Open Access
Add to BibSonomy
Abstract
Several algorithms have been proposed to detect floating macroalgae blooms in the global ocean. However, some of them are difficult or even impossible to routinely apply by non-experts because of performing a sophisticated atmospheric correction scheme or due to the mismatch in spectral bands from one sensor to another. Here, a generic, simple and effective method, referred to as the Floating Green Tide Index (FGTI), was proposed to detect floating green macroalgae blooms (GMB). The FGTI was defined as the difference between greenness and wetness features extracted from digital number (DN) observation through Tasseled Cap Transformation analysis, providing the advantage of bypassing the atmospheric correction procedure. Through cross-index and cross-sensor comparisons, the FGTI showed similar performance to the existing VB-FAH (Virtual-Baseline Floating macroAlgae Height) and FAI (Floating Algae Index) algorithms but proved more robust than the traditional NDVI (Normalized Difference Vegetation Index) in terms of response to perturbations by environmental conditions, viewing geometry, sun glint, and thin cloud contamination. Given the requirement for spectral bands in the current and planned satellite sensors, the FGTI design can easily be extended to any satellite sensor, and therefore provide an excellent data resource for studying GMB in any part of the global ocean.
© 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
Various types of macroalgae blooms have been reported in both literature and international news media during the past decades [1–3]. Because of their significant ecological and economic impacts, macroalgae blooms including harmful algal blooms (HABs), have become the focus of ecological research throughout the world [4–7]. In the Yellow Sea (YS), the world’s largest green macroalgae blooms (GMB) of Ulva prolifera (also called Green Tide; a type of HABs) on record was reported in the summer 2008 [8]. This event caused significant adverse impact on local environment and economy and even threatened the successful holding of the 2008 Beijing Olympics sailing games off Qingdao, China [9–11]. Subsequent studies have reported excessive amount of floating GMB every summer in this region since 2008 [12]. Such extensive and long-lasting macroalgae blooms devastate the coastal marine ecological systems (e.g., affecting the benthic algae growth due to blocked sunlight), and have a negative impact on maritime traffic, tourism, and other marine activities [13,14]. In addition, degrading algae both in coastal waters and washed onto the beaches cause serious economic and ecological issues [1]. For instance, they must be physically removed in a timely manner, thereby increasing economic burden to local management. Therefore, accurate and timely information on the spatial and temporal distribution of GMB plays an important role in understanding their ecology and impact on higher trophic level, and in making management decisions. Unlike in situ observations that are limited in both space and time, satellite remote sensing data can provide synoptic observations at high frequency, and have been widely used to monitor GMB from space.
Recently, various algorithms have been developed for remote detection of GMB using optical traits [15–17] or microwave backscattering features [18,19]. These algorithms rely on the assumption that the unique spectral traits of green macroalgae in the visible, infrared, and microwave domains are distinct from those of normal seawaters. Compared to microwave data, satellite optical data have been used frequently to study GMB in previous studies [20], including MERIS (Medium Resolution Imaging Spectrometer), Aqua and Terra/MODIS (Moderate Resolution Images Spectroradiometer), Landsat-5/TM (Thematic Mapper), Landsat-8/OLI (Operational Land Imager), and HJ-1/CCD (Charge-Coupled Device). More importantly, the availability of these satellite optical data at no cost further promotes their widespread use.
Taking into account the distinct spectral characteristics between macroalgae and water body, several algorithms have been proposed for effectively monitoring GMB using satellite optical data [21–23]. Many indices designed originally for land vegetation were used for the GMB detection in oceanic waters [10,24]. This is partly motivated by the underlying principle that all forms of vegetation have a generally similar reflectance spectra with a typical “red-edge” reflectance near 700 nm [9]. Among them, the concepts of NDVI (Normalized Difference Vegetation Index) [25] and EVI (Enhanced Vegetation Index) [26] are perhaps the most commonly used for GMB detection. However, the NDVI and EVI values of both floating macroalgae and surrounding waters are sensitive to changes in environmental and observing conditions [22,27]. Based on MODIS data, Shi and Wang [1] designed the NDAI (Normalized Difference Alga Index) to address the above limitation by analyzing the development and vanishing process of the 2008 GMB event in the YS. Hu [22] proposed a simple FAI (Floating Algae Index) based on the Rayleigh-corrected reflectance (Rrc) difference between near-infrared (NIR) band and a baseline formed by red and short-wave infrared (SWIR) bands. This index was successfully applied to several satellite sensors (e.g., MODIS, Landsat-5/TM, and Landsat-7/ETM +) for studying algae blooms in both oceans and wetlands [28,29]. However, the FAI approach cannot be extended to other satellite sensors that are not equipped with a SWIR band, such as the high-resolution HJ-1A/B and GF series. In the absence of SWIR band, Xing and Hu [30] introduced the VB-FAH (Virtual-Baseline Floating macroAlgae Height) index using a virtual baseline reflectance height technique. The VB-FAH was applied to HJ-CCD and Landsat-5/TM and −7/ETM + for studying the history and causes of GMB in the Yellow and East China Seas.
Although many spectral-based algorithms have been proposed, some of them are difficult if not even impossible to apply on specific satellite sensors due to lack of the required spectral bands. For example, the FAI index is not applicable in the HJ-1 and GF satellite data due to lack of a SWIR band, as mentioned above. On the other hand, some algorithms (e.g., NDVI and VB-FAH) are derived from surface reflectance information, and thus require application of an accurate and sophisticated atmospheric correction scheme to the raw satellite data, adding a complexity in the implementation process of the method. Atmospheric correction requires auxiliary information, including the scene center location, sensor type and altitude, flight date and time, geometry parameters, atmospheric models, and visibility that relied on atmospheric condition. In addition, there is no reliable atmospheric correction available for various sensors to detect the ocean green macroalgae [11,22]. Therefore, there is a need to develop a simple and effective method using satellite digital number (DN), with the expectation that the new method could be practically applied to sensors with different spectral characteristics without performing any atmospheric correction.
Tasseled cap transformation (TCT) approach is a powerful tool to capture vegetation feature, and has been widely used in vegetation studies [31–33]. For example, Dymond et al. [33] used the phenological differences in tasseled-cap features to improve deciduous forest classification based on multi-resource satellite DN images. Similarly, Cheng et al. [34] proposed a fixed-threshold approach to extract vegetation information from various IKONOS DN images using combined TCT analysis and an image fusion method. These studies imply that the TCT analysis may be also an effective tool for extracting image feature of floating green macroalgae from satellite DN, and therefore for detecting and quantifying GMB in the ocean.
In this study, we adopted the TCT analysis to extract and strengthen the macroalgae feature from satellite DN signal without performing any atmospheric correction; and then proposed a simple index (namely Floating Green Tide Index: FGTI) for mapping GMB using the extracted tasseled-cap features. Our proposed FGTI method can be easily implemented on various satellite sensors. Here, several examples of using the FGTI method to detect GMB in the Yellow Sea based on various satellite sensor data were presented.
2. Study area and material
2.1 Study area
The study area is located in the Yellow Sea, a marginal sea of the Pacific Ocean influenced by the East Asian monsoon regime [Fig. 1(a)]. It is bounded by approximately 32-38 °N and 119-125 °E, where green macroalgae blooms of Ulva prolifera have occurred every year in the past decade [29]. A sample of satellite “true-color” composite image at 03:11 GMT on 8 July 2015 [Fig. 1(b)] and a photo taken on 14 July 2016 [Fig. 1(c)] show Ulva prolifera blooms in the coastal waters off Shandong Province, China.
Fig. 1 Location of the study area showing the YS, and the boxes with different colors show the coverage of different satellite images used in this study, respectively (a). A sample of the satellite Red-Green-Blue “true-color” composite image recorded by GF-1/WFV (8 July 2015) acquired over the Qingdao coastal waters, showing floating green macroalgae bloom (b). A photo of the Ulva prolifera macroalgae mat was taken 14 July 2016 in the Qingdao coast (c).
2.2 Satellite data and processing
Data from four different satellite sensors were used in this study, including (1) the Wide Field View Multispectral Camera (WFV) onboard the Chinese GF-1 satellite, (2) the CCD onboard the Chinese HJ-1A/B satellite, (3) the Enhanced Thematic Mapper Plus (ETM +) onboard the Landsat-7 satellite from the United States, and (4) the Geostationary Ocean Color Imager (GOCI) onboard Communication, Ocean, and Meteorological Satellite (COMS) from South Korea. Each of the satellite sensors has its unique spatial and spectral resolution, and revisit frequency. The GF-1 satellite, which carries four WFV sensors, was launched by the China Centre for Resources Satellite Data and Application (CRESDA) on 26 April 2013. The spatial and temporal resolutions of GF-1/WFV are 16 m and 4 days, respectively, which facilitate spatiotemporal analysis for monitoring GMB. The HJ-1B satellite is a member of the Chinese HJ-1A/B satellite constellation launched by the CRESDA on 6 September 2008. The HJ-1 A/B CCD data provides ground surface spectral information with a high spatial resolution of 30 m and a short revisit period of 2 days. For Landsat-7/ETM +, its spatial resolution of 30 m (except thermal infrared band) is the same as that of the HJ-1A/B CCD, whereas its revisit period (16 days) is 8 times less frequent than HJ-1A/B CCD. Compared to the above-mentioned polar-orbiting satellites, the GOCI is the world’s first ocean color geostationary satellite sensor, providing near real-time monitoring of marine environments in northeast Asia. It records satellite images at hourly intervals up to 8 times a day from 0:15 to 7:15 (GMT); however, it has a coarse spatial resolution of 500 m. Additionally, the spectral resolution of these satellite sensors is summarized in Table 1. The GF-1/WFV and HJ-1A/B CCD have four wavebands, namely, blue, green, red, and near-infrared (NIR) bands, which are similar to those of the IKONOS sensor. Their wavelength settings are similar to the first four bands of Landsat-7/ETM +. The GOCI sensor has six visible bands and two NIR bands.
Table 1. The wavelength settings (μm) of the satellite sensors used in the current study. Here, B, G, R, NIR, and SWIR are blue, green, red, near infrared, and short-wave infrared bands, respectively. “-” means no band or no Visible-NIR-SWIR band. “*” means the GOCI wavebands used for the TCT analysis.
In this study, a total of five satellite images with low cloud cover were used for the development and validation of the proposed FGTI, which coverage and overpass time are shown in Fig. 1(a). The two GF-1/WFV images and one HJ-1B/CCD image were obtained from the CRESDA (http://www.cresda.com/cn/). The Landsat-7/ETM + and GOCI images were downloaded from the United Stated Geological Survey (USGS, https://earthexplorer.usgs.gov/) and the Korea Ocean Satellite Center (KOSC, http://kosc.kiost.ac.kr/), respectively. Note that the Landat-7/ETM + image had wedge-shaped gaps due to the failure of the scan-line corrector (SLC) of the Landsat-7/ETM + sensor since 2003. Thus, in this study, a local linear histogram-matching method [35] was used to fill in the missing-data gaps for further analysis. Among these satellite images, the GF-1/WFV image on 8 July 2015 was mainly used to elaborate the FGTI concept (see details in section 3.2).
The satellite images used in this study were geometrically corrected and clouds were masked (discussed later for details of cloud mask). To assess the performance of the FGTI method, three previously published algorithms were used, including the NDVI, VB-FAH algorithms, which were applied to the GF-1/WFV data, and the FAI, which was applied to the Landsat-7/ETM + data. These published algorithms were derived from sea surface reflectance (hereafter called Ref). Therefore, an atmospheric correction scheme was required here, even though our proposed FGTI method is only based on satellite DN signal and bypass the atmospheric correction step. The ENVITM FLAASH (Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes) atmospheric correction module was applied to the top-of-atmosphere reflectance measured by the satellite mentioned above to obtain Ref information. The FLAASH approach has been widely used for the atmospheric correction of multi-band or hyperspectral images over waters [15,36,37]. Xing and Hu [30] applied the FLAASH approach to HJ-CCD and Landsat-7/EMT + data to obtain Ref and thereby produce VB-FAH for detecting floating macroalgae blooms happened in the YS.
2.3 Approach to assess the accuracy of GMB detection
Unlike conventional practices to validate ocean color observations using in situ measurements such as chlorophyll-a concentration and sea-surface temperature, it is very difficult to directly validate the GMB detection from satellite data using field measurements. This is because the macroalgae is very patchy, and it is difficult to obtain near-timely field measurements of the GMB distribution. Alternatively, the relative spatial pattern and coverage area of GMB presented in satellite composite image should remain valid from one approach to another one. Thus, without concurrent field data, this study used two alternative ways to verify the satellite-detected GMB. In a qualitative way, floating macroalgae slicks were delineated using the FGTI approach, which was visually compared to its corresponding satellite RGB composite images (i.e., true-color). In essence, the visually-determined algae slicks can be used as the truth [29]. In a quantitative way, the FGTI performance was assessed by cross-index comparison with three previously published indices (NDVI, VB-FAH, and FAI). The three indices can be inferred from the Ref signal as follows:
where Rgreen, Rred, Rnir, and Rswir are the reflectance values in the green, red, near-infrared, and short-wave infrared bands, respectively, λi are the wavelength with subscript band name. For GF-1/WFV and HJ-1/CCD data, λgreen = 560 nm, λred = 660 nm, λnir = 830 nm; for Landsat-7/ETM + data, λgreen = 560 nm, λred = 662 nm, λnir = 835 nm, λswir = 1648 nm.Once the satellite images of the aforementioned indices were established, the macroalgae pixels were identified from macroalgae-free pixels through the use of an optimal threshold. This was achieved through several iterations for each satellite image. In practice, the identified macroalgae slicks were always visually inspected to assure the reliability of the results [29]. Thus, the index threshold was varied gradually until the identified GMB slicks (i.e., whose values of the index were above one threshold) agreed with both the index image and the RGB composite image when all images were linked together in the ENVI software. It should be noted that although the threshold selection could be optimized by using several sophistical methods [10,38], the goal of this study was to provide a simple and effective index for floating macroalgae detection. After identifying macroalgae pixels from macroalgae-free pixels, the cross-comparison of the identified GMB patterns was conducted using different algorithms. Meanwhile, the area covered by all macroalgae pixels was used as an evaluation indicator to quantitatively assess the performance of the FGTI method. In this study, the macroalgae coverage of individual satellite scene (Aa; in km2) was calculated as a simple summation of all macroalgae pixels [29]:
where NA is the total number of the classified macroalgae pixels in an individual satellite scene. SR is the spatial resolution of satellite sensor (for GF-1/WFV, SR = 0.016 km; for HJ-1/CCD and Landsat-7/ETM +, SR = 0.03 km).3. Development of FGTI from DN signal
3.1 Tasseled cap transformation
Tasseled cap transformation is a transform method that can extract and strengthen image information. The TCT approach was originally developed by Kauth and Thomas [39] by analyzing spectral features of wheat growth in fields, and was applied to the Landsat/MSS (multi-spectral scanner) sensor. This transform method has been widely adapted to modern sensors such as QuickBird, IKONOS, Landsat series, HJ-1, and MODIS [40–43]. The TCT approach is essentially a linear transformation of a variable X, which can be defined as:
where Y is the transform vector; X is the signal vector of spectral space; c is the orthogonal transform matrix; and a is an offset vector introduced to avoid negative values in the transformed data (i.e., Y).The green macroalgae floating on the water surface has a generally similar spectral property to land vegetation in the visible and NIR wavelengths with a typical red-edge signal (see Fig. 1(c) in Xing and Hu [30]). With this underlying principle, the floating macroalgae signal can be strengthened through TCT and differentiated from surrounding waters. In this study, the TCT coefficients from the IKONOS sensor were used for GF1-WFV, HJ-CCD, Landsat-7/ETM + images. The IKONOS TCT coefficients were developed by Horne [44], who analyzed a set of 195 IKONOS images for various environments and geometric conditions. The use of IKONOS TCT coefficients in the current study was motivated by the facts that (1) the IKONOS TCT coefficients are based on DN signal, and (2) the wavelength setting of IKONOS sensor is similar to those of the GF-1/WFV and HJ-1/CCD sensor and the first four bands of Landsat-7/ETM + (see Table 1 for details). Therefore, combining Eq. (6) and the IKONOS TCT coefficients, the TCT analysis used in this study can be expressed in the matrix form:
where DNi represents the DN value at i waveband. The four components are generated from the TCT analysis, namely, brightness (TCB), greenness (TCG), wetness (TCW), yellowness (TCY), respectively.Generally, the first three components (i.e., TCB, TCG, and TCW) explain over 90% of the spectral variance of individual scenes [34,45]. The first component (TCB) responds to the physical property that influences total reflectance, which is very similar to the panchromatic image. The second component (TCG) is responsive to the characteristic of green vegetation that is the combination of high absorption of chlorophyll in the visible band and high reflectance in the near-infrared band. In term of the TCG component, the floating macroalgae tend to have a strong signal in the green, while seawaters tend to have a weak signal. The third component (TCW) can be regarded as red minus blue, and responds to the amount of moisture being held by the water and vegetation [46]. The fourth component (TCY) has a low variance, usually as small as noise, with about 0.2% of the total variance. The TCY was rarely used in most studies due to the strong contamination by “noise” in this component [44,47]. As discussed above, this study focused on analyzing the image features of floating macroalgae and seawater using the first three TCT components (i.e., TCB, TCG, and TCW) and the overall objective is to build a robust method to distinguish GMBs, as described in the following section.
3.2 Development of the floating green tide index
In this study, the FGTI index was developed using a GF-1/WFV image acquired at 03:11 GMT on 8 July 2015 over the YS. In order to better illustrate the FGTI design, we chose three different satellite data subsets from the entire GF-1/WFV image [Fig. 2(a)]. These regions encompassed various environmental conditions, including turbid waters [Fig. 2(b)], relatively clear waters [Fig. 2(c)], and partially cloud-covered clear waters [Fig. 2(d)]. The TCT analysis [Eq. (6)] was applied to the three satellite-DN data sets to generate the TCT components. The TCT features of the floating macroalgae and water pixels were selected from the three satellite scenes (4525, 7555, 1009 macroalgae pixels and 5272, 8403, 1413 water pixels for Region 1-3, respectively) using ENVI Region of Interest (ROI) tool by visual inspection [Fig. 3]. Floating macroalgae showed higher TCG and TCB components than over seawaters in both turbid and clear waters, and had similar TCW to turbid waters and slightly different TCW to clear waters [Fig. 3]. However, the difference between TCG and TCW would still hold in turbid and clear waters and could maximize the contrast between macroalgae and seawater. Hence, the Floating Green Tide Index (FGTI) was defined as:
Fig. 2 GF-1/WFV “true-color” composite image over the western YS near Shandong coast at 03:11 GMT on 8 July 2015 (a). The three small regions (700 × 700 pixels) encompassed various environmental conditions, including the turbid waters (b), relatively clear waters (c), and partially cloud-covered clear waters (d).
Fig. 3 The TCT component features of floating macroalgae (circles) and seawaters (squares) from three different regions shown in Figs. 2(b)-2(d), respectively. Error bars represent one standard deviations of the means.
Note that the TCB signal was not used for the FGTI design here, although the TCB values of macroalgae were also obviously higher than those of seawater [Fig. 3]. The use of the TCB signal to derive the FGTI (hereafter called FGTI3c) masked the contrast between clouds and macroalgae pixels, especially for thin clouds (data not shown here) such that the TCG was favoured to avoid misclassification of cloud pixels as macroalgae pixels.
The FGTI method [Eq. (7)] was applied to satellite DN over three regions presented in Figs. 2(b)-2(d), respectively, to produce their corresponding FGTI images for different water types [Figs. 4(a)-4(c)]. For the different water types (turbid and relatively clear), the FGTI range of variation for floating macroalgae pixels (red color) was positive while it remains negative for water pixels (blue color). This contrast also held under thin clouds, as marked by the dashed circles in Fig. 4(c). Figure. 4(d) shows an example of histogram for demonstrating the FGTI performance. The FGTI had highly negative with a histogram peak between −100 and −50 for the turbid waters, and showed slightly negative with a peak values between −5 and 0 for the clear waters. However, for the macroalgae pixels, the FGTI values were mainly distributed between about 20 and 500. It was also evident in Fig. 4(e) showing the different FGTI features between macroalgae and water. Overall, these results indicated that the FGTI method derived from satellite DN signal through TCT analysis is able to detect and quantify floating green macroalgae from seawaters.
Fig. 4 The FGTI images (a-c) in three small regions presented in Figs. 2(b)-2(d). Histogram results in the FGTI distribution (d) for the macroalgae and water pixels analyzed in Fig. 3, and their mean and standard deviation of FGTI values (e).
It should be noted that the FGTI method was not able to detect the macroalgae slicks or patches under thick clouds [Fig. 4(c)], since the satellite optical sensor cannot observe the marine signal under opaque clouds [48]. We believe that this limitation occurs across all methods based on optical properties, including the NDVI, FAI, and VB-FAH approaches. Hence, it is desired for an approach to automatically discard heavy cloud pixels while retaining other pixels. In this study, the cloudy pixels were masked using a simple cloud-screening algorithm by determining a threshold of DN values at blue band (i.e., DNblue >C for heavy cloud pixels; C is the threshold value), considering that their DNblue values are high compared to cloud-free pixels (data not shown here). Note that the requirement for absolute accuracy of cloud mask is not critical here because the FGTI values of macroalgae are different from those of both thin- and thick-cloud pixels [Figs. 4(c) and 4(e)].
4. Results
4.1 Validation of FGTI for macroalgae detection
The quantitative performance and monitoring accuracy of the FGTI method was evaluated in two ways: by comparison (1) against the published VB-FAH index using the same GF-1/WFV image and (2) between the FGTI index from GF-1/WFV and the published FAI index from Landsat-7/ETM + over the same region. Note that the image pair (GF-1 and Landsat-7) was acquired on the same day about forty minutes apart.
4.1.1. Comparison between FGTI and VB-FAH using GF-1/WFV data
The GF-1/WFV data over three different regions of the same image were used for comparison between the FGTI and VB-FAH indices [Figs. 2(b)-2(d)]. The RGB “true-color” composite images [Fig. 5(a)] reveal qualitative information on floating macroalgae slicks. First, the FGTI method [Eqs. (6) and (7)] was applied to the three regions of interest [Fig. 5(b)]. Second, the FLAASH model was applied to retrieve the surface reflectance information (Ri in Eq. (2)) and third, the VB-FAH algorithm was applied [Fig. 5(c)]. The FGTI patterns were very similar to those of VB-FAH under all environmental conditions. Even under thin clouds, the floating macroalgae pixels were easily visualized and delineated for further quantification in the FGTI image (see dashed circles in Region 3 of Fig. 5(b)). Furthermore, the strong linear relationships between FGTI and VB-FAH indices were observed for both water and macroalgae pixels in the three regions [Fig. 6], with R2 values of 0.96, 1.0, and 0.91 (p < 0.001) for Region 1 (N = 490000), Region 2 (N = 490000), and Region 3 (N = 266580), respectively. Note that only pixels with FGTI > −50 in Region 3 were used to analyze the relationship between the two indices, for the purpose of artificially removing the cloud pixels. The strong linear relationships between the FGTI and VB-FAH indices [Fig. 6] attest that the FGTI is robust to changes in environmental conditions.
Fig. 5 The GF-1/WFV RGB “true-color” composite images in three small regions (a). Their FGTI results were derived from DN signal (b); and their VB-FAH results were derived from Ref signal (c). Comparison of mapping GMB detected by the FGTI and VB-FAH indices.
Following the approach of threshold determination described in section 2.3, we first set a threshold range (e.g., from 0 to 15 for FGTI in Fig. 5(b)) by observing the index pattern. Then, the threshold was varied gradually within the threshold range until the identified GMB slicks agreed with both the index image and the corresponding RGB composite image. To do this, the FGTI threshold presented in Fig. 5(b) was artificially set to 7.0. Meanwhile, it was clear from Fig. 4 that the FGTI values for the macroalgae-free pixels (i.e., water and cloud pixels) was lower than this threshold. Similarly, according to the aforementioned approach of threshold determination, the VB-FAH threshold for the same GF-1/WFV data was set to 0.02 in this study, which was very close to the threshold of 0.025 reported in Xing et al. [30]. All the pixels visually classified as microalgae in the RGB “true-color” composite images [Fig. 5(a)] were also detected by both the FGTI and VB-FAH approaches (green color in Fig. 5(d)), even under thin clouds (dashed circles in Region 3 of Fig. 5(d)). Both the FGTI and VB-FAH indices detected the spatial distribution of floating macroalgae patches [Fig. 5(d)] were in agreement with the visual identification [Fig. 5(a)]. The FGTI-derived GMB surface area in the three regions of the satellite scenes (Table 2) was very close to surface area computed using the VB-FAH index. There was a slight difference in GMB distribution detected by two indices (i.e., the macroalgae pixel was detected only by one single index), and occurred mainly in cloud-covered area and along the edge of macroalgae patches [Fig. 5(d)]. This is likely caused by the minor interference in the signal due to submerged vegetation and proximity to clouds. Overall, the results of Fig. 5 and Table 2 indicated that the performance of the proposed FGTI method had a high degree of consistency with the VB-FAH index, and that the FGTI could successfully detect GMB in different water types (turbid and relatively clear waters) and under thin clouds.
Table 2. Comparison between the GMB coverage [Fig. 5(d)] detected by the FGTI and VB-FAH indices.
4.1.2. Comparison between GF-1/WFV and Landsat-7/ETM +
To further assess the performance of the proposed FGTI index, we selected concurrent GF-1 and Landsat-7 images of the same region from over the coastal waters of the YS [Fig. 7(a) and 7(d)]. In this analysis, the 16-m resolution image from GF-1/WFV was re-projected on a 30 m grid same to Landsat-7/ETM +. Then, the FGTI method was applied to both images [Figs. 7(b) and 7(e)]. In addition, the Landsat-7/ETM + were further processed to derived surface reflectance and the FAI following the method of Hu [22] [Fig. 7(h)]. For both GF-1 and Landsat-7, the macroalgae slicks shown as bright pixels with high FGTI or FAI were in good agreement with each other, and matched the corresponding slicks in the “true-color” RGB composite images. Similar to the relationship between FGTI and VB-FAH [Fig. 6], a strong linear correlation was found between FGTI and FAI for the entire Landsat-7/ETM + image (including water and macroalgae pixels but except for the missing-data gaps) (R2 = 0.9, N = 57931; Fig. 8).
Fig. 7 The satellite “true-color” RGB composite images by GF-1/WFV (a) and Landsat-7/ETM + (d), respectively, both acquired on 8 July 2015. Their locations are shown by the red boxes in the middle-right figure. The GF-1 FGTI (b), Landsat-7 FGTI (e), and Landsat-7 FAI (h) images with floating macroalgae shown as bright pixels. The mapping GMB results were detected by the GF-1 FGTI (c), Landsat-7 FGTI (f), and Landsat-7 FAI (l), respectively.
Fig. 8 The relationship between FGTI and FAI shown in Figs. 7(e) and 7(h) over the Landsat-7/ETM + image. The black line is the linear-fitting line.
Following the threshold determination approach (see section 2.3), the FGTI threshold of 7.0 for GF-1/WFV, the FGTI threshold of 2.0 for Landsat-7/ETM +, and the FAI threshold of 0.02 for Landsat-7/ETM + were selected. For each satellite image, the macroalgae pixels were identified and their coverage area were calculated (right panel; Fig. 7). For the GF-1/WFV data, the FGTI-identified GMB pattern [Fig. 7(c)] matched the “true-color” RGB composite image in Fig. 7(a), and showed good agreement with the FAI-identified GMB pattern using Landsat-7 data [Fig. 7(l)]. Furthermore, the macroalgae surface area was close between the FGTI-detected GMB using GF-1 data (15156 pixels; Aa = 13.64 km2) and the FAI-detected GMB using Landsat-7 data (14383 pixels; Aa = 12.95 km2). The small difference in GMB area between the two images is likely caused by the data lost for Landsat-7/ETM + missing-data gaps and the cross-sensor mismatch between changing location of macroalgae slicks due to forty minutes apart for the two different sensors. The GMB detection using the Landsat-7 FGTI was in good agreement with that using the Landsat-7 FAI [Fig. 7(l)]. For the same Landsat-7/ETM + data, the FGTI-identified macroalgae coverage area (14392 pixels; Aa = 12.95 km2) was the same that using the FAI index. These results indicate that the proposed FGTI index had similar good performance to the FAI, and that the FGTI was consistent between GF-1 and Landsat-7 sensors. Clearly, our results suggest that the FGTI approach based on satellite DN can efficiently detect and classify floating green macroalgae from seawaters, and that it performed similarly to published indices derived from surface reflectance such as VB-FAH and FAI.
4.2 Sensitivity analysis of FGTI
The above analysis revealed that the FGTI method works effectively at detecting GMB, and it shows a strong linear relationship with the VB-FAH and FAI approach. We hypothesize that the FGTI remains stable against changing water types and observing conditions for both water and macroalgae pixels in a similar way the VB-FAH and FAI indices. The sensitivity of the FGTI to environmental conditions was investigated and its performance was compared to the NDVI, VB-FAH, and FAI approaches in the following sections.
4.2.1 Viewing geometry
The sensitivity analysis of FGTI to viewing geometry was carried out using GF-1/WFV data acquired on July 8th 2015 during a GMB event in the YS [Fig. 9(a)]. In this analysis, the comparison with NDVI and VB-FAH indices was also performed. The GF-1/WFV data were thus processed to Ref, VB-FAH, and NDVI. Two hatching lines with wide span were selected in the zonal and meridional directions (yellow lines in Fig. 9(a)), including macroalgae, turbid water, and clear water. The tight linear relationships were observed between FGTI and VB-FAH both along the zonal (R2 = 0.97, 9797 pixels) and meridional lines (R2 = 0.98, 7525 pixels) [Fig. 9(b)]. Furthermore, Figs. 9(c) and 9(d) show that both FGTI and VB-FAH of macroalgae-free pixels along the zonal and meridional lines had low variation, suggesting that these two indices were robust to changes in viewing geometry. The spikes were caused by macroalgae pixels. In contract, the corresponding NDVI along the same transect lines showed much higher variation [Figs. 9(e) and 9(f)]. Such a result indicated that the proposed FGTI was equivalent to VB-FAH and better than NDVI in terms of tolerance to changes in the observing geometry.
Fig. 9 The GF-1/WFV “true-color” RGB composite image, and two yellow lines are the artificial transect lines in the zonal and meridional directions (a). The relationships between FGTI and VB-FAH and NDVI for all pixels along the transect lines (b). Variations of FGTI and VB-FAH for all pixels along the zonal line (c) and meridional line (d). Variations of FGTI and NDVI for all pixels along the zonal (e) and meridional lines (f). The spikes in (c-f) show the presence of floating macroalgae.
4.2.2 Sun glint
As reported in Xing and Hu [30], sun glint might be one of major factors causing uncertainties in classifying macroalgae pixels when using the NDVI, VB-FAH, and FAI methods. The sun glint has significant impact on the magnitude of surface reflectance [49], especially for the satellite images with high resolution [50], such as GF-1 (16 m) and HJ-1 and Landsat-7 (30 m) used in this study. Thus, to investigate the sensitivity of the FGTI to variation in sun glint, the GF-1/WFV image acquired on 13 June 2016, containing macroalgae and high dynamic range sun glint, was used in this study [Fig. 10(a)]. We also analyzed the comparison of FGTI with NDVI and VB-FAH in term of their sensitivity to the variation in sun glint. Similar to the above analysis, we derived the FGTI result for satellite DN, and also obtained the NDVI and VB-FAH results from satellite surface reflectance.
Fig. 10 GF-1/WFV Pseudo-RGB composite image acquired on 13 June 2016, showing macroalgae in red color (a). A testing area with high dynamic range of glint and macroalgae (b). Surface reflectance of macroalgae and seawater with high sun glint (SW #1) and low sun glint (SW #2) (c), corresponding to the annotations in (b), respectively. The FGTI image for a small region (d). The cloud pixels were masked as shown in white color.
Figure 10(b) shows the satellite composite image for a small area containing both high and low sun glint. To better show the effect of sun glint on the sea surface Ref, we selected the seawater pixels with high glint (SW #1) and low glint (SW #2) from this small area. As shown in Fig. 10(c), the difference of Ref values between SW #1 and SW #2 (i.e., SW #1 – SW #2) was assumed to be caused by sun glint. Here, the perturbations possibly caused by the water constituents, viewing geometry, and aerosol were negligible in the small area. Figure 10(d) shows that the FGTI contract between macroalgae and macroalgae-free pixels was obvious over rough sea surface with variation in glint, which was similar to the result of Fig. 4 over sea surface with less variation in glint. Such a result indirectly implies that the FGTI appeared stable for both macroalgae and water even under significant sun glint.
We selected a transect line with narrow span including 5501 pixels (blue line in Fig. 10(a)), to further analyze the sensitivity of the FGTI to sun glint in a quantitative manner. The pixels along the transect contained both high and low sun glint. The use of a narrow-span transect was motivated by the following consideration: in the short distance transect, there was a high dynamic range of glint intensity while the effects of viewing angles, aerosols, and water compositions remain negligible. As shown in Fig. 11(a), for the macroalgae-free pixels along the transect line, the FGTI values showed low variation similar to the VB-FAH. Furthermore, the FGTI and VB-FAH showed a strong linear relationship (R2 = 0.84, N = 5501), especially for the macroalgae pixels with high FGTI values [Fig. 11(b)]. In contrast, the NDVI had much higher variation than FGTI along the same transect line [Fig. 11(c)], and showed a monotonical nonlinear relationship with FGTI [Fig. 11(d)]. Clearly, the comparison of the FGTI with other indices derived from Ref signal (i.e., VB-FAH and NDVI) showed comparable results in terms of tolerance to variation in sun glint.
Fig. 11 Variations of FGTI and VB-FAH (a) and NDVI (c) along the transect line shown in Fig. 10(a). The relationship between FGTI and VB-FAH (b) and NDVI (d) along the same transect line. The spikes in (a and c) show the presence of floating macroalgae.
4.3 Extension of FGTI to other satellite sensors
The FGTI method developed in this study requires satellite DN observations at four bands, namely blue, green, red, and NIR bands. To the best of our knowledge, such a requirement of these four bands is available in many existing satellite optical sensors, such as the HJ-1 A/B CCD, Landsat-7/ETM +, and GOCI. As shown in Table 1, the wavelength settings in the visible and NIR bands of HJ-1B/CCD and Landsat-7/ETM + were similar to those of the IKONOS sensor, which TCT coefficients were used in this study. We expected that the FGTI could be extended to these sensors. Here we discussed the extension of the FGTI approach to three other sensors, i.e., HJ-1B/CCD, Landsat-7/ETM +, and GOCI. It is noteworthy that GOCI has different wavelength specifications than the IKONOS sensor, namely narrow bands (20 nm for GOCI versus 60-90 nm for IKONOS), the objective of this analysis is to assess the application of the FGTI method to all type of sensors, including multispectral ocean colour satellite. The results presented in Figs. 7(d)-7(f) indicated that the FGTI method can be successfully applied to the Landsat-7/ETM + data. In this analysis, we chose two regions of interest [Fig. 12(b)] from individual satellite scenes (Fig. 12(a); their locations and overpass times shown in Fig. 1(a)). These regions for HJ-1/CCD and GOCI sensors contain 500 × 500 and 120 × 120 pixels, respectively.
Fig. 12 The satellite images of HJ-1B/CCD (top) and GOCI (bottom) (a), whose locations and overpass times were shown in Fig. 1(a). Two regions with floating macroalgae (b) were selected from (a). The FGTI pattern (c) and the detected GMB results (d) from individual satellite sensors.
The FGTI method [Eqs. (6) and (7)] were applied to the satellite DN of HJ-1/CCD and GOCI [Fig. 12(c)]. The classification of macroalgae in each images was generated by using the image-independent threshold for two regions, as shown in Fig. 12(d). Here, 7.0 was used as the threshold for HJ-1/CCD and 0.7*107 for GOCI data. The FGTI contrast between macroalgae and water pixels were obvious for both the HJ-1B/CCD and GOCI data, with high FGTI values for macroalgae pixels [Fig. 12(c)], and were similar to the GF-1 FGTI images [Fig. 5]. In other words, the FGTI features of macroalgae are clearly revealed in the HJ-1B/CCD and GOCI images. Visual comparison between Figs. 12(b) and 12(d) indicated a good agreement of the GMB classification with the slick morphology in the corresponding RGB composite images. The results of Figs. 7 and 12 suggest that the proposed FGTI method could be successfully extended to the GF-WFV, HJ-1 A/B - CCD, Landsat-7/ETM +, and GOCI data.
5. Discussion
5.1 Rationality and limitation of FGTI
Green macroalgae blooms have already become an ecological problem in the global coastal zones because of their significant socio-economic impacts. Effective and timely monitoring is one of the key tasks of coping with this ecological disaster [19]. The most important advantage of satellite remote sensing data is the ability to provide information on the location and extent of GMB at synoptic scale [30,51]. Many methods have been proposed to detect GMB using various satellite sensor. However, coarse-resolution satellite sensors (e.g., the best MODIS with 250 m; GOCI with 500 m) are unable to resolve small slicks with a surface area significantly smaller than the pixel footprint. This can also lead to mixed-pixel problems; in other words, the limit of spatial resolution results in single pixel covering multiple ground objects [11]. In contrast, the satellite data with fine-resolutions (e.g., GF-1/WFV with 16 m, HJ-1/CCD and Landsat-7/ETM + with 30 m) could detect small macroalgae slicks that would not be seen in the coarse-resolution image. As discussed above, however, the technical difficulties in monitoring GMB using existing methods are mainly due to the lack of required spectral bands in some sensors (e.g., SWIR in Landsat) or the need to perform atmospheric correction scheme. To circumvent these limitations, a simple FGTI method was developed based on satellite DN and TCT analysis [Eqs. (6) and (7)]. More importantly, the FGTI approach did not require any atmospheric correction scheme.
The TCT analysis, an image transform method, can compress spectra data into a few bands that are associated with physical scene characteristics [45]. In this study, the output of the TCT analysis consists of four components: brightness, greenness, wetness, and yellowness. Among them, the greenness component is defined in the direction of vegetation signature on an axis that is orthogonal to the brightness vector, and is strongly correlates with variation in the vigor of green vegetation [52]. The wetness component can reveal the information of the water body as the background in satellite image. More importantly, the TCT image features of macroalgae in the greenness and wetness were obviously different with those of surrounding waters [Fig. 3]. Therefore, these two components were used to design the simple FGTI proposed here [Eq. (7)] for the GMB detection.
It was found that the significant contrast of the FGTI values between macroalgae and waters was observed under both turbid and relatively clear water conditions, even under thin clouds [Fig. 4]. This is mainly because the greenness component contains and enhances most vegetation information, while it reduces other types of background information such as water, suspended sediment, clouds, and sun glint. Thus, our new FGTI method showed high performance on classifying and quantifying green macroalgae under various environmental conditions [Figs. 5 and 7]. The FGTI index was also able to effectively extract the GMB information under high and low sun glint [Fig. 10(d)]. In addition, similar to the VB-FAH and FAI indices, the FGTI index showed more tolerance to perturbation induced by viewing geometry [Fig. 9], sun glint [Fig. 11], and thin clouds [Figs. 4 and 5]. This is evident by the tight linear relationships between FGTI and FAI [Fig. 8] and VB-FAH (Fig. 6 for various environmental conditions, Fig. 9(b) for the zonal and meridional lines, and Fig. 11 for high dynamic range of sun glint). As shown in Fig. 4(c), pixels with thin cloud had lower FGTI values that the macroalgae-containing pixels, and thus were easily differentiated (see Region 3 in Fig. 5). Macroalgae pixels under thin clouds kept a strong remote sensing features in the FGTI image, which was similar to those of macroalgae pixels with cloud-free sky [Fig. 4(c)]. Unfortunately, there is no existing cloud-masking algorithm that identifies all cloud pixels while keeping all valid macroalgae pixels. Thus, we chose a simple cloud masking algorithm to mask the pixels contaminated by opaque cloud. It is also noted that the use of the simple cloud-masking algorithm here is motivated by the fact that even without the cloud mask step, the FGTI index alone can rule out thin clouds and detect floating macroalgae. In other words, for both the thin and opaque cloud pixels that are missed in the cloud mask step, the FGTI index can classify these pixels to macroalgae-free pixels in the subsequent steps, due to the significant contrast of the FGTI values between macroalgae and cloud pixels [Figs. 4(c) and 4(e)].
Considering that the optical signature of floating macroalgae closely resembles that of land vegetation, it appears logical to use TCT coefficients designed were designed for land application. The wavebands specifications of the satellite sensors used in this study were generally similar to those of IKONOS sensor in the visible-near infrared wavelengths such that we used the IKONOS TCT coefficients with confidence [Eq. (6)]. In the near future, we expect that macroalgae-specific TCT coefficients will be developed to improve the detection of floating macroalgae in the ocean. There are many methods which are used for deriving TCT coefficients, such as principal components analysis (PCA) [53] and Gram-Schmidt orthogonalization (GSO) method [54]. For example, Chen et al. [43] developed the reflectance-based TCT coefficients in HJ-1/CCD data using the combines of PCA and GSO approaches.
5.2 Continuity considerations and other applications
The FGTI method proposed in our study requires four bands in the blue, green, red, and NIR bands. Fortunately, these four bands as input of TCT analysis can always be met for many satellite optical sensors. Thus, the FGTI method may be extendable to these sensors, for example, Landsat series, GF1-4, MODIS, and VIIRS (Visible Infrared Imaging Radiometer Suite). Indeed, the FGTI method showed high performance on both GF-1/WFV, Landsat-7/ETM +, HJ-1 A/B CCD, and GOCI sensors [Figs. 5, 7, and 12]. However, for the GOCI data, the FGTI value ranges of both macroalgae and water pixels were very different to those of other satellite data, suggesting the sensor-specific thresholds are needed when applying the FGTI method to different satellite sensors. Several geostationary satellite sensor (e.g., GOCI and GF-4/VNIR) can provide high temporal resolution (hourly intervals) and medium resolution (≤ 500 m per pixel) data [21,55], which facilitates real-time observation of the dynamics and drift of GMB.
In addition to Ulva prolifera, green tides are also caused by excessive growth of various species of green benthic algae such as Enteromorpha, Chaetomorpha, and Cladophara [56]. Meanwhile, previous studies have reported that green tides occurred in many ocean regions of the world, including Europe, South America, Australia, and Japan [57–59]. For the TCT analysis, the second tasseled-cap component (i.e., greenness) mainly reveals the characteristic of green vegetation, and is typically used as an index of photosynthetically-active vegetation [45,60]. Therefore, the FGTI design through TCT analysis are expected to detect green algae blooms of other species than Ulva prolifera. However, the performance of the FGTI approach to monitor various types of algae blooms in other ocean regions remain to be demonstrated. If successful, combination of the FGTI approach with public access of global satellite data will provide an excellent data source to study floating algae in any part of the global oceans.
6. Conclusion
A simple and effective method, namely the Floating Green Tide index (FGTI), was developed in this study to detect floating green macroalgae blooms in the Yellow Sea. The FGTI was defined as the difference between greenness and wetness generated from TCT analysis based on satellite DN observation without performing any atmospheric correction, which made this simple FGTI straightforward and easy to implement. Comparison with other commonly used indices derived from surface reflectance showed that the FGTI can successfully detect GMB under various water types and thin clouds, and that similar to the VB-FAH and FAI, FGTI appeared robust to perturbations in viewing geometry and sun glint. Additionally, the FGTI index was successfully extended to HJ-1B/CCD, Landsat-7/ETM +, and GOCI sensors. Because of the availability of the four spectral bands (i.e., blue, green, red, and near infrared) in other existing and planned satellite sensors with high and medium-resolutions, the FGTI design may be extended to more satellite sensors such as GF series, Landsat series, QuickBird, VIIRS, and MODIS. This could provide the efficient and important record of floating macroalgae for exploring its causes and consequences, even for timely routine detection of floating macroalgae blooms in any part of the global oceans.
Funding
The National Key Research and Development Program of China (2016YFC1400901); the National Natural Science Foundation of China (41876203, 41576172, and 41506200); the Jiangsu Provincial Programs for Marine Science and Technology Innovation (HY2017-5); the Provincial Natural Science Foundation of Jiangsu in China (BK20161532); the Research and Innovation Project for College Graduates of Jiangsu Province (KYLX16_0952).
Acknowledgments
Sincere thanks are given to the USGS for providing the Landsat data, the KOSC for providing the GOCI data, and the CRESDA for providing the GF and HJ data. The authors are grateful for the helpful comments from two anonymous reviewers.
References
1. W. Shi and M. Wang, “Green macroalgae blooms in the Yellow Sea during the spring and summer of 2008,” J. Geophys. Res. 114(C12), C12010 (2009). [CrossRef]
2. M. Wang and C. Hu, “Mapping and quantifying Sargassum distribution and coverage in the Central West Atlantic using MODIS observations,” Remote Sens. Environ. 183, 350–367 (2016). [CrossRef]
3. B. E. Lapointe, J. M. Burkholder, and K. L. Van Alstyne, “Harmful Macroalgal Blooms in a Changing World: Causes, Impacts, and Management,” Harmful Algal Blooms: A Compendium Desk Reference, (John Wiley & Sons, 2018), pp: 515–560.
4. D. A. Lyons, C. Arvanitidis, A. J. Blight, E. Chatzinikolaou, T. Guy-Haim, J. Kotta, H. Orav-Kotta, A. M. Queirós, G. Rilov, P. J. Somerfield, and T. P. Crowe, “Macroalgal blooms alter community structure and primary productivity in marine ecosystems,” Glob. Change Biol. 20(9), 2712–2724 (2014). [CrossRef] [PubMed]
5. J. Gower, C. Hu, G. Borstad, and S. King, “Ocean color satellites show extensive lines of floating Sargassum in the Gulf of Mexico,” IEEE Trans. Geosci. Remote Sens. 44(12), 3619–3625 (2006). [CrossRef]
6. M. Hiraoka, M. Ohno, S. Kawaguchi, and G. Yoshida, “Crossing test among floating Ulva thalli forminggreen tide’in Japan,” Hydrobiologia 512(1–3), 239–245 (2004). [CrossRef]
7. Q. Xing, L. Wu, L. Tian, T. Cui, L. Li, F. Kong, X. Gao, and M. Wu, “Remote sensing of early-stage green tide in the Yellow Sea for floating-macroalgae collecting campaign,” Mar. Pollut. Bull. 133, 150–156 (2018). [CrossRef] [PubMed]
8. Q. Xing, C. Hu, D. Tang, L. Tian, S. Tang, X. Wang, M. Lou, and X. Gao, “World’s Largest Macroalgal Blooms Altered Phytoplankton Biomass in Summer in the Yellow Sea: Satellite Observations,” Remote Sens. 7(9), 12297–12313 (2015). [CrossRef]
9. C. Hu and M. X. He, “Origin and offshore extent of floating algae in Olympic sailing area,” Eos (Wash. D.C.) 89(33), 302–303 (2008). [CrossRef]
10. D. Liu, J. K. Keesing, Q. Xing, and P. Shi, “World’s largest macroalgal bloom caused by expansion of seaweed aquaculture in China,” Mar. Pollut. Bull. 58(6), 888–895 (2009). [CrossRef] [PubMed]
11. Y. Xiao, J. Zhang, and T. Cui, “High-precision extraction of nearshore green tides using satellite remote sensing data of the Yellow Sea, China,” Int. J. Remote Sens. 38(6), 1626–1641 (2017). [CrossRef]
12. X. Liu, Z. Wang, and X. Zhang, “A review of the green tides in the Yellow Sea, China,” Mar. Environ. Res. 119, 189–196 (2016). [CrossRef] [PubMed]
13. I. Valiela, J. McClelland, J. Hauxwell, P. J. Behr, D. Hersh, and K. Foreman, “Macroalgal blooms in shallow estuaries: controls and ecophysiological and ecosystem consequences,” Limnol. Oceanogr. 42(5part2), 1105–1118 (1997). [CrossRef]
14. R. H. Charlier, P. Morand, C. W. Finkl, and A. Thys, “Green tides on the Brittany coasts,” in US/EU Baltic International Symposium, (IEEE, 2006), pp: 1–13.
15. H. Zhang, D. Sun, J. Li, Z. Qiu, S. Wang, and Y. He, “Remote Sensing Algorithm for Detecting Green Tide in China Coastal Waters Based on GF1-WFV and HJ-CCD Data,” Acta Optica Sinica 36(6), 0601004 (2016). [CrossRef]
16. Q. Zhang, Q. Liu, Z. Kang, R. Yu, T. Yan, and M. Zhou, “Development of a fluorescence in situ hybridization (FISH) method for rapid detection of Ulva prolifera,” Estuar. Coast. Shelf Sci. 163, 103–111 (2015). [CrossRef]
17. IOCCG, “Phytoplankton functional types from Space,” in Reports of International Ocean-Colour Coordinating Group, S. Sathyendranath, ed. (IOCCG, 2014), pp. 1–156.
18. H. Shen, W. Perrie, Q. Liu, and Y. He, “Detection of macroalgae blooms by complex SAR imagery,” Mar. Pollut. Bull. 78(1-2), 190–195 (2014). [CrossRef] [PubMed]
19. T. Cui, J. Zhang, L. Sun, Y. Jia, W. Zhao, Z. Wang, and J. Meng, “Satellite monitoring of massive green macroalgae bloom (GMB): imaging ability comparison of multi-source data and drifting velocity estimation,” Int. J. Remote Sens. 33(17), 5513–5527 (2012). [CrossRef]
20. Y. Qiu and J. Lu, “Advances in the monitoring of Enteromorpha prolifera using remote sensing,” Acta Ecol. Sin. 35(15), 4977–4985 (2015).
21. Y. B. Son, J.-E. Min, and J.-H. Ryu, “Detecting massive green algae (Ulva prolifera) blooms in the Yellow Sea and East China Sea using Geostationary Ocean Color Imager (GOCI) data,” Ocean Sci. J. 47(3), 359–375 (2012). [CrossRef]
22. C. Hu, “A novel ocean color index to detect floating algae in the global oceans,” Remote Sens. Environ. 113(10), 2118–2129 (2009). [CrossRef]
23. Z. Qiu, Z. Li, M. Bilal, S. Wang, D. Sun, and Y. Chen, “Automatic method to monitor floating macroalgae blooms based on multilayer perceptron: case study of Yellow Sea using GOCI images,” Opt. Express 26(21), 26810–26829 (2018). [CrossRef] [PubMed]
24. Q. Xu, H. Zhang, and Y. Cheng, “Multi-sensor monitoring of Ulva prolifera blooms in the Yellow Sea using different methods,” Front. Earth Sci. 10(2), 378–388 (2016). [CrossRef]
25. J. W. Rouse Jr, R. Haas, J. Schell, and D. Deering, “Monitoring vegetation systems in the Great Plains with ERTS,” 309–317 (1974).
26. A. Huete, C. Justice, and W. Van Leeuwen, “MODIS vegetation index (MOD13),” Algorithm theoretical basis document 3, 213 (1999).
27. A. P. Trishchenko, J. Cihlar, and Z. Li, “Effects of spectral response function on surface reflectance and NDVI measured with moderate resolution satellite sensors,” Remote Sens. Environ. 81(1), 1–18 (2002). [CrossRef]
28. K. Yu and C. Hu, “Changes in vegetative coverage of the Hongze Lake national wetland nature reserve: a decade-long assessment using MODIS medium-resolution data,” J. Appl. Remote Sens. 7(1), 073589 (2013). [CrossRef]
29. L. Qi, C. Hu, Q. Xing, and S. Shang, “Long-term trend of Ulva prolifera blooms in the western Yellow Sea,” Harmful Algae 58, 35–44 (2016). [CrossRef] [PubMed]
30. Q. Xing and C. Hu, “Mapping macroalgal blooms in the Yellow Sea and East China Sea using HJ-1 and Landsat data: Application of a virtual baseline reflectance height technique,” Remote Sens. Environ. 178, 113–126 (2016). [CrossRef]
31. A. M. West, P. H. Evangelista, C. S. Jarnevich, S. Kumar, A. Swallow, M. W. Luizza, and S. M. Chignell, “Using multi-date satellite imagery to monitor invasive grass species distribution in post-wildfire landscapes: An iterative, adaptable approach that employs open-source data and software,” Int. J. Appl. Earth Obs. Geoinf. 59, 135–146 (2017). [CrossRef]
32. S. P. Healey, W. B. Cohen, Y. Zhiqiang, and O. N. Krankina, “Comparison of Tasseled Cap-based Landsat data structures for use in forest disturbance detection,” Remote Sens. Environ. 97(3), 301–310 (2005). [CrossRef]
33. C. C. Dymond, D. J. Mladenoff, and V. C. Radeloff, “Phenological differences in Tasseled Cap indices improve deciduous forest classification,” Remote Sens. Environ. 80(3), 460–472 (2002). [CrossRef]
34. W. C. Cheng, J. C. Chang, C. P. Chang, Y. Su, and T. M. Tu, “A Fixed-Threshold Approach to Generate High-Resolution Vegetation Maps for IKONOS Imagery,” Sensors (Basel) 8(7), 4308–4317 (2008). [CrossRef] [PubMed]
35. USGS, “Phase 2 gap-fill algorithm: SLC-off gap-filled products gap-fill algorithm methodology,” https://landsat.usgs.gov/sites/default/files/documents/L7SLCGapFilledMethod.pdf, Available online at (accessed 25 June 2018) (2004).
36. B. Gao, M. J. Montes, C. O. Davis, and A. F. Goetz, “Atmospheric correction algorithms for hyperspectral remote sensing data of land and ocean,” Remote Sens. Environ. 113, S17–S24 (2009). [CrossRef]
37. T. Kutser, “The possibility of using the Landsat image archive for monitoring long time trends in coloured dissolved organic matter concentration in lake waters,” Remote Sens. Environ. 123, 334–338 (2012). [CrossRef]
38. R. A. Garcia, P. Fearns, J. K. Keesing, and D. Liu, “Quantification of floating macroalgae blooms using the scaled algae index,” J. Geophys. Res. Oceans 118(1), 26–42 (2013). [CrossRef]
39. R. J. Kauth and G. Thomas, “The tasselled cap–a graphic description of the spectral-temporal development of agricultural crops as seen by Landsat,” in LARS Symposia, 159 (1976).
40. M. Lu, E. Hamunyela, J. Verbesselt, and E. Pebesma, “Dimension Reduction of Multi-Spectral Satellite Image Time Series to Improve Deforestation Monitoring,” Remote Sens. 9(10), 1025 (2017). [CrossRef]
41. C. Huang, B. Wylie, L. Yang, C. Homer, and G. Zylstra, “Derivation of a tasselled cap transformation based on Landsat 7 at-satellite reflectance,” Int. J. Remote Sens. 23(8), 1741–1748 (2002). [CrossRef]
42. M. H. A. Baig, L. Zhang, T. Shuai, and Q. Tong, “Derivation of a tasselled cap transformation based on Landsat 8 at-satellite reflectance,” Remote Sens. Lett. 5(5), 423–431 (2014). [CrossRef]
43. C. Chen, P. Tang, and Z. Bian, “Tasseled cap transformation for HJ-1A/B charge coupled device images,” J. Appl. Remote Sens. 6(1), 63575 (2012). [CrossRef]
44. J. H. Horne, “A tasseled cap transformation for IKONOS images,” in ASPRS 2003 Annual conference proceedings, (2003)
45. E. P. Crist and R. C. Cicone, “A physically-based transformation of Thematic Mapper data—The TM Tasseled Cap,” IEEE Transactions on Geoscience and Remote sensing (3), 256–263 (1984). [CrossRef]
46. L. Qingsheng, L. Gaohuan, H. Chong, L. Suhong, and Z. Jun, “A tasseled cap transformation for Landsat 8 OLI TOA reflectance images,” in 2014 IEEE International, 541–544 (2014).
47. R. D. Jackson, “Spectral indices in n-space,” Remote Sens. Environ. 13(5), 409–421 (1983). [CrossRef]
48. M. Moran, A. Vidal, D. Troufleau, J. Qi, T. Clarke, P. Pinter Jr., T. Mitchell, Y. Inoue, and C. Neale, “Combining multifrequency microwave and optical data for crop management,” Remote Sens. Environ. 61(1), 96–109 (1997). [CrossRef]
49. T. Kutser, E. Vahtmäe, B. Paavel, and T. Kauer, “Removing glint effects from field radiometry data measured in optically complex coastal and inland waters,” Remote Sens. Environ. 133, 85–89 (2013). [CrossRef]
50. S. Kay, J. D. Hedley, and S. Lavender, “Sun glint correction of high and low spatial resolution images of aquatic scenes: a review of methods for visible and near-infrared wavelengths,” Remote Sens. 1(4), 697–730 (2009). [CrossRef]
51. H. Zhang, S. Wang, Z. Qiu, D. Sun, J. Ishizaka, S. Sun, and Y. He, “Phytoplankton size class in the East China Sea derived from MODIS satellite data,” Biogeosciences 15(13), 4271–4289 (2018). [CrossRef]
52. P. M. Mather and M. Koch, Computer processing of remotely-sensed images: an introduction (John Wiley & Sons, 2011), Chap. 6.
53. E. P. Crist and R. C. Cicone, “Application of the tasseled cap concept to simulated thematic mapper data,” Ann Arbor 1001, 48107 (1984).
54. L. D. Yarbrough, G. Easson, and J. S. Kuszmaul, “QuickBird 2 tasseled cap transform coefficients: a comparison of derivation methods,” Pecora 16 Global Priorities in Land Remote Sensing, 23–27 (2005).
55. Y. Yuan, Z. Qiu, D. Sun, S. Wang, and X. Yue, “Daytime sea fog retrieval based on GOCI data: a case study over the Yellow Sea,” Opt. Express 24(2), 787–801 (2016). [CrossRef] [PubMed]
56. R. Fletcher, “The occurrence of “green tides”—a review,” in Marine benthic vegetation (Springer, 1996), pp. 7–43.
57. J. Blomster, S. Bäck, D. P. Fewer, M. Kiirikki, A. Lehvo, C. A. Maggs, and M. J. Stanhope, “Novel morphology in Enteromorpha (Ulvophyceae) forming green tides,” Am. J. Bot. 89(11), 1756–1763 (2002). [CrossRef] [PubMed]
58. M. Merceron, V. Antoine, I. Auby, and P. Morand, “In situ growth potential of the subtidal part of green tide forming Ulva spp. stocks,” Sci. Total Environ. 384(1-3), 293–305 (2007). [CrossRef] [PubMed]
59. T. Yabe, Y. Ishii, Y. Amano, T. Koga, S. Hayashi, S. Nohara, and H. Tatsumoto, “Green tide formed by free-floating Ulva spp. at Yatsu tidal flat, Japan,” Limnology 10(3), 239–245 (2009). [CrossRef]
60. S. Jin and S. A. Sader, “Comparison of time series tasseled cap wetness and the normalized difference moisture index in detecting forest disturbances,” Remote Sens. Environ. 94(3), 364–372 (2005). [CrossRef]
