivers not only represents a major part of global material cycle but also indicates the influence of human activity and environmental change on local marine ecosystems [3]. Suspended sediment, the main source of suspended particulate matter (SPM) in estuarine waters, in the water column can affect primary production by influencing light penetration and nutrient availability [4]. Suspended sediment also plays an important role in controlling the structure and functioning of ecosystems [5]. Furthermore, sediment is one of the environmental stressors for marine protected areas (MPAs) that are being implemented with an increasing pace around the world to protect marine biodiversity and ecosystems and mitigate over-fishing [6]. PR is not exceptional. The Pearl River Estuary White Dolphin Reserve is one of the environmental initiatives to protect endangered marine species. Therefore, accurate quantification of suspended sediment in the water column is of great use and of urgent need.
Satellite remote sensing presents a valuable and attractive tool for observations of marine environments. With the merits of low cost and synoptic view over large spatiotemporal scales, satellite remote sensing has widely been used for monitoring of suspended sediment in turbid coastal waters [7]. There are mainly three types of algorithms for the retrieval of suspended sediment from remotely sensed data, namely empirical model [8], semi-analytical model [9, 10], and physical model [11]. The first two models are commonly utilized. Although empirical models are simple to implement, they are lack of physical foundation and their applicability is limited to areas with similar optical properties to those used for algorithm development. The semi-analytical models rely on accurate knowledge of optical properties of optically active components that cannot be measured accurately at present. The capabilities of different sensors to observe suspended matter were evaluated including Moderate Resolution Imaging Spectroradiometer (MODIS) onboard the Terra and Aqua satellites [12–15], Sea-Viewing Wide Field-of-view Sensor (SeaWiFS) onboard the Orbview-2 satellite [16], Geostationary Ocean Color Imager (GOCI) onboard the Communication, Ocean and Meteorological Satellite (COMS) satellite [17], MEdium Resolution Imaging Spectrometer (MERIS) onboard the Envisat satellite [18], and Hyperspectral Imager for the Coastal Ocean (HICO) onboard the International Space Station (ISS) [19]. Nevertheless, a common method that is applicable to different satellite sensors is still challenging [20].
With respect to the PR, several algorithms were established to estimate suspended sediment concentration. Based on in situ measured data, Fenfen Liu, et al. [21] developed an empirical band ratio algorithm to obtain suspended sediment for two ranges and applied to MERIS imagery. By comparing different combinations of wavebands, Dazhao Liu, et al. [22] found that the three-band model gave the best results. Xi and Zhang [23] proposed an empirical band ratio algorithm by using MERIS channels. Xing, et al. [24] analyzed the correlations between different combinations of wavebands and suspended sediment and proposed an empirical model that was then applied to Hyperion data. Ye, et al. [25] elucidated the variations of suspended sediment after the passage of Typhoon Vicente. It is clear that most of those proposed approaches are empirical. Although variations of suspended sediment in the PR were studied, diurnal dynamics of suspended sediment in the PR is scarcely reported. Furthermore, the mechanisms regulating the variations of suspended sediment behind the scene are still unknown.
The objectives of this study are three-fold: (1) to develop an inversion method for SPM in the turbid PR with improved accuracy; (2) to illustrate the diurnal dynamics of SPM and reveal the controlling mechanisms in the PR; (3) to apply the proposed algorithm to hyperspectral satellite imagery. To achieve the goals, in situ measured reflectance and SPM data are used for algorithm calibration and validation. By implementing the developed algorithm to buoy-measured reflectance data in the PR estuary (PRE), the diurnal dynamics of SPM in the study area are revealed. Finally, the established algorithm is applied to satellite scenes collected by HICO.
2. Data and method
In this study, data from three cruises were utilized. The first cruise was carried out in October-November 2003 along the South China coast. The second cruise was conducted between August 14 and 28 2007 when one station was occupied with a moored optical buoy in the PRE. The third cruise was done on June 5 2012 in the PRE. All stations are shown in Fig. 1. The Waglan Island station where meteorological and tide observations were made by the Hong Kong Observatory is also annotated in Fig. 1.
Fig. 1 Map of the study area. Bathymetry data were obtained from GEBCO with a spatial resolution of 1 km. The red pentagons denote the location where an optical buoy was deployed between August 14 and 28 2007. The green triangle presents the station where meteorological and tide observations were made by the Hong Kong Observatory. The blue pluses indicate stations visited during the cruise on June 5 2012. The magenta circles show stations where concurrent apparent optical properties (AOP) and SPM measurements were made during the cruise carried out between October and November 2003. The blue and red lines display transects whose SPM profiles from satellite scenes are plotted below.
2.1 In situ sampling
At each station for the 2003 and 2012 cruises, water samples were collected from the surface layer less than 50 cm deep with a plastic bucket. For the 2007 field survey, surface water samples were collected in the same way as for the other cruises at 10:00, 12:00, and 15:00 Beijing Time (BJT) in the vicinity of the optical buoy when the sea state was not high. In total, 55 samples were collected, 6 from the 2003 cruise, 33 from the 2007 cruise, and 16 from the 2012 cruise. The whole data set was randomly separated to calibration and validation data sets, with 38 data points for the former and 17 for the latter.
Once back in the laboratory, water samples were filtered under low vacuum onto pre-weighed polycarbonate filters with a pore size of 0.45 μm. The filters were then rinsed with 30-50 ml of Milli-Q water to eliminate salts and avoid cell lysis caused by osmotic imbalance. At 40 °C, the filters were dried for 48 hours. Under room temperature, the filters were reweighed. SPM concentration was obtained by subtracting the weight of blank filters from that of corresponding filters which particulate attached to.
2.2 In situ data processing
During the 2003 cruise, a Satlantic MultiPro profiler was deployed at stations shown in Fig. 1. The profiler measured upwelling radiance and downwelling irradiance at depths simultaneously. A cosine irradiance sensor was installed on deck to remove potential effects of cloud cover. The sensors measured radiance or irradiance at seven channels centered at 412, 443, 490, 520, 555, 620, and 683 nm. Remote sensing reflectance (Rrs) was obtained following the NASA protocol. Since the profiler does not have the 645 channel, Rrs(645) was calculated from the following equation based on the buoy observations in 2007:
For the 2007 field campaign, Rrs data were measured with an optical buoy. It was equipped with several commercial hyperspectral radiometers (USB2000, Ocean Optics, Inc.). One radiometer mounted on the top of the buoy was used to measure the incident spectral irradiance above the sea surface (Es). At nominal depths of 0.32 and 2.30 m below the sea surface, two radiometers were mounted at each depth, one for underwater downwelling irradiance (Ed) and the other for upwelling radiance (Lu). A pressure sensor was integrated to obtain the depth signal. Thus we can record the exact depth of radiometers and the depth fluctuation caused by the buoy’s vertical wave-follow-motion can be corrected. A tilt sensor was also outfitted to record the pitch of the buoy. Other ancillary sensors included wind sensor, GPS, etc. These optical and ancillary data were automatically measured eleven times from 8 to 18 BJT at hourly interval. Data at 02:00 BJT in the midnight were also recorded as dark signals of Ed (or Lu). All measured data were sent to the land-based laboratory via the GPRS or CDMA network in real time. The spectral range of radiometers was from 350 to 900 nm with a spectral resolution of 0.38 nm and radiometers were calibrated in laboratory with a vigorous adherence to the hyperspectral radiometric calibration protocols of the National Institute of Metrology of China [26]. Solar panels were used to support power supply.
Rrs is defined as the ratio of water-leaving radiance (Lw) to incident irradiance (Es). Since Es was measured directly by the buoy, Lw is the only unknown. It was obtained from steps below:
- (1) Attenuation coefficient for Lu (Ku) was calculated using the following equation according to [27],
where Lu(0.32) and Lu(2.3) are radiances measured at 0.32 and 2.3 m below the sea surface, respectively, and 1.98 is the vertical distance between the two radiance sensors.
- (2) Lu just beneath the sea surface (Lu(0-)) was extrapolated by using Ku. The equation is expressed as,
where Z denotes the depth.
where ρ is the Fresnel reflectance of the air-sea interface and n is the refractive index of seawater. The ratio on the right side of the above equation indicates the Lu transmittance of the sea surface for nadir-viewing geometries. In this study, an empirical constant of 0.543 [28] was used.
During the 2012 cruise, in situ Rrs data were collected by using a hyperspectral radiometer (USB4000, Ocean Optics, Inc.) following the NASA ocean optics protocol [29]. The radiance signals reflected from water surface (Lwater), sky (Lsky), and a grey plaque (Lp) were measured. For each target, 15 scans were made and their average was used for Rrs calculation. Lw and Es were calculated from
Then Rrs = Lw/Es can be obtained.2.3 Satellite data processing
HICO scenes for November 29 2011, September 12 2012, November 10 2012, June 21 2013, and June 29 2013 were acquired from National Aeronautics and Space Agency (NASA) ocean color data archive. The MODIS/Aqua scene for November 29 2011 was also obtained from the same source. The SeaWiFS Data Analysis System (SeaDAS, version 7.4) developed by NASA was used for satellite data processing. For Aqua imagery, the shortwave infrared (SWIR) approach [30] was utilized to correct atmospheric contribution to satellite-measured radiance, which has successfully been applied for waters with high loads of sediment like the Bohai Sea [31]. The Management Unit Mathematical Models (MUMM) scheme [32] that was also developed for turbid coastal waters was exploited for atmospheric correction of HICO scenes since the HICO sensor has no SWIR wavebands. When the MUMM method was used for the MODIS imagery, negative Rrs was obtained over waters of high turbidity. Therefore the SWIR method was chosen for the MODIS imagery instead. Rrs at 555 and 645 nm for Aqua and at 553 and 644 nm for HICO were generated. The standard flags of the SeaDAS were output and checked, which include atmospheric correction failure, cloud, sun glint, navigation error, low water-leaving radiance, high top-of-atmosphere radiance, stray light, and excessive viewing zenith angle, and solar zenith angle [33].
2.4 Ancillary data acquisition
Ancillary data consist of wind speed and direction, accumulated rainfall, tide height, and rainfall rate. The first three data were observed at the Waglan Island station operated by the Hong Kong Observatory. The last one was acquired from the NASA TRMM data archive. TRMM-derived rainfall rate over the area bounded by 112°-115° N and 21.8°-24° E were used. All ancillary data were daily mean values.
2.5 Performance assessment
To evaluate the accuracy of algorithms, mean absolute percentage error (MAPE) and root mean square deviation (RMSD) as well as linear-regressed R2, slope and intercept were used. MAPE was defined by the following equations:
where yp and ym denote predicted and measured values, respectively, and N is the number of samples. RMSD was calculated from3. Results
3.1 In situ measurements
The histogram of in situ measured SPM and the corresponding cumulative percentage are shown in Fig. 2(a). SPM concentration varied in the range of 0.675-25.7 mg L−1, 91% of which is less than 10 mg L−1. Rrs spectra for different levels of SPM are displayed in Fig. 2(b). They indicated typical characteristics for case II waters. Rrs shows a peak 570 nm, except for the case of 0.675 mg L−1 which demonstrates a broad peak between 540 and 560 nm. Two spectral troughs around 620 and 670 nm are apparent and caused by the strong absorption of phycocyanin and chlorophyll-a, respectively [34]. The shoulders around 650 nm are related to the fluorescence of phycocyanin [35]. The peaks around 685 nm are due to chlorophyll-a fluorescence. As SPM concentration increases, the peak around 685 nm shifts toward longer wavelengths. It is associated with the combined effect of chlorophyll-a fluorescence and particulate backscattering. Furthermore, Rrs(555) and Rrs(645) present prominent changes as SPM enhances.
Fig. 2 (a) Histogram and cumulative percentage for in situ measured SPM. (b) Remote sensing reflectance (Rrs) spectra for difference SPM concentrations. The y axis on the right side corresponds to Rrs spectra plotted with dashed lines. The y axis on the left side is for Rrs spectra plotted with solid lines. Two black dashed lines indicate the positions for 555 and 645 nm, respectively.
3.2 Algorithm calibration and validation
In this study, HICO channels centered at 553 and 644 nm with a spatial resolution of 90 m are exploited for SPM inversion. Similar wavebands are also outfitted on the MODIS sensor and the center wavelengths are 555 and 645 nm with spatial resolutions of 500 and 250 m, respectively. Since differences in the corresponding center wavelengths of the two sensors are ignorable, Rrs(555) and Rrs(645) are used for SPM calculation hereinafter. As shown by previous studies, reflectance band ratio or difference indicates significant correlations with SPM. Both of the two approaches are taken into account in this work, which demonstrates better results than each of them is used alone (data now shown). 38 out of 55 data points were used for model calibration, as plotted in Fig. 3(a). Different statistical methods were tested including power decay, polynomial, exponential decay, and rational. The best results were obtained with a rational function model. The best fitted relationship can be expressed as
Fig. 3 (a) Model development using the calibration data set. (b) Comparisons between in situ measured and model predicted SPM using the calibration and validation data sets. (c) Performance assessment of different SPM algorithms. Empirical models compared here include those proposed by Ye et al. (2014), Zhao et al. (2011), Sipelgas et al. (2006), Kuster et al. (2007). The black dashed lines in (b) and (c) denote the 1:1 relationship.
To evaluate the accuracy of the proposed algorithm, model estimated SPM is plotted versus in situ measured SPM in Fig. 3(b) using both the calibration and validation data sets. The statistical results are shown in Table 1. It can be seen that the scatter plot lies closely to the 1:1 line. Model estimated SPM demonstrated good agreement with in situ measurement as indicated by linear-regressed R2s of 0.97 and 0.88 from the calibration and validation data sets, respectively. MAPEs are < 30%. Slopes are very close to 1 and absolute values of intercepts are < 0.2.
Table 1. Statistics for the proposed algorithm using the calibration and validation data sets
3.3 Comparison with published models for turbid coastal waters
The algorithm proposed in this study was evaluated against representative ones developed by Kutser, et al. [36], Sipelgas, et al. [37], Ye, et al. [25], and Haihong Zhao, et al. [38]. The comparisons are shown in Fig. 3(c) and results from the statistical analysis are summarized in Table 2. Apparently, Kutser’s algorithm induces a plateau at the low end of SPM concentration. Sipelgas’s and Zhao’s algorithms produce similar patterns. Model-predicted SPM versus measured values show large scatter and deviate significantly from the 1:1 line as indicated by the relatively small slope and large intercept. MAPEs are > 60% and RMSDs are > 2.8 mg L−1 for the three algorithms. On one hand, it demonstrates that their study areas have higher loads of SPM than the PR. On the other hand, it suggests that optical properties in turbid coastal waters are very complex and may differ from one region to another. In contrast, Ye’s algorithm produced better results than the above-mentioned three ones with a R2 of 0.93, a slope of 0.97, and an intercept of 0.051. This is explainable since they also used data collected in the PR for algorithm development. The algorithm established in this study produced the best agreement with in situ measurements. The scatter points are distributed tightly along the 1:1 line. A linear regression between model estimated and in situ measured data resulted in a R2 of 0.97, a slope of 0.97, and an intercept of 0.058. A MAPE of 26.08% and a RMSD of 0.98 are both lower than those from other algorithms.
Table 2. Performance of different empirical algorithms developed for different waters around the world. The whole data set was used.
3.4 Diurnal variations of SPM in the Pearl River estuary
Based on the proposed algorithm in this study, SPM time series were generated from the buoy-measured Rrs data in the PRE. SPM from 8:00 to 18:00 BJT with hourly interval between August 14 and 28 2007 is shown in Fig. 4(a). Daily mean rainfall and wind velocity and direction are illustrated in Figs. 4(b)-4(d). No significant trend can be found for the whole time series of SPM. But some short-term patterns are detectable. Between August 14 and 16, the highest SPM was found between 15 and 18 BJT and the lowest values were observed in the morning. And the diurnal cycle of SPM was reversed to the tidal cycle. During these three days, SPM demonstrated an increasing trend. There was heavy rain on August 16 and the daily rainfall reached 54 mm. From August 17 to 18, SPM was much lower than previous days and decreased by > 90%. There were two peak values for August 17, one at 11:00 and the other at 18:00. For August 18, the highest SPM was observed in the morning. From August 19 to 22, strong southwesterly winds prevailed the study region with the speed over 6 m s−1 and were favorable for upwelling. Mann-Kendall test analysis indicated that SPM showed a statistically significant increasing trend with p < 0.00001. In the period of August 20-24, a bloom event occurred with chlorophyll-a concentration > 6 mg m−3 [39]. For the first three days during the bloom period, the highest SPM was observed in the afternoon and SPM showed a negative relationship with the tidal height with large values during ebb tide and low values during flood tide, which is consistent with the pre-bloom conditions. However, the highest SPM was observed in the afternoon for the latter two days during the bloom period and the behavior of SPM followed that of tide. The highest SPM for the entire time series happened on August 22. Since August 24, SPM decreased gradually and returned to the normal level. During the post-bloom period, the highest SPM was found at 8:00 BJT for August 15 and 26 and at 17:00 BJT for August 27 and 28. The wind speed on August 27 and 28 was also > 6 m s−1 while the dominant winds were northeasterly. And the rainfall on August 27 was 10.5 mm. These factors together resulted in no enhancement in SPM on those days.
Fig. 4 (a) Time series of buoy-derived hourly SPM (black solid point) and tide height (red line) between August 14 and 28 2007. (b)-(d) show the daily mean rainfall and wind velocity and direction during the observation period. The meteorological and tide data were acquired from the Waglan Island station operated by the Hong Kong Observatory. All dates start from 00 h Beijing Time.
4. Discussion
4.1 Application to HICO imagery
The algorithm proposed in this study was applied to a HICO scene collected on November 29 2011 as shown in Fig. 5(a). SPM in the dark red region of the map reached 50 mg L−1. It is clear that SPM on the western side of the PRE is higher than on the eastern side. This is associated with the fact that the eight outlets of the PRE are located on the western side, through which the major tributaries of the PR discharge freshwater with high loads of SPM to the PRE [40]. MODIS/Aqua derived SPM for the same day is shown in Fig. 5(b). Although the general pattern is similar for the two maps, discrepancies do exist. SPM on the west side along the coast is higher in MODIS/Aqua scene than in the HICO scene. This is very likely caused by the resuspension of bottom sediment in shallow waters. On the other hand, SPM on the eastern side in the HICO scene is higher than in the MODIS/Aqua scene, which is probably associated with the settlement of suspended sediment in the area when MODIS/Aqua overpassed. SPM fronts are apparent in both scenes in the south of the estuary. Note that the SPM front in the HICO scene extended much more southward then in the MODIS/Aqua scene. This is probably related to the tidal current that the ebb duration is much longer than the flood duration on the western side of the estuary [2].
Fig. 5 SPM maps observed by HICO (a) and MODIS/Aqua (b) on November 29 2011. The algorithm proposed in this study was applied. The satellite overpass time was 2:10:50 GMT for HICO and 5:45:00 GMT for Aqua, respectively. The area outlined with a purple box in (a) is enlarged in (c) Two transects in black and green colors are annotated whose Rrs spectra are shown in (d).
The area outlined with a purple box in Fig. 5(a) is enlarged in Fig. 5(c). Two transects through apparent SPM gradients are shown in green and black in Fig. 5(c), respectively. Rrs spectra along the two transects are displayed in Fig. 5(d). Rrs is generally higher in the green to infrared domain along the green track than along the red track while Rrs is similar in the blue to green domain for the two tracks. This is associated with the higher SPM along the green track than along the red track, which resulted in higher backscattering, and with the high CDOM concentration for both tracks, which caused similar Rrs in the blue to green domain. Rrs spectral shapes of are similar to those observed in Fig. 2(b) during high SPM and those shown by Dazhao Liu, et al. [41].
To further examine the difference of SPM from the two scenes, two representative profiles of satellite-derived SPM for November 29 2011 are shown in Fig. 6. The blue transect is through the SPM front. Except the peak with SPM > 13 mg L−1 along the northern end of this transect, HICO-measured SPM is generally higher than Aqua-measured SPM. Satellite-derived SPM along this transect from both sensors is smooth as indicated by the small standard deviations from the 3 × 3 window where data points were extracted. In contrast, satellite-measured SPM along the red transect presents big fluctuations, especially for Aqua-derived SPM. Two apparent peaks appear in the HICO-measured SPM profile, one around 113.82° E and the other around 113.93° E. These two peaks cannot be seen along the Aqua-measured SPM profile. But the Aqua-derived SPM along the red transect demonstrates peaks that are not observable by the HICO-derived SPM profile, for example around 113.875° E. Since the red transect is close to the coast line, the complicated characteristics of SPM along the red transect is likely due to disturbance by busy marine traffic passing-by. Small scale features can be much better captured by HICO with a higher spatial resolution as indicated by the smoother shape of the profile.
Fig. 6 Satellite-derived SPM on November 29 2011 along the red (a) and blue (b) transects shown in Fig. 1. The green and magenta lines indicate SPM from the HICO and Aqua scenes, respectively. The shaded colors demonstrate the standard deviations for the 3x3 window where data points were extracted.
Four HICO-derived SPM maps over the PRE with appropriate sky conditions are displayed in Fig. 7. The major pattern with high SPM on the western side and low SPM on the eastern side of the estuary is very clear. For the two scenes collected in 2012 that are nearly two months apart, significant differences can be found as displayed in Figs. 7(a) and 7(b). SPM is higher during the wet season with high discharge of freshwater while SPM is lower during the dry season with low discharge of freshwater. The spatial extension of the SPM front is more southward during the wet season than during the dry season. HICO-measured SPM for June 21 and 29 2013 are shown in Figs. 7(c) and 7(d). SPM as high as 80 mg L−1 can be observed south of Hongqimen and Hengmen. In both scenes, the plumes caused by freshwater discharges from Modaomen and Jimingmen can be remarkably distinguished. The two outlets account for 28.3% and 6.1% of the total river discharge of the PRE, respectively. SPM in the upper stream of the PR during June is much higher than during September and November. Although the two scenes were collected eight days apart, prominent differences in SPM in coastal waters can be observed. For example, in Shenzhen Bay satellite-derived SPM on June 21 2013 was as twice as that on June 29 2013. As discussed below, wind and tide are two major factors affecting SPM in the PRE. The examination of wind velocity data showed no significant difference for the two days with values of 3.54 and 4.28 m s−1, respectively. Therefore, the difference of SPM for the two days were mainly caused by tide.
Fig. 7 HICO-derived SPM maps for September 12 2012 (a), November 10 2012 (b), June 21 2013 (c), and June 29 2013 (d). The locations of Jimingmen, Modaomen, Hongqimen, Hengmen, and Shenzhen Bay are annotated.
4.2 Merit and limitation of the proposed algorithm
Comparisons between model predicted and in situ measured SPM suggest the high accuracy of the proposed algorithm in this study. Relative errors of SPM defined by the difference between model estimated and field measured data scaled by filed measured data was examined against field measurements (data not shown), which indicated no significant correlation and suggested the robustness of the algorithm. Furthermore, the empirical algorithm is very simple and fast to implement. Despite channels exactly centered at 555 and 645 nm may not be equipped, Rrs data for these two bands can be conveniently inverted from those for similar bands on retired or on-orbit satellite sensors, such as SeaWiFS, MEIRS, Visible Infrared Imaging Radiometer Suite (VIIRS) onboard the Suomi National Polar-Orbiting Partnership (Suomi NPP) satellite and, and the Ocean and Land Color Instrument (OLCI) onboard Sentinel 3A satellite. Thus historical and present spatiotemporal variations of SPM can be investigated and then effects of human activities and climate change on the marine environment can be studied. With several satellite imagery collected on the same day, the dynamics of SPM in the turbid estuarine coastal waters can be revealed and model simulated transport of SPM can be validated and assessed.
It should also be noted that limitations of the proposed algorithm do exist. The SPM concentration used for algorithm development ranged between 0.675 and 25.7 mg L−1. Higher levels of SPM may happen, especially in wet season. For example, SPM reported by [24] amounted to 140 mg L−1 in the PRE. Therefore, more field surveys during both wet and dry seasons are needed to cover a larger dynamic range of SPM for algorithm refinement. When quantifying SPM from satellite imagery, a proper correction for atmospheric effect and thus an accurate retrieval of reflectance are essential for implementing the proposed model [42]. In highly turbid coastal waters, the default black pixel assumption which assumes that water-leaving radiance in the near-infrared (NIR) are negligible fails. Great efforts have been made for the development of atmospheric correction methods over turbid coastal waters, such as the Management Unit of the North Sea Mathematical Models (MUMM) [32] and shortwave infrared (SWIR) [43]. The latter one has been demonstrated to be able to accurately retrieve radiance in turbid waters [31]. And its capability was evaluated for investigating impacts of land use and land cover on water quality in the PRE [44]. In this work, MODIS-derived Rrs based on the SWIR scheme was used as a benchmark. Since the HICO sensor does not have channels in the SWIR domain, the MUMM approach was applied instead. HICO and MODIS/Aqua measured Rrs at 555 and 645 nm are compared in Fig. 8. It can be seen that HICO-derived Rrs agrees well with MODIS/Aqua-derived Rrs, as indicated by the slopes close to 1 and R2 of 0.53 and 0.66 for the two wavebands. Therefore, HICO-derived Rrs based on the MUMM method and thus HICO-measured SPM maps are reliable. On the other hand, for systematic evaluation of HICO-derived Rrs, concurrent in situ measurements are required.
Fig. 8 Comparisons between HICO and MODIS/Aqua measured Rrs at 555 and 645 nm on November 29 2011. The MUMM atmospheric correction approach was used for the HICO imagery and the SWIR method was used for the MODIS/Aqua imagery. The best fitted linear regression results, R2, and the number of data samples are annotated. The red lines show the best fitted correlations and the dashed black lines denote the 1:1 relationship.
4.3 Factors affecting SPM variations
As demonstrated above, SPM presented complicated characteristics during the observation period. This corroborates the finding that the hydrodynamic conditions in the PRE are regulated by different factors. To further elucidate the driving forces for variations of SPM, the daily mean SPM is plotted against wind speed and direction, accumulated rainfall, tide height, and TRMM-measured rainfall rate with SPM leading by 0 and 1 day in Figs. 9(a) and 9(b), respectively. Determination coefficients from linear regressions between SPM and meteorological parameters are given in Table 3. When no lag was considered, there are no apparent trends between SPM and accumulated rainfall, while positive correlations are found between SPM and wind speed and direction and TRMM-measured rainfall rate and a significant negative correlation is observed between SPM and tide height. SPM shows the highest correlation with tide height with a linear regressed R2 of 0.65. When one day lag was considered, SPM versus wind speed and direction and TRMM-measured rainfall rate indicate positive correlations and SPM versus tide height and accumulated rainfall demonstrated negative correlations. R2s from linear regressions between SPM and wind speed and direction increased from 0.14 to 0.6 and from 0.16 to 0.18, respectively. The correlation between SPM and rainfall rate weakened with R2 decreasing from 0.25 to 0.1. Moreover, the correlation between SPM and accumulated rainfall becomes statistically significant.
Fig. 9 Correlations between daily mean SPM and meteorological and hydrodynamic parameters including daily wind speed (black) and direction (red), accumulated rainfall (green), tide height (blue), and TRMM rainfall rate (pink). All ancillary data except TRMM rainfall rate were acquired from the Waglan Island station. The TRMM rainfall rate was averaged over the area defined by 112°-115° N and 21.8°-24° E. The daily mean SPM was calculated by averaging buoy-derived hourly SPM based on our proposed algorithm. Panels (a) and (b) illustrate the relationships with SPM leading by 0 and 1 day, respectively. The R2s from linear regressions between SPM and different parameters are also annotated with the corresponding colors.
Table 3. R2s from linear regression between daily SPM and several meteorological parameters including wind speed and direction, accumulated rainfall at the Waglan Island station, tide height, and TRMM-derived rainfall rate. Negative values indicate negative correlations.
Apparently, SPM increases as wind increases either in velocity or in angle of direction. The strongest wind was observed during the bloom period. Meanwhile, the dominant wind was from the southwest. This is favorable for upwelling that transports CDOM-rich bottom water to the surface layer. When the northwesterly wind prevailed, CDOM-rich water from the upstream of the PR flew southward. The negative effect of tide on SPM can be associated with the fact that the outflow of CDOM-rich water from the inner PR during neap tide increases SPM while the inflow of oceanic water during spring tide decreases SPM. The convection of CDOM-rich water during ebb tide after upwelling was formed also contributed to the increase of SPM. The negative relationship between SPM and the accumulated rainfall observed at the Waglan Island station is related to the dilution of surface water by rainfall. However, SPM increases as the aggregated rain over the PR basin increases, which is caused by the increase of river runoff that carries high loads of CDOM to the PRE. Although the effect of circulation on SPM is not evaluated here, it is acknowledged that circulations in the PRE influence the PR plume. Thus circulations also have effects on SPM. Therefore, the diurnal behavior of SPM in the PRE was influenced by several factors including wind, tide, ocean circulation, and terrestrial runoff. This corroborates the previous finding that the PR plume was under the combined effects of those factors [2]. It is also noteworthy that wind and tide have stronger influence on SPM than other factors.
5. Conclusion
In this work, an empirical band ratio algorithm was developed for the retrieval of SPM from remotely sensed data. Linear regressions between model estimated and in situ measured SPM indicated the robustness of the proposed algorithm. The algorithm was also evaluated against published ones for turbid coastal waters and demonstrated the highest accuracy. By implementing the algorithm to HICO imagery, the spatial distribution of SPM in the PRE was characterized with high values on the western side while low values on the eastern side. Diurnal dynamics of SPM in the PRE was revealed by applying the algorithm to buoy-measured data. With MODIS/Aqua derived Rrs based on the SWIR atmospheric correction scheme as a benchmark, HICO-derived Rrs indicated great reliability. Analysis suggests that the combined effects of wind, tide, rainfall, and circulation regulated the variations of SPM in the PRE.
Remote sensing cannot only provide synoptic views of SPM in turbid coastal waters over large spatial and temporal scales but also assist in better interpretation of effects of anthropogenic activities on variations on SPM. With region-specific algorithms, SPM in turbid coastal waters can be approximated with high accuracy. Thus dynamics of SPM can be described in detail. Furthermore, satellite-measured SPM can be used as initial conditions for sediment transport simulation models to improve their accuracy. And the effects of terrestrial runoff on the marine environments can be well established. More efficient policies can be made to explore the marine resources in a sustainable way.
Funding
Startup funding from Sun Yat-sen University (52601106); National Natural Science Foundation of China (41776044; 41406205; 41776045; 41576030); Science and Technology Planning Project of Guangzhou City, China (201504010034); Open Project Program of the State Key Laboratory of Tropical Oceanography (LTOZZ1602).
Acknowledgments
We express our appreciations to other colleagues in the optics laboratory, who contributed to the design of the bio-optical buoy and supported the field campaign. We give our great thanks to anonymous reviewers for their thoughtful and careful suggestions. We are also indebted to NASA for providing HICO and MODIS data and the SeaDAS software.
References and links
1. S. Dai, S. Yang, and A. Cai, “Variation of sediment discharge of the Pearl River Basin from 1995-2005,” Acta Geogr. Sin. 62, 545–554 (2007).
2. L. A. Wong, J. C. Chen, H. Xue, L. X. Dong, J. L. Su, and G. Heinke, “A model study of the circulation in the Pearl River Estuary (PRE) and its adjacent coastal waters: 1. Simulations and comparison with observations,” J. Geophys. Res. Oceans 108(C5), 3156 (2003). [CrossRef]
3. Z. Y. Wu, Y. Saito, D. N. Zhao, J. Q. Zhou, Z. Y. Cao, S. J. Li, J. H. Shang, and Y. Y. Liang, “Impact of human activities on subaqueous topographic change in Lingding Bay of the Pearl River estuary, China, during 1955-2013,” Sci. Rep. 6(1), 37742 (2016). [CrossRef] [PubMed]
4. P. Davies, “Nutrient processes and chlorophyll in the estuaries and plume of the Gulf of Papua,” Cont. Shelf Res. 24(19), 2317–2341 (2004). [CrossRef]
5. J. E. Cloern, A. D. Jassby, T. S. Schraga, E. Nejad, and C. Martin, “Ecosystem variability along the estuarine salinity gradient: Examples from long-term study of San Francisco Bay,” Limnol. Oceanogr. 62(S1), S272–S291 (2017). [CrossRef]
6. B. S. Halpern, K. A. Selkoe, C. White, S. Albert, S. Aswani, and M. Lauer, “Marine protected areas and resilience to sedimentation in the Solomon Islands,” Coral Reefs 32(1), 61–69 (2013). [CrossRef]
7. M. Zhang, J. Tang, Q. Dong, Q. Song, and J. Ding, “Retrieval of total suspended matter concentration in the Yellow and East China Seas from MODIS imagery,” Remote Sens. Environ. 114(2), 392–403 (2010). [CrossRef]
8. S. Chen, L. Han, X. Chen, D. Li, L. Sun, and Y. Li, “Estimating wide range Total Suspended Solids concentrations from MODIS 250-m imageries: An improved method,” ISPRS J. Phogramm. 99, 58–69 (2015). [CrossRef]
9. J. Chen, E. D’Sa, T. Cui, and X. Zhang, “A semi-analytical total suspended sediment retrieval model in turbid coastal waters: A case study in Changjiang River Estuary,” Opt. Express 21(11), 13018–13031 (2013). [CrossRef] [PubMed]
10. 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]
11. R. Doerffer and J. Fischer, “Concentrations of chlorophyll, suspended matter, and gelbstoff in case II waters derived from satellite coastal zone color scanner data with inverse modeling methods,” J. Geophys. Res. Oceans 99(C4), 7457–7466 (1994). [CrossRef]
12. L. Feng, C. Hu, X. Chen, and Q. Song, “Influence of the Three Gorges Dam on total suspended matters in the Yangtze Estuary and its adjacent coastal waters: Observations from MODIS,” Remote Sens. Environ. 140, 779–788 (2014). [CrossRef]
13. K. Shi, Y. Zhang, G. Zhu, X. Liu, Y. Zhou, H. Xu, B. Qin, G. Liu, and Y. Li, “Long-term remote monitoring of total suspended matter concentration in Lake Taihu using 250m MODIS-Aqua data,” Remote Sens. Environ. 164, 43–56 (2015). [CrossRef]
14. M. Ondrusek, E. Stengel, C. S. Kinkade, R. L. Vogel, P. Keegstra, C. Hunter, and C. Kim, “The development of a new optical total suspended matter algorithm for the Chesapeake Bay,” Remote Sens. Environ. 119, 243–254 (2012). [CrossRef]
15. K. Shi, Y. Zhang, Y. Zhou, X. Liu, G. Zhu, B. Qin, and G. Gao, “Long-term MODIS observations of cyanobacterial dynamics in Lake Taihu: Responses to nutrient enrichment and meteorological factors,” Sci. Rep. 7, 40326 (2017). [CrossRef] [PubMed]
16. Z. Mao, J. Chen, D. Pan, B. Tao, and Q. Zhu, “A regional remote sensing algorithm for total suspended matter in the East China Sea,” Remote Sens. Environ. 124, 819–831 (2012). [CrossRef]
17. 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]
18. Z. Qiu, “A simple optical model to estimate suspended particulate matter in Yellow River Estuary,” Opt. Express 21(23), 27891–27904 (2013). [CrossRef] [PubMed]
19. Q. Xing, M. Lou, D. Yu, R. Meng, P. Shi, F. Braga, L. Zaggia, and L. Tosi, “Features of turbid waters from Hyperspectral Imager for the Coastal Ocean (HICO): Preliminary results at the Yellow River Delta and the Bohai Sea,” in 2012 4th Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS) (2012), paper 1–4. [CrossRef]
20. B. Nechad, K. G. Ruddick, and Y. Park, “Calibration and validation of a generic multisensor algorithm for mapping of total suspended matter in turbid waters,” Remote Sens. Environ. 114(4), 854–866 (2010). [CrossRef]
21. F. Liu, C. Chen, S. Tang, and D. Liu, “A piecewise algorithm for retrieval of suspended sediment concentration based on in situ spectra data by MERIS in Zhujiang River estuary (in Chinese),” J. Tropical Oceanogr. 28, 9–14 (2009).
22. D. Liu, C. Zhang, D. Fu, and C. Shen, “Hyperspectral data-based remote-sensing inversion model for suspended sediment in surface waters at the Pearl River Estuary (in Chinese),” Mark. Sci. 34, 77–80 (2010).
23. H. Xi and Y. Zhang, “Total suspended matter observation in the Pearl River estuary from in situ and MERIS data,” Environ. Monit. Assess. 177(1-4), 563–574 (2011). [CrossRef] [PubMed]
24. Q. Xing, M. Lou, C. Chen, and P. Shi, “Using in situ and Satellite Hyperspectral Data to Estimate the Surface Suspended Sediments Concentrations in the Pearl River Estuary,” IEEE J-STARS 6, 731–738 (2013).
25. H. Ye, C. Chen, S. Tang, L. Tian, Z. Sun, C. Yang, and F. Liu, “Remote sensing assessment of sediment variation in the Pearl River Estuary induced by Typhoon Vicente,” Aquat. Ecosyst. Health 17(3), 271–279 (2014). [CrossRef]
26. T. Li, “The Absolute Irradiant Calibration of the Spectrograph,” Ocean Tech. 20, 26–28 (2001).
27. D. Antoine, M. Chami, H. Claustre, F. D’Ortenzio, A. Morel, G. Bécu, B. Gentili, F. Louis, J. Ras, E. Roussier, A. J. Scott, D. Tailliez, S. B. Hooker, P. Guevel, J. F. Desté, C. Dempsey, and D. Adams, “BOUSSOLE: a joint CNRS-INSU, ESA, CNES and NASA Ocean Color Calibration And Validation Activity. NASA Technical memorandum N° 2006 - 214147,” (2006).
28. R. W. Austin, “The remote sensing of spectral radiance from below the ocean surface,” in Optical Aspects of Oceanography, N. G. Jerlov and E. S. Nielsen, eds. (Academic Press, 1974), pp. 317–344.
29. J. L. Mueller, A. Morel, R. Frouin, C. Davis, R. Arnone, K. Carder, Z. Lee, R. G. Steward, S. Hooker, C. D. Mobley, S. McLean, B. Holben, M. Miller, C. Pietras, K. D. Knobelspiesse, G. S. Fargion, J. Porter, and K. Vos, “Radiometric measurements and data analysis protocols,” (Goddard Space Flight Center, Greenbelt, MD, 2003).
30. M. Wang and W. Shi, “The NIR-SWIR combined atmospheric correction approach for MODIS ocean color data processing,” Opt. Express 15(24), 15722–15733 (2007). [CrossRef] [PubMed]
31. W. Shi and M. Wang, “Satellite views of the Bohai Sea, Yellow Sea, and East China Sea,” Prog. Oceanogr. 104, 30–45 (2012). [CrossRef]
32. 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]
33. S. W. Bailey and P. J. Werdell, “A multi-sensor approach for the on-orbit validation of ocean color satellite data products,” Remote Sens. Environ. 102(1-2), 12–23 (2006). [CrossRef]
34. I. Ogashawara, D. Mishra, S. Mishra, M. Curtarelli, and J. Stech, “A Performance Review of Reflectance Based Algorithms for Predicting Phycocyanin Concentrations in Inland Waters,” Remote Sens. 5(10), 4774–4798 (2013). [CrossRef]
35. G. Dall’Olmo and A. A. Gitelson, “Effect of bio-optical parameter variability on the remote estimation of chlorophyll-a concentration in turbid productive waters: experimental results,” Appl. Opt. 44(3), 412–422 (2005). [CrossRef] [PubMed]
36. T. Kutser, L. Metsamaa, E. Vahtmae, and R. Aps, “Operative monitoring of the extent of dredging plumes in coastal ecosystems using MODIS satellite imagery,” J. Coast. Res. 50, 180–184 (2007).
37. L. Sipelgas, U. Raudsepp, and T. Kõuts, “Operational monitoring of suspended matter distribution using MODIS images and numerical modelling,” Adv. Space Res. 38(10), 2182–2188 (2006). [CrossRef]
38. H. Zhao, Q. Chen, N. D. Walker, Q. Zheng, and H. L. MacIntyre, “A study of sediment transport in a shallow estuary using MODIS imagery and particle tracking simulation,” Int. J. Remote Sens. 32(21), 6653–6671 (2011). [CrossRef]
39. J. Zhao, W. Cao, Y. Yang, G. Wang, W. Zhou, and Z. Sun, “Measuring natural phytoplankton fluorescence and biomass: A case study of algal bloom in the Pearl River estuary,” Mar. Pollut. Bull. 56(10), 1795–1801 (2008). [CrossRef] [PubMed]
40. P. J. Harrison, K. Yin, J. H. W. Lee, J. Gan, and H. Liu, “Physical–biological coupling in the Pearl River Estuary,” Cont. Shelf Res. 28(12), 1405–1415 (2008). [CrossRef]
41. D. Liu, D. Fu, B. Xu, and C. Shen, “Estimation of total suspended matter in the Zhujiang (Pearl) River estuary from Hyperion imagery,” Chin. J. Oceanology Limnol. 30(1), 16–21 (2012). [CrossRef]
42. G. Zheng and P. M. DiGiacomo, “Uncertainties and applications of satellite-derived coastal water quality products,” Prog. Oceanogr. 159, 45–72 (2017). [CrossRef]
43. M. Wang, S. Son, and W. Shi, “Evaluation of MODIS SWIR and NIR-SWIR atmospheric correction algorithms using SeaBASS data,” Remote Sens. Environ. 113(3), 635–644 (2009). [CrossRef]
44. Y. Zhang, Y. Wang, Y. Wang, and H. Xi, “Investigating the Impacts of Landuse-landcover (LULC) Change in the Pearl River Delta Region on Water Quality in the Pearl River Estuary and Hong Kong’s Coast,” Remote Sens. 1(4), 1055–1064 (2009). [CrossRef]
