Information on red tide events, hydrometeorology, the marine environment and socio-economic development indicators, which are closely related to long-term red tide changes in the Zhejiang coast waters, was systematically collected and organized. Due to multiple reforms of local institutions, the weakness of monitoring techniques and insufficient funding for scientific research, information on the restoration of red tide events is the most complex among these data.

In this study, the red tide event was defined as discolored water caused by a high biomass of phytoplankton in a marine habitat, and the monitoring dataset included time, location, maximum area and other elements. A red tide time-series dataset from 1981 to 2006 was collected and reorganized by the Zhejiang Ocean Monitoring and Forecasting Center, which is managed by the Zhejiang Natural Resources Department. The dataset from 2007 to 2015 was provided by the Zhejiang Ocean Monitoring and Forecasting Center as well. The monitoring dataset from 2016–2018 was provided by the East China Sea Monitoring Center, which is managed by the East China Sea Bureau of the Ministry of Natural Resources. Local county annals and other references (Zhao, 2010) were also used for data collection, compilation, quality control and verification. Thus, overall, a dataset including 711 red tide events from 1981 to 2018 in the Zhejiang coastal waters was constructed.

The hydrometeorological factors used in this study include the ENSO index and SST. The ENSO indexes during 1981–2018 originate from NOAA. The SST from 1981–2018 recorded at the south and north ends of the Zhejiang coastal waters originated from Nanji Station (27.41°N, 121.1°E) and Shengshan Station (30.7°N, 122.81°E), respectively, of the National Global Ocean Stereoscopic Observation Network of Donghai Bureau. The SST refers to the sea water temperature between the sea surface and 0.5 m depth and is expressed by the average observation data values at 8:00, 11:00, 14:00 and 17:00 from the artificial observation station according to the Specification for Coastal Observation (GBT 14914−2006).

The marine water quality related to the “proportion of class IV to classes less than IV” from 2000 to 2018 that was published in the Zhejiang Marine Environmental Bulletin is released by the Department of Ecology and Environment of Zhejiang Province.

Finally, gross domestic product (GDP), industrial added value index and mariculture area were introduced as indicators with a high relationship between red tide occurrence and social and economic development of Zhejiang Province. The datasets were produced based on the statistical yearbook released by the Ningbo and Wenzhou. Table 1 presents a statistical summary of the data.

Both rank correlation and multivariate nonlinear regression analysis were used to identify and examine the sector-based effects of red tide occurrence and were performed using the 25th version of Statistical Product and Service Solutions (SPSS) software program.

To understand the multistage temporal variation, the evolution characteristics of red tide over the past 40 years were analyzed and are shown in Fig. 1. The annual occurrences and area of red tide in the Zhejiang coast waters fluctuated year by year. However, three distinct stages, namely, low frequency, explosive growth and fluctuating decline, were exhibited in different time periods. The three stages correspond to the periods from 1981 to 1999, 2000 to 2009 and 2010 to 2018, respectively. The transition point between increasing and decreasing red tide was the year 2003 in the second phase, which reached a peak with a maximum value of 79 in the last 40 a. The inter-annual variability of the red tide area is similar to that of the annual occurrence, although it fluctuates more in the second stage. Since the 21st century, the explosive growth of red tide has made the annual occurrence and area of red tide reach the highest value in history, with an annual average incidence of 45. At the same time, the area increased rapidly to approximately 13.361×10⁴ km², which is 44 times that in the last century. The annual occurrences reached a historical peak of 79 km² times in 2003, while the area increased to 2.734×10⁴ km² in 2000. After the peak, the number of annual red tide events decreased gradually year by year, but the large-scale red tide events with a large area of more than 1000 km² increased to a total of 24 times in this stage. However, in the third stage after 2010, the annual occurrence and area both declined slowly and showed stable fluctuation, with values significantly lower than those in the previous 10 a. The annual occurrence remained 18 events every year, while the total area of red tide dropped to 1.78×10⁴ km².

In the coastal area of Zhejiang Province, the red tide outbreaks have obvious spatial distribution characteristics of an east-west and north-south trend. Most red tide events in the east-west direction occurred side by side with the coastline and isobaths, 95.4% of which were located to the west of the 50 m isobaths (as illustrated in Fig. 2). A total of 4.2% of the total events occurred between the 50 m and 100 m isobaths. However, due to the limited nutrient input from the land, only three red tides occurred to the east of the 100 m isobath. The above analysis is basically consistent with a previous study (Cao, 1986). There is a long and narrow submarine slope in the northeast-southwest direction between the 10 m and 50 m isobaths of the Zhejiang coastal waters, and the coastline is actually in line with the isobaric lines and island trends. As a result, the plan-view distribution of nutrients is also in line with the isobaths, which is affected by terrestrial inputs, such as diluted water from the Changjiang River and Qiantang River. The nutrients decrease gradually from the coastal to offshore. Additionally, affected by the coastal current and upwelling, an abnormal structure of inversion and low salinity occurred to the west of the 50 m isobaths along the Zhejiang coastal waters, which can strengthen full vertical mixing and provide the necessary nutrients for a red tide outbreak. Previous studies have shown that the temporal and spatial variation of upwelling is consistent with that of red tide outbreak, which begins in May, intensifies in June, and dissipates gradually from September to October each year (Lou, 2010).

Over the past 39 a, the high-incidence area of red tide in the Zhejiang coastal waters has obviously expanded southward in the north-south direction. Figure 3 shows the frequency of the total red tide occurrences in four coastal cities (as illustrated in Fig. 2) in the three stages, and the change trend of each city is consistent with the three-stage pattern in Fig. 1.

Before the 21st century, the frequency in every city was generally low, while the frequency in Zhoushan was the highest, with a percentage of 17.5%. In the second stage (2000–2009), the frequency showed an upward trend in all the cities, with Ningbo showing the highest at 71.8%. In the third stage (2010–2018), the frequency in all cities began to decrease, but the frequency in Taizhou and Wenzhou continued to increase significantly. The frequency in Taizhou increased by up to 29.5%, and the high-incidence areas gradually shifted from the waters of Sanmen Bay in the 1990s to Dachen Island and Wenling-Shitang in the 21st century. The frequency of Wenzhou increased from the lowest in the first stage to the highest now, which accounts for 30.8%. The above analysis indicates that the high-incidence area of red tide constantly expanded southward in the second and third stages. Additionally, areas such as the Yushan-Jiushan Islands, Dachen Islands, and Dongtou Islands were newly added, in addition to the conventional high-incidence areas of the Zhoushan Islands, Xiangshangang Bay, Sanmen Bay and Nanji Islands.

Red tide is a complex ecological anomaly, and previous studies (Mu, 2011; Fu et al., 2012) have shown that eutrophication provides the material basis and necessary conditions for red tide. Land-based emissions and mariculture are the main sources of nutrients. The occurrences of red tide and the succession of biological populations are regulated by hydrological and meteorological conditions at different temporal and spatial scales. Therefore, nutrient and hydrometeorological conditions are considered simultaneously in this paper. We introduce seven factors, namely, the sea surface temperature (X _sst), mariculture area (X _mar), growth rate of primary industry (X _pri), growth rate of secondary industry (X _sec), growth rate of tertiary industry (X _ter), ENSO index (X _Niño3.4), and proportion of water quality class IV to classes less than IV (X _wq). The statistical analysis and physical mathematical model were adopted to analyze the main factors influencing the annual occurrence of red tide (Y _AORT) in different stages. In addition, we conducted causal analysis and provided a mechanistic discussion of the inter-annual variation and southward expansion.

First, all the datasets were analyzed by SPSS, and a histogram with a normal curve was drawn to determine whether the data had a bell-shaped distribution. Second, the skewness and kurtosis were calculated, and the nonparametric Shapiro-Wilk (W) test was adopted. The results indicate that X _sst, X _Niño3.4, X _mar, and X _wq exhibit normal or partial normal distributions and that Y _AORT, X _pri, X _sec, and X _ter exhibit non-normal distributions (Table 2). In scenarios where certain dependent variables diverged from the normal distribution or the overall distribution was unknown, the Spearman correlation coefficient was used to analyze the correlation between Y _AORT and the impact factors in this study. It was found that the factors X _sst, X _mar, X _pri, X _sec, and X _ter were positively correlated with Y _AORT, among which the correlation coefficient between X _mar and Y _AORT was the highest at 0.743 (Table 3).

Among the nutrient indexes, moderate positive correlations existed between X _mar, X _sec and Y _AORT, while low positive correlations existed between X _pri, X _ter and Y _AORT. Hence, X _mar and X _sec were selected as the representative indicators. Although X _Niño3.4 had no significant correlation with Y _AORT based on the 39 a time-series dataset, the correlation between X _sst and Y _AORT in the Zhejiang coastal waters increased, with a correlation coefficient of 0.326. Thus, X _sst was selected as the representative temperature indicator.

Because any function can be approximated by polynomials within an appropriate range, the regression equation is suitable for the analysis through adjusting the degree of a polynomial. In this study, based on the multivariate nonlinear regression process provided by SPSS, a mathematical model considering X _mar, X _sec, and X _sst was established by curve fitting and the iterative calculation by adopting the Levenberg-Marquardt estimation method. The specific equations are shown below:

The goodness of fit coefficient of this equation is 0.907, indicating that the proportion of variables that can be explained by the independent variables in the total variation of the dependent variables is as high as 90.7% (Table 4), which can be used to conduct causal analysis and provide a mechanistic discussion of the spatiotemporal variation and southward expansion of red tide events.

The difference in the influence of the local SST and ENSO teleconnection on red tide outbreak is significant. Although previous studies (Deng et al., 2017; He and Yang, 2009) revealed that the annual occurrence of red tide is correlated with the ENSO index, they have not discussed the three distinct inter-annual stages of red tide in the Zhejiang coastal waters. In this paper, the data analysis shows a high correlation (r=0.85, p=0.01) between Y _AORT and X _Niño3.4 from 2000 to 2009, a moderate correlation (r=0.55, P=0.05) between these factors from 2010 to 2018 and a weak correlation (r=0.31, p=0.05) between them from 1981 to 1999. These findings are consistent with the strong correlation reported between Y _AORT and X _Niño3.4 over a certain period of time (He and Yang, 2009). Moreover, X _Niño3.4 was not significantly correlated with Y _AORT based on the 39-year time-series dataset, although the correlation between X _sst and Y _AORT in the Zhejiang coastal waters increased, with a correlation coefficient of 0.326. The variation characteristics of SSTA_10a in the three stages were consistent with that of red tide events, with correlation coefficients as high as 0.7. As shown in Fig. 4, the local SSTA_10a increased in steps from 1981 to 2009, reached a peak value of 0.36°C in the second stage (2000–2009) and decreased to 0.002°C in the third stage (2010–2018). These findings indicated that the influence of SSTs on the annual occurrence of red tide can be revealed over an interdecadal-scale period. However, the local SST was more representative because the effect of El Niño on the local SST in the Zhejiang coastal waters presented a lag of 8 to 10 months via the teleconnection (Zhong et al., 2004).

In addition to natural environmental factors, human activities also play a key role in the outbreak of red tides. Figure 5 shows a comparison of two fitting values to quantitatively analyze the contribution of each factor. In this case, Y ₁ is the fitting value obtained by comprehensively considering X _sst, X _mar, and X _sec, with a goodness of fit coefficient of 0.907; Y ₂ is the fitting value in which only X _sst and X _mar are considered, and the goodness of fit coefficient is 0.687. The values of Y ₂ in 2006, 2007 and 2011 were obviously smaller than that of Y _1. It is noteworthy that the mariculture areas in 2006 and 2007 decreased to approximately half of that in 2005, with a value of 60×10⁴ km². Additionally, the SST suddenly decreased from 17.33°C in 2010 to 16.8°C in 2011, while the mariculture area was stable. However, it is interesting to note that the growth rate of the secondary industry increased from 13.27% in 2005 to 19.54% in 2006 and 23.84% in 2007, and it increased suddenly to 32.76% in 2011. Thus, the fitting value Y ₁, in which X _sec was considered, is highly consistent with the actual value (Fig. 5). Therefore, 68.7% of the variation in the annual red tide occurrence was due to the SST and mariculture area; however, secondary industry growth could compensate for the nutrient deficiency caused by the sharp mariculture area reduction and SST decline.

We can provide a detailed account of the main controlling mechanism of the distinct inter-annual variation rule of red tide in three stages based on the regression model of Y ₁ from the perspective of nutrient supply and SST changes. From 1981 to 1999, the Zhoushan sea area had the most red tide events, which may be related to the abundant nutrients brought by the fresh water of the Changjiang River, while the central Zhejiang sea area had fewer red tide occurrences. During this period, the mariculture area maintained a slow growth rate of less than 2%, and the growth rate of the secondary industry was less than 7%. The annual occurrences of red tide fluctuated between 2 and 5 times in most years. However, the first small peak was 21 times, which occurred in 1990. Because the local SST rose to 17.9°C in 1990, which was the highest value between 1981 and 1993, we can conclude that the local SST was the main controlling factor in this low-frequency stage.

After the 21st century, the obvious “roller coaster” structure of red tide occurrence appeared (Fig. 1). The occurrence of red tide increased sharply and reached a historical peak of 79 times in 2003 and then declined quickly. The annual average was 45.8 times per year. The high-incidence areas of red tide gradually spread from the northern to the central and southern areas of Zhejiang coastal waters. The large-scale red tide events broke out frequently, and six large-scale red tides with an area of 3000 km² or more occurred from 2001–2005. These red tides mainly broke out in the upwelling area, which was distributed between 122°E and 124°E longitude within the 30 m to 50 m isobaths, and they exhibited a sinuous and swinging strip-like pattern that paralleled the coastlines. The local SST maintained a positive anomaly and rose rapidly during the same period. SST_10a reached 18.11°C, which was the highest value to date. The mariculture area gradually expanded from 9.7×104 km² in 1981 to 118.3×104 km² at the peak in 2004. Environmental pollution was caused by the agglomeration and development of coastal cities in Zhejiang. As a result, the proportion of water quality class IV to classes less than IV increased, reaching a peak of 85.7% in 2002. Moreover, studies have shown that in the past 30 years, the PDO (Pacific Decadal Oscillation) changed from low value to high value in the late 1990s, which led to a stronger Kuroshio intrusion and enhanced coastal upwelling (Cloern et al., 2007). The East Asian summer monsoon also showed an increasing trend at the interdecadal scale, especially after 2000, which led to stronger atmospheric convergence of the surface layer (925 hPa) along the coast of Zhejiang-Fujian; moreover, the convective activity was intensified, the cyclonic vorticity was enhanced, and the coastal upwelling was strengthened (Xu et al., 2013). The enhancement of upwelling increased the nutrient concentration (Wooster and Zhang, 2004), which promoted the outbreak of red tide. The unique period of large-scale and frequent outbreaks of red tide occurred because of the combined effect of the abundant nutrient supply and appropriate hydrological and meteorological conditions. However, the ecological effects caused by ocean-atmosphere interactions and climate change should be thoroughly investigated.

In the third stage (2010–2018), the SST decreased slightly and showed a negative anomaly and SST_10a decreased to 17.75°C. The mariculture area decreased slightly but remained stable over 86×10⁴ –93×10⁴ km². Although the net value of secondary industry production continued to increase, the growth rate slowed down. Moreover, all types of land-based pollution emissions were effectively controlled due to environmental protection programs, such as “clear water and green mountains”. Consequently, the annual occurrence of red tide decreased significantly and remained at a stable level of approximately 20 times per year. Small and medium-sized red tide events accounted for the majority and the southward expansion of high-incidence areas was still obvious. Since 2008, the frequency of red tide was higher in Wenzhou than in Ningbo. Figure 6 shows that the variation tendency of SST in the Wenzhou and Ningbo coastal areas were the same, with a stable difference of 1.4°C, while the difference in GDP between the two cities changed dramatically. Before 1992, the GDP index of Wenzhou was significantly lower than that of Ningbo, and the maximum difference was –114.9. After 1992, Wenzhou’s economy developed rapidly and has surpassed that of Ningbo City, with the difference reaching the maximum value of 487 in 2007. In particular, the index difference of the secondary industry increased from –84.4 in 1989 to 6447.4 in 2014. The rapid economic development of southern cities, especially the secondary industry of these cities, inevitably led to an increase in land-based pollution emissions, which was one of the reasons for the continuous southward expansion of high-incidence areas. Over time, the nutrients carried by the runoff of the Changjiang River and Qiantang River were continuously transported to the south by the Zhejiang-Fujian Coastal Current (ZFCC) and deposited in the southern muddy sea area, resulting in the continuous accumulation of nutrients at the bottom, which may also partially explain the southward expansion of the high incidence area.