Identifying the spatio-temporal variations of Ulva prolifera disasters in all life cycle

NDVI

normalized difference vegetation index

FAI

floating algae index

SAI

scaled algae index

VB-FAH

virtual baseline floating macroalgae height

EVI

enhanced vegetation index

ARVI

atmospherically resistant vegetation index

RVI

ratio vegetation index

DVI

difference vegetation index

SYS

Southern Yellow Sea

U. prolifera

Ulva prolifera

NDVIM

NDVI-mean

CA

coverage area

SST

sea surface temperature

SWF

sea wind field

PCC

Pearson correlation coefficient

AA

affected area

ATAA

average sea surface temperature of the affected area

ATSYS

average sea surface temperature of the Southern Yellow Sea

Since 2007, the Southern Yellow Sea (SYS) of China has witnessed millions of tons of biomass and Ulva prolifera (U. prolifera) disasters spread across thousands of square kilometers every year from May to July. By 2019, the disaster coverage area (CA) had exceeded 20,000 km², becoming the most severe U. prolifera disaster in the world (Hu et al. 2017; Zhang et al. 2017). Particularly, in 2008, the U. prolifera disaster unprecedentedly affected the Olympic sailing competition scheduled to be conducted in the sea area of Qingdao. In addition, the annual disaster of U. prolifera caused substantial economic losses to aquaculture, coastal tourism, maritime transportation, maritime activities, and other related industries. Accordingly, the disaster mitigation cost (e.g., clean-up and transportation) and aquaculture losses exceeded 2 billion only in 2008 (Ye et al. 2011; Wang et al. 2013) and reached 3.5 billion dollars by 2015 (Hu et al. 2017), respectively. The disasters have caused widespread ecological and environmental impacts, e.g., on phytoplankton biomass, water clarity, and zooplankton community in coastal waters (Tiyasha et al. 2020; Deng et al. 2021), which is a new ecological disaster along the China coast (Liu et al. 2013; Wang et al. 2016; Hu et al. 2017). Thus, timely detection and tracking of U. prolifera are important to prevent and manage marine environmental pollution (Hu et al. 2018).

Discovering and identifying possible U. prolifera disasters in the ocean is the primary task in monitoring their growth. The remote sensing technology has become an essential method to study U. prolifera disasters owing to its large-scale, fast, and dynamic observation advantages over the field observation method (Ke et al. 2015; Chao et al. 2020). A series of indices has been developed and applied to identify U. prolifera disasters, such as the floating algae index (FAI) (Hu 2009), normalized difference vegetation index (NDVI) (Cui et al. 2012), scaled algae index (SAI), and virtual baseline floating macroalgae height (VB-FAH) (Xing & Hu 2016). The FAI and VB-FAH are baseline subtraction methods, are less sensitive to changes in environmental and observation conditions (aerosol type and thickness, solar/viewing geometry, and sunglint), and can ‘see’ through thin clouds. The SAI is an improved image-based semiautomatic algorithm, based on the local ocean index value of a given pixel in an NDVI or FAI image, and scaled to generate a scaled image, to improve the accuracy. The NDVI is a vegetation index commonly used to determine the vegetation growth state and vegetation coverage and to eliminate some radiation errors. It is widely used because of its simple calculation process and good stability. Wang et al. (2014) compared the accuracy of five different vegetation indices (NDVI, enhanced vegetation index, atmospherically resistant vegetation index, ratio vegetation index, and difference vegetation index) based on MODIS data and concluded that the NDVI was most accurate and stable. Therefore, the NDVI is widely used as an indicator to express the spatio-temporal distribution of U. prolifera in previous researches. Yet, they rarely focused on accurately evaluating the correlation between the NDVI characteristics of U. prolifera and its distribution. Exploring whether there is a time lag or change regularity between them can help provide a scientific basis for predicting U. prolifera disasters.

U. prolifera disasters can be characterized by a wide distribution area, large coverage, and long duration. The key to controlling such disasters is to monitor the spatio-temporal distribution characteristics and drift trajectory accurately. Existing studies have made significant achievements in terms of understanding the origin of U. prolifera disaster (Xing et al. 2018), its spatio-temporal distribution (Qi et al. 2016; Yang et al. 2017), drift monitoring (Zhang et al. 2018a, 2018b; Chen et al. 2020), biomass estimation (Hu et al. 2017; Xiao et al. 2017), and environmental factors (Feng et al. 2012; Zhang et al. 2020a). Among the representative studies, Gower et al. (2006) used 300 m resolution data from MERIS to monitor a large area of Sargasso seaweed in the Gulf of Mexico, laying a foundation for the monitoring of U. prolifera disasters based on remote sensing data. Xing et al. (2011) monitored U. prolifera using multisource remote sensing data, found that the sudden occurrence of a large-scale green tide was observed in 2007, and the increasing trend of such disasters since this year could be attributed to the seaweed aquaculture in a specific mode at specific locations. So, they proposed the controllability of green tide and the key for its monitoring. Qiao et al. (2011) and Fangli et al. (2011) studied the drift path of U. prolifera under the combined effect of the wind field and surface current field using a quasi-operational wave-current-circulation coupled numerical model for offshore regions of China. However, these studies focused on describing the overall distribution characteristics of U. prolifera. The corresponding analysis of its spatio-temporal changes was general, which could not help determine the periodic characteristics of U. prolifera in all life cycles and the drift trend in the different growth stages. Moreover, there is an evident difference in the ecological impact of U. prolifera in the different growth stages and the corresponding disaster mitigation approaches, but these were ignored in previous studies. Therefore, comparing the regional distribution characteristics of U. prolifera at different growth stages is essential for the formulation of targeted governance measures.

Influenced by multiple factors such as the temperature, geographical conditions, aquaculture, and economic development (Wu et al. 2018; Wang et al. 2020), the formation mechanism of U. prolifera has multidimensional and comprehensive characteristics. The sea surface temperature (SST) is proven to be an essential factor influencing the growth of U. prolifera (Zhang et al. 2020b); the sea wind field (SWF) is another important environmental factor driving its drift (Sun et al. 2020; Zhang et al. 2020b). However, few studies have analyzed the effects of SST and SWF on U. prolifera in the different growth stages. Hence, exploring the influence of SST and SWF in the different stages on U. prolifera disasters is an urgent problem to be solved.

This study solves three problems. First, we determine whether there is a time lag or change regularity between the NDVI-mean (NDVIM) and the CA of U. prolifera disaster. Second, the spatio-temporal characteristics during the life cycle and the variations in the influencing factors in the different growth stages of U. prolifera were analyzed. Finally, the issue of poor continuity of the available optical remote sensing data for U. prolifera monitoring was resolved, and the growth curve of U. prolifera was fitted. MODIS data (2015–2019) corresponding to the SYS were selected as the monitoring data source for U. prolifera (Section ‘Remote sensing data’). The correlation between the NDVIM and CA was obtained (Section ‘Correlation between NDVIM and CA’). Subsequently, the life cycle of U. prolifera was divided on the basis of the CA curve (Section ‘Division of life cycle of U. prolifera’), and its spatio-temporal distribution characteristics and drift trajectory in the different stages were analyzed (Section ‘Spatio-temporal distribution and drift trajectory of U. prolifera in different stages’). Finally, the differences in the effects of the SST and SWF on U. prolifera in the different stages were discussed to provide a phased management plan and theoretical reference for preventing U. prolifera disasters (Section ‘Discussion’).

The SYS is an area with a high occurrence of U. prolifera, which is connected to the Bohai Sea in the north and the South China Sea in the south. The SYS is a typical semi-closed sea area in the Western Pacific and is characterized by a cold and dry climate in winters and a warm and humid climate in summers. The water temperature varies between 13 and 24 °C, and the salinity of the seawater is in the range of 33–34 g/kg, which provides good living conditions for the growth of U. prolifera (Hu 2009). The study area SYS (119°–123°E, 32°–36°N) is selected to consider the continuous occurrence of U. prolifera for many years, including the coastal areas of Jiangsu and Shandong Provinces (Figure 1).

MODIS data have high spectral and temporal resolutions, with advantages in extracting information regarding U. prolifera disasters and monitoring their spatio-temporal characteristics, as confirmed by existing research (Wang et al. 2016, 2018). MODIS data have three resolutions, namely 250, 500, and 1,000 m, among which the 250 m products cover the red and near-infrared bands.

The period with the highest occurrence of U. prolifera in the SYS is from May to August every year based on the existing studies (Qiu & Lu 2015; Liu et al. 2016). Therefore, we selected EOS MODIS 1B (Terra/Aqua) remote sensing data products from May to August for the SYS (2015–2019) as the data source, including MOD02QKM and MOD02HKM, which have resolutions of 250 and 500 m, respectively. The former is the main data source for U. prolifera extraction, and the latter was used as auxiliary data for visual interpretation. 2015–2019 was chosen, because it was the most serious year of U. prolifera disaster in the past 10 years, and the overall harmfulness change of U. prolifera disaster first decreased and then increased, which has more research value and significance (An 2020; Zhang 2020). Table 1 presents the basic band information of MODIS, such as the wavelength range, central wavelength, and corresponding band resolution. MODIS data come from the NASA data sharing website (https://ladsweb.modaps.eosdis.nasa.gov/search (downloaded on 15 June 2020)). Cloudless data were downloaded, and the time range included the cycle of U. prolifera from occurrence, outbreak, to final extinction. The SWF data are QuickSCAT wind field data from remote sensing systems (http://www.remss.com), and the SST data are from the SST of NOAA Earth System Research Laboratory (http://www.esrl.noaa.gov/psd).

Figure 2 shows the flowchart methodology employed for U. prolifera monitoring. The research contents of this study can be divided into four parts: (A) remote sensing data correction, (B) data pre-processing (calibration, atmospheric correction, and clipping), (C) U. prolifera extraction (land detection and elimination, cloud detection and elimination, NDVI calculation, threshold segmentation, and visual interpretation), and (D) spatio-temporal pattern and feature analysis (CA, drift trajectory, and average NDVI calculation). Thus, our study comprised four parts: downloading data/coordinate transformation, data pre-processing, U. prolifera extraction, and drift law analysis. We used the dynamic threshold method to extract U. prolifera. For brevity, this method is not discussed herein, as it has been thoroughly described by Cao et al. (2019).

The extracted image by threshold segmentation was compared with the false-color composite image to verify the accuracy of the extraction method. Due to the impact of islands, the thin clouds, tidal flats, and other ground features were first removed to avoid misjudgment.

The ‘barycenter’ is a physical concept that refers to the point of action of the resultant force generated by gravity at various positions in an object. In recent years, it has been gradually applied to study the space economy and population (He et al. 2019). The concept of barycenter in physics has been used to study the drift path of U. prolifera (Wu et al. 2013; Hu et al. 2017). U. prolifera is considered a set of geometric blocks with a uniform mass, and its barycenter is the point of the drift stress. The drift trajectory processing based on the ArcGIS can be divided into six steps (Figure 3): (A) converting the plane attribute, (B) extracting the gravity center and area, (C) converting the geographical coordinates into plane coordinates, (D) calculating the average gravity center, (E) converting the plane coordinates into geographical coordinates, and (F) calculating the drift trajectory.

In practical applications, M_i represents the attribute values having different meanings (regional GDP, population, arable land, and energy). In this article, M_i represents the extracted U. prolifera. We plotted the draft trajectory of U. prolifera by calculating its barycenter coordinates in different periods.

Based on a given level of significance α and degrees of freedom d_f = n−2 and through the t-distribution to find the critical value ⁠, if ⁠, the null hypothesis H₀ is rejected, indicating a significant linear relationship.

We selected U. prolifera corresponding to the NDVIM from 2015 to 2019 as one set of variables (the number of samples is 38) and the CA of U. prolifera from 2015 to 2019 as another set of variables (the number of samples is 38). The daily correlation coefficient was obtained based on the Pearson coefficient of the NDVIM and CA. Similarly, a linear interpolation was performed to calculate the correlation coefficients between NDVIM and CA at 5, 10, and 15 days.

The NDVIM and CA curves of U. prolifera were drawn. Both curves have similar shapes and changing trends, including a first increasing and then decreasing trend (Figure 4). In 2015, 2016, and 2019, the NDVI was maximum at the end of May and the beginning of June, and the CA of U. prolifera was maximum in the middle of June (Figure 4). In 2017 and 2018, the NDVI was maximum in mid-June, and the CA was maximum in late June. The NDVI peak appeared earlier than the maximum CA peak (Figure 4).

Moreover, the PCCs of the NDVIM and CA lagging 0, 5, 10, and 15 days all passed the 0.01 significance level test (Table 2). Table 2 presents the correlation coefficients between the average NDVI of U. prolifera and the CA in the same period and earlier period in the study area from 2015 to 2019. The correlation coefficients between the NDVI and CA passed the significance level of 0.01.

In particular, the NDVIM has the strongest response to CA when the time interval is 5 days, indicating that the CA has an evident lay effect on NDVIM, which means that when the overall change trends in both A and B are similar, the trend in A is earlier than that in B, that is, B has a lag effect relative to A. Therefore, the NDVIM of the same period was not sufficiently accurate to reflect the change trend in CA. Accordingly, the NDVIM could predict the trend in CA after 5 days. In addition, the NDVIM still had a greater impact on CA after 10 days, and the effect was the least after 15 days.

The life cycle curve of U. prolifera was obtained by polynomial fitting (R² = 0.7702) based on the normalization method to the data (Figure 5). We analyzed the time and CA of U. prolifera after normalization, and a scatter diagram was fitted. The growth of U. prolifera was divided into five different stages by selecting feature points. The time and CA before the disaster were set to 0, and at the end of the disaster, these values were 1 and 0, respectively. A theoretical hypothesis is made by combining statistical and literature data because the starting and ending points of U. prolifera cannot be obtained from remote sensing. Based on the five characteristic points (‘one extreme point and four change rate characteristic points’) in all life cycle curves, the one extreme point (0.441) was the time of maximum CA of U. prolifera, and the four change rate characteristic points were the time points with change rates of 1 and −1 (0.053, 0.382, 0.499, and 0.963). The change rates gradually increased from 0 to 1 (corresponding to the emergence stage), first increased, and then decreased from 1 to −1 (development stage), decreased from 1 to −1 (outbreak stage), first decreased and then increased from −1 to 1 (decline stage), and finally increased from −1 to 0 (extinction stage), respectively.

The earliest time of U. prolifera emergence was from the end of April to early May, with a low CA and affected area (AA) in terms of the time. U. prolifera entered a growth period from late May to early June. It increased exponentially and reached the outbreak stage in mid- and late June after the growth period. The first 10 days of July was the decline period, and its CA and AA decreased. In mid- and late July, it began to die out and gradually reached the shore. The life cycle was approximately 90 days.

The growth of U. prolifera conformed to the characteristics of the entire life cycle in 2015, 2016, 2018, and 2019. In 2017, although there were ‘two growth and two decay’ phenomena, the growth and decay rates in the first period were relatively low, and the second period was more in line with the growth characteristics of ‘rapid growth, rapid extinction.’

The occurrence of U. prolifera was sporadic near the Subei Shoal, covering small CA and AA, presenting a strip shape, and generally lasting for 4–5 days (Figure 6). The spatial effects of U. prolifera in the different periods can be divided into five stages, denoted by A, B, C, D, and E, which represent the emergence, development, outbreak, decline, and extinction stages. The emergence stage in 2017 was spatially distributed in the southeast of Yancheng (123°–124°E, 32°–33°N), slightly different from that observed in the other years.

The development stage occurred in the eastern part of Jiangsu and Shandong coastline (west of the central part of the Yellow Sea), which lasted for approximately 25–29 days. In the development stage, U. prolifera gradually drifted to the north and gathered to form large-scale patches in the growth process. The outbreak lasted for approximately 9 days, mainly in the eastern part of Jiangsu and Shandong coastline (110–123°E, 33.5–37°N). Compared with the development stage, the CA in the outbreak stage was the largest (averaging at 1,724 km²), approximately 200 times that in the emergence stage. At this stage, the distance between the patches narrowed, the spatial distribution was more concentrated, and the harm to marine ecology was the most serious.

In the decline stage, the patches of U. prolifera mainly occurred near the coastline of Shandong Province, and the western part was close to the coastline. The longest duration was approximately 35–40 days. The extinction stage lasted for approximately 3–4 days. Except for a few scattered regions in the sea, the extinction mainly occurred near Qingdao. In the first 10 days of August, satellite remote sensing data showed that large areas of U. prolifera could not be monitored.

The spatio-temporal distribution of U. prolifera not only exhibits phase characteristics, but more importantly also constantly drifts the growth process. The drift trajectory was obtained by extracting the barycenter of U. prolifera in the different stages of the year, which shows the position of its center of gravity in the different stages (Figure 7). The barycenter of U. prolifera emerged in the southeast of Lianyungang (120.88°E, 34.82°N) in 2017 and the northeast of Yancheng (123.32°E, 32.55°N) in 2019. The location moved to the northeast, and the barycenter in the development stage was relatively concentrated in the northeast of Lianyungang (121.00–121.50°E, 34.60–34.80°N), from the development stage to the outbreak stage, in 2015, 2018, and 2019. The barycenter moved to the northwest, which was northeast (121.42°E, 34.32°N) in 2016 and southeast (122.83°E, 33.51°N) of Yancheng in 2017.

In the outbreak stage, the barycenter shifted to the northwest, and it was in the east and north of Lianyungang (121.16–121.62°E, 34.60–35.88°N), and consequently, the decline phase drifted to the northwest. The extinction stage in 2017 and 2018 was in the coastal area of Qingdao. Thus, the overall drift direction of U. prolifera was northwest.

From 2015 to 2019, the main direction is north in terms of the shift direction of the center of gravity of U. prolifera. The occurrence stages in 2015, 2016, 2018, and 2019 were near the northern Jiangsu shoal, slightly different from that seen in 2017, in the southeast of Yancheng (120.88°E, 34.82°N).

The development stage was in the middle of the South Yellow Sea (east of Lianyungang), the outbreak stage was in the east of Rizhao, the decline stage was in the offshore waters of Shandong (south of Qingdao and northeast of Rizhao), and the extinction stage was in the offshore waters of Qingdao. Among them, the migration track was longest in 2017, the development stage was southeast relative to that observed in the other years, and there was no evident location difference between the outbreak, decline, and extinction stages and other years.

During U. prolifera disasters, their CA and AA were largest in the outbreak period. Thus, this stage has the greatest impact on the coastal landscape, economy, and marine ecological environment (Wu et al. 2013). The results showed significant spatial differences in the NDVIM of U. prolifera in the different growth stages. When the NDVIM reached the maximum value, it was the beginning of the outbreak period that lasted for approximately 9–10 days. In addition, the normal distribution curve between the CA of the U. prolifera patches and the NDVIM exhibited an evident lag effect. Moreover, the correlation reached the strongest at the 5-day lag period. This indicates that the change trend in the NDVIM could be used to predict the change trend in CA of U. prolifera in 5 days. Thus, the NDVIM was used as an indicator to monitor the outbreak period of U. prolifera, and the outbreak period of U. prolifera could be predicted based on the time difference between the peak points on the NDVIM and CA distribution curves. This is helpful for decision-makers to take disaster prevention and control measures in advance.

The NDVI is an important index to monitor the growth of vegetation, and the vegetation coverage is a standard index to measure the volume of vegetation. Generally, the community growth period is the growth rate greater than the decline rate and decline period is the growth rate less than the decline rate. When the growth rate is equal to the decline rate, the community reaches the maximum. That is, the NDVIM reaches the maximum first, and U. prolifera covers the largest area. There is a certain time difference between the two peaks. This shows that the response of the CA of U. prolifera to the growth has a certain time-delay effect.

Unlike previous studies (Hu 2009; Chen et al. 2020), this study realized the quantitative division of the life cycle of U. prolifera based on the five characteristic points in its life cycle curve. We obtained the time nodes of each stage of U. prolifera. Accordingly, the growth of U. prolifera could be divided into five stages: emergence, development, outbreak, decline, and extinction. The data assimilation was realized using the normalization method in time and CA, and the problem of incomplete monitoring data of U. prolifera in a particular year was solved.

The staged distribution of U. prolifera not only manifested differences temporally, but also dynamic changes in its spatial location. Most studies have shown that U. prolifera originated from Subei Shoal (Xing et al. 2018; Zhang et al. 2018a); however, we concluded that this area may be not the only source. In 2017, U. prolifera emerged in a different location from that seen in other years, which may arise by ‘floating germination’ of the microscopic propagules of green algae in the coastal waters (Feng et al. 2012; Song et al. 2018). U. prolifera entered the breeding period in the development stage, and its CA and AA gradually increased. The outbreak stage was mainly reflected in the short period (9-day) and wide distribution (the average CA was 1,723.66 km²). In late June, U. prolifera peak appeared in the western coastal area of the Yellow Sea, and its AA was relatively concentrated, which was distributed in the eastern sea area of the Jiangsu and Shandong coastlines.

During the decline stage, the time of U. prolifera disaster was 35–40 days. Compared with the outbreak stage, it was closer to the coastline, and the patches were gradually broken and reduced accordingly. The extinction stage lasted for 3–4 days, mainly distributed along the Qingdao coast, Shandong Province, and some scattered in the central Yellow Sea. From late July to August, the green tides in the coastal waters of the Shandong Peninsula disappeared.

The CA of U. prolifera first decreased and then increased during the years studied. In 2015, the CA was the largest in 5 years. In 2019, the CA showed an increasing trend. Compared with the previous 4 years, U. prolifera gathered in significant amounts and floated in the waters near Jiangsu shoal in early 2019, growing significantly compared with that in 2018. This may be related to the fact that eutrophication, rainfall, and light were more conducive to its growth in Northern Jiangsu shoal in 2019 (Tiyasha et al. 2020; Deng et al. 2021). U. prolifera originated in the sea near the shoal of Jiangsu Province, consistent with existing research conclusions (Cao et al. 2019).

The average SSTs in the different growth stages were obtained based on MODIS level-3 SST data from https://oceancolor.gsfc.nasa.gov/l3/ (downloaded on 15 June 2020), as shown in Figure 8, which is divided into three parts, in which (a) includes two parts: (1) the average temperature of the South Yellow Sea in different stages within the influence range of U. prolifera and (2) the influence range and CA of the influence range of U. prolifera in different stages. (b) Effect of inter-annual temperature on the change in the area of U. prolifera. The average SSTs in the different stages of U. prolifera were 16.3, 19.6, 21.9, 25.4, and 27.4 °C, respectively. Taylor et al. (2001) showed that its growth has wide adaptability to temperature (5–30 °C). U. prolifera has a significant growth rate in the temperature range of 10–25 °C (Fan et al. 2015; Liu et al. 2015), and its most suitable temperature range for growth is 20–26 °C (Taylor et al. 2001; Ye et al. 2011). When the temperature exceeds the optimum temperature range, its growth rate decreases (Hu et al. 2010). The average SST in the SYS meets the overall requirement for U. prolifera growth from May to August every year. The average SST was over 13 °C in all life cycle of U. prolifera. The suitable temperature environment is a prerequisite for the emergence of U. prolifera. The average SST gradually increased in the development stage and was within its optimal temperature range in the outbreak stage. At this time, the average SST was just in the most suitable temperature range for its growth, which further proved the correctness of the division of the life cycle of U. prolifera made in this study. In the extinction stage, the SST exceeded 26 °C and the SST was no longer suitable for its growth.

The temperature difference between the SYS and the CA of U. prolifera was statistically obtained (Figure 8(b)). The temperature difference between them first decreased and then increased. Among them, the average SST in 2017 was unique, and the average SST in the entire life cycle of U. prolifera was lower than that in the other years. Moreover, the difference between ATSYS and ATAA was the largest. When discussing the influence of SST on U. prolifera, the SST in 2017 was found to be slightly different from that in the other years, consistent with the results obtained by Zhang et al. (2020a). However, the temperature difference between the area where U. prolifera may occur and the area where U. prolifera occurs is the highest. Moreover, the temperature is an important factor affecting the growth of U. prolifera (Wang et al. 2018). The significant temperature difference in the region in 2017 may be an important factor affecting the growth of U. prolifera and may be a reason for the ‘double peaks’ in its life cycle curve in 2017 (Cao et al. 2019).

Previous studies showed that the disaster of U. prolifera is closely related to sea temperature, nutrients, and photosynthetic rate (Liu et al. 2016; Shamshirband et al. 2019), and our results also confirmed the effect of temperature on the growth of U. prolifera. Under ideal conditions and sufficient nutrition, a suitable temperature can promote the growth of U. prolifera (Zhang et al. 2020c). The temperature was not a limiting factor; nevertheless, optimizing the temperature gradually increases the intensity of the green tide (Fan et al. 2018; Zhang et al. 2020c). Therefore, a suitable temperature condition is necessary for the disaster of U. prolifera, particularly in its outbreak, decline, and extinction stages (Song et al. 2015).

Figure 9 shows the drift direction of U. prolifera and its corresponding average wind field distribution. The drift direction of the corresponding stage of U. prolifera under the background of the wind field in different stages was introduced (the red arrow indicates the drift direction in this stage). In 2017, U. prolifera emerged in the southeast of the Yellow Sea and drifted to the northwest due to the southeast wind (Figure 9). In 2019, it emerged in the shoal of Northern Jiangsu Province and gradually drifted to the northeast under the influence of the southwestern wind.

In the development stage, U. prolifera exhibited two drift directions: northwest direction (2016, 2017, and 2019) and northeast direction (2015 and 2018). The wind direction in 2019 deviated from the drift direction. In the bursting stage, the drift and wind directions were consistent. In the decline stage (except 2015), its drift direction in the other years was consistent with the wind direction. In the extinction stage, the sea wind direction was mainly affected by southeast and south winds. Finally, it drifted to the Northern Shandong Peninsula and died off the coast near Qingdao.

Previous studies (Qiao et al. 2011; Sun et al. 2020) have shown that the drift direction of U. prolifera in the different stages and the wind direction are consistent, indicating that its main driving force is the horizontal drift from the SWF. From the overall drift direction, the emergence (starting point) was generally in Subei Shoal, and the extinction (ending point) was generally near Qingdao waters. Its overall drift direction was northwest, indicating that the SWF is the power source driving the drift of U. prolifera. However, the wind field direction was not always consistent with the drift direction. For example, there was a certain offset between the two directions in 2015, which also showed that the power source of U. prolifera migration was not only the influence of SWF. The results of this study are consistent with those of existing research and have a strong correlation with the prevailing southeast wind and south wind in the SYS (Cao et al. 2019). The track of U. prolifera will change due to ocean current, artificial external interference, and other factors in the process of drift. Moreover, the wind field is the main force driving its horizontal movement, which was confirmed again and is consistent with existing research results (Zhao et al. 2018; Sun et al. 2020).

The spatio-temporal distribution characteristics of U. prolifera in different stages of its life cycle were identified and extracted using the dynamic threshold method. Based on its growth curve, U. prolifera disaster can be divided into five different stages, namely emergence, development, explosion, decline, and extinction, which lasted for approximately 90 days. The ‘outbreak period’ is the stage with the highest CA and AA and the highest growth rate of U. prolifera. In addition, the NDVIM and CA exhibited an evident lag effect, and the correlation reached the strongest at a 5-day lag. Therefore, the NDVIM can provide a scientific theoretical reference for predicting the future outbreak period of U. prolifera.

There were evident regional differences in the spatio-temporal distribution of U. prolifera in the different stages. U. prolifera emerged near Subei Shoal, developed in the middle of the Yellow Sea, burst in the eastern sea area of Shandong and Jiangsu Provinces, declined in the sea area of Shandong Province, and finally disappeared near Qingdao. Moreover, the drift of U. prolifera in five periods was regular, and the overall direction was northwest. Finally, it is concluded that the SST is the necessary condition for the occurrence of U. prolifera disaster, and the SWF is the main driving force for its horizontal drift. Therefore, we should pay more attention to these factors and devise appropriate and timely disaster prevention and control measures in future research.

The factors influencing the process of U. prolifera disaster are diverse and have a complex relationship. In future research, the environmental factors affecting U. prolifera disaster, such as the light intensity, salinity, and chlorophyll concentration, should be considered when building a pre-measurement model of U. prolifera. The applicability of the method of constructing the model for other U. prolifera areas should be verified.

This work was jointly supported by the project of Major Scientific and Technological Innovation in Shandong Province (Y9338301). The authors are grateful for the helpful comments from the anonymous reviewers and MJEditor (www.mjeditor.com) for its linguistic assistance during the preparation of this manuscript.

All relevant data are included in the paper or its Supplementary Information.