Science of the Total Environment 407 (2009) 4012–4021

Contents lists available at ScienceDirect

Science of the Total Environment

j ourna l homepage: www.e lsev ie r.com/ locate /sc i totenv

Occurrence and potential risks of harmful algal blooms in the East China Sea

Jinhui Wang a,c, Jianyong Wu b,⁎a East China Sea Monitoring Center, State Oceanic Administration, Shanghai, 200137, Chinab School of Public Health, University of North Carolina at Chapel Hill, NC, 27514, USAc School of Environmental Science and Engineering, Shanghai Jiaotong University, Shanghai, 200240, China

Abbreviations: DSP, diarrhetic shellfish poisoninggeographic information system; HAB, Harmful algal bpoisoning.⁎ Corresponding author. Tel.: +1 919 208 1666; fax: +

E-mail address: w_jianyong@hotmail.com (J. Wu).

0048-9697/$ – see front matter © 2009 Elsevier B.V. Adoi:10.1016/j.scitotenv.2009.02.040

a b s t r a c t

a r t i c l e i n f o

Article history:Received 23 October 2008Received in revised form 6 February 2009Accepted 18 February 2009Available online 5 May 2009

Keywords:Harmful algal bloomShellfish toxinsRisk assessmentGeographic information systemKernel density estimationNearest neighbor analysis

Harmful algal blooms (HABs) have drawn great attention in coastal areas worldwide in the past decadesbecause of their multiple effects on marine ecosystems as well as public health. This study utilizedgeographic information system (GIS) techniques to analyze the primary data on HABs, as well as shellfishtoxins data, in the East China Sea from 2000 to 2006. The frequency of HABs was mapped by kernel densityestimation, and the relative risk posed by HABs was assessed based on their physical–chemicalcharacteristics. In addition, the spatial patterns and the trend of HAB events were examined by nearestneighbor analysis and time series analysis, respectively. The results revealed that HAB events not only had anincreasing trend and significant seasonality, but also were clustered in space and time. HAB events displayeda higher frequency and a higher risk in Zhejiang coastal waters, particularly in the Zhoushan Archipelago, thelargest marine fishery in China. Shellfish toxins were detected in areas with high HAB risk, but were notcorrelated with the risk. This paper provides a novel method to assess the relative risk caused by HABs andsome useful information for HAB monitoring and management and aquaculture development.

© 2009 Elsevier B.V. All rights reserved.

1. Introduction

Harmful algal blooms (HABs), commonly known as “red tides” inChina, have increased in frequency of occurrence and expanded inspatial extent worldwide in the recent decades in coastal waters(Hallegraeff, 1993; Anderson, 1997). HABs have multiple impacts onmarine ecosystems as well as human health. One major concern is thetoxins produced by certain species of algae or phytoplankton. Thesealgal toxins can be accumulated by shellfish, such as mussels andclams when they filter water for food, leading to some health prob-lems when contaminated shellfish are consumed (Anderson et al.,2001). According to the symptoms observed in human intoxications,algal toxins are classified as several types: paralytic shellfish poisoning(PSP), diarrhetic shellfish poisoning (DSP), amnesic shellfish poison-ing (ASP), neurotoxic shellfish poisoning (NSP), Ciguatera fishpoisoning, and azasporacid poisoning (AZP) (Daranas et al., 2001).PSP toxins are the most common algal toxins found in coastal watersaround the world, and produced by Alexandrium spp., Gymnodiniumcatenatum and Pyrodinium bahamense (Shimizu, 1993). The maincompounds that cause paralytic shellfish poisoning isolated from

; ECS, East China Sea; GIS,loom; PSP, paralytic shellfish

1 919 966 7911.

ll rights reserved.

these species were identified as saxitoxins, a group of chemicals thatcan block sodium channels in the nervous system of bioorganisms(Noda et al., 1989). These toxins have resulted in approximate 2000cases of human intoxication with a mortality rate of 15% on a globalscale (Hallegraeff, 1993). Studies also revealed that PSP toxins wereinvolved in the deaths of birds and humpback whales (Nisbet, 1983;Geraci et al., 1989). Detailed information about health effects of PSPtoxins and others has been described in some literature (Van Dolah,2000; Daranas et al., 2001).

HABs have threatened seafood safety and public health in Chinesecoastal waters since the first HAB event was documented in Zhejiangcoast in 1933 (Fei, 1952). At least 512 documented HAB eventsoccurred from 1952 to 2002 in the coastal areas of the Chinese main-land (Yan and Zhou, 2004). According to annual reports from the StateOceanic Administration (SOA) of China, the East China Sea (ECS) is thearea most seriously affected by HABs (SOA, 2001). The East China Seais a marginal sea over a broad continental shelf located between thelargest continent (Asia) and the largest ocean (the Pacific). Marinefishery industries and products account for a large proportion of theeconomic growth in this area (Chen and Li, 1997). Therefore, studyingthe characteristics of the HAB events and assessing their risks in theEast China Sea are of particular importance. To date, a number ofstudies have been conducted in this area (Wang and Huang, 2003;Tang et al., 2006). These studies focused more on the description ofthe HABs events, such as the location, the area, and the frequency.however, few papers explored the spatial patterns displayed by HABevents, or the spatial and temporal associations between HAB events.In addition, assessing risks caused by HABs is still a challenge because

4013J. Wang, J. Wu / Science of the Total Environment 407 (2009) 4012–4021

the hazards are undefined and the exposure is hard to measure (VanDolah et al., 2001). Therefore, developing a method to assess therelative risk of HABs is necessary and valuablewhen a quantitative riskassessment guideline is unavailable.

A geographic information system (GIS) is a computer-based sys-tem for making and using spatial information. Recently, GIS has beencomprehensively applied into many fields in relation to public healthproblems, such as determining geographic distribution of diseases,analyzing spatial patterns and mapping population at risk (Richetts,

Fig. 1. The study area in the East China Sea. The specific monitoring districts were alphabetiD (the Zhoushan Islands), E (the Xiangshan Bay), F (the Taizhou Islands), G (the Dongtou IsIsland), L (the Meizhou Bay), M (the Xiamen Islands) and N (the Dongshan Bay).

2003). Though GIS-based spatial analysis and spatial statistics haveapplied into some environmental (Oguchi et al., 2000) or epidemio-logical studies (Lai et al., 2004), their application in studying harmfulalgal blooms is still few.

The objectives of this study are to identify the spatial and temporalpatterns associated with the HAB events using GIS techniques; and todevelop amethod to assess the relative risk caused by HABs. The studyprovides useful information for HAB monitoring and management, aswell as aquaculture development.

cally marked. A (the Yangtze Estuary), B (the Hangzhou Bay), C (the Shengshi Islands),land), H (the Nanji Islands), I (the Funing Bay), J (the Minjiang Estuary), K (the Pingtan

4014 J. Wang, J. Wu / Science of the Total Environment 407 (2009) 4012–4021

2. Materials and methods

2.1. Study area

Our study area is the East China Sea, including the coastal areas ofShanghai municipality, Zhejiang province and Fujian province (Fig. 1).Since the East China Sea consists of various areas with differentgeographic characteristics, to facilitate the description of the occur-rence of HABs, we marked the specific districts where HABs wereroutinely monitored. These specific districts are A (the Yangtzeestuary), B (the Hangzhou Bay), C (the Shengshi Islands), D (theZhoushan Islands), E (the Xiangshan Bay), F (the Taizhou Islands), G(the Dongtou Island), H (the Nanji Islands), I (the Funing Bay), J (theMinjiang Estuary), K (the Pingtan Island), L (theMeizhou Bay), M (theXiamen Islands) and N (the Dongshan Bay). Among these districts,the Zhoushan Archipelago, comprising of the Zhoushan Islands andthe Shengshi Islands, is the largest fishery in China; and the XiangshanBay is a key shellfish harvesting area.

2.2. Data collection

2.2.1. HABs dataSince 2001, a network coordinated by the East China Monitoring

Center was set out to monitor and manage HABs in the East China Sea.Based on the different administrative authority, the network encom-passed four-tier monitoring systems, including the East China Seamonitoring Center, four province-level monitoring centers, five regionalmonitoring centers and nine county-level monitoring stations, whichdistribute along the coastline of the East China Sea and haveresponsibility to investigate and report HABs in their administrativedistricts. A variety of tools were applied to collect information aboutHABs. Among them, ship-tracking played a key role in HAB monitoring.Meanwhile, remote sensing from aircraft and satellites andwater qualitydata from autonomous moorings provided additional information.Furthermore, we also collaborated closely with trained volunteers(fisherman) to make sure that all the HAB events were detected.

Seven years (2000–2006) of data about the HAB events in the EastChina Sea were analyzed. When a HAB event occurred, the locationwas measured by Global positioning system (GPS), the causativespecies were identified and the cell concentrations were measuredunder microscope by a trained staff following the standard operationprocedures (SOA, 2005). For each event, the detailed informationwascompiled, which included the location, the start time, the end time,the scale (area extent), the causative species, the concentration ofalgal cells and the specific coastal area.

2.2.2. Shellfish toxins dataBesides the HABs data, we also analyzed shellfish toxins data in the

East China Sea, which were collected from two sources. Primarily, thedata were collected from our own monitoring projects, includingroutine monitoring action and emergency action. In routine monitor-ing action, shellfish samples were collected 1 time each month fromMay to October. The sampling frequency was increased to 2 times in2005 and 2006. Ten sites were selected for shellfish toxins monitoringbecause where HABs frequently occurred (Fig. 1), and a variety ofshellfish species were collected based on their abundance in thesampling area. Totally, more than 200 samples were collected (24samples in 2002, 40 samples in 2003, more than 50 samples in 2004,2005 and 2006). The emergency action was carried out after theoccurrence of large scale (more than 1000 km2) HABs, for example, in2005, we investigated algal toxins in shellfish when a large scale HAB(about 7000 km2) occurred in Zhoushan coastal waters (Wang andWu, 2006). The samples were analyzed immediately or stored at −20 °C when delivered to the lab. For each sample, PSP and DSP toxinswere primarily monitored by mouse bioassay and then confirmed byHigh Performance Liquid Chromatography (HPLC) according to the

previous methods (Wu et al., 2005). In addition, we collected shellfishtoxins data from literature (Jiang et al., 2003; Dai et al., 2005). Theliteraturewas searched from Chinese Science Citation database and ISI(Institute for Scientific Information) Citation database. The compat-ibility of the data from each literature was evaluated based on itsmethodology description, such as study area, sampling time, types ofshellfish toxins and detection methods.

2.3. Data management with GIS

Though the shape of each HAB event is a polygon, we used a point(the centroid of polygon) to represent each HAB event for GIS analysis.We chose point pattern analysis because it is difficult to measure theexact shape of a HAB event, which is irregular, and varies constantly. Inaddition, to date, techniques (GIS tools) for point analysis are betterdeveloped than those for polygon pattern analysis. Each HAB eventwas given a geographic coordinate. If the geographic coordinate of aHAB event was not provided, we assigned a geographic coordinate tothe HAB event according to the description of the area where the HABoccurred. A point layer was created to demonstrate HAB events in eachyear. all the collected information on HAB events was stored in theattribute table of each layer. A basemap of the East China Sea wasacquired from the ESRI website (http://www.esri.com). We selectedGeographic Coordinate Systems-World Geodetic System of 1984(GCS-WGS-1984) as the projection for all the GIS maps, because it isfrequently used in world GIS maps and also served as referencecoordinate system for Global Positioning System. ArcGIS 9 withArcMap 9.2 version (ESRI, Redlands, CA) was used in this study.

2.4. Hot spots identification

When 7 layers of HAB events are overlaid on the basemap, the areawith a high density of points (HAB events) has a higher frequency ofoccurrence of HABs. Based on this assumption, the frequency ofoccurrence of HABs in the study area was mapped by kernel densityestimation (KDE). KDE is a method to generate a map that shows thedensity of the eventsmodeled as continuousfield by taking the value ofa specific point and spreading it across a predefined area (Gatrell et al.,1996). In KDE, a kernel, usually a circle with a constant radius, orbandwidth, is moved across the study area. The events dispersedwithin the circle are weighted according to their distance from thecenter of the circle. Events near the center have a higherweight. In thisway, the density of the point at center is estimated (Spencer andAngeles, 2007). We tested with different values for the radius andchose 0.5° as the radius to get a smoothed map. When a continuoussurfacewas generated, the frequencywas classified into five categoriesby natural breaks: higher, high, medium, low and lower. Areasindicating a higher frequency are hot spots of occurrence of HABs.

2.5. Risk assessment of HABs

A standard procedure for risk assessment consists of four steps:hazard identification, dose–response assessment, exposure assess-ment, and risk characterization. Simply, risk is expressed as theproduct of toxicity and exposure. The simplest approach assessing therelative risk only considers toxicity but ignores exposure assessment(Zhang et al., 2001). In terms of harmful algal blooms, to ourknowledge, no standards or guidelines for risk assessment can bereferenced. In addition, calculating the specific value of risk isimpractical because of the difficulty of exposure assessment. There-fore, we brought forward a model to assess the relative risk of HABs,by which we can know where HABs pose a higher risk and whereHABs pose lower risk. The model is expressed as below:

Risk = f Tð Þ × f Cð Þ × f Að Þ × f Dð Þ= f Lð Þ

4015J. Wang, J. Wu / Science of the Total Environment 407 (2009) 4012–4021

Where, T represents the toxicity, the product of toxicity equivalentfactor and the concentration of toxins produced by an algal cell; Crepresents the cell density of causative species; A represents the areaof a HAB event; D represents the duration, namely, how long a HABevent lasted; L represents the length, the shortest distance from thepoint of a HAB event to the coastline. f(T), f(C), f(A), f(D), and f(L) arethe functions of T,C, A, D, and L, respectively.

In thismodel, we assumed that the risk of a HAB eventwas positivelyassociatedwith the toxicity, the concentration, the area and theduration,but inverselyassociatedwith thedistance. To calculate the risk ofHABs inthe East China Sea, we simplified this model. We assumed f(i)=α×i

Fig. 2. The HAB events occurred in the East China Sea fr

(i=T, C, A, D or L, α is the coefficient). Thus, the model can be simplywritten as: Risk=α×T×C×A×D/L. We assumed that Twas equal to 1 ifthe causative species in a HAB event were identified as toxin-producingalgae, otherwise, T was zero. C, A and D were recorded in the attributetableduringdata collection. Since the concentrationof cells is avery largenumber, we used the value of its log transformation. Lwas measured byGIS tools. For the eventwhich informationwas not recorded completely,the default values of C, A and D were 106 cell per liter, 1 km2 and 1 day,respectively. When the relative risk of each HAB event was calculated, itwas classified into five categories (Lower, Low, medium, High andHigher) in light of geometrical interval.

om 2000 to 2006. Each point denotes a HAB event.

4016 J. Wang, J. Wu / Science of the Total Environment 407 (2009) 4012–4021

2.6. Spatial statistics

Spatial statistics was applied to understand the spatial relation-ships in the HAB data. The spatial pattern of the HAB events wasexamined by nearest neighbor analysis. Nearest neighbor analysis wasoriginally developed by Clarke and Evans (1954). It is a method todetermine whether a set of points in space is distributed in adispersed, random or clustered pattern on basis of the value of thenearest neighbor index (NNI), a ratio of the average nearest neighbordistance to the expected nearest neighbor distance if the pattern israndom. In nearest neighbor analysis, the null hypothesis is that all the

Fig. 3. The frequency of HABs in the East China Sea. The frequency of HABs in the study areawthe frequency was classified into five categories. The areas with higher frequency are the h

points are distributed randomly. Then the hypothesis is tested by Zstatistic. A negative value of Z score indicates the points are clustered,while a positive value of Z score indicates they are dispersed ordistributed randomly (Getis, 1964). Z statistic is calculated using thefollowing equations:

Z =dm − deffiffiffiffiffiffiffiffiffiffiffiffiffiffiffivar dið Þp ; and dm =

XN

i

di =N:

Where, dm is the average nearest neighbor distance, de is theexpected value of the nearest neighbor distance in a random pattern,

as mapped by kernel density estimation based on the HAB data from 2000 to 2006. Then,ot spots of HAB occurrence.

Table 1The causative species of HABs and their frequency in the East China Sea.

Causative species Total 2000 2001 2002 2003 2004 2005 2006

Akashiwo sanguinea 1 1Alexandrium catenellaa 4 3 1Alexandrium sp.a 1 1Alexandrium tamarensea 3 1 2Asterionella japonica 1 1Ceratium furca 1 1Chaetoceros costatus 1 1Chaetoceros curvisetus 11 1 3 3 4Chaetoceros debilis 2 2Chaetoceros gracilis 2 1 1Chaetoceros sociaIis 2 1 1Chaetoceros sp. 9 1 1 5 2Cyanobacteria 1 1Dactyliosolen mediterraneus 2 2Gonyaulax polygramma 2 1 1Gonyaulax sp. 1 1Gonyaulax spinifera 1 1Gymnodinium catenatum 1 1Gymnodinium brevea 3 1 1 1Gymnodinium mikimotoia 16 13 3Gymnodinium sanguineuma 1 1Gymnodinium sp. a 5 2 3Heterocapsa circularisquamaa 1 1Karenia mikimotoia 36 1 18 17Leptocylindrus danicus 1 1Melosira sulcata 1 1Mesodinium rubrum 8 3 1 2 1 1Nitzschia closterium 1 1Noctiluca scintillans 32 3 13 6 6 4Prorocentrum dentatum 120 6 8 33 23 15 35Prorocentrum minimum 1 1Prorocentrum sp. 7 4 3Prorocentrum triestinum 4 4Rhizosolenia delicatula 1 1Scrippsiella trochoidea 3 1 2Skeletonema costatum 57 2 8 11 4 14 18Thalassiosira nordenskioldii 1 1Thalassiosira rotula 1 1Thalassiosira sp. 3 1 2Thalassiosira subtilis 1 1Unknown 94 12 24 10 19 4 6 19

a Toxin-producing algal species.

Table 2Spatial patterns of HABs in the East China Sea examined by nearest neighbor analysis.

Year NN index Z score Significance level (p) Spatial pattern

2000 1.93 6.16 0.01 Dispersed2001 1.39 4.34 0.01 Dispersed2002 0.62 −4.05 0.01 Clustered2003 0.45 −9.77 0.01 Clustered2004 0.45 −6.79 0.01 Clustered2005 0.84 −2.16 0.05 Clustered2006 0.50 −9.14 0.01 ClusteredTotal 0.29 −25.17 0.01 Clustered

4017J. Wang, J. Wu / Science of the Total Environment 407 (2009) 4012–4021

N is the number of points, di is the nearest neighbor distance for point i,and var(di) is the variation of di.

The calculation of the nearest neighbor index and Z statistic wasexecuted by the Average Nearest Neighbor tool in Spatial Statisticstoolbox provided by ArcMap 9.2 software. The boundary for nearestneighbor analysis was set as our study area.

2.7. Time series analysis

The frequency of HAB events in each month from 2000 to 2006was calculated and analyzed for its time series patterns. Beforeanalysis, a serial number was assigned to each month, which wasstarted on January 2000 as 1 and ended on December 2006 as 84.Generally, time series patterns can be described by two basic types ofcomponents: trend and seasonality. For trend analysis, an ARIMAmodel (autoregressive integratedmoving averagemodel) was createdby an expert modeler to fit the frequency in each month, and a linearregression line was generated to show the trend. Seasonality of thedata was analyzed via autocorrelation. Both trend analysis andseasonality analysis were conducted with SPSS 15.0 software (SPSSInc., Chicago, Illinois).

3. Results and discussion

After 2000, monitoring HABs became a crucial obligation forgovernment agencies responsible for marine environmental monitor-

ing. Since then, a report on the occurrence of HABs in the coast ofChina was issued annually by the State Oceanic Administration ofChina. The frequency, location, scale and causative species of HABs inseveral major coastal areas were described briefly in these reports.Therefore, the above information is skipped in this paper. We areinterested in the spatial and temporal patterns displayed by HABs andtheir relative risk.

3.1. Occurrence of HABs

Each HAB event from 2000 to 2006 represented as a point wasdemonstrated in Fig. 2. Most events were distributed in the coastalwaters band, about 60 kmwide from the shore. The figure also showedthe HABs occurred frequently along Zhejiang coast and some districtsof Fujian coast. According to kernel density estimation (Fig. 3), thehighest frequency of HAB occurrence was observed in the Zhoushancoastal area (C and D), the coastal area of the Naji islands (H), theFuning bay (I) and the Xiamen islands (M). High frequency was alsoobserved in the coastal areas of the Xiangshan Bay (E), the TaizhouIslands (F), the Dongtou island (G).

The causative species in each event were listed in Table 1. In total,40 causative species were identified in the East China Sea from 2000to 2006. Among them, Prorocentrum dentatum is the most commonspecies (dominating in 120 events), followed by Skeletonema costatum(dominating in 57 events), Karenia mikimotoi (dominating in 36events) and Noctiluca scintillans (dominating in 32 events). Thenumbers of causative species in each year varied. From 2002 to 2006,10, 18, 11, 12 and 20 causative species were observed respectively. Ineach year, several new causative species were identified. Gymnodi-nium mikimotoi and K. mikimotoi draw special attention because theyare the principal toxin-producing species that emerged in this arearecently. G. mikimotoi emerged suddenly as a causative species in 13events in 2003, and re-emerged in 2006. K. mikimotoi was firstlyobserved as a causative species in 2004 in one event, but the eventscaused by this species increased dramatically in 2005 and 2006.

The results showed that the frequency of HAB occurrence in theEast China Sea had increased over time, with a notable increase after2000, which can be explained by two reasons: 1. the HAB eventsincreased due to the deterioration of marine environmental quality(Lu and Zhou, 2004); 2. the HAB monitoring network has taken intoeffect. As for causative species, new species were identified in eachyear. Exotic species should be of particular concern because they oftenhave harmful effects on biodiversity (Galil, 2007) and may cause thecomplexity of HAB behaviors (Smayda, 2007).

3.2. Spatial patterns and trend of HABs

Nearest neighbor analysis revealed the distribution of the HABevents was clustered from 2002 to 2006 but dispersed in 2000 and2001(Table 2). The overall events were distributed in a clusteredpattern with the NNI as 0.29, Z score as −23.47 SD (standarddeviation), and the significance level at 0.01. The clustered pattern isexpected because HABs were frequently observed in several keymonitoring districts, such as the Yangtze Estuary, Xiangshan Bay and

Fig. 5. The seasonality of the frequency of HAB occurrences demonstrated byAutocorrelogram. When the lag number is 1, 11 or 12. ACF(Autocorrelation Function)has positive values, which are 0.463, 0.276 or 0.533. The results show that the frequencyof HAB occurrence has a moderate autocorrelation when lag number is 12, whichsuggests the occurrence of HABs has a periodicity by year.

4018 J. Wang, J. Wu / Science of the Total Environment 407 (2009) 4012–4021

Xiamen coastal waters in recent years, which was also demonstratedby kernel density estimation (Fig. 3). This clustered pattern mighthave resulted from the high concentration of nutrients (especially fortotal nitrogen) in these districts, because HABs had a strongcorrelation with eutrophication in coastal waters (Anderson et al.,2002).

The trend of the frequency of HAB events in each month wassimulated by the ARIMA model (Fig. 4). The linear regression lineshowed the frequency of HAB events increased slightly by month. Inaddition, the peak of the frequency in each year appeared in May andJune. The autocorrelogram showed the seasonality of the occurrenceof HABs (Fig. 5). As shown in this figure, the autocorrelation functioncoefficient has the highest value when the lag number is 12, whichsuggested the occurrence of HABs showed similar characteristics overa year period.

There are some implications in these findings. Understanding thespatial and temporal patterns of HAB events can help policy makers tooptimize public resources and make approximate strategies, sincemonitoring HABs is costly and time-consuming in China due to thelong coastline. Our results revealed that the HAB events weresignificantly clustered in space, and more likely occurred in certainmonths. In combination this information with the hot spots identifiedby kernel density estimation, the optimal locations where HABsoccurred frequently or clustered can be chosen as key monitoringdistricts. The similar trend and seasonality of the HABs occurrencemight also exist in the coast of China. As summarized in the nationalreport of HAB in China (Lu and Zhou, 2004), the frequency of HABoccurrence in coastal China have increased remarkably after 2000.

3.3. Potential risks of the HABS

From 2001 to 2006, 67 HAB events were caused or co-caused bytoxin-producing algae, and 8 toxic species were identified, includingAlexandrium catenella, Alexandrium sp., Alexandrium tamarense, Gym-nodinium breve, G. mikimotoi, Gymnodinium sanguineum, Gymnodi-nium sp., Heterocapsa circularisquama, and K. mikimotoi. As mentionedabove, G. mikimotoi and K. mikimotoi posed serious risks in this area.We evaluated the relative risk of HAB events using the simple modeldeveloped according to the physical–chemical characteristics of HABs.As shown in Fig. 6, the HABs with the highest risk were detected in

Fig. 4. The trend of the frequency of HAB occurrence. The red curve is the observedfrequency of HAB occurrence. The blue curve is the expected frequency of HABoccurrence fit by ARIMA model. The black line is the linear tendency line. The values ofARIMA model parameters are shown as following: for constants, estimate=4.115,p=0.002; for lag number=1, estimate=−0.399, p=0.001; for lag number=12,estimate=−0.271, p=0.038. (For interpretation of the references to colour in thisfigure legend, the reader is referred to the web version of this article.)

Zhoushan coast (D) and the Xiangshan Bay (E). In the Funing Bay (I),Naji coast (H) and Xiamen coast (M), HABs also caused high risk.

The occurrence of HABs poses risks not only to marine fishery,aquiculture, but also to recreational water or beaches. However,presently, there are no guidelines to assess the risk posed by HABs.Some European countries use cell concentrations of HAB species asaction limits. For example, in Denmark, if cell concentration of A.tamarense is above 500 cell/L, shellfisheries should be intensivelymonitored or closed (Anderson et al., 2001). This action limit based oncell concentration is easily implemented. However, this regulationdoes not reflect any risks posed by HABs. In this study, not only cellconcentration, but also the duration, the scale and the distance(location) of HABs were considered. the association of these factorswith the risk of the HABs is easy to understand. For example, mostshellfish harvesting areas are close to the coastal line, the risk is higherwhen the distance (from the HAB event to the coastal line) is shorter.Though the risk was not quantified, the assessment of the relative riskis still helpful. First, it provides a reference with clear information topolicy makers to make decisions. Second, the area with relatively lowrisk of HABs is primarily under consideration to make a strategy formarine fishery or aquiculture, given that the HAB is a naturalphenomenon and is difficult to predict. As the result showed, areasat high risk of HABs distributed along Zhejiang coastal water,including Zhoushan fishery, the largest fishery in China, and theaquaculture area in Xiangshan Bay, which supplies a large amount ofshellfish products to the world. Thus, mitigating the risk of HABs is acritical task for developing marine aquiculture in these areas.

We attempted to connect the shellfish toxins data with the HABrisk. In general, we assume that shellfish would be potentiallycontaminated by algal toxins when HABs occur in that area. Asshown in Table 3, PSP and DSP toxins were frequently detected in theshellfish (Mytilus edulis) collected in Shengshi coastal area (C) andSandu bay (the southwest of I district). In addition, PSP or DSP toxinswere detected in the shellfish samples collected in the Hangzhou Bay,Zhoushan coastal area, Naji coastal area, Xiamen coastal area and theDongshan Bay. No shellfish toxins were detected in the samplescollected from the East Chongming beach (near A distict), theXiangshan Bay and the Funing Bay. The results partly confirmed ourassumption. However, the accumulation of algal toxins in shellfishdepends on shellfish species, sampling time and location, and so on(Sagou et al., 2005; Shumway et al., 2006). For example, 15 sampleswere collected after a large scale HAB occurred in 2005, no algal toxinswere detected in these samples (Wang and Wu, 2006). Algal toxins

Fig. 6. The relative risk of HAB events. The value of the risk was calculated according to the model developed in this study, then classified into five categories. The larger circlerepresents the higher risk.

4019J. Wang, J. Wu / Science of the Total Environment 407 (2009) 4012–4021

were also detected in a few samples in the low risk area, whichindicated that shellfish accumulated algal toxins even before toxin-producing algae forming blooms. It suggests that the safety of shellfishconsumption cannot be guaranteed even if there is no harmful algalbloom occurred in that area. Additionally, pathogens and other toxicchemicals also threaten the safe consumption of shellfish in coastalareas with poor water quality or affected by sewage disposal fromnearby urban and industrial centers (Refnstam-Holm and Hernroth,2005). it is necessary to monitor shellfish toxins as well as otherhazards routinely regardless of the HAB occurrence.

3.4. Considerations for GIS application

We used two GIS techniques: kernel density estimation andnearest neighbor analysis. With respect to kernel density estimation,the selection of radius (bandwidth) is a key issue. In this work, weselected a relatively large radius (0.5°) in light of several considera-tions. First, each HAB event represented as a point actually is polygon,most of which range from 0.1 km2 to 1000 km2 in area. A large radiusallows one or more events being covered in a kernel. Second, a smallradius produced an irregular bumpy map, while a larger radius

Table 3PSP and DSP toxins detected in shellfish samples from the East China Sea.

Time Species Type of toxins Toxins content Location descriptiona2002, 6 Mytilus coruscus PSP <170 mu/100 g D district, Zhejiang coasta2002, 8 Mytilus coruscus PSP <170 mu/100 g C district, Zhejiang coasta2002,8 Cultellus attenuatus PSP <170 mu/100 g C district, Zhejiang coast2002, 8 Mytilus edulis DSP 20 mu/100 g H district, Zhejiang coast2002, 8 Scapharca subcrenata DSP 5 mu/100 g B district, Zhejiang coast2003, 5 Mytilus edulis DSP 5 mu/100 g C district, Zhejiang coast2003, 6 Mytilus edulis DSP 5 mu/100 g C district, Zhejiang coastb2003, 7 Mytilus edulis DSP 5 mu/100 g Sandu bay, (Southwest of I district), Fujian coastb2003, 8 Mytilus edulis DSP 5 mu/100 g Sandu bay, (Southwest of I district), Fujian coast2003, 9 Mytilus coruscus DSP 5 mu/100 g C district, Zhejiang coast2003, 9 Perna viridis DSP 5 mu/100 g N district, Fujian coastb2003, 10 Mytilus edulis DSP 10 mu/100 g Sandu bay, (Southwest of I district), Fujian coast2004, 6 Mytilus edulis DSP 5 mu/100 g Sandu bay, (Southwest of I district), Fujian coast2004, 6 Mytilus edulis PSP 166 mu/100 g Near C district, Zhejiang coast2004, 6 Saccostrea cucullata PSP <175 mu/100 g M district, Fujian coast2004, 6 Saccostrea cucullata DSP 5 mu/100 g M district, Fujian coast2004, 6 Saccostrea cucullata PSP <175 mu/100 g M district, Fujian coast2004, 6 Saccostrea cucullata DSP <6 mu/100 g M district, Fujian coast2005, 6 Mytilus edulis PSP 4.4 µg/100 gc H district, Zhejiang coast2005, 8 Thais luteostoma DSP 5 mu/100 g C district, Zhejiang coast2005, 8 Mytilus edulis DSP 5 mu/100 g C district, Zhejiang coast

a Cited from Jiang et al., 2003.b Cited from Dai et al., 2005.c Measured by HPLC.

4020 J. Wang, J. Wu / Science of the Total Environment 407 (2009) 4012–4021

generated a smoothed map (Chiu, 1991). Considering that regularlydistributed districts instead of sparsely distributed are more preferredtomonitor HABs in the East China Sea, we chose a relative large radiusto get a smoothed map. Furthermore, this estimation is well coincidedwith our long-time observation of HABs. In terms of nearest neighboranalysis, spatial patterns are significantly affected the boundary ofstudy area. The points may be clustered in large area extent butdispersed in small area extent. In this study, we chose the study area asthe boundary; thus, the spatial clustering of the HABs is valid withinthe scope of our monitoring area but may not in small area extent.

3.5. Limitations and future research

There are several caveats that should be mentioned. First, we tookeach event as a point which should be a polygon. The polygon patternanalysis might be better to reflect the real situation than the pointpattern analysis (Wong and Lee, 2005). Second, we only consideredseveral basic physical–chemical characteristics of a HAB event in riskassessment. Actually, when a HAB occurs, many other variables, suchas meteorological factors, tide and current, can affect the risk. Inaddition, we simply assumed that the variables we considered have alinear relationship with the risk, which is not verified. We also simplyassumed that the toxicity was 1 if toxin-producing algae wereidentified. However, the concentration and toxicity of toxins producedby algae might vary greatly in different species. If a quantitative riskassessment is put forward, these factors should be considered. In thefurther study, the concrete relationship of the risk of HABs with thedistance, the scale and the duration will be investigated; thevariability of toxin concentrations produced by different species willbe analyzed. Once these questions are resolved, a deterministic modelcombining with GIS techniques can be used to accurately calculate therisk caused by a HAB event, which will be of great help to mitigateeconomic loss and health problems due to harmful algal blooms.

4. Conclusions

The results of sevenyears of observation indicated that theoccurrenceof HABs in the East China Sea had an increasing trend and significantseasonality. In addition, new causative species, especially toxin-produ-cing species, emerged annually. The GIS and statistical analyses revealedthat the HAB events occurred in clustered patterns in space and time.

The study developed a simple method to assess the risk caused byHABs. This method considered several variables of a HAB event and isbetter than the others which only use cell concentration. The result ofthe assessment showed that most areas of Zhejiang coastal waterswere potentially affected by the high risk of HABs. PSP and DSP toxinsin shellfish were detected in the high risk area as well as the low riskarea, suggesting that shellfish toxins and the risk of HABsmight not becompletely correlated.

The study also demonstrated that GIS was a powerful tool in HABmonitoring and prediction, seafood safety management and aqua-culture development.

Acknowledgements

The authors thank all the people involved in this project and thevolunteers who provided information to us. This project wassupported by the Science and Technology Commission of ShanghaiMunicipality (062358101) and the Marine Science Fund for YoungScientists (2007129).

References

Anderson DM. Turning back the harmful red tide. Nature 1997;388:513–4.Anderson DM, Andersen P, Bricelj VM, Cullen JJ, Rensel JE. Monitoring and management

strategies for harmful algal blooms in coastal waters. APEC #201-MR-01.1, AsiaPacific Economic Program, Singapore and Intergovernmental OceanographicCommission Technical Series No. 59, Paris; 2001. p. 24.

Anderson DA, Glibert PM, Burkholder JM. Harmful algal blooms and eutrophication:nutrient sources, composition and consequences. Estuaries 2002;25:704–26.

Chen WZ, Li CS. A review of the fisheries resource status in East China Sea. J Fish China1997;4:39–43 in Chinese).

Chiu ST. Bandwidth selection for kernel density estimation. Ann Stat 1991;19:1883–905.Clarke PJ, Evans FC. Distance to nearest neighbor as a measure of spatial relationships in

population. Ecology 1954;35:445–53.Dai H, Li NJ, Chen GB. Paralytic shellfish poisoning and diarrhetic shellfish poisoning in

red tide monitoring area of Sandu Bay, Fujian. Mar Environ Sci 2005;24:44–7 (inChinese).

Daranas HA, Norte M, Fernandez JJ. Toxic marine microalgae. Toxicon 2001;39:1101–32.Fei H. The cause of red tides. Sci Art 1952;22:1–3 (in Chinese).Galil BS. Loss or gain? Invasive aliens and biodiversity in the Mediterranean Sea. Mar

Pollut Bull 2007;55:314–22.Gatrell AC, Bailey TC, Diggle PJ, Rowlingson BS. Spatial points pattern analysis and its

application geographical epidemiology. Trans Inst Br Geogr 1996;21:256–74.Geraci JR, Anderson DM, Timperi RJ, St. Aubin DJ, Early GA, Prescott JH, et al. Humpback

whales fatally poisoned by dinoflagellate toxin. Can J Fish Res Aquat Sci1989;46:1895–8.

4021J. Wang, J. Wu / Science of the Total Environment 407 (2009) 4012–4021

Getis A. Temporal land-use pattern analysis with the use of nearest neighbor andquadrant methods. Ann Assoc Am Geogr 1964;54:391–9.

Hallegraeff GM. A review of harmful algal blooms and their apparent global increase.Phycologia 1993;32:79–99.

Jiang TJ, Chen JF, Zou YL, Liu JS, Yang WD. Paralytic shellfish toxins in shellfish from thecoast of high frequent harmful algae blooms occurrence areas in East China Sea andSouth China Sea. Chin J Appl Ecol 2003;14:1156–60 (in Chinese).

Lai PC, Wong CM, Hedley AJ, Lo SV, Leung PY, Kong J, et al. Understanding the spatialclustering of Severe Acute Respiratory Syndrome(SARS) in Hong Kong. EnvironHealth Perspect 2004;112:1550–6.

Lu SH, Zhou MJ. ational report on HABs in China; 2004. p. 5–7.Nisbet IC. Paralytic shellfish poisoning: effects on breeding terns. Condor 1983;85:338–45.Noda M, Suzuki H, Numa S, StuÈhmer W. A single point mutation confers tetrodotoxin

and saxitoxin insensitivity on the sodium channel II. FEBS Lett 1989;259:213–6.Oguchi T, Jarvie HP, Neal C. River water quality in the Humber catchment: an introduction

using GIS-based mapping and analysis. Sci Total Environ 2000;251:9-26.Refnstam-Holm AS, Hernroth B. Shellfish and public health: a Swedish perspective.

Ambio 2005;34:139–44.Richetts TC. Geographic information systems and public health. Annu Rev Public Health

2003;24:1–6.Sagou R, Amanhir R, Taleb H, Vale P, Blaghen M, Loutfi M. Comparative study on

differential accumulation of PSP toxins between cockle (Acanthocardia tubercula-tum) and sweet clam (Callista chione). Toxicon 2005;46:612–8.

Shimizu Y. Microalgal metabolites. Chem Rev 1993;93:1685–98.Shumway SE, Sherman SA, Cembella AD, Selvin R. Accumulation of paralytic shellfish

toxins by surfclams, Spisula solidissima (Dillwyn, 1897) in the Gulf of Maine:seasonal changes, distribution between tissues, and notes on feeding habits. NatToxins 2006;2:236–51.

Smayda TJ. Reflections on the ballast water dispersal—harmful algal bloom paradigm.Harmful Algae 2007;6:601–22.

SOA. The public report on marine environmental quality of China; 2001.SOA. Technical specification for red tide monitoring, HY/T 069-2005; 2005.Spencer J, Angeles G. Kernel density estimation as a technique for assessing availability

of health services in Nicaragua. Health Serv Outcomes Res Method 2007;7:145–57.Tang DL, Di BP, Wei GF, Ni IH, Oh IS, Wang SF. Spatial, seasonal and species variations of

harmful algal blooms in the South Yellow Sea and East China Sea. Hydrobiologia2006;568:245–53.

Van Dolah FM. Marine algal toxins: origins, health effects, and their increasedoccurrence. Environ Health Perspect 2000;108:133–41.

Van Dolah FM, Roelke D, Greene RM. Health and ecological impacts of harmful algalblooms: risk assessment needs. Hum Ecol Risk Assess 2001;7:1329–45.

Wang J, Huang X. Ecological characteristics of Prorocentrum dentatum and the cause ofharmful algal bloom formation in China Sea. Chin J Appl Ecol 2003;14:1065–9 (inChinese).

Wang JH, Wu JY. Marine algal toxins in shellfish from shanghai markets and originalsources following a large scale red tide occurrence. Bull Environ Contam Toxicol2006;77:164–70.

Wong DW, Lee J. Statistical analysis of geographic information with ArcView GIS andArcGIS. Hoboken, NJ: John Wiley & Sons, Inc; 2005.

Wu JY, Zheng L, Wang JH. Contamination of shellfish from Shanghai seafood marketswith paralytic shellfish poisoning and diarrheic shellfish poisoning toxinsdetermined by mouse bioassay and HPLC. Food Addit Contam 2005;22:647–51.

Yan T, Zhou MJ. Environmental and health effects associated with harmful algal bloomand marine algal toxins in China. Biomed Environ Sci 2004;17:165–76.

Zhang Q, Crittenden JC, Mihelcic JR. Does simplifying transport and exposure yieldreliable results? An analysis of four risk assessment methods. Environ Sci Technol2001;35:1282–8.