Phytoplankton Size Class in the East China Sea Derived from MODIS Satellite Data

Home , East China Sea

Biogeosciences, 15, 4271–4289, 2018 https://doi.org/10.5194/bg-15-4271-2018 © Author(s) 2018. This work is distributed under the Creative Commons Attribution 4.0 License.

Phytoplankton size class in the East China Sea derived from MODIS satellite data

Hailong Zhang1,2, Shengqiang Wang1,2, Zhongfeng Qiu1,2, Deyong Sun1,2, Joji Ishizaka3, Shaojie Sun4, and Yijun He1,2 1School of Marine Sciences, Nanjing University of Information Science & Technology, Nanjing, Jiangsu, China 2Jiangsu Research Centre for Ocean Survey Technology, NUIST, Nanjing, Jiangsu, China 3Institute for Space-Earth Environmental Research, Nagoya University, Nagoya, Japan 4College of Marine Science, University of South Florida, St. Petersburg, Florida, USA

Correspondence: Zhongfeng Qiu ([email protected])

Received: 28 November 2017 – Discussion started: 5 February 2018 Revised: 28 May 2018 – Accepted: 29 June 2018 – Published: 13 July 2018

Abstract. The distribution and variation of phytoplankton sonal variations of the PSC in the ECS were probably af- size class (PSC) are key to understanding ocean biogeo- fected by a combination of the water column stability, up- chemical processes and ecosystems. Remote sensing of the welling, sea surface temperature, and the Kuroshio. Addi- PSC in the East China Sea (ECS) remains a challenge, al- tionally, human activity and riverine discharge might also though many algorithms have been developed to estimate influence the PSC distribution in the ECS, especially in the PSC. Here based on a local dataset from the ECS, a re- coastal region. gional model was tuned to estimate the PSC from the spectral features of normalized phytoplankton absorption (aph) using a principal component analysis approach. Before applying the refined PSC model to MODIS (Moderate Resolu- 1 Introduction tion Imaging Spectroradiometer) data, reconstructing satellite remote sensing reflectance (Rrs) at 412 and 443 nm was Phytoplankton size class (PSC) is fundamentally important critical through modeling them from Rrs between 469 and for ocean biogeochemical processes and ecosystems, espe- 555 nm using multiple regression analysis. Satellite-derived cially for photosynthesis efficiency (Bouman et al., 2005; PSC results compared well with those derived from pig- Uitz et al., 2008), primary production, and the carbon trans- ment composition, which demonstrated the potential of satel- port (Kiørboe, 1993; Guidi et al., 2009; Hirawake et al., lite ocean color data to estimate PSC distributions in the 2011). Thus, knowledge of the PSC dynamics can contribute ECS from space. Application of the refined PSC model to to the improvement of our understanding of marine ecolog- the reconstructed MODIS data from 2003 to 2016 yielded ical and biogeochemical cycles. The classical size fractions the seasonal distributions of the PSC in the ECS, suggest- of phytoplankton proposed by Sieburth et al. (1978) include ing that the PSC distributions were heterogeneous in both three classes, namely, micro- (> 20 µm), nano- (2–20 µm), temporal and spatial scales. Micro-phytoplankton were dom- and pico-phytoplankton (< 2 µm). Among the methods to inant in coastal waters throughout the year, especially in the measure PSC from water samples, including microscopy Changjiang estuary. For the middle shelf region, the seasonal (Montagnes et al., 1994), the Coulter counter method (Shel- shifts from the dominance of micro- and nano-phytoplankton don and Parsons, 1967), and flow cytometry (Sun et al., in the winter and spring to the dominance of nano- and 2000), pigment concentration by high-performance liquid pico-phytoplankton in the summer and autumn were ob- chromatography (HPLC) is the most systematic and quality- served. Pico-phytoplankton were especially dominant in the controlled method (Van Heukelem and Hooker, 2011). How- Kuroshio region in the spring, summer, and autumn. The sea- ever, these methods are time-consuming and methodologi- cally complex. Furthermore, large spatial and temporal vari-

Published by Copernicus Publications on behalf of the European Geosciences Union. 4272 H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data abilities make it difficult to continuously monitor PSC using that the correlation between the variation patterns of the the field sampling methods. PSC and total Chl a was not valid, and pointed out that the Realistically, satellite ocean color data can provide syn- “abundance-based” methods for estimating PSC were proba- optic observations, which are ideal for investigating PSC at bly not applicable in the ECS. Therefore, Wang et al. (2015) large spatial and temporal scales. In recent years, various al- proposed a model to estimate the PSC in the ECS using gorithms have been designed to estimate PSC using in situ the spectral shape of normalized aph(λ) through principal data and ocean color data on both global and regional scales component analysis (PCA). This model showed good perfor- (IOCCG, 2014). Most algorithms can be partitioned into two mance for estimating the PSC from both in situ measured categories, namely, “abundance-based” and “spectral-based” aph and Rrs. However, this model was developed using a methods. The “abundance-based” methods are based on the field dataset mainly from offshore waters of the ECS and off statistical relationship between phytoplankton size fraction the coast of Japan; more importantly, the Wang et al. (2015) and phytoplankton abundance using measurements such as model has not been implemented in satellite data yet. chlorophyll a concentration (Chl a) (refer to Bracher et al., Therefore, the goals of this study were to (1) refine the 2017, Table 2). These approaches rely on the assumption that Wang et al. (2015) model for regional application in the high and low Chl a waters are dominated by large and small ECS using an extensive dataset covering highly varied wa- phytoplankton, respectively. The “spectral-based” methods ter conditions and various seasons, (2) apply the refined PSC utilize the relationship between the variations in inherent op- model to Moderate Resolution Imaging Spectroradiometer tical properties with changes in the PSC using measurements (MODIS) satellite data, and (3) then preliminarily investigate such as phytoplankton absorption (aph), remote sensing re- previously unknown seasonal and spatial variation patterns flectance (Rrs), and particulate backscattering (bbp) (refer to of the PSC in the ECS. Bracher et al., 2017, Table 2). The East China Sea (ECS) is the base of the marine fish- ery resources in China and is one of the most productive 2 Materials and methods ocean areas in the world (Furuya et al., 1996). Ascertain- ing the distribution of PSC can provide valuable informa- 2.1 Study area and sampling stations tion on the state of the marine ecosystem and primary production in the area. Recent efforts have been focused on in- The East China Sea is one of the largest marginal seas in vestigating the phytoplankton community and size classes the western North Pacific and is bounded by China, Korea, in the ECS and have suggested that the PSC exhibited ob- and Japan (Fig. 1a). Nearly 70 % of the ECS is occupied by a vious spatiotemporal heterogeneity in this region (Li et al., continental shelf shallower than 200 m. Numerous rivers flow 2007; Luan et al., 2007; Jiang et al., 2014). For instance, into the ECS from mainland China, including the Changjiang Chen (2000) investigated the PSC and primary productivity (Yangtze) River which provides nearly 90 % of the riverine in the marginal regions of the southern ECS using field data. discharge to the ECS (Zhang et al., 2007). In addition, the The results showed that the phytoplankton size structure and ECS experiences strong currents and multiple water masses, their contributions to primary production displayed signifi- such as Changjiang diluted water (CDW), shelf mixed wa- cant spatial differences in the shelf waters, upwelling waters, ter, and the Kuroshio (Ichikawa and Beardsley, 2002; Su and and Kuroshio water. Furuya et al. (2003) presented the phy- Yuan, 2005). Here we analyzed the mean shape and coef- toplankton dynamics in the ECS in the spring of 1994 and the ficient of variation (CV) of in situ Rrs(λ) collected in the summer of 1996 using HPLC-derived pigment signatures. A ECS, to better show the ocean color variability in the ECS distinct horizontal heterogeneity in phytoplankton composi- which covers many water types (Fig. 2). The in situ Rrs(λ) tion was observed in the spring, and a “two-layer” distribu- of all samples exhibited large variability in both magnitudes tion of phytoplankton appeared both off and on the shelf in and spectral shapes (Fig. 2a and b). For the samples in the the summer. Liu et al. (2016) used 7-year (2006–2012) field coastal region of Zhejiang (Zhe) and Fujian (Min), all 10 measurements to investigate the seasonal and spatial varia- wavebands showed larger variability in Rrs(λ) magnitude, tions of major phytoplankton groups in the ECS, and found with CV larger than 55 % (Fig. 2c). For the samples in south- that monsoon forcing was a key factor in impacting phyto- ern Jeju, CV varied from 20 to 60 %, with a minimum around plankton dynamics at the seasonal scale. 531 nm and 547 nm (Fig. 2d). Overall, they showed a large Note that previous investigations on the PSC in the ECS dynamic range with significant variability. Because of highly have been conducted based on field observations, which may variable environmental conditions, the ECS exhibits complex not reflect the real variation patterns of PSC. To our knowl- marine biogeochemical processes and ecosystems. edge, no study has attempted to examine the PSC distribu- The field measurements used in this study were collected tions in the ECS at synoptic scales from satellite observa- from approximately 10 cruises over the last decade. These tions. Consequently, the PSC dynamics in the ECS at dif- sampling stations were distributed irregularly in the ECS and ferent spatial and temporal scales and their mechanisms are a few were in the Tsushima Strait (Fig. 1a). The field dataset still poorly understood. In the ECS, Wang et al. (2014) found encompassed various seasons and environmental conditions

Biogeosciences, 15, 4271–4289, 2018 www.biogeosciences.net/15/4271/2018/ H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data 4273

Figure 1. Distribution of in situ and matchup datasets and locations of the selected subareas (black boxes) (a), namely MCJR (mouth area of the Changjiang River), MSR (middle shelf region), and KR (Kuroshio region); locations of sampling stations collected in the North Pacific and North Atlantic oceans from the NASA SeaBASS archive (b); the average satellite Rrs(λ) spectra from 2003 to 2016 in the Kuroshio region, North Pacific Ocean, and North Atlantic Ocean (blue circles in b) (c). Error bars represent standard deviations of the means. of the ocean, including turbid waters in the mouth area of eas were selected based on geographical locations and driv- the Changjiang River, less turbid coastal water, and clear ing forces. Within each subarea, the averages of all valid val- water away from the coast. This field dataset consisted of ues were calculated for further analysis. in situ measured aph(λ), measured Rrs(λ) data, and phytoplankton pigments measured by HPLC. In total, 69 samples 2.2 In situ measurements with synchronous measurements of pigments, aph, and Rrs data, 101 samples with coincident pigments and measured Surface water samples (0–3 m) were collected with Niskin aph data, and 27 samples with only measured Rrs were avail- samplers mounted on a CTD rosette or a clean bucket. These able, and Fig. 1a shows the spatial distribution of samples. water samples were used for measurements of aph(λ) and The Kuroshio water in our study area suffered from a paucity pigment concentrations. of in situ Rrs data. Hence, in addition to the regional dataset, 227 in situ Rrs samples collected in the North Pacific and 2.2.1 Measurement and analysis of HPLC-derived PSC North Atlantic oceans (Fig. 1b) from the NASA SeaBASS archive were used as a supplementary dataset. The SeaBASS For pigment analysis, seawater samples were filtered onto dataset was only used for algorithm development to recon- 47 mm Whatman GF/F glass fiber filters under gentle pres- struct satellite Rrs data, along with our regional field dataset sure (< 0.01 MPa), and then stored initially on board in (see Sect. 2.4). The average spectral shapes of the 14-year liquid nitrogen (−70 ◦C) for later analysis in the labora- (2003–2016) MODIS Rrs data in the North Pacific and North tory. Briefly, the concentrations of 19 pigments were deter- Atlantic oceans were similar to that in the Kuroshio water mined by reverse-phase HPLC following Van Heukelem and (Fig. 1c). Thus, in situ measured Rrs data collected in the Thomas (2001). To remove measurements with lower pre- North Pacific and North Atlantic oceans were used in the cision, the quality control (QA) process was applied to the present study, although the distribution regions of these data pigment dataset by HPLC according to the rules of Aiken were beyond our study area. et al. (2009). In our study, the diagnostic pigment analysis Meanwhile, three specific subareas were selected for fur- (DPA) was applied to compute the PSC values from HPLC ther investigation in this study, including the mouth area pigment data (hereafter called the HPLC-derived PSC). In of the Changjiang River (MCJR, 122.3–123.5◦ E and 31– brief, the DPA approach uses seven diagnostic pigment con- 32◦ N), middle shelf region (MSR, 123.5–125◦ E and 28– centrations to obtain the HPLC-derived PSC, including fu- ◦ ◦ 0 29 N), and Kuroshio region (KR, 126–127.1 E and 25.2– coxanthin (Cf), peridinin (Cp), 19 -hexanoyloxyfucoxanthin ◦ 0 26.2 N), as marked by black boxes in Fig. 1a. These subar- (Ch), 19 -butanoyloxyfucoxanthin (Cb), alloxanthin (Ca), chlorophyll b (CCb), and zeaxanthin (Cz). The DPA approach www.biogeosciences.net/15/4271/2018/ Biogeosciences, 15, 4271–4289, 2018 4274 H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data

Figure 2. Rrs(λ) spectra at MODIS wavelengths collected in the coastal region of Zhe–Min (a) and southern Jeju (b); mean spectra and coefficient of variation (CV) of Rrs(λ) in the coastal region of Zhe–Min (c) and southern Jeju (d). The CV is derived as the standard deviation (SD) over the mean. was originally proposed by Vidussi et al. (2001), and subse- cording to the formula quently improved by Uitz et al. (2006). In addition, Hirata et X al. (2008) used the improved DPA approach to account for WiPi =1.41Cf + 1.41Cp + 0.60Ca + 0.35Cb the occurrence of CCb in the nano-phytoplankton class, be- + 1.27Ch + 0.86Cz + 1.01CCb. (4) −3 cause CCb was most abundant at high Chl a (> 0.25 mg m ) and was a minor pigment at lower Chl a. Subsequently, 2.2.2 Measurement of aph Brewin et al. (2010) and Hirata et al. (2011) further refined To obtain aph data, we used the quantitative filter technique the DPA approach to account for ambiguity of the Cf signal (QFT) via a series of processes (Mitchell, 1990). Water in diatoms and the occurrence of the Ch signal in picophy- toplankton. In this study, the HPLC-derived PSC was then samples were filtered through 25 mm Whatman GF/F glass given by fiber filters under gentle pressure, and immediately frozen on board in liquid nitrogen. In this study, the “transmittance” X approach was used for the samples collected from southern f = 1.41C + 1.41C / W P , (1) micro f p i i Jeju and the Tsushima Strait (hereafter referred to as dataset- fnano = (0.60Ca + 0.35Cb + 1.01CCb 1). The optical density (OD) values of total particles were X measured using a dual-beam multi-purpose spectrophotome- +x × 1.27Ch)/ WiPi, (2) X ter between 350 and 750 nm at 1 nm resolution. Similarly, fpico = (0.86Cz + y × 1.27Ch)/ WiPi. (3) we measured the OD values of the detritus after extract- ing phytoplankton pigments in methanol for at least 24 h. where fmicro, fnano, and fpico denote the size fractions of Meanwhile, a blank filter saturated with pure seawater was micro-, nano-, and pico-phytoplankton, respectively. x and used as the reference filter. Then, the absorption coefficients y are the proportions of nano- and pico-phytoplankton in of total particles ap(λ) and detritus ad(λ) were calculated Hex, respectively. When Chl a > 0.08 mg m−3, x = 1 and from the corresponding OD values based on a correction y = 0; when Chl a is between 0.001 and 0.08 mg m−3, x = of Cleveland and Weidemann (1993). The “transmittance– P 12.5 Chl a and y = 1–12.5 Chl a. WiPi is the weighted reflectance” approach was performed on the samples col- sum of the seven diagnostic pigments (Uitz et al., 2006), ac- lected from the coastal and offshore regions of Zhejiang, Fu-

Biogeosciences, 15, 4271–4289, 2018 www.biogeosciences.net/15/4271/2018/ H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data 4275 jian, and Jiangsu (hereafter referred to as dataset-2). The op- (25–35◦ N and 118–132◦ E) was extracted from these global tical densities of the total particles, detritus, and reference coverage datasets. These regional Rrs products were pro- filter were obtained in both transmission mode and reflection cessed using the MathWorks MATLAB software to obtain mode between 250 and 850 nm at 1 nm resolution using a the satellite-derived PSC. Additionally, daily Level 2 Rrs data PerkinElmer lamda650s. Then, we converted these OD val- from the MODIS sensor (1 km resolution) were downloaded ues into ap(λ) and ad(λ) values using the method of Tassan from the NASA Ocean Color website. and Ferrari (1995, 2002). Finally, the aph data were obtained Samples were matched to daily Rrs data to assess the ac- as the difference of ap(λ) − ad(λ) at all sampling stations. curacy of satellite-derived aph and PSC results. To ensure the validity of satellite data before the matchup analysis, the 2.2.3 Measurement of Rrs following constraints were applied to the matchup dataset: (1) the matchup dataset only included satellite data with an To obtain Rrs data in dataset-1, the PRR-800/811 was used overpass time window within 5 h before and after the field to measure the vertical profiles of the downwelling irradiance measurements; (2) to reduce the effect of outliers, the median Ed(λ, z) and upwelling radiance Lu(λ, z) at 13 spectral chan- Rrs value for a window of size 3 centered on the sampling sta- nels (380, 412, 443, 465, 490, 510, 532, 555, 565, 589, 625, tion coordinates was defined as satellite Rrs data; (3) negative 665, and 683 nm). The water-leaving radiance Lw(λ) was MODIS Rrs data were eliminated from the matchup analysis. then determined from the profile of Lu(λ, z) (Hirawake et Based on these criteria, 21 satellite matchups with coinci- al., 2011). The above-water surface downwelling irradiance + dent measured Rrs, and 22 satellite matchups with coincident Ed(λ, 0 ) was simultaneously measured by a cosine collec- measured PSC and aph, were available, as shown in Fig. 1a. tor. Then, Rrs(λ) data were calculated as the ratio of Lw(λ) + to Ed(λ, 0 ). For the purpose of consistency with satellite 2.4 Model accuracy assessment observations that characterize the oceanic surface layer, our analysis exclusively considered the near-surface Rrs data. To evaluate the consistency between the derived and mea- For dataset-2, Rrs data were collected under suitable so- sured values, the Pearson correlation coefficient (R), root lar illumination (generally between 09:00 and 15:00 local mean square error (RMSE), and mean absolute percentage time) using an ASD FieldSpec spectroradiometer in the spec- error (MAPE) were used. Statistical assessments were per- tral range of 350–1050 nm with 1.5 nm increments. The ra- formed in log10 space for the phytoplankton absorption co- diance spectra of water, sky and a gray reference panel were efficient and in linear space for the phytoplankton size class. measured following the above-water measurement approach These statistical indicators can be written as v (Mueller et al., 2003). For each of the three targets, 10 spectra u n were collected and then averaged after removing abnormal 1 uX 2 RMSE = t xi, derived − xi, field /xi, field , (6) n spectra. According to the Ocean Optics Protocol (Mueller et i=1 al., 2003), the Rrs(λ) data were obtained as n 1 X MAPE(%) = xi, derived − xi, field /xi, field = − · · n Rrs(λ) Lt γ Lsky / Lp π/ρp , (5) i=1 × 100%, (7) where Lt , Lsky, and Lp correspond to the radiance values measured from the water, sky, and reference panel, respec- where n is the number of samples. xi, derived and xi, field are tively. ρp is the diffuse reflectance of the reference panel the derived and measured data for the i-th sampling station, provided by the manufacturer. γ is the surface Fresnel re- respectively. flectance related to wind speed (2.6–2.8 % for 10 m s−1 wind, 2.5 % for < 5 m s−1 wind, 2.2 % for calm weather) (Tang et 2.5 Modifying the Wang et al. (2015) model for al., 2004). retrieving PSC All Rrs and aph data were resampled at the centers of MODIS wavebands (i.e., 412, 443, 469, 488, 531, 547, 555, Wang et al. (2015) developed a spectral-based PSC model 645, 667, and 678 nm) using the spectral response function to quantify the size fractions of three phytoplankton classes of the MODIS sensor. using the spectral shape of aph(λ) through the PCA approach. Details of the development and parameterization of 2.3 Satellite data the model were described in Wang et al. (2015). In brief, to reduce the biomass effects, the normalized aph(λ) (hereafter std The global standard monthly MODIS remote sensing re- called aph (λ)) was computed by the ratio of aph(λ) to their flectance, chlorophyll a concentration, and sea surface tem- wavelength mean values in the spectral range between 412 std perature (SST) products (Level 3, about 4 km resolution) and 547 nm. Then, the PCA approach was applied to the aph from 2003 to 2016 were provided by the NASA Ocean (λ) to capture the spectral variation in phytoplankton absorp- Color website (http://oceancolor.gsfc.nasa.gov/, last access: tion related to cell size. The input of PCA is a m × N ma- std 1 May 2017). The dataset corresponding to our study area trix constituted of aph (λ), where m and N are the number www.biogeosciences.net/15/4271/2018/ Biogeosciences, 15, 4271–4289, 2018 4276 H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data of input wavelengths and samples, respectively. The output the PSC model. The established parameters and associated R of PCA comprises two terms, i.e., principal component (PC) and RMSE values for each of the fits are shown in Table 1. scores and PC weights (also called loading factors). The PC Figure 3 shows the strong linear relationships between the in std scores were assumed to correlate with the size class. There- situ aph -derived PSC and HPLC-derived results, with R val- fore, the relationships between the size fractions of micro- ues of 0.89, 0.70, and 0.84 and RMSE values of 0.11, 0.11, and pico-phytoplankton and PC scores were established us- and 0.11 for micro-, nano-, and pico-phytoplankton, respec- ing a logistic-type regression model (Hosmer Jr et al., 2013), tively. The samples were close to the 1 : 1 line, with most of as follows: the samples within the ±20 % fraction range. Using 69 measurements of R and associated HPLC- " k !# rs X derived PSC and in situ measured aph, we also examined ft =1/ 1 + exp −β0 − βiSi , i=1 the feasibility of the PSC model for satellite observations by m coupling QAA. First, we used QAA version 5 (QAA_v5) to X std retrieve a from measured R . To assess the performance of Si = wij aph (λj ), (8) ph rs j=1 QAA_v5, negative retrievedaph values were eliminated, and the remainder were compared with the measured values at where f denotes the phytoplankton size fraction (t = micro t all MODIS wavebands (Fig. 4). The retrieved aph values by or pico). β0 and βi are the regression coefficients between QAA_v5 show reasonably good agreement with the in situ f and PC scores. k is the number of PC scores (k = 4 in t measuredaph at short wavelengths from 412 to 547 nm, with this study). wij refers to the loading factor for the i-th PC. high R values and low RMSE and MAPE values. In con- m is the number of wavelengths. Similar to previous studies trast, the performance of QAA_v5 was poor and produced (Brewin et al., 2010; Hirata et al., 2011), f was calcu- nano large overestimation of aph at long wavelengths, especially at lated as 1-fmicro-fpico, by considering that the sum of three 645, 667, and 678 nm, consistent with previous findings (Lee phytoplankton size fractions was 1. et al., 2014; Tiwari and Shanmugam, 2014). These results The aph(λ) at MODIS wavelengths were derived from clearly demonstrated that QAA_v5 can produce accurate es- R (λ) data using the quasi-analytical algorithm (QAA) pro- rs timates ofaph at 412, 443, 469, 488, 531, and 547 nm. There- posed by Lee et al. (2002). QAA was used in this study before, only aph values at these bands were used to calibrate cause it does not suppose a fixed shape for aph(λ) (Lee et al., the PSC model, as previously stated. Then, the PSC values 2002, 2009). Because QAA could give satisfactory retrievals were inferred from the retrieved aph using Eq. (8) with the of aph(λ) at the first six MODIS wavebands (i.e., 412, 443, established parameterizations. As shown in Fig. 5, the QAA 469, 488, 531, and 547 nm), as shown later for details, only std aph (λ)-derived PSC values were consistent with the HPLC- aph(λ) data at these wavebands were used for the PSC model derived results, and almost all of the points fell within the development in this study (i.e., m = 6 in Eq. 8). ±20 % fraction range. The R and RMSE values were 0.79 To improve the accuracy of MODIS Rrs(λ) at short wave- and 0.13 for micro-phytoplankton, 0.43 and 0.12 for nano- lengths (see details in Sect. 3.3), the reconstruction approach phytoplankton, and 0.80 and 0.13 for pico-phytoplankton, (Lee et al., 2014; Sun et al., 2015) was used to reconstruct respectively. These results suggested that the refined PSC satellite Rrs(λ) at 412 and 443 nm before applying the refined model for the ECS coupling QAA_v5 is able to accurately PSC model to satellite data. In our study, satellite R (412) rs estimate the PSC from remote sensing reflectance Rrs. and Rrs(443) were quantified as a multivariable linear relationship using R data from 469 to 555 nm, as follows: rs 3.2 Comparison of satellite Rrs with in situ n measurements rc X Rrs (λ) = KiRrs (λi) + K0, (9) i=1 Before applying the PSC model to MODIS data, we as- sessed the accuracy of MODIS R using the synchronous rc rs where Rrs (λ) is the reconstructed Rrs data at wavelength λ in situ measurements. Table 2 shows the statistical results of (412 or 443 nm); Rrs(λi) are the input Rrs data at five MODIS the comparison between satellite Rrs and in situ values for wavebands (λi = 469, 488, 531, 547, and 555 nm);K0 and Ki MODIS wavebands. For MODIS Rrs data, a reasonably good are the coefficients determined from multivariant regression. consistency was found at green and red bands (from 469 to 555 nm), with R values within 0.85–0.97 and MAPE val- 3 Results ues within 14.9–27.25 %. Although the R values were above 0.85, the MAPE values were high (> 54 %) at 645, 667, and 3.1 Regional tuning of the PSC model for the ECS 678 nm. This is probably caused by the lower Rrs values at these bands due to strong absorption of water itself. In addi- Following Wang et al. (2015), Eq. (8) was fitted to 170 pairs tion, a low accuracy was observed at 412 and 443 nm. The of the HPLC-derived PSC and in situ measured aph data us- R values were 0.46 and 0.73, and the MAPE values were ing a nonlinear least-squares fitting procedure for developing 47.33 and 36.90 % at 412 and 443 nm, respectively. The high

Biogeosciences, 15, 4271–4289, 2018 www.biogeosciences.net/15/4271/2018/ H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data 4277

std Figure 3. Comparison between in situ aph (λ)-derived and HPLC-derived PSC for micro- (a), nano- (b), and pico-phytoplankton (c). Dashed lines represent the ±20 % fraction range relative to the 1 : 1 line.

Table 1. Parameter βi values for the PSC model development.

Class NR RMSE β0 β1 β2 β3 β4 Micro 170 0.89 0.11 1.05 3.48 −4.34 −13.09 16.04 Nano 170 0.70 0.11 – – – – – Pico 170 0.84 0.11 −2.56 −1.52 1.24 −25.87 −1.86

noise and low accuracy at these two wavebands were sug- sults indicated that the accuracy of the satellite Rrs data at gested to be caused by the uncertainty of the atmospheric cor- 412 and 443 nm could be improved through reconstruction rection procedures and significant band degradation (Meis- using the selected MODIS wavebands. ter, 2011; Hu et al., 2013). Considering the importance of Rrs(412) and Rrs(443) to the QAA algorithm, the poor ac- 3.4 Validation of satellite-derived aph and PSC with in curacy of satellite Rrs at these bands may introduce uncer- situ measured data tainty into the retrieved aph data, and further increase the uncertainty of satellite-derived PSC. Thus, an accurate assess- rc Based on the above analysis, we used the satellite Rrs (412) ment of satellite-derived PSC requires the improved quality rc and R (443) data rather than original satellite Rrs data to of satellite Rrs data. In this study, the reconstruction approach rs compute aph using QAA_v5. Figure 6 shows the comparison was used to fulfill this objective. rc of the derived aph data from satellite Rrs (hereafter called rc 3.3 Reconstruction of MODIS Rrs data aph) and the derived aph from original satellite Rrs with in situ measurements at the first six MODIS wavebands. Table 5 The reconstruction function (Eq. 9) was applied to the re- summarized their corresponding statistical comparisons, i.e., gional field dataset and SeaBass dataset to obtain the regres- R, RMSE, MAPE, and percentage of valid points (PVP). sion coefficients. The resulting relationships between the in Here PVP is defined as the ratio of the number of positive situ measured and modeled Rrs show strong agreement, with satellite-derived values (n) to the total number of matchups 2 = × rc high R and low RMSE and MAPE values (Table 3). For 412 (N) (as PVP n/N 100 %). For the satellite-derived aph, and 443 nm, the R2 values were close to 1.0, with a signifi- the R values were above 0.80, except at 547 nm (R = 0.69), cance level of p < 0.001. The MAPE values were both lower and were significantly higher than those for the satellite- than 9.0 %. The reconstruction functions with the established derived aph (with most of the values below 0.7). The statis- rc coefficients were applied to the original MODIS Rrs data to tics (RMSE, MAPE, and PVP) for the satellite-derived aph rc rc obtain the reconstructed satellite Rrs (412) and Rrs (443) data. were also generally better than those for the satellite-derived Table 4 shows the comparison of the original satellite Rrs aph. Compared with the satellite-derived aph, the PVP for the rc rc and satellite Rrs data with in situ measured Rrs at 412 and satellite-derived aph significantly increased with an average rc 443 nm. The satellite Rrs data were in better agreement with of 23.48 %. Meanwhile, Fig. 6 also shows that the satellite- rc the in situ measured Rrs data than the original satellite Rrs, derived aph had more valid samples and were more clustered especially at 412 nm. At 412 nm, the values of R, RMSE, and around the 1 : 1 line than the satellite-derived aph. Overall, rc MAPE reached 0.70, 0.0019, and 35.15 % for the satellite Rrs both Table 5 and Fig. 6 indicated that the satellite-derived data, respectively, while these values were 0.46, 0.0026, and aph had poor accuracy and low PVP values, whereas the ac- rc 47.33 % for the original satellite data, respectively. These re- curacy of satellite-derived aph can be significantly improved www.biogeosciences.net/15/4271/2018/ Biogeosciences, 15, 4271–4289, 2018 4278 H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data

Figure 4. Comparison of aph derived from Rrs using QAA_v5 with in situ measured aph data.

Figure 5. Comparisons of the PSC modeled using aph derived from Rrs with the HPLC-derived values for micro- (a), nano- (b), and pico- phytoplankton (c).

rc with more valid samples through the reconstruction of satel- from reconstructed Rrs data and compared it with the HPLC- lite Rrs data. derived values (Fig. 7b). The satellite-derived PSC from re- rc The reﬁned PSC model was applied to the satellite-derived constructed Rrs data agreed well with the HPLC-derived re- aph data from original satellite Rrs to estimate PSC (Fig. 7a). sults. Their R values of 0.68, 0.46, and 0.64 and RMSE It can be seen that the satellite-derived PSC from origi- values of 0.13, 0.13, and 0.19 were observed for micro-, nal satellite Rrs was inconsistent with the HPLC-derived nano-, and pico-phytoplankton, respectively. Almost all of results, showing obvious underestimations and overestima- the samples fell within the ±20 % fraction range, although tions of the retrieved PSC for most of the samples. Their a slight underestimation of pico-phytoplankton size fraction R values were all below 0.27 (Fig. 7a). For comparison, occurred in a few samples. rc we also estimated the PSC from the satellite-derived aph

Biogeosciences, 15, 4271–4289, 2018 www.biogeosciences.net/15/4271/2018/ H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data 4279

Table 2. Results of the matchup comparison of satellite Rrs(λ) with in situ measurements.

Wavelength (nm) 412 443 469 488 531 547 555 645 667 678 N 21 21 21 21 21 21 21 21 21 21 R 0.46 0.73 0.85 0.88 0.95 0.96 0.97 0.90 0.86 0.85 RMSE 0.0026 0.0019 0.0016 0.0016 0.0011 0.0011 0.0012 0.00081 0.00077 0.00079 MAPE (%) 47.33 36.90 27.25 19.92 16.39 14.90 18.41 54.86 91.39 111.3

Table 3. Statistical parameters and coefﬁcients for the algorithm to reconstruct Rrs data.

Wavelengths NR2 RMSE MAPE Constant coefﬁcients Kj , j =0,1,. . . n 412 nm 341 0.99 6.3 × 10−4 8.50 % 4.43 × 10−4, 3.91, −3.19, 0.20, 0.72, −0.69 443 nm 341 0.99 2.2 × 10−4 3.13 % 7.39 × 10−5, 2.50, −1.59, −0.36, 1.22, −0.77

Additionally, to further examine the performance of the 0.066, and 0.53 and RMSE values of 0.2, 0.14, and 0.18 were refined PSC model in our study, our refined PSC model observed for micro-, nano-, and pico-phytoplankton, respec- was compared with another two published PSC models (the tively (Fig. 7c). For the retuned Sun et al. (2017) model, the Brewin et al., 2015, model and the Sun et al., 2017, model) R values were 0.36, −0.042, and 0.5 for micro-, nano-, and (Fig. 7c and d). Here, we regionally tuned these published pico-phytoplankton, when the corresponding RMSE values models using a standard nonlinear least-squares method were 0.25, 0.17, and 0.18, respectively (Fig. 7d). The retuned based on our field dataset collected in the ECS. It should be Brewin et al. (2015) model and the retuned Sun et al. (2017) noted here that the two retuned models were used to better model had relatively poor performance in the ECS. These assess the performance of our refined PSC model only, al- comparison results indicated that the performance of our re- though these “abundance-based” models may not perform fined PSC model using the reconstructed satellite data was well in the ECS (data not shown), as suggested by Wang better than those of the retuned Brewin et al. (2015) model et al. (2014). In this study, the retuned Brewin et al. (2015) and the retuned Sun et al. (2017) model in our study region. rc model for the ECS was expressed as Overall, these results suggested that the use of satellite Rrs could significantly improve the performance of the refined fpico = 0.19 1 − exp(−3.6Chl a) /Chl a, PSC model on satellite observations and yielded reasonable fp,n = 1.0 1 − exp(−1.0Chl a) /Chl a, (10) satellite-derived PSC results, which were better than those fnano = fp, n − fpico and fmicro = 1 − fp, n, derived from original satellite observations. Therefore, we further investigated the spatiotemporal variability of the PSC where f is the sum of the nano- and pico-phytoplankton p, n in the ECS based on satellite-derived products from the re- size fraction. And, the retuned Sun et al. (2017) model for constructed satellite remote sensing reflectance. the ECS was expressed as 3.5 Seasonal distribution patterns of the PSC in the −1 2 0.16 fpico = 0.66Chl a 1 − exp −Chl a × Rrs (680) , ECS −1 2 0.32 fnano = 4.17Chl a 1 − exp −Chl a × Rrs (680) , fmicro = 1 − fnano − fpico, To describe the seasonal variability of the PSC in the ECS, (11) the refined PSC model was applied to 14 years (2003–2016) of MODIS monthly Rrs data to obtain monthly PSC products. where Rrs (680) is the remote sensing reflectance at 680 nm. Then, seasonal composite PSC images were generated by av- The scatter distributions of the satellite-derived PSC using eraging the monthly PSC products over a 3-month period for our refined PSC model were closer to the 1 : 1 line than those each season (Fig. 8a–l). In this study, spring, summer, au- of the other two models. According to the statistical indica- tumn, and winter were defined as March to May, June to Au- tors, our refined PSC model had the best performance, with gust, September to November, and December to February of higher R values and lower RMSE values (Fig. 7b). For the the next year, respectively. Meanwhile, to better understand retuned Brewin et al. (2015) model, the R values of 0.58, the spatiotemporal variations of PSC, we analyzed the sea- www.biogeosciences.net/15/4271/2018/ Biogeosciences, 15, 4271–4289, 2018 4280 H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data

rc Table 4. Comparison of the original satellite Rrs and reconstructed Rrs with in situ measured data at 412 and 443 nm.

original satellite R reconstructed satellite Rrc wavelength N rs rs R RMSE MAPE (%) R RMSE MAPE (%) 412 nm 21 0.46 0.0026 47.33 0.70 0.0019 35.15 443 nm 21 0.73 0.0019 36.90 0.80 0.0017 34.53

Table 5. Results of the matchup comparison for Fig. 6.

satellite-derived arc satellite-derived a wavelength N ph ph R RMSE MAPE (%) PVP (%) R RMSE MAPE (%) PVP (%) 412 nm 22 0.80 0.27 15.35 100 0.58 0.56 58.15 72.73 443 nm 22 0.83 0.23 14.87 100 0.48 0.67 63.95 72.73 469 nm 22 0.85 0.21 11.62 100 0.62 0.55 48.35 81.82 488 nm 22 0.84 0.22 11.52 100 0.78 0.37 31.93 90.91 531 nm 22 0.80 0.31 20.82 86.36 0.86 0.73 45.57 54.55 547 nm 22 0.69 0.43 20.38 90.91 0.69 0.52 35.71 63.64

gion (> 2.0 mg m−3). The tongue-shaped structure extended outward the southeast along the 50 m isobath during autumn and along the 70 m isobath during winter. Seasonal variation patterns of the PSC (Fig. 8a–l) indicated that the phytoplankton size classes in the ECS were heterogeneous in both temporal and spatial scales. Their general distribution patterns were consistent with results reported from ﬁeld measurements by other researchers (Chen, 2000; Furuya et al., 2003; Wang et al., 2015). In the spring (Fig. 8a–c), the higher fmicro values (0.45–0.85) were found on the ECS shelf sea with lower values in offshore waters. Relatively high fnano (0.4–0.6) were clearly observed on offshore shelf and in southern Japan. However, pico- phytoplankton were the dominant size class over the south- eastern ECS (fpico = 0.50–0.75). During summer (Fig. 8d– Figure 6. Comparison of the satellite-derived aph (crosses) and f), the micro-phytoplankton size fractions were still high rc satellite-derived aph (open circles) with in situ measured data at in coastal waters. The high fmicro tongue-shape structure 412, 443, 469, 488, 531, and 547 nm. near the Changjiang Bank extended toward southeast along the 30 m isobath. High nano-phytoplankton proportions occurred in the ECS shelf sea with water depths of 30–200 m. sonal distributions of Chl a in the ECS for four seasons, as The pico-phytoplankton contributions to Chl a were rela- shown in Fig. 8a–d. tively high around the ECS shelf break. Pico-phytoplankton Seasonal distributions of Chl a (Fig. 8a–d) illustrated that represented the most abundant size class in the areas deeper −3 Chl a was higher (0.4–3.0 mg m ) on the ECS shelf than than 200 m (fpico > 0.6), which was similar to the results − in the Kuroshio water (< 0.4 mg m 3), and the Chl a values from the ﬁeld measurements by Chen (2000). In the autumn in the Changjiang River mouth were particularly high (3.0– (Fig. 8g–i), the fmicro remained high in coastal waters and − − 25 mg m 3). During spring, the high Chl a (> 1.0 mg m 3) extended over the area shallower than 50 m isobath. The pro- was found on the ECS shelf, and the tongue-shaped structure portion patterns of nano- and pico-phytoplankton in the au- was unclear because of the increase in Chl a in the surround- tumn were broadly similar to those in the summer. However, − ing areas. During summer, Chl a values above 1.0 mg m 3 high nano-phytoplankton proportions were also in the north- were observed in the coastal region. The higher Chl a ern Japan. In the winter (Fig. 8j–l), high fmicro were mainly −3 (> 3.0 mg m ) was limited to the regions at depths shal- distributed on the ECS shelf. The regions with higher fmicro lower than 30 m isobath, including the Changjiang mouth. (> 0.5) extended outward, and connected to the area around In the autumn, the Chl a remained high in the coastal re-

Biogeosciences, 15, 4271–4289, 2018 www.biogeosciences.net/15/4271/2018/ H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data 4281

Throughout the year, micro-phytoplankton comprised 60– 80 % of the Chl a, with the maximum value in April and relatively low fractions from summer to early autumn (June– September). Nano-phytoplankton comprised 18–30 % of the Chl a, while the contributions of pico-phytoplankton to Chl a were below 10 % throughout the year (Fig. 9a). In the MSR, mean Chl a in this region domain was lower than that in the MCJR, with a peak in the spring (April) (Fig. 9b). The micro-phytoplankton proportions were slightly larger than pico-phytoplankton in the winter and spring, while the opposite was found in the summer and autumn. The pico-phytoplankton in the MSR were highest in August and September, with a peak in the summer and early autumn (June–September). Nano-phytoplankton were dominant (40–50 %) for most of the year in this region (Fig. 9b). In contrast to the MCJR and MSR, the mean Chl a was much lower in the Kuroshio region throughout the year (< 0.3 mg m−3) (Fig. 9c). The KR domain showed a predom- inance of pico-phytoplankton (40–90 %) throughout the year, with higher proportions observed in the summer. The nano- phytoplankton proportions (about 40 %) were slightly lower than pico-phytoplankton in the winter and early spring, while their proportions became low (< 20 %) in the rest of the year. The micro-phytoplankton size fractions in the KR remained low (< 23 %) throughout the year (Fig. 9c).

4 Discussion

4.1 Satellite application of the reﬁned PSC model

The most important advantage of satellite ocean color data is Figure 7. Comparison of the HPLC-derived PSC with the satellite- the ability to provide information on the spatiotemporal vari- derived PSC values from the original satellite Rrs (a) and the recon- ability of the PSC. However, remote sensing of the PSC in the rc structed satellite Rrs (b) using the refined PSC model in this study; ECS is still a challenging task, although many “abundance- using the retuned Brewin et al. (2015) model (c); using the retuned based” and “spectral-based” algorithms have been designed Sun et al. (2017) model (d). Solid lines denote the 1 : 1 lines. using field measurements and satellite data in the global scale. Taking into account the optical property in the ECS, Wang et al. (2014) reported that the “abundance-based” ap- the Korean coast. The distributions of the size fractions of proaches are not necessarily applicable in the ECS, and the nano- and pico-phytoplankton were broadly similar to those absorption spectra of phytoplankton could instead be used to in the spring. obtain the PSC in the ECS. More than 80 % of the variability in the spectral shape of phytoplankton absorption was highly 3.6 Regional difference in the monthly climatological related to the changes in the size classes (Ciotti et al., 2002; PSC in the ECS Bricaud et al., 2004). Therefore, in this study we refined the Wang et al. (2015) model for deriving PSC in the ECS from Since the East China Sea is extensive, with a number of dif- the spectral variation of aph. However, the application of this ferent environmental conditions and ecosystems, three sub- refined PSC model to original MODIS data has hampered, as areas were selected for further investigation as shown in showed in Fig. 7a. This may be related to the low accuracy of Fig. 1a. Within each of the subareas, this study investi- the MODIS Rrs at 412 and 443 nm (Table 2), which can in- gated averages of the monthly climatological PSC, as well troduce additional uncertainties into the satellite-derived aph as chlorophyll a concentration (Fig. 9). from original Rrs (Fig. 6; Table 5), and thereby affect the es- In the MCJR, higher Chl a was observed throughout the timation accuracy of the satellite-derived PSC (Fig. 7a). To year (> 3.0 mg m−3), and two Chl a peaks occurred in the solve this problem, the multivariable linear relationship was spring (May) and summer (June), respectively (Fig. 9a). employed to reconstruct MODIS Rrs(412) and Rrs(443) val- www.biogeosciences.net/15/4271/2018/ Biogeosciences, 15, 4271–4289, 2018 4282 H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data

Figure 8. Seasonal distributions of the PSC (a–l) and Chl a (a–d, right panel) in the ECS during 2003–2016.

ues using satellite Rrs from 469 to 555 nm. Previous studies 4.2 Spatial and temporal variations of the PSC in the have reported that the use of multiple spectral bands could ECS successfully reconstruct hyperspectral Rrs data (Lee et al., rc 2014; Sun et al., 2015). In our study, the use of satellite Rrs As described in the results Sects. 3.5 and 3.6, the seasonal rc distributions of the PSC and Chl a in the East China Sea improved the accuracy and PVP of the satellite-derived aph data using QAA_v5 (Fig. 6; Table 5), and dramatically im- (Figs. 8 and 9) had great variability spatially and temporally. proved the accuracy of the satellite-derived PSC (Fig. 7b). In general, micro-phytoplankton are favored under environ- The R and RMSE values for all size fractions derived from mental conditions with stronger mixing and high nutrients, the reconstructed satellite Rrs data were 0.7 and 0.15 respec- while pico-phytoplankton are dominant in low-nutrient wa- tively, compared to the values of 0.064 and 0.38 respectively ters (IOCCG, 2014; Lamont et al., 2018). Here, we discussed for those derived from original satellite Rrs data. Overall, the regional-scale characterization of the full seasonal cycle this study successfully estimated the PSC in the ECS from in stellate-derived PSC and Chl a and the related physical the reconstructed MODIS remote sensing reflectance. The and biochemical effects for helping to understand the spa- findings presented here complement recent studies that have tiotemporal variability of the PSC in the ECS. demonstrated that satellite ocean color data can be used to retrieve the PSC in the ECS (Wang et al., 2015; Sun et al., 4.2.1 The coastal region 2017). However, it should be noted that there was no assessment of the credibility of satellite-derived PSC results In the coastal region, such as the coast of Zhejiang and in the Kuroshio waters due to lack of field dataset in this re- Jiangsu (including the mouth area of the Changjiang River), gion, and further investigations focusing on the applicability the combined effects of variable wind forcing, riverine dis- of the reconstruction algorithm and the refined PSC model in charge, and vertical mixing of the water column promote Kuroshio waters and other regions are still required. phytoplankton growth, resulting in high biomass levels and the presence of larger-sized phytoplankton (Zhou et al., 2008; Wang et al., 2014). Seasonal distributions and vari-

Biogeosciences, 15, 4271–4289, 2018 www.biogeosciences.net/15/4271/2018/ H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data 4283

Figure 9. Monthly climatological PSC and Chl a from 2003 to 2016 in the mouth area of the Changjiang River (MCJR) (a), middle shelf region (MSR) (b), and Kuroshio region (KR) (c). Error bars indicate standard deviations of the means. ability of Chl a (Figs. 8a–d and 9a) in the coastal region pre- phytoplankton proportion (Fig. 8d). In the mouth area of sented in this study generally agreed well with the patterns the Changjiang River, a long-lasting summer Chl a maxi- reported by previous studies of satellite Chl a (Yamaguchi et mum form May to August was found. This may be related al., 2012; He et al., 2013). to the enhanced nutrient concentrations from river and estu- In the spring, increased solar radiation and air temperature arine discharges, e.g., the Changjiang River, Qiantangjiang gradually warm up SST, which can reduce the vertical mix- River, and Minjiang River (Guo et al., 2014). Due to an- ing of the water column. At the same time, weak wind stress thropogenic activities such as various agricultural and in- can retain mixing of the water column, which transports nu- dustry activities, nutrient-rich waters discharge into the East trients to the upper layer from the nutrient-rich deep layer China Sea, especially in the summer monsoon rainy season (Behrenfeld et al., 2006; Boyce et al., 2010). Meanwhile, (Siswanto et al., 2008). This is especially evident in higher coastal nutrient transporting to the inner shelf of the ECS Chl a concentrations and higher micro-phytoplankton pro- can be enhanced under the northwesterly wind action (Liu portions observed in the MCJR (Fig. 9a). Meanwhile, the lit- and Wang, 2013). These physical processes can allow phy- toral currents, e.g., the Zhe–Min Coastal Current (ZMCC) toplankton to live longer in the upper euphotic layer in the and Yellow Sea Coastal Current (YSCC), may play a key sufficient nutrient and light conditions (Zhang et al., 2017), role in the transport of nutrient from riverine discharge. Pre- resulting in the spring bloom in the coastal region and in- vious studies reported that much of sediments is transported ner shelf of the ECS (Fig. 8a), consistent with a previous southward along the Zhejiang–Fujian coast by the ZMCC study by Liu et al. (2016) based on the field measurement (Liu et al., 2007). In addition, the coastal region is relatively of Chl a. This phenomenon was also clearly seen in Fig. 9a shallow, and the water body therefore has a weak stratifica- showing the monthly climatological Chl a in the MCJR with tion of water column in the summer. The hydrodynamic in a local maximum in April and May. These enhanced nutri- coastal waters is dominated by the variation of wind-tide- ent conditions favor the presence of micro-phytoplankton in thermohaline circulations (Guan, 1994). These physical con- the Changjing bank and coastal waters (Figs. 8a and 9a) and ditions may lead to the increase in nutrient and thereby influ- nano-phytoplankton in the offshore region (Fig. 8b). This ence the phytoplankton size structure in the coastal region. was consistent with previous studies which showed that the Nano-phytoplankton were found to dominate the inner part large cell sizes, such as diatoms and Prorocentrum dong- of the ECS shelf (Fig. 8e), likely due to the increased nu- haiense, were dominant on the ECS shelf sea in the spring trient concentrations in offshore waters resulting from the (Furuya et al., 2003; Lou and Hu, 2014; Liu et al., 2016). coastal region by strong convection currents. The study of In the summer, the coastal region with shallower than 40 m Yamaguchi et al. (2012) revealed that the CDW takes ap- isobath displayed higher Chl a (Fig. 8b) and higher micro- proximately 2 months to move from the Changjiang River www.biogeosciences.net/15/4271/2018/ Biogeosciences, 15, 4271–4289, 2018 4284 H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data

Figure 10. The scatterplots showing the relationships between the monthly phytoplankton size fractions and SST from 2003 to 2016 for the MCJR (a), MSR (b), and KR (c). mouth to the Tsushima Strait. Therefore, the nutrient supply chain-forming diatoms and dinoflagellates in coastal waters from riverine discharge may be a major controlling factor in throughout the year. the large cell sizes (micro- and nano-phytoplankton) in the coastal region in the summer. These findings were consistent 4.2.2 The middle shelf and shelf break of the ECS with the study of Jiang et al. (2015) based on field investigations who reported that the micro-sized diatoms and di- Similar to the coastal region, the middle shelf of the ECS noflagellates dominated the Changjiang estuary and adjacent exhibited the spring bloom with a peak of Chl a occur- areas in the summer in response to available nutrients. ring in April, and micro-phytoplankton dominance (Fig. 9b), During autumn and winter, as wind stress strengthens and mainly due to mixing processes of the water column in the temperature decreases, convectional mixing of the water col- spring. These results agree with a previous study reported umn increases and the stratification weakens, which bring by Liu et al. (2016) that during springtime, the contribu- nutrients upward from the underlying layer. The mixing pro- tions of dinoflagellates and diatoms (micro) to total Chl a cesses through internal waves, tides, and winds, as well as the were relatively higher in the middle shelf region and par- terrestrial nitrate from runoff, provide a high nutrient condi- ticularly in the river plume. During summer and early au- tion, which promotes the phytoplankton growth and the pres- tumn, due to surface warming and low wind stress, the re- ence of larger-sized phytoplankton, as suggested by Taylor duced mixing and stronger thermal stratification result in less and Joint (1990). Guo et al. (2014) also observed that the ni- nutrient supply to the surface layers. As reported by Guo trate concentrations were high in coastal waters of the ECS et al. (2014), a nitracline formed in the middle shelf water during autumn and winter. Thus, the larger-sized phytoplank- in summer, and no nitracline formed in autumn and win- ton (micro and nano) dominance was clearly observed in the ter due to strong water mixing. This region is also affected coastal region (Fig. 8g–h and j–k), and high Chl a was found by ocean currents carrying warm waters, e.g., the Kuroshio in this region (Fig. 8c and d), consistent with previous stud- Branch Current to the north of Taiwan and the Yellow Sea ies by Guo et al. (2014) and Wang et al. (2014), who sug- Coastal Current (Ichikawa and Beardsley, 2002), which can gested that the most dominant phytoplankton groups were enhance the water column stability. These oligotrophic conditions can favor the presence of pico-phytoplankton. Mean- while, the coastal nutrients are transported to the middle

Biogeosciences, 15, 4271–4289, 2018 www.biogeosciences.net/15/4271/2018/ H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data 4285 shelf region by convection currents, but the nutrient concen- firmed by a significant positive correlation between SST trations in the middle shelf are not as high as those in the and pico-phytoplankton proportions in the KR (Fig. 10c). coastal region. This may be one reason that nano- and pico- In addition, temperature and salinity are implicated as im- phytoplankton size classes overlapped during summer and portant ecological determinants for some small-sized pho- early autumn (Figs. 8e–f and 9b). In the winter, mixing of tosynthetic bacteria, e.g., Prochlorococcus and Synechococ- the water column increases due to strong winds (Guo et al., cus. Prochlorococcus are largely confined to the warm waters 2014), allowing nutrients to enter the surface layer. This con- and almost absent in coastal waters in the winter (Jiao et al., dition favors the increase in micro- and nano-phytoplankton 2005). Some previous studies also showed that for the abun- in the middle shelf of the ECS (Fig. 9b). Previous studies dant Prochlorococcus in surface waters, its lower boundaries have shown that micro-phytoplankton dominated the shelf of temperature and salinity were 15.6 and 33.5 ◦C in the win- regions in wind-driven upwelling and mixing systems, where ter, respectively, and 26.4 and 29.1 ◦C in the summer, respec- nutrient concentrations are high and seawater temperatures tively (Jiao et al., 2005; Liu et al., 2016). This is particularly are lower (Hirata et al., 2009; Sun et al., 2017; Lamont et al., clear in the Kuroshio region where pico-phytoplankton were 2018). These larger-sized phytoplankton-dominated commu- dominant throughout the year, except in winter and early nities can support higher rates of photosynthesis because of spring when nano-phytoplankton size fractions were slightly their larger photosynthetic rates per unit volume (Hirata et more elevated (Fig. 8b and k; Fig. 9c). The slight increase al., 2009). Additionally, upwelling usually occurs at the shelf in nano-phytoplankton proportions during winter and early break of the ECS, transposing nutrient-rich waters from the spring may be related to the increased nutrient concentrations subsurface layer to the upper layer (Chen et al., 2009). This that result from vertical mixing due to stronger wind stress condition can promote the nano-phytoplankton growth. Ad- during this period, as reported by Liu et al. (2016) that mean −+ − vective processes in the upwelling system are regarded as surface concentrations of nutrients (NO3 NO2 ) in the off- an important force, as well as the biochemical forces such shore Kuroshio region were higher in the winter than in the as nutrients, in controlling the phytoplankton size structure summer, and the mixed-layer depth was much deeper in the and species composition (Smith et al., 1983). Malone (1975) winter than in the summer due to strong vertical mixing in reported that small nano-phytoplankton were selectively re- the winter. moved from upwelling regions by mass transport to the dis- tance as a result of their low sinking rates. 4.3 Response of phytoplankton size class to sea surface temperature 4.2.3 The Kuroshio region and open ocean It has previously been suggested that sea surface tempera- In comparison to the ECS shelf sea, the Kuroshio region ture is one of the important factors that influence the PSC and open ocean generally exhibited relatively low chloro- dynamic (Chen, 2000; Barnes et al., 2010; IOCCG, 2014). phyll a concentrations (Figs. 8a–d and 9c). These phyto- Based on the 14-year (2003–2016) time series of the monthly plankton biomass levels are controlled by a variety of forc- SST and satellite-derived PSC data, we investigated the cor- ing factors, among which the key factors are water column relations between SST and PSC in the three subareas of stability and the availability of light and nutrients (Behren- the ECS (Fig. 10), aiming to discuss the PSC response feld, 2010; Yamaguchi et al., 2012). The Kuroshio region is to the SST change under different hydrological conditions. largely influenced by the Kuroshio, in addition to solar ir- In the Kuroshio region, significant negative correlation be- radiance that governs light availability and also influences tween nano-phytoplankton size fraction and SST was found the water column stability. The mainstream of the Kuroshio (R = −0.66 < −0.5 and p < 0.001), and weak negative cor- strongly flows northeastward along around the 200 m isobath relation was found for micro-phytoplankton (R = −0.31). (Ichikawa and Beardsley, 2002), carrying warm and low- Significant positive correlation between pico-phytoplankton nutrient waters (Jiao et al., 2005). High surface temperature size fraction and SST was identified (R = 0.64 > 0.5 and could strengthen the water column stability, thereby prevent- p < 0.001) (Fig. 10c). Similarly, Chen (2000) reported that ing the nutrient supply to the upper layer from the deeper there was a significant positive correlation between in situ layer (Lovelock, 2007). Thus, these oligotrophic conditions measured pico-phytoplankton proportion and water temper- lead to the low phytoplankton biomass and promote the ature. Several studies have found that surface warming can growth of pico-phytoplankton, which are better adapted to weaken vertical mixing due to the increase in water col- take advantage of such light and nutrient-depleted conditions umn stability (Behrenfeld et al., 2006; Boyce et al., 2010), (Finkel et al., 2009). Liu et al. (2016) found that the impor- which causes less nutrient supply to the surface layers tance of larger phytoplankton (diatoms and dinoflagellates) from the underlying nutrient-rich waters. In addition, the decreased appreciably in the offshore waters, and their con- Kuroshio water is characterized by high salinity, high tem- tributions were partially replaced by small-sized phytoplank- perature, and low nutrients (Jiao et al., 2005). These olig- ton (e.g., Synechococcus, Prochlorococcus, chrysophytes, otrophic conditions favor the presence of smaller-sized phy- and prymnesiophytes). On the other hand, this was also con- toplankton (pico) and restrict the growth of larger-sized phy- www.biogeosciences.net/15/4271/2018/ Biogeosciences, 15, 4271–4289, 2018 4286 H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data toplankton (micro and nano). It offers an explanation to ronmental factors, e.g., wind speed, riverine discharge, and help us understand the correlation between the increasing monsoon forcing. trend of SST and decreasing trend of micro- and nano- phytoplankton size fraction and increasing trend of pico- phytoplankton size fraction. Similar to the KR, a nega- 5 Conclusions tive correlation between micro-phytoplankton proportion and SST (R = −0.76 and p < 0.001) and a positive correlation In this study, the PSC model was regionally tuned for ap- for pico-phytoplankton (R = 0.65 and p < 0.001) were ob- plication to the ECS using extensive in situ measured data served in the MSR (Fig. 10b). Different environmental con- covering various seasons and environmental conditions in ditions in the two subareas showed similar responses of the the ECS. When the refined model was applied to MODIS variability of micro- and pico-phytoplankton size fractions observations, there was a critical step to reconstruct satel- to SST. However, the increasing trend of SST and increasing lite remote sensing reflectance at blue wavebands. It led to trend of nano-phytoplankton showed a weak positive corre- reliable performance of the refined PSC model on MODIS lation (R = 0.34 and p < 0.001) (Fig. 10b), which was differ- observation, which showed good agreement with the HPLC- ent to the Kuroshio region. The weak correlation suggested derived PSC results, with almost all of the samples falling ± that nano-phytoplankton in this region may be affected by within the 20 % fraction range. Along the way, our present other factors (e.g., grazing-nitrogen rate) other than SST (Fu- study preliminarily estimated spatial distributions of the PSC ruya et al., 2003). For instance, Barlow et al. (2016) re- in the ECS from space. The refined PSC model was applied ported that nano-phytoplankton (e.g., flagellates) were dom- to satellite data from MODIS during 2003 and 2016 to in- inant in warmer shelf regions, because they are better at uti- vestigate the PSC distribution at the seasonal scale. The ob- lizing the increase in nutrient concentrations after upwelled tained results showed that the PSC in the ECS varied across water has warmed. In the mouth area of the Changjiang both spatial and temporal scales. The seasonality of the PSC River, the SSTs were negatively (R = −0.59 and p < 0.001), in the ECS was likely to be related to the vertical structure positively (R = 0.58 and p < 0.001), and positively (R = of the water column, upwelling, sea surface temperature, and 0.54 and p < 0.001) correlated with micro-, nano-, and pico- the Kuroshio. It was also affected by riverine discharge and phytoplankton size fraction, respectively (Fig. 10a). The wa- human activity, especially for coastal waters. The interannual ter body in the coastal region mixes well in winter with low and longer-term variations in phytoplankton size class in the SST and has a weak stratification of water column in sum- East China Sea and their mechanisms need to be investigated mer with high SST, as the hydrodynamic in coastal water is in the future. dominated by the variation of wind–tide–thermohaline circulations (Guan, 1994). In some degree, increasing SST could result in a decrease in larger-sized phytoplankton (micro and Data availability. The phytoplankton absorption and size class data are available upon request to the lead author. nano) and an increase in small-sized phytoplankton (pico). However, the trend of rising SST and the increasing nano- phytoplankton size fraction in the MCJR were observed. This Author contributions. HLZ analyzed the data and wrote the may be related to the optimum temperature for the growth manuscript; SQW and ZFQ contributed to the design of this study of different algal groups. Additionally, previous studies have and interpretation of the results; DYS and JI revised the draft; shown that the nutrient structure in the ECS was altered by SJS and YJH provided comments and suggestions to improve the the Changjiang discharge, especially for the Changjiang estu- manuscript. ary and adjacent area (Zhang et al., 2007; Wang et al., 2014). The change in nutrient structure (increase in the N/P ratio) may play an important role in regulating the phytoplankton Competing interests. The authors declare that they have no conflict community structure in the MCJR (Guo et al., 2014). These of interest. results suggested the interannual variability of PSC in coastal waters is more complicated than in offshore waters. The de- tailed study focusing on the mechanism of the PCS change Acknowledgements. This research was jointly supported by the in the ECS is still required. National Key Research and Development Program of China Overall, the correlations between PSC and SST (Fig. 10) (2016YFC1400904), the National Natural Science Foundation of indicated SST is an important factor influencing the PSC dy- China (41506200, 41576172, and 41276186), the Natural Science Foundation of the Jiangsu Higher Education Institutions of China namic in the ECS. The interannual variations of phytoplank- (15KJB170015), the Provincial Natural Science Foundation of ton size classes in the ECS were complicated and could not Jiangsu in China (BK20150914, BK20151526, and BK20161532), be fully explained by the individual factor. Further investi- the National Program on Global Change and Air-sea Interaction gations therefore are required to understand the interannual (GASI-03-03-01-01), the Public Science and Technology Research variability of the PSC in the ECS and its response to envi- Funds Projects of Ocean (201005030), a project funded by “the Priority Academic Program Development of Jiangsu Higher Edu-

Biogeosciences, 15, 4271–4289, 2018 www.biogeosciences.net/15/4271/2018/ H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data 4287 cation Institutions (PAPD)”, the Research and Innovation Project Chen, C., Gong, G., and Jan, S.: Seasonal and episodic effects (dust for College Graduates of Jiangsu Province (KYLX16_0952), and storm) on nutrient dynamics in the upwelling across shelf break the China Scholarship Council. on the East China Sea to the Kuroshio Water, northeastern Tai- wan, AGU Spring Meeting, 24–27 May 2009, 2009. Edited by: Anja Rammig Chen, Y. L. L.: Comparisons of primary productivity and Reviewed by: two anonymous referees phytoplankton size structure in the marginal regions of southern East China Sea, Cont. Shelf Res., 20, 437–458, https://doi.org/10.1016/S0278-4343(99)00080-1, 2000. Ciotti, A. M., Lewis, M. R., and Cullen, J. J.: Assessment References of the relationships between dominant cell size in natural phytoplankton communities and the spectral shape of Aiken, J., Pradhan, Y., Barlow, R., Lavender, S., Poulton, A., the absorption coefficient, Limnol. Oceanogr., 47, 404–417, Holligan, P., and Hardman-Mountford, N.: Phytoplankton pig- https://doi.org/10.4319/lo.2002.47.2.0404, 2002. ments and functional types in the Atlantic Ocean: A decadal Cleveland, J. and Weidemann, A. D.: Quantifying absorp- assessment, 1995–2005, Deep-Sea Res. Pt. II, 56, 899–917, tion by aquatic particles: A multiple scattering correc- https://doi.org/10.1016/j.dsr2.2008.09.017, 2009. tion for glass-filters, Limnol. Oceanogr., 38, 1321–1327, Barlow, R., Gibberd, M., Lamont, T., Aiken, J., and Holligan, P.: https://doi.org/10.4319/lo.1993.38.6.1321, 1993. Chemotaxonomic phytoplankton patterns on the eastern bound- Finkel, Z. V., Beardall, J., Flynn, K. J., Quigg, A., Rees, T. A. ary of the Atlantic Ocean, Deep-Sea Res. Pt. I, 111, 73–78, V., and Raven, J. A.: Phytoplankton in a changing world: cell https://doi.org/10.1016/j.dsr.2016.02.011, 2016. size and elemental stoichiometry, J. Plankt. Res., 32, 119–137, Barnes, C., Irigoien, X., De Oliveira, J. A., Maxwell, D., https://doi.org/10.1093/plankt/fbp098, 2009. and Jennings, S.: Predicting marine phytoplankton commu- Furuya, K., Kurita, K., and Odate, T.: Distribution of phytoplankton nity size structure from empirical relationships with re- in the East China Sea in the winter of 1993, J. Oceanogr., 52, motely sensed variables, J. Plankton Res., 33, 13–24, 323–333, https://doi.org/10.1007/BF02235927, 1996. https://doi.org/10.1093/plankt/fbq088, 2010. Furuya, K., Hayashi, M., Yabushita, Y., and Ishikawa, A.: Phyto- Behrenfeld, M. J., O’Malley, R. T., Siegel, D. A., McClain, C. R., plankton dynamics in the East China Sea in spring and sum- Sarmiento, J. L., Feldman, G. C., Milligan, A. J., Falkowski, mer as revealed by HPLC-derived pigment signatures, Deep- P. G., Letelier, R. M., and Boss, E. S.: Climate-driven trends Sea Res. Pt. II, 50, 367–387, https://doi.org/10.1016/S0967- in contemporary ocean productivity, Nature, 444, 752–755, 0645(02)00460-5, 2003. https://doi.org/10.1038/nature05317, 2006. Guan, B.: Patterns and structures of the currents in Bohai, Huanghai Behrenfeld, M. J.: Abandoning Sverdrup’s critical depth hy- and East China Seas, in: Oceanology of China seas, Springer, pothesis on phytoplankton blooms, Ecology, 91, 977–989, Dordrecht, the Netherlands, 17–26, 1994. https://doi.org/10.1890/09-1207.1, 2010. Guidi, L., Stemmann, L., Jackson, G. A., Ibanez, F., Claustre, H., Bouman, H., Platt, T., Sathyendranath, S., and Stuart, V.: De- Legendre, L., Picheral, M., and Gorskya, G.: Effects of phyto- pendence of light-saturated photosynthesis on temperature and plankton community on production, size, and export of large ag- community structure, Deep-Sea Res. Pt. I, 52, 1284–1299, gregates: A world-ocean analysis, Limnol. Oceanogr., 54, 1951– https://doi.org/10.1016/j.dsr.2005.01.008, 2005. 1963, https://doi.org/10.4319/lo.2009.54.6.1951, 2009. Boyce, D. G., Lewis, M. R., and Worm, B.: Global phyto- Guo, S., Feng, Y., Wang, L., Dai, M., Liu, Z., Bai, Y., and Sun, plankton decline over the past century, Nature, 466, 591–596, J.: Seasonal variation in the phytoplankton community of a https://doi.org/10.1038/nature09268, 2010. continental-shelf sea: the East China Sea, Mar. Ecol. Prog. Ser., Bracher, A., Bouman, H. A., Brewin, R. J., Bricaud, A., Brotas, 516, 103–126, https://doi.org/10.3354/meps10952, 2014. V., Ciotti, A. M., Clementson, L., Devred, E., Di Cicco, A., and He, X., Bai, Y., Pan, D., Chen, C.-T. A., Cheng, Q., Wang, D., and Dutkiewicz, S.: Obtaining phytoplankton diversity from ocean Gong, F.: Satellite views of the seasonal and interannual vari- color: a scientific roadmap for future development, Front. Mar. ability of phytoplankton blooms in the eastern China seas over Sci., 4, 55, https://doi.org/10.3389/fmars.2017.00055, 2017. the past 14 yr (1998–2011), Biogeosciences, 10, 4721–4739, Brewin, R. J., Sathyendranath, S., Hirata, T., Lavender, https://doi.org/10.5194/bg-10-4721-2013, 2013. S. J., Barciela, R. M., and Hardman-Mountford, N. J.: Hirata, T., Aiken, J., Hardman-Mountford, N., Smyth, T. J., and Bar- A three-component model of phytoplankton size class low, R. G.: An absorption model to determine phytoplankton size for the Atlantic Ocean, Ecol. Model., 221, 1472–1483, classes from satellite ocean colour, Remote Sens. Environ., 112, https://doi.org/10.1016/j.ecolmodel.2010.02.014, 2010. 3153–3159, https://doi.org/10.1016/j.rse.2008.03.011, 2008. Brewin, R. J. W., Sathyendranath, S., Jackson, T., Barlow, R., Bro- Hirata, T., Hardman-Mountford, N. J., Barlow, R., Lamont, T., tas, V., Airs, R., and Lamont, T.: Influence of light in the mixed- Brewin, R., Smyth, T., and Aiken, J.: An inherent optical prop- layer on the parameters of a three-component model of phy- erty approach to the estimation of size-specific photosynthetic toplankton size class, Remote Sens. Environ., 168, 437–450, rates in eastern boundary upwelling zones from satellite ocean https://doi.org/10.1016/j.rse.2015.07.004, 2015. colour: an initial assessment, Prog. Oceanogr., 83, 393–397, Bricaud, A., Claustre, H., Ras, J., and Oubelkheir, K.: Natural vari- https://doi.org/10.1016/j.pocean.2009.07.019, 2009. ability of phytoplanktonic absorption in oceanic waters: Influ- Hirata, T., Hardman-Mountford, N. J., Brewin, R. J. W., Aiken, ence of the size structure of algal populations, J. Geophys. Res.- J., Barlow, R., Suzuki, K., Isada, T., Howell, E., Hashioka, T., Oceans, 109, 45–50, https://doi.org/10.1029/2004JC002419, Noguchi-Aita, M., and Yamanaka, Y.: Synoptic relationships be- 2004. www.biogeosciences.net/15/4271/2018/ Biogeosciences, 15, 4271–4289, 2018 4288 H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data

tween surface Chlorophyll a and diagnostic pigments specific Liu, D. and Wang, Y.: Trends of satellite derived chloro- to phytoplankton functional types, Biogeosciences, 8, 311–327, phyll a (1997–2011) in the Bohai and Yellow Seas, https://doi.org/10.5194/bg-8-311-2011, 2011. China: Effects of bathymetry on seasonal and inter- Hirawake, T., Takao, S., Horimoto, N., Ishimaru, T., Yamaguchi, annual patterns, Prog. Oceanogr., 116, 154–166, Y., and Fukuchi, M.: A phytoplankton absorption-based primary https://doi.org/10.1016/j.pocean.2013.07.003, 2013. productivity model for remote sensing in the Southern Ocean, Liu, J., Xu, K., Li, A. E. A., Milliman, J., Velozzi, D., Xiao, Polar Biol., 34, 291–302, https://doi.org/10.1007/s00300-010- S., and Yang, Z.: Flux and fate of Yangtze River sediment de- 0949-y, 2011. livered to the East China Sea, Geomorphology, 85, 208–224, Hosmer Jr, D. W., Lemeshow, S., and Sturdivant, R. X.: Applied https://doi.org/10.1016/j.geomorph.2006.03.023, 2007. logistic regression, John Wiley & Sons, Hoboken, New Jersey, Liu, X., Xiao, W., Landry, M. R., Chiang, K. P., Wang, L., and 2013. Huang, B.: Responses of Phytoplankton Communities to Envi- Hu, C., Feng, L., and Lee, Z.: Uncertainties of SeaWiFS and ronmental Variability in the East China Sea, Ecosystems, 19, MODIS remote sensing reflectance: Implications from clear 832–849, https://doi.org/10.1007/s10021-016-9970-5, 2016. water measurements, Remote Sens. Environ., 133, 168–182, Lou, X. and Hu, C.: Diurnal changes of a harmful algal bloom in https://doi.org/10.1016/j.rse.2013.02.012, 2013. the East China Sea: Observations from GOCI, Remote Sens. En- Ichikawa, H. and Beardsley, R. C.: The current system in viron., 140, 562–572, https://doi.org/10.1016/j.rse.2013.09.031, the Yellow and East China Seas, J. Oceanogr., 58, 77–92, 2014. https://doi.org/10.1023/A:1015876701363, 2002. Lovelock, J.: The revenge of gaia: Earth’s climate crisis & the fate IOCCG: Phytoplankton functional types from Space, in: Reports of humanity, Basic Books, New York, 2007. of International Ocean-Colour Coordinating Group, edited by: Luan, Q. S., Sun, J., Song, S. Q., Shen, Z. L., and Sathyendranath, S., no 15, 1–156, Dartmouth, Canada: IOCCG, Yu, Z. M.: Canonical correspondence analysis of sum- 2014. mer phytoplankton community and its environment in the Jiang, Z., Liu, J., Chen, J., Chen, Q., Yan, X., Xuan, J., and Yangtze River Estuary, China, J. Plant Ecol., 31, 445–450, Zeng, J.: Responses of summer phytoplankton community https://doi.org/10.17521/cjpe.2007.0054, 2007 (in Chinese). to drastic environmental changes in the Changjiang (Yangtze Malone, T. C.: Environmental control of phytoplank- River) estuary during the past 50 years, Water Res., 54, 1–11, ton cell size, Limnol. Oceanogr., 20, 490–490, https://doi.org/10.1016/j.watres.2014.01.032, 2014. https://doi.org/10.4319/lo.1973.18.4.0511, 1975. Jiang, Z., Chen, J., Zhou, F., Shou, L., Chen, Q., Tao, B., Yan, Meister, G.: Calibration and characterization adjustments to the X., and Wang, K.: Controlling factors of summer phytoplank- MODIS ocean color band by the OBPG, Proc. MODIS Sci. Team ton community in the Changjiang (Yangtze River) Estuary and Meeting, 18–20 May 2011, 1–23, 2011. adjacent East China Sea shelf, Cont. Shelf Res., 101, 71–84, Mitchell, B. G.: Algorithms for determining the absorption coef- https://doi.org/10.1016/j.csr.2015.04.009, 2015. ficient for aquatic particulates using the quantitative filter tech- Jiao, N., Yang, Y., Hong, N., Ma, Y., Harada, S., Koshikawa, H., nique, Orlando’90, 16–20 April, 137–148, 1990. and Watanabe, M.: Dynamics of autotrophic picoplankton and Montagnes, D. J., Berges, J. A., Harrison, P. J., and Taylor, F.: Esti- heterotrophic bacteria in the East China Sea, Cont. Shelf Res., mating carbon, nitrogen, protein, and chlorophyll a from volume 25, 1265–1279, https://doi.org/10.1016/j.csr.2005.01.002, 2005. in marine phytoplankton, Limnol. Oceanogr., 39, 1044–1060, Kiørboe, T.: Turbulence, phytoplankton cell size, and the struc- https://doi.org/10.4319/lo.1994.39.5.1044, 1994. ture of pelagic food webs, Adv. Mar. Biol., 29, 1–72, Mueller, J. L., Fargion, G. S., Mcclain, C. R., Pegau, S., Zanefeld, J. https://doi.org/10.1016/S0065-2881(08)60129-7, 1993. R. V., Mitchell, B. G., Kahru, M., Wieland, J., and Stramska, M.: Lamont, T., Brewin, R., and Barlow, R.: Seasonal varia- Ocean Optics Protocols for Satellite Ocean Color Sensor Valida- tion in remotely-sensed phytoplankton size structure around tion, Revision 4, Volume IV: Inherent Optical Properties: Instru- southern Africa, Remote Sens. Environ., 204, 617–631, ments, Characterizations, Field Measurements and Data Analysis https://doi.org/10.1016/j.rse.2017.09.038, 2018. Protocols, 2003. Lee, Z., Carder, K. L., and Arnone, R. A.: Deriving inherent op- Sheldon, R. and Parsons, T. R.: A practical manual on the use of the tical properties from water color: a multiband quasi-analytical Coulter Counter in marine research, Coulter Electronics Sales, algorithm for optically deep waters, Appl. Opt., 41, 5755–5772, 1967. https://doi.org/10.1364/AO.41.005755, 2002. Sieburth, J. M., Smetacek, V., and Lenz, J.: Pelagic ecosystem struc- Lee, Z., Lubac, B., Werdell, J., and Arnone, R.: An update of the ture: heterotrophic compartments of the plankton and their rela- quasi-analytical algorithm (QAA_v5), International Ocean Color tionship to plankton size fractions, Limnol. Oceanogr., 23, 1256– Group Software Report, 1–9, 2009. 1263, https://doi.org/10.4319/lo.1978.23.6.1256, 1978. Lee, Z., Shang, S., Hu, C., and Zibordi, G.: Spectral interdepen- Siswanto, E., Nakata, H., Matsuoka, Y., Tanaka, K., Kiyomoto, dence of remote-sensing reflectance and its implications on the Y., Okamura, K., Zhu, J., and Ishizaka, J.: The long-term design of ocean color satellite sensors, Appl. Opt., 53, 3301– freshening and nutrient increases in summer surface water in 3310, https://doi.org/10.1364/AO.53.003301, 2014. the northern East China Sea in relation to Changjiang dis- Li, Y., Li, D. J., Tang, J. L., Wang, Y. M., Liu, Z. G., Ding, P. X., charge variation, J. Geophys. Res.-Oceans, 113, 2601–2613, and He, S. Q.: Phytoplankton Distribution and Variation in the https://doi.org/10.1029/2008JC004812, 2008. Yangtze River Estuary and Its Adjacent Sea, Environ. Sci., 28, Smith, W. O., Heburn, G. W., Barber, R. T., and O’Brien, J. 719–729, 2007 (in Chinese). J.: Regulation of phytoplankton communities by physical pro-

Biogeosciences, 15, 4271–4289, 2018 www.biogeosciences.net/15/4271/2018/ H. Zhang et al.: Phytoplankton size class in the East China Sea derived from MODIS satellite data 4289

cesses in upwelling ecosystems, J. Mar. Res., 41, 539–556, Van Heukelem, L. and Hooker, S.: The importance of a https://doi.org/10.1357/002224083788519777, 1983. quality assurance plan for method validation and min- Su, J. L. and Yuan, Y. L.: Coastal Hydrology in China, Ocean Press, imizing uncertainties in the HPLC analysis of phyto- Beijing, 2005 (in Chinese). plankton pigments, Phytoplankton Pigments, 5, 195–256, Sun, D., Hu, C., Qiu, Z., and Wang, S.: Reconstruction of hyper- https://doi.org/10.1017/CBO9780511732263.009, 2011. spectral reflectance for optically complex turbid inland lakes: test Van Heukelem, L. and Thomas, C. S.: Computer-assisted of a new scheme and implications for inversion algorithms, Opt. high-performance liquid chromatography method develop- Exp., 23, A718–A740, https://doi.org/10.1364/OE.23.00A718, ment with applications to the isolation and analysis of 2015. phytoplankton pigments, J. Chromatogr. A, 910, 31–49, Sun, D., Huan, Y., Qiu, Z., Hu, C., Wang, S., and He, Y.: https://doi.org/10.1016/S0378-4347(00)00603-4, 2001. Remote-Sensing Estimation of Phytoplankton Size Classes Vidussi, F., Claustre, H., Manca, B. B., Luchetta, A., and Marty, From GOCI Satellite Measurements in Bohai Sea and J. C.: Phytoplankton pigment distribution in relation to up- Yellow Sea, J. Geophys. Res.-Oceans, 122, 8309–8325, per thermocline circulation in the eastern Mediterranean Sea https://doi.org/10.1002/2017JC013099, 2017. during winter, J. Geophys. Res.-Oceans, 106, 19939–19956, Sun, S., Lu, J., and Zhang, L.: Applications of flow cytometer https://doi.org/10.1029/1999JC000308, 2001. in ecological studies of nano- and pico-phytoplankton, Chinese Wang, S., Ishizaka, J., Hirawake, T., Watanabe, Y.,Zhu, Y.,Hayashi, Journal of Ecology, 19, 72–78, 2000 (in Chinese). M., and Yoo, S.: Remote estimation of phytoplankton size frac- Tang, J., Tian, G., Wang, X., Wang, X., and Song, Q.: The meth- tions using the spectral shape of light absorption, Optics Exp., ods of water spectra measurement and analysis I: above-water 23, 10301–10318, https://doi.org/10.1364/OE.23.010301, 2015. method, J. Remote Sens., 8, 37–44, 2004. Wang, S. Q., Ishizaka, J., Yamaguchi, H., Tripathy, S. C., Hayashi, Tassan, S. and Ferrari, G. M.: An alternative approach M., Xu, Y. J., Mino, Y., Matsuno, T., Watanabe, Y., and Yoo, S. J.: to absorption measurements of aquatic particles re- Influence of the Changjiang River on the light absorption proper- tained on filters, Limnol. Oceanogr., 40, 1358–1368, ties of phytoplankton from the East China Sea, Biogeosciences, https://doi.org/10.4319/lo.1995.40.8.1358, 1995. 11, 1759–1773, https://doi.org/10.5194/bg-11-1759-2014, 2014. Tassan, S. and Ferrari, G.: A sensitivity analysis of the Yamaguchi, H., Kim, H. C., Son, Y. B., Sang, W. K., Oka- “Transmittance–Reflectance” method for measuring light ab- mura, K., Kiyomoto, Y., and Ishizaka, J.: Seasonal and sum- sorption by aquatic particles, J. Plankt. Res., 24, 757–774, mer interannual variations of SeaWiFS chlorophyll a in the Yel- https://doi.org/10.1093/plankt/24.8.757, 2002. low Sea and East China Sea, Prog. Oceanogr., 105, 22–29, Taylor, A. H. and Joint, I.: A steady-state analysis of the “mi- https://doi.org/10.1016/j.pocean.2012.04.004, 2012. crobial loop” in stratified systems, Mar. Ecol. Prog., 59, 1–17, Zhang, H., Qiu, Z., Sun, D., Wang, S., and He, Y.: Seasonal https://doi.org/10.3354/meps059001, 1990. and Interannual Variability of Satellite-Derived Chlorophyll a Tiwari, S. P. and Shanmugam, P.: An evaluation of models for (2000–2012) in the Bohai Sea, China, Remote Sensing, 9, 582, the satellite-estimation of phytoplankton absorption coefficients https://doi.org/10.3390/rs9060582, 2017. in coastal/oceanic waters, IEEE J. S. Top. Appl., 7, 364–371, Zhang, J., Liu, S., Ren, J., Wu, Y., and Zhang, G.: Nutrient gra- https://doi.org/10.1109/JSTARS.2013.2252151, 2014. dients from the eutrophic Changjiang (Yangtze River) Estuary Uitz, J., Claustre, H., Morel, A., and Hooker, S. B.: Vertical dis- to the oligotrophic Kuroshio waters and re-evaluation of bud- tribution of phytoplankton communities in open ocean: An as- gets for the East China Sea Shelf, Prog. Oceanogr., 74, 449–478, sessment based on surface chlorophyll, J. Geophys. Res.-Oceans, https://doi.org/10.1016/j.pocean.2007.04.019, 2007. 111, C8005, https://doi.org/10.1029/2005JC003207, 2006. Zhou, M. J., Shen, Z. L., and Yu, R. C.: Responses of a coastal Uitz, J., Huot, Y., Bruyant, F., Babin, M., and Claustre, H.: Re- phytoplankton community to increased nutrient input from the lating phytoplankton photophysiological properties to commu- Changjiang (Yangtze) River, Cont. Shelf Res., 28, 1483–1489, nity structure on large scales, Limnol. Oceanogr., 53, 614–630, https://doi.org/10.1016/j.csr.2007.02.009, 2008. https://doi.org/10.4319/lo.2008.53.2.0614, 2008.

www.biogeosciences.net/15/4271/2018/ Biogeosciences, 15, 4271–4289, 2018