Study Area

The Yellow Sea (31°40′N–39°50′N, 119°10′E–126°50′E) is a marginal sea of the western Pacific Ocean, lying between mainland China to the west and north and the Korean Peninsula to the east, and linking the Bohai Sea in the northwest with the East China Sea in the south. The Yellow Sea is usually separated into the North Yellow Sea and the South Yellow Sea by a line running from Chengshantou on the Shandong Peninsula to Changshan. The South Yellow Sea is where most green tides have broken out in recent years [15, 21]. The study area is within the range of 32°N–38°N, 119°E–123°E in the South Yellow Sea (Fig. 1), where MABs have occurred for 17 years since 2007.

Remote Sensing Data and Processing

The HJ-2 satellite data used in this work were obtained by the CCD sensor with 16 m spatial resolution, including the five bands of blue, green, red, red edge, and near infrared (NIR). The GF-1 satellite data used were obtained by the WFV sensor with 16 m spatial resolution, including four bands of blue, green, red, and NIR. The spatial resolution, width, and revisit period of these two satellites are consistent, and the blue, green, red, and NIR band ranges are the same. The specific parameters are shown in Table 1.

The HJ-2 and GF-1 data used in this work were Level-1 product data, which are shown as digital number (DN) values. All images were converted into TOA radiance by radiometric calibration (Eq. (1)) to avoid the influence caused by different sensors. They were transformed into the WGS_1984_UTM_Zone_51N coordinate system.

Fig. 1.

Map of the study area. The true-color image is a composite of the China & Brazil Earth Resource Satellite (CBERS)-04 wide field imager’s (WFI) band 4 (red), band 3 (green), and band 2 (blue), acquired on 25 June, 2022. The green regions show the spatial distribution of Ulva. prolifera macroalgae bloom, and the red regions show the current area affected by macroalgae.

where L(λ) is the TOA radiance (w•m⁻² •sr⁻¹•μm⁻¹) and Gain and Bais are the gain and offset values, respectively, which can be obtained from the Header File. The Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes (FLAASH) atmospheric correction module, implemented via ENVI5.6 software (Exelis Visual Information Solutions, Inc., Boulder, CO, USA), was applied to process HJ-2 and GF-1 images to derive the reflectance values (R, unitless). Subsequently, all images were transformed into the WGS_1984_UTM_ Zone_51N coordinate system after projection.

Macroalgae Detection Method and MABs Area Statistics

Many macroalgae detection algorithms have been proposed based on the distinct spectral difference between natural seawater and macroalgae-covered seawater in the red and NIR bands. Among them, difference index algorithms, such as the Floating Algae Index (FAI), Difference Vegetation Index (DVI), and Virtual-Baseline Floating Macroalgae Height (VB-FAH), are relatively unaffected by sunlight and aerosol changes [26, 31, 34]. Because HJ-2 lacks a short-wave infrared band and the detection effects of DVI and VB-FAH were similar [24], the DVI index (Eq. (2)) was selected for detecting pixels containing macroalgae in this study. This index has been widely used and demonstrated high accuracy for macroalgae detection in numerous studies [22, 24, 35].

where R_NIR and R_Red are the reflectance at the NIR band and red band, respectively.

Given the significant variability in atmospheric turbidity, ocean background, and sun glint, a dynamic threshold of DVI was used to detect the macroalgae [20-22, 26]. DVI images were segmented into small windows, each measuring 100×100 pixels. A set of thresholds (dependent on the optical conditions of the water surface) was used to classify, on a window-by-window basis, which pixels contained macroalgae. Pixels with DVI values exceeding the applicable threshold were considered to contain macroalgae [20, 26]. In this study, these thresholds ranged from -0.02 to 0.02. The daily coverage area of MABs (CA, km²) was derived by multiplying the pixel size of the satellite image (PS, km²) by the total number of pixels (N) that were identified as macroalgae [30], as shown in Eq. (3).

Analysis Method for Macroalgae Detection Capability of HJ-2 Satellite Images

GF-1 imagery has been widely used in MAB monitoring and comparison analysis of coarse resolution macroalgae detection capability [31, 36]. We used a GF-1 image as a reference to evaluate HJ-2’s macroalgae detection capability through three indicators.

The first indicator was the average gradient of the DVI image (G̅ , defined by Eq. (4)), which reflects the image’s ability to express small, detailed changes. It can not only evaluate the image clarity but also reflect the contrast of small details and texture changes. The greater the average gradient is, the better the image clarity and detail expression are [37-38]. G̅ was compared using image pairs selected over the same macroalgae region with the same acquisition date.

where M and N are the numbers of rows and columns of the image, respectively, and Z (i, j) is the DVI value of row i and column j.

The second indicator was the macroalgae visibility on the DVI image (V, defined by Eq. (5)), consisting of the mean DVI value difference between a pixel containing macroalgae and the background (i.e., seawater). The larger the difference, the greater the macroalgae visibility, which is more conducive to macroalgae detection. V was compared with the same image pairs as G̅ .

where Mean_macroalge and Mean_background represent the average value of macroalgae and the seawater background on the DVI image, respectively.

The third indicator was the mean relative difference (MRD, defined by Eq. (6)) between the macroalgae patch area detected in the GF-1 image and the HJ-2 image. The smaller the MRD, the more consistent the macroalgae detecting capability of these two images. MRD was calculated using pairs of macroalgae with the same or similar distribution selected on the same acquisition date.

where N is the number of image pairs.

Four satellite images (listed in Table 2) were used to analyze the above three indicators, within a time interval of 90 minutes.

Analysis of the MABs’ Spatiotemporal Variation in 2022

HJ-2 images from 2022 with fewer clouds were selected to monitor MABs, with a total of 12 periods. The macroalgae were extracted using the method described in Materials and Methods Section and analyzed to determine their spatiotemporal variation while considering three characteristics.

The first characteristic was the variation in macroalgae daily coverage area (CA, detailed in Materials and Methods Section). The second characteristic was the variation of the area affected by MABs (AA). The AA acquisition method was to use the Aggregate Polygons Tool in ArcGIS software to aggregate macroalgae patches within a certain range and generate the polygons enveloping macroalgae patches. This polygon area was designated AA (Fig. 1). The third characteristic was the degree to which MABs were affecting the Yellow Sea, which was represented by the annual distribution density (ADD). ADD was obtained in two steps: (1) the distribution density of macroalgae (D_m) was calculated based on Eq. (7); (2) ADD was obtained by calculating the average D of all images based on Eq. (8). The larger the ADD value, the more serious the effect [21]. ADD values were divided into five equal intervals designated as L1 to L5 (low to high) based on their annual range [21].

where m is the mth window; A_m and A_s are the macroalgae coverage area and the total area of the region in the mth window, respectively. The window size was 4 km × 4 km, which is consistent with our previous study [21]. N_s is the number of windows in an image, and N_image is the number of selected remote sensing images.

Comparison of Macroalgae Detection Capability between HJ-2 CCD and GF-1 WFV

Fig. 2 shows that the HJ-2 CCD images and GF- 1WFV images from two semi-synchronous periods in the same region on 11 June 2022 and 18 July 2022 were enhanced by a linear 2% stretch. From the perspective of visual interpretation, the macroalgae information on HJ-2 images is more obvious, and the visual interpretation effect of HJ-2 is slightly better than that of GF-1.

To objectively evaluate the macroalgae detection capability of HJ-2 satellite images, 10 pairs of sub-images were selected from each of these two groups of semi-synchronous images according to differences in water turbidity. Two pairs were located near the coast of Jiangsu Province, where the water is turbid, as shown in T1 and T2 in Fig. 2a) and c), and three pairs were located in the middle of the Yellow Sea, where the water is clear, as shown in Y1, Y2, and Y3 in Fig. 2a) and c). Five pairs were located near the coast of Shandong Province, where the water is relatively clear, as shown in S1-S5 in Fig. 2b) and d). In these 10 pairs of sub-images, there were some regions with a greater clustering of macroalgae, in which the macroalgae were mostly distributed in strips and patches (Y1, Y2, Y3, S1, S2, and S3). There were also several regions with less macroalgae clustering; in these regions, the macroalgae mostly exhibited a scattered distribution (T1, T2, S4, and S5). The average gradient and macroalgae visibility of these sub-images of the DVI images were counted based on the methods described in Materials and Methods Section. The results showed that the average gradient of the HJ-2 DVI images was close to or greater than that of the GF-1 DVI images, regardless of whether the water was clear or not. The average mean of the gradient of all the sub-images of the HJ-2 DVI images was about 1.87 times that of the GF-1 DVI images (Fig. 3a), indicating that the image clarity and detail display of HJ-2 were better than that of GF-1. Additionally, the macroalgae visibility of HJ-2 DVI images was comparable to that of GF-1 DVI images (Fig. 3b), suggesting that the detection capability of HJ-2 was equivalent to that of GF-1.

Fig. 2.

a) and b) are false color images composed of GF-1 band 4 (NIR), band 3 (red), and band 2 (green), acquired on June 11 and July, 18 2022, respectively. c) and d) are false color images composed of HJ-2 band 5 (NIR), band 3 (red), and band 2 (green), acquired on June 11 and July 18, 2022, respectively.

Fig. 3.

Average gradient a) and macroalgae visibility b) of 10 sub-images.

Fig. 4.

Comparison of MAB monitoring results of HJ-2 and GF-1. a) and b) show the spatial distribution of the floating macroalgae monitored by remote sensing images and the locations of the similar patches that were selected, and c) shows a comparison of the coverage area of the selected patches.

Fourteen pairs of macroalgae with the same or similar distribution were selected to investigate consistency in the retrieved macroalgae coverage area. The results showed that the coverage area estimated by HJ-2 was consistent with that estimated by GF- 1. The MRD on 11 June and 18 July 2022 were 2 1.76% and 20.27%, respectively (Fig. 4). The MRD might be smaller if the time interval were smaller. This result shows that the macroalgae monitoring results derived from HJ-2 and GF-1 are close to each other.

Variation in the Area of MABs in the Yellow Sea in 2022

According to the monitoring results from HJ-2 images, the duration of MABs in the Yellow Sea in 2022 was 71 days. Fig. 5 shows the variation in area. The development trend of the CA and AA (Area Affected by MABs, defined in Materials and Methods Section) of MABs was similar to that of previous years, with four stages: emergence, development, outbreak, and dissipation [9]. The maximum values of CA and AA (3.50×10² km² and 1.92×10⁴ km², respectively) occurred on the same day (25 June 2022). In the development and outbreak phase (from May 27 to June 25), CA and AA increased by about 11.98 km²/day and 5.59×10² km²/day on average, respectively. In the dissipation phase (from June 25 to August 3), CA and AA decreased by about 8.61 km²/day and 4.57×10² km²/day on average, respectively. Even when CA was relatively small (less than 1.00×10² km²), the AA was higher than 5.00×10³ km², indicating that the influence of MABs was still serious in the early and late stages of the MABs.

Spatiotemporal Variation and Distribution Density Variation in the Yellow Sea in 2022

Fig. 6 shows the spatiotemporal distribution map of MABs in the Yellow Sea in 2022, reflecting the distribution and variation of MABs. Macroalgae were found near the Subei Shoal of Jiangsu Province on 23 May 2022. MABs were concentrated in the seawaters near the Subei Shoal until 31 May. After that, MABs began to move north. CA and AA increased to their maximum on 25 June (Fig. 5). During this period, MABs began to affect the offshore waters of the Shandong Peninsula. Then, MABs gradually decreased, and by 3 August, the CA was less than 15.00 km². At this point, the MABs were at the end of the dissipation phase. During the dissipation phase, MABs were concentrated in the coastal waters of the Shandong Peninsula. After 3 August, macroalgae were not detected in the remote sensing images, indicating that MABs in 2022 were coming to an end.

Fig. 7 illustrates the annual distribution density (ADD) of MABs in the Yellow Sea in 2022 during the whole phase (Fig. 7a)) and the dissipation phase (Fig. 7b)). The ADD of the coastal waters of Shandong Peninsula and the central waters of the Yellow Sea was significantly higher than that of other sea areas, indicating that the effect of MABs on these areas was higher than on other sea areas (Fig. 7a)). In particular, the ADD on the coastal waters of Shandong Peninsula was higher than in other sea areas during the MABs dissipation phase, indicating that the effect of MABs on the coastal waters of Shandong Peninsula was more serious (Fig. 7b)).

Fig. 5.

Variation in the area of MABs in 2022. a) The macroalgae daily coverage area (CA). b) The area affected by MABs (AA). The vertical axis label adopts scientific notation, such as 1.00E+04 representing 1.00×104.

Fig. 6.

Spatiotemporal variation of MABs in the Yellow Sea in 2022.

Fig. 7.

Annual distribution density (ADD) of MABs during all MAB stages a) and during the dissipation phase b).