Interannual variability of dimethylsulf ide in the Yellow Sea*

Sijia WANG , Qun SUN ,**, Siyu LI , Jiawei SHEN , Qian LIU , Liang ZHAO

1 College of Marine and Environmental Sciences, Tianjin University of Science and Technology, TEDA, Tianjin 300222, China

2 School of Marine Science and Technology, Tianjin University, Tianjin 300072, China

Abstract The 22-year (1998-2019) surface seawater dimethylsulf ide (DMS) concentrations in the Yellow Sea (YS) were hindcasted based on satellite sea surface temperature (SST) and chlorophyll- a (Chl- a) data using a generalized additive mixed model (GAMM). A continuous monthly dataset of DMS concentration in the YS was obtained after using the data interpolation empirical orthogonal function (DINEOF) to reconstruct missing information in the dataset. Then, the interannual DMS variability in the YS was analyzed. The results indicated that the monthly climatological DMS concentration in the YS was 3.61 nmol/L. DMS concentrations in the spring and summer were signif icantly higher than those in the autumn and winter. DMS concentrations were highest in coastal YS waters and lowest primarily in off shore YS waters. Interannual DMS variability between 1998 and 2019 was subdivided into two inverse phases: with the exception of the central YS, DMS increased before the turning point and decreased after. The turning point in interannual DMS variation was earlier in the inshore YS as compared to the central YS. Spectrum analysis identif ied some signif icant patterns of interannual variation in the DMS anomaly in the YS. Chl a appeared to be the main factor inf luencing interannual trends in DMS in the YS. Interannual DMS variability was under the joint control of Chl a and SST. However, short-term interannual DMS variation (2-3 years) was primarily related to SST, while longer term interannual DMS variation (6-8 years) was signif icantly correlated with Chl a and SST.

Keyword: interannual variability; dimethylsulf ide; hindcasting; generalized additive mixed modelling;Yellow Sea

1 INTRODUCTION

The most important, abundant, and widespread volatile biogenic sulfur compound in surface seawater is dimethylsulf ide (DMS) (Lovelock et al., 1972;Andreae and Barnard, 1984; Andreae et al., 1985).DMS in surface seawater is the source of more than 50% of all natural sulfur emitted into the global atmosphere (Andreae, 1986), and represents a large portion of marine organic volatile emissions (Andreae,1990). DMS emitted into the atmosphere is rapidly oxidized by OH, NO3, and halogen-oxidized radicals including BrO, IO, and ClO, forming various sulfurcontaining products; these sulfur-containing particles are an important source of cloud condensation nuclei(CCN) (Charlson et al., 1987). CCN indirectly aff ect the Earth’s radiative balance by altering the radiative properties of clouds (Nguyen et al., 1992; Andreae and Rosenfeld, 2008), strongly inf luencing global climate change. Indeed, it has been shown that decreases in sulfur f lux as a result of ocean acidif ication and climate change further exacerbate global warming(Six et al., 2013). Therefore, to better understand climate change, it is important to explore interannual variations in DMS concentration.

Many studies since the 1990s have attempted to hindcast DMS levels in surface seawater based on insitu observations or using various models in order to analyze interannual variations in DMS concentrations.For example, Song et al. (2019) used an empirical DMS algorithm, and reported that global surface seawater DMS concentrations were stable between 2000 and 2007, but DMS concentrations decreased between 2008 and 2015. In addition, based on the global warming scenarios from 1860 to 2100 outlined by Houghton (2001), Kloster et al. (2007) suggested that global annual mean DMS sea surface concentration and DMS f lux will decrease 10%between 1861-1890 and 2061-2090. These studies indicate that the interannual variability in global surface seawater DMS concentration is decreasing.

The DMS emissions from continental shelves form a non-negligible proportion of total global DMS emissions (Gypens and Borges, 2014; Jian et al.,2018; Xu et al., 2019). Although all oceanic continental shelves together represent only 7.5% of total ocean area, the DMS emissions from shelf waters represent about 15.7% of global oceanic DMS emissions (Sciare et al., 2002). River-inf luenced DMS emissions, which are considered the major factor inf luencing DMS emission levels in coastal waters,account for about 10% of all DMS emissions from shelf waters (Sciare et al., 2002). In particular,although the European shelves represent only 0.9% of global ocean area, these shelf waters are associated with 2.3% of global oceanic DMS emissions (Uher,2006). Due to the eff ects of anthropogenic activities,shelf sea regions are areas of high primary production,with concomitantly high surface seawater DMS concentrations (Yang et al., 2000). The Yellow Sea(YS), which is located at the edge of the northwest Pacif ic Ocean, is an important shelf sea in China.Yearly variations in DMS concentration in the YS have been extensively investigated. For example,surface seawater DMS concentration in the YS was greater in 2013 than in 2006 (Yang et al., 2011b,2014); spring DMS concentration in 2006 was also greater than that in 1994 (Yang et al., 2012). Indeed,DMS concentration in the western YS was shown to have increased 2.0-2.8 fold between 2005 and 2012(Li et al., 2016).

It is diffi cult to determine the continuous interannual variations in surface seawater DMS concentrations in the YS based on diff erences in DMS between specif ic years. Thus, to better clarify ecological and environmental changes in the China Sea, it is important to analyze the characteristics and interannual variations of surface seawater DMS with continuous data. Previously, surface seawater DMS concentrations in the East China Sea (ECS) have been calculated using statistical algorithms (Qu et al.,2020a) and a biogeochemical model (Qu et al.,2020b). To perform a similar long-term continuous surface seawater DMS hindcast in the YS, this study aimed to utilize a generalized additive mixed model(GAMM), in conjunction with satellite sea surface temperature (SST) and local modif ied chlorophyll-a(LMC) data, to estimate DMS. The interannual variations and trends in DMS, and the factors inf luencing these trends, are discussed in Section 3.

Fig.1 Study region

2 DATA AND METHODOLOGY

The surface seawater DMS concentrations in the YS from 1998 to 2019 were hindcasted using a 22-year GAMM based on satellite SSTs and locally modif ied Chl-adata. Because DMS could not be estimated with a simple linear model, a GAMM was established with historical in situ data and then used to model the temporal and spatial structures of DMS.The study area covered the entire YS (119°E-127°E,23°N-40°N), as shown in Fig.1.

2.1 SST and Chl- a satellite data

The 1/4° daily optimum interpolation sea surface temperature (OISST, version 2.1) dataset, produced by the Advanced Very High Resolution Radiometer(AVHRR) from the NOAA National Climatic Data Center (NCDC, http://www.ncdc.noaa.gov/oisst),was processed to obtain the monthly mean SST data.In-situ SST observation data (1998-2001) from two Marine Environment Monitoring Stations,Xiaochangshan station (122.45°E, 39.14°N) and Chengshantou station (122.70°E, 37.23°N), were compared with the SST in AVHRR. The correlation coeffi cient between the AVHRR SST and in-situ observation data was 0.97.

Fig.2 Estimated smoothing curves for Chl a (a) and SST (b)

Standard mapped satellite images from the Ocean Colour Climate Change Initiative dataset (OC-CCI,Version 4.2), produced by the European Space Agency(ESA, http://www.esa-oceancolour-cci.org), were used to derive surface Chl-aconcentrations in the YS.The daily LMC in the YS was calculated by establishing a modif ied model based on OC-CCI and in-situ chlorophyll data; in this modif ied model, the inf luence of turbidity was removed (Hao et al., 2019).The correlation coeffi cient between the LMC and in situ observation data was 0.76.

2.2 GAMM of DMS

To investigate the long-term trends in surface seawater DMS concentration, a GAMM of DMS was established based on previously published observational data. This model determines temporal and spatial correlations between the observations and the nested data structures (Wood, 2017). We collected 1 885 surface layer measurements obtained during 48 scientif ic cruises from 2005 to 2017, with the exception of data from 2015; data from this year were used for external validation. In the GAMM, DMS concentration was set as the response variable and all other parameters (e.g., longitude, latitude, month,SST, and Chl-aconcentration) were set as explanatory variables. The GAMM was constructed as follows:

whereαwas a grand mean,eiwas a random intercept,s(·) was a one-dimensional nonlinear function based on a thin plate regression spline, andζwas the residuals of the model. The spatial eff ects on the residuals were considered. The relationships between DMS and Chla, and between DMS and SST, are shown in Fig.2.Chlawas signif icantly correlated with DMS when Chlawas greater than 0.3 μg/L (P＜0.05). DMS increased with SST when SST was less than 22 ° C, but remained stable at SSTs above 22 ° C.

After we established the GAMM, the correlations between all the in-situ DMS data and the simulated DMS data were determined. The correlation coeffi cient between these two datasets was 0.68(Fig.3a). To more precisely evaluate our model, we separated the spatial data into four subregions, divided by 40-m isobaths. We found that the in-situ and simulated data were well-correlated in coastal areas than in central areas: the correlation coeffi cient in the coastal YS was 0.70, while the correlation coeffi cient in the central YS was 0.62. Similarly, the correlation coeffi cient of the coastal ECS (0.71) was greater than that of mid-shelf ECS (0.63). This indicated that our GAMM simulated DMS more accurately in coastal areas, and that the correlation with observed data was acceptable in the YS. To further assess the reliability of the model, external validation was performed using the DMS data collected in 2015; this dataset included 186 data points collected over three seasons (spring,summer, and autumn) in the ECS. The correlation coeffi cient for this external validation dataset was 0.62.

Using our GAMM, we performed a 22-year surface seawater DMS concertation hindcast based on daily satellite SST and Chl-adata, with a spatial grid resolution unif ied to 1/4°.

Fig.3 Validation in the East China Sea

2.3 DMS reconstruction

Due to clouds, rain, or malfunctions, satellite images have gaps. To f ill these missing data in our 22-year DMS concentrations, we used the Data Interpolation Empirical Orthogonal Function(DINEOF) method (Beckers and Rixen, 2003; Alvera-Azcárate et al., 2005; Beckers et al., 2006). The DINEOF is a self-consistent, parameter-free EOFbased method for gappy data f ields, especially for large data matrices with many gaps.

After DINEOF reconstruction of the period 1998-2019, the time series for the monthly percentage anomalies in DMS, Chl-a, and SST levels were obtained as follows:

whereswas the monthly mean value andcwas corresponding monthly climatological value derived by averaging the values for each month over the period 1998-2019.

To clearly demonstrate the interannual variation in DMS, a 24-month low-pass f ilter was applied to remove seasonal signals. Here the low-pass f ilter was used to pass signals with low frequency as well as block with high frequency than a preferred cut-offfrequency. The standard deviation of the 22-year monthly mean DMS concentrations was used as a proxy for the amplitude of interannual variation. A linear regression of time series of the monthly mean DMS variation was calculated; the slope and intercept of the best f it line were calculated using the least squares method. To determine the point at which the linear relationship changed abruptly, the cumulative DMS departure from the mean DMS anomaly was used to delineate trends in DMS variation.Periodogram analysis (Bisina and Azeez, 2017) was used to determine the periodicity of the interannual DMS variation. Finally, the power spectral density was estimated by performing a Fast Fourier Transformation (FFT) on the sampled data.

3 RESULT

3.1 Spatial and seasonal variations in DMS concentrations

Fig.4 Seasonal distributions of surface seawater DMS concentration

The annual mean concentration of surface seawater DMS in the YS was 3.61 nmol/L during the period 1998-2019. The average DMS concentrations in the North Yellow Sea (NYS) and the South Yellow Sea(SYS) were 4.48 and 3.39 nmol/L, respectively. Over the same period (1998-2019), the annual mean DMS concentration in the YS was 3.64 nmol/L. This is higher than the global annual mean DMS sea surface concentration for 1998-2017 (2.26 nmol/L; Lana et al., 2011).

The average DMS concentrations in the spring(Mar.-May), summer (Jun.-Aug.), autumn (Sep.-Nov.), and winter (Dec.-Feb.) were 4.22, 4.88, 3.25,and 2.11 nmol/L, respectively (Table 1). The seasonal and spatial distributions of DMS in the YS, based on monthly mean surface seawater values, are shown in Fig.4. There were signif icant seasonal variations in DMS, with the highest DMS concentrations appearing in the summer, followed by spring, autumn, and winter. Across all seasons, DMS concentrations were greatest in the coastal waters of the NYS coastal water. The maximum DMS concentrations in the nearshore NYS in spring, summer, autumn, and winter were 14.40, 19.33, 8.40, and 6.47 nmol/L,respectively. On average, DMS concentrations were lowest in the southern SYS; DMS concentrations in the southern SYS in spring, summer, autumn, and winter were 2.05, 2.36, 2.26, and 1.11 nmol/L,respectively. The maximum standard deviation of mean nearshore DMS concentration was 5.86 nmol/L,which was greater than the minimum standard deviation in the SYS (0.14 nmol/L). In summary, the seasonal variations in surface seawater DMS concentration were greater in coastal waters than in off shore areas.

Table 1 Annual mean DMS concentration (nmol/L)

3.2 Interannual variation in DMS

The contours in Fig.4 show the amplitudes of interannual variations in DMS. The amplitudes of interannual variations in DMS were generally greater in spring and summer, and lower in autumn and winter. The amplitudes in coastal waters were greater than those in off shore areas. The greatest interannual variations in amplitude were found along the coasts of the Shandong Peninsula, Subei Shoal, and NYS during the summer, and the lowest interannual variations in amplitude were found in the central YS during the autumn and winter. The long-term annual trends and anomalies of the annual DMS means for the period 1998-2019 are shown in Fig.5a. Annual mean DMS increased signif icantly from 1998 to 2012, and then decreased between 2013 and 2019.This indicated that interannual DMS variability varied by region, and that trends in interannual DMS variability were not monotonic.

The gridded turning point of the annual trends in DMS was calculated using cumulative departure(Fig.5b). The results indicated that turning points diff ered signif icantly among regions. Monthly DMS trends changed in 2014 or 2015 in the NYS; in 2017 is the west coastal and central SYS; and in 2012 in Subei Shoal, the southern SYS and the coastal SYS.

Fig.5 Monthly percentage anomalies in DMS in the YS (a) and gridded turning points in DMS trends (b)

To further clarify regional diff erences, four subregions were established to represent the diff erent identif ied turning points in DMS trends: domain A was located in the NYS, domain B was located in the western SYS, domain C was located in the central SYS, and domain D was located in Subei Shoal. To obtain a long-term DMS time series without the interference of the seasonal signal, seasonal signals were removed by applying a 24-month low-pass f ilter to the mean DMS anomaly percentage. The interannual variations in DMS anomaly percentages in domains A, B, C, and D are shown in Fig.6.

Obvious turning points in DMS anomaly trends were observed between 1998 and 2019 in each of the four subregions, consistent with the analysis shown in Fig.5b. In domain A, surface seawater DMS anomalies increased signif icantly until 2015 (a 10% increase from 1998 to 2015), and then remained stable. Similar results were observed in domain B, where DMS anomalies increased 10.46% between 1998 and 2015,but decreased after 2015. However, unlike the insignif icant variation observed in domain A, the DMS anomalies in domain B decreased by approximately 2.69% between 2015 and 2019.

Domain C is the only subregion in which DMS anomalies increased both before and after the turning point (2017), although the increase became more gradual after than the increase in domain A and domain B. In this region, DMS increased 6.44% from 1998 to 2017. The DMS increase may have slowed due to the Yellow Sea Cold Water Mass in this subregion. Domain D in Subei Shoal exhibited the greatest increase in DMS (13.94%) before the turning point; after the turning point, DMS decreased by 4.74%.

All of the trends reported above were statistically signif icant at the 95% conf idence level. Overall, the earlier the DMS-trend turning point occurred in a given YS domain, the faster the DMS concentration increased before the turning point.

Figure 7 shows DMS anomaly trends in the YS before and after each turning point. Based on the turning points in the YS, we subdivided the original DMS anomaly percentage time series into two phases exhibiting inverse trends. Prior to the turning point,DMS anomaly percentages were positive across the entire YS, signifying an upward trend (Fig.7a). The subregions exhibiting larger increases were mainly located in the shallow regions of the YS, especially in Subei Shoal. Conversely, after the turning point, DMS anomaly percentages were negative over most of the YS (with the exception of the central YS), indicating a downward trend (Fig.7b).

Fig.6 Interannual variations in surface seawater DMS anomalies for four chosen domains in the YS

Fig.7 Distributions of the long-term DMS anomaly trends before (a) and after (b) the turning point

To identify periodic patterns in the long-term series of DMS, Chl-a, and SST anomalies, spectrum analyses were performed to estimate the power spectra of the three time series in each of the four chosen domains. All monthly data were f iltered with a 24- to 120-month band-pass f ilter prior to powerspectrum estimation. The power spectral densities ref lected the frequencies of three time series, and decomposed the raw data into diff erent frequencies(Fig.8).

In subregion A, the power spectra of the DMS and SST anomalies peaked at 31, 38, 88, and 106 months,while the power spectrum of the Chl-aanomalies peaked signif icantly at 33, 44, and 88 months. Similar results were observed in domains B and C. In domain A, DMS and SST exhibited similar patterns of interannual variability, with periods of 31, 38, and 88-106 months. However, the patterns of interannual variation in the Chl-aanomalies diff ered from those of DMS and SST in both subregions. Spectrum peaks were identif ied at 31, 48, and 88 months in subregion B, and there was a possible time lag of 2-4 months between Chlaand DMS.

Fig.8 Power spectra of interannual DMS, Chl- a, and SST anomalies in the four subregions

In domain C (central SYS), the patterns of interannual variation in DMS and SST were similar to the 31- and 38-month periods observed in subregions A and B. However, the interannual variation period of DMS was much longer (106 months), and the power spectrum was non-signif icant. There were only two variation periods in Chla, one of approximately 35 months and one of approximately 48 months.

For domain D (Subei Shoal), DMS and Chlaseemed to have similarly long periods of interannual variation. Two signif icant peaks were observed in the power spectra of both variables, one at 48 months(4 years) and a second at 75 months (6.25 years). The SST anomalies in domain D exhibited three periods of interannual variation: 31, 38, and 75 months.

4 DISCUSSION

The seasonal and interannual variability in SST and Chlain the four closed subregions will be discussed in the following subsections. By comparing interannual DMS patterns with those of these two key environmental variables, we will assess whether DMS trends and cycles in each subregion can be explained by changes in SST and Chla.

4.1 Eff ects of Chl a on interannual variation in DMS

Figure 9 shows the zone-averaged interannual variations in DMS, Chl-a, and SST anomaly percentages in the four subregions for the period 1998-2019. The long-term trends of the DMS and Chl-aanomalies were similar: both variables had the same turning point in subregions B, C, and D. In subregion A, the turning point of Chlawas two years before that of DMS. These results indicated that the interannual variability in Chlahad a greater inf luence on interannual trends in DMS than SST in the YS.Correlation analysis identif ied a signif icant positive correlation between DMS and Chlaover the longterm in all four subregions. The coeffi cients of determination (R2) were 0.72, 0.83, 0.67, and 0.94 in subregions A, B, C, and D, respectively. However,SST was poorly correlated with DMS in the YS.Previous studies have identif ied a signif icant correlation between DMS and phytoplankton biomass in the surface seawater of the YS. In addition, based on DMS and Chl-ameasurements taken during a variety of cruises, several studies have reported statistically signif icant positive correlations between DMS and Chla(Yang et al., 2009, 2011b, 2012, 2014,2015).

Fig.9 Anomalies in DMS, Chl a, and SST for four subregions in the YS after the application of a 24-month low-pass f ilter

During the period 1998-2019, Chl-aconcentration increased until the turning point and then decreased(Fig.9). This increase in Chlafrom 1998 to 2012 was consistent with previous results (Yamaguchi et al.,2012; Liu and Wang, 2013). It has been argued that the spatio-temporal variability of Chl-aconcentration in the surface layer is associated with nutrient levels(Wang and Jiao, 1999; Liu et al., 2013b). Both riverine inputs and atmospheric deposition strongly inf luence the biogeochemistry of nutrient species in coastal and shelf oceans (Zhang et al., 2007a). Previous studies have shown that dissolved inorganic nitrogen (DIN)and the N/P ratio along the western coast of domain A increased during the period 1998-2008 (Yang et al.,2018). In domain B, DIN and the N/P ratio also appeared to increase (Wei et al., 2015), concomitant with the increase in Chl-alevels. Due to the greater riverine input from the Guanhe River and Sheyang River in domain D, the DIN and N/P ratio along the Subei Shoal also increased between 1996 and 2014(Li et al., 2017).

In domain C, the increased use of nitrogen fertilizers since 1998 might be partially responsible for an increase in riverine DIN inputs into the YS (Wu et al., 2019). As one of the essential nutrient sources for the central YS (Guo et al., 2006; Zhang et al.,2007b; Yang et al., 2011a; Li et al., 2017), the Kuroshio intrusion might have an important inf luence on the interannual variability of Chla. The variability of Kuroshio onshore intrusion across 200-m isobaths is roughly congruent with the interannual variations in Chl-aconcentration in the central YS (Liu et al.,2014, 2015).

In domains A and B, the period of interannual variability for Chlaand DMS was 88 months (about 7.3 years). Both variables exhibited 48-month (4-year) and 75-month (6.25-year) cycles in domain D.In domain C, Chlaand DMS both peaked at 106 months (about 8.8 years), although these peaks were not signif icant.

4.2 Eff ects of SST on the interannual variation in DMS

Surface seawater DMS concentrations were regulated by a subtle dynamic equilibrium between the rates of DMS production and of DMS consumption.Some studies have shown that SST directly or indirectly physically inf luences DMS concentrations in the surface water. In addition, SST was the principal factor aff ecting the effi ciency of zooplankton grazing,which might represent a potential DMS sink (Yu et al., 2015). Enzyme activity increases with strong zooplankton grazing, and this process aff ects the source of surface seawater DMS (Wolfe et al., 1997).By inf luencing bacterial activity, SST indirectly controls DMS consumption (Zhao, 2007).

SST exhibited similar interannual patterns across all four subregions: the SST anomaly decreased signif icantly between 1998 and 2012, and then increased from 2013 to 2019. These results indicated that DMS and SST were consistently correlated. In subregions A, B, and C, the interannual variations were shown to be 31 and 38 months (about 2.6 and 3.17 years, respectively) during the 1998-2019 period(Fig.8). Additionally, longer periods of interannual variability were identif ied: 88-106 months in subregions A and B, 106 months in subregion C, and 75 months in subregion D. This may explain why SST is the main driver of interannual DMS variability over shorter timescales (2-3 years), while both SST and Chladrive interannual DMS variability over longer timescales (7-8.8 years).

SST anomalies were strongly correlated with the meridional and zonal components of the winter monsoon in the YS during the period 1976-2008. The period of interannual variability of SST was 2-3 years.El Niño-Southern Oscillation (ENSO) events might aff ect wind f ields in the YS, leading to alterations in the pathway and extent of the Yellow Sea Warm Current (Wang et al., 2008). The water temperature in section the Pollution Nagasaki (PN) section was inf luenced by the invasion of Kuroshio and was signif icantly correlated with the Pacif ic Decadal Oscillation index (Hu et al., 2018). Liu et al. (2013a)found that, between 1950 and 2007, patterns of interannual SST variability in the China Sea corresponded to the Niño-3 index, with correlations being the highest when a lag of 10 months was allowed in periods of 2-7 years. In addition, Wang et al. (2020) found that the periodicity of the interannual variability of the SST anomaly was about 2.5 years during the period 1985-2017; allowing a 6-month lag, this variation was well correlated with the Niño-3.4 index. Regular, periodic, strong Pacif ic ENSO events may regulate the interannual variability of SST in the YS (Zhang et al., 2009; Wang et al., 2020).

5 CONCLUSION

In this study, the long-term (1998-2019) trends in surface seawater DMS concentrations in the YS were characterized using a continuous 22-year DMS dataset. This dataset was obtained using a DMS GAMM based on the historical in situ sampling, as well as satellite SST and modif ied Chl-adata.

The DMS concentration in YS surface seawater was 3.61 nmol/L. DMS concentrations in the spring and summer were signif icantly higher than those in the autumn and winter. In the YS, DMS values were highest nearshore and were generally lowest off shore.Beginning in 1998, DMS concentration increased signif icantly in the YS until the turning point. The earlier the turning point occurred, the more quickly DMS concentration increased before the transition and the greater the decrease after the transition. Chlaappears to be the main factor inf luencing interannual patterns of DMS in the YS. Interannual cycles of DMS were under the joint control of Chlaand SST:short-term variations in DMS (2-3 years) were strongly aff ected by SST, while longer term variations(6-8 years) were aff ected by both Chlaand SST. The variability in Chlawas mainly aff ected by anthropogenic eutrophication and natural processes,while SST was mainly associated with natural processes.

6 DATA AVAILABILITY STATEMENT

The data that support the f indings of this study are available from the corresponding author upon reasonable request.

7 ACKNOWLEDGMENT

The authors are grateful to the captain and crew of the R/VDongFangHong2for help and cooperation during the study. We are thankful to Dr. Qiang HAO(Second Institute of Oceanography, Ministry of Natural Resources, China) for providing locally modif ied chlorophyll dataset. We also wish to thank the colleagues in our laboratory for the discussion of this study.