Remote silicate supply regulates spring phytoplankton bloom magnitude in the Gulf of Maine

Spring phytoplankton blooms in the Gulf of Maine (GoM) are sensitive to climate‐related local and remote forcing. Nutrient supply through the slope water intrusion has been viewed as critical in regulating the GoM spring blooms, with an assumption that nitrogen is the primary limiting nutrient. In recent years, this paradigm has been challenged, with silicate being recognized as another potential limiting nutrient, but the source of silicate and its associated water mass remain difficult to be determined. In this study, a time series of spring bloom magnitude was constructed using a self‐organizing map algorithm, and then correlated with the fluctuation of water composition in the deep Northeast Channel. The results reveal the importance of silicate supply from previously less‐recognized deep Scotian Shelf Water inflow. This study offers a new hypothesis for spring bloom regulation, providing a better understanding of mechanisms controlling the spring bloom magnitude in the GoM.

The Gulf of Maine (GoM) is a biologically productive shelf ecosystem, receiving water flows from the Scotian Shelf and the northwest Atlantic Ocean (Bigelow 1927;Brown and Irish 1993;Smith et al. 2001). There has been strong evidence suggesting that nutrients in the GoM are supplied by Slope Water intrusion through the Northeast Channel (NEC; Ramp et al. 1985;Townsend 1998;Townsend et al. 2006). The Slope Water entering the GoM involves Warm Slope Water (WSW) from the Gulf Stream and North Atlantic Central Water and Labrador Slope Water (LSW) from the Labrador Current system (Gatien 1976). Both WSW and LSW have higher concentrations of nitrate than silicate: nitrate concentration in WSW is $ 24 μM, which is about 8 μM higher than that in LSW; silicate concentration, however, ranges from 10 to 14 μM in both WSW and LSW (Petrie and Yeats 2000;Townsend et al. 2006). Another important deep-water mass entering the GoM is deep Scotian Shelf Water (dSSW; Hannah et al. 2001). Compared with WSW and LSW, dSSW has a high silicate concentration due to the mixing with terrestrial freshwater and the supply of regenerated silicate from the Scotian Shelf (Townsend et al. 2010). Once entering the GoM, the admixture of three water masses spills into the deep basins ( Fig. 1), and it is brought to the surface by vertical mixing, Ekman upwelling, and convective overturning (Thomas et al. 2003;Rebuck and Townsend 2014;Townsend et al. 2015).
Spring phytoplankton bloom is responsible for a substantial part of the annual primary production and plays an important role in regulating the biogeochemical cycle and the productivity of higher trophic levels. In the GoM, the spring bloom timing and magnitude directly affect multiple food web components, including zooplankton (Durbin et al., 2003;Pershing and Stamieszkin, 2020), larval fish (Platt et al., 2003), and demersal fish (Townsend and Cammen, 1988). Many previous studies have confirmed that diatoms are the dominant phytoplankton group during the spring bloom in the GoM (Townsend et al. 2005(Townsend et al. , 2006Pan et al. 2011;Turner et al. 2021;Zang et al. 2021). Canonically, the intrusion of nitrate-rich Slope Water and the ratio of LSW to WSW are viewed as critical in regulating the nutrient supply and spring bloom magnitude, with an assumption that nitrogen is the primary limiting nutrient (Schlitz and Cohen 1984;Ramp et al. 1985;Townsend 1991;Thomas et al. 2003). More recent studies, however, challenge this viewpoint and suggest the importance of silicate in regulating productivity in the GoM (Townsend et al. 2010;Switzer et al. 2020). Given roughly equal proportions of nitrate and silicate required by diatoms from waters with disproportionally lower concentrations of silicate than nitrate (Brzezinski 1985), silicate could be more limiting than nitrate for diatoms in the GoM. Thus, understanding the variations of silicate supply due to the changing deep-water components is essential in exploring the nutrient regimes in the GoM and its role in the development of the spring bloom (Townsend et al. 2006;Smith et al. 2012). Previous studies have suggested that changing water mass properties and nutrient regimes might influence phytoplankton blooms in the GoM (Thomas et al. 2003;Rebuck and Townsend 2014;Saba et al. 2015), but some key questions remain, including (1) What is the major water mass responsible for the variability of silicate supply during spring blooms? (2) Is remote supply more important than local regeneration of silicate? In this study, we integrate data from a hydrodynamic model, field measurements, and satellite observations from 2001 to 2016, aiming to quantitatively investigate the relationship between the deep-water composition and the spring bloom magnitude in the GoM.

Sea surface chlorophyll, nutrients, and physical variables
Surface chlorophyll data used in this study are the 8-day composite of MODIS-terra chlorophyll products. The L3 data with a 4-km spatial resolution were extracted from 42 N to 44.5 N and 65.5 W to 71 W and interpolated onto a 0.025 Â 0.025 grid (Fig. 1). The chlorophyll data were reconstructed using the Data INterpolation Empirical Orthogonal Function (DINEOF) to fill the spatial gaps (Beckers and Rixen 2003;Alvera-Azc arate et al. 2005, 2007. Readers are referred to the Supporting Information Section S1 for more details about DINEOF. Concentrations of silicate and nitrate at the surface 20 m in the GoM from 2001 to 2016 were extracted from the Gulf of Maine Region Nutrient and Hydrographic Database. The physical fields used in this study are model results of Finite Volume Community Ocean Model-Gulf of Maine Version 3 (FVCOM-GOM3; Chen et al. 2003Chen et al. , 2011Chen et al. , 2021a. The model domain covers the entire GoM and adjacent regions. FVCOM-GOM3 assimilates mooring and ship measurements of temperature, salinity, and current profiles to improve the quality of model outputs (Chen et al. 2009). The results of FVCOM-GOM3 have been calibrated and applied in previous studies (Chen et al. 2011(Chen et al. , 2021bSun et al. 2013Sun et al. , 2016Li et al. 2015;Ji et al. 2017;Zang et al. 2021). Simulated temperature, salinity, and current velocity were interpolated onto the satellite data grid to facilitate further analysis.

Self-organizing map and spring bloom identification
We applied the self-organizing map (SOM) to identify spring bloom from the satellite data. As an effective tool for data clustering, the artificial neural network SOM performs a non-linear projection from the high-dimensional input data onto a low-dimensional grid (Kohonen 1982(Kohonen , 2001Liu and Weisberg 2011). Here we used a flat rectangular lattice, and the size of map was specified as 3 Â 4 following Liu and Weisberg (2005). The node weight vectors were linearly initialized, and the batch training algorithm was used to enhance the training efficiency (Liu et al. 2006(Liu et al. , 2009(Liu et al. , 2016. The chlorophyll data were log 10 normalized to alleviate the influence of coastal water with high chlorophyll (Telszewski et al. 2009). By estimating the Euclidean distance between the satellite data and the SOM patterns, a best matching unit (BMU) with the minimum distance could be found for each frame (Kohonen 2001). Therefore, all the input data were clustered into 12 patterns, and the time series of BMU showed the temporal evolution of these patterns (Fig. 2). The SOM patterns were classified into bloom and non-bloom patterns using the spatial averaged chlorophyll concentration, and those patterns with high values (> 1 mg m À3 ) were defined as bloom patterns. To minimize the impact of highly productive coastal areas, we excluded the nearshore regions with chlorophyll > 5 mg m À3 from the estimation. Since the bloom patterns also existed in the second half year due to the occurrence of fall bloom, only those bloom patterns appearing in the first half year were used in our analysis.

Estimations of deep-water composition in the NEC and dSSW silicate concentration
We employed the temperature-salinity triangle diagram following Mountain (2012) to quantify the proportions of three deep-water masses entering the GoM through the NEC. The spatial mean temperature and salinity in the NEC were based on the FVCOM-GOM3 model results from January to May (1-3 months prior to the spring bloom; Townsend et al. 2015;Feng et al. 2016;Du et al. 2021). The details of this method are described in Mountain (2012) and the Supporting Information Section S2.
Silicate concentration of dSSW in the NEC is essential in estimating the contribution of dSSW to total new silicate flux into the GoM. However, most previous dSSW silicate measurements were conducted on the Scotian Shelf (Townsend et al. 2010), and directly applying it to our study might underestimate dSSW-related silicate influx because regeneration can enhance silicate concentration from the Scotian Shelf to the NEC (Smith et al. 2012). In this study, we estimated mean  dSSW silicate concentration (Si OH ð Þ ssw 4 ) using the following equation: is mean Slope Water silicate concentration (12 μM; Petrie and Yeats 2000;Townsend et al. 2006), and SW% represents the proportion of Slope Water (WSW plus LSW) estimated by the triangle diagram.

SOM patterns of surface chlorophyll
The 12 SOM patterns of surface chlorophyll are shown in the upper panels of Fig. 2, and patterns g, h, j, and k were identified as bloom patterns. Pattern j represented the strongest bloom due to its higher spatial averaged chlorophyll concentration (g: 1.42 mg m À3 ; h: 1.20 mg m À3 ; j: 1.91 mg m À3 ; k: 1.54 mg m À3 ). The time series indicated that the spring bloom began in late March/early April, and the peak (pattern j) mainly occurred in April and early May (Fig. 2). Both the time series of BMU and bloom patterns demonstrated the profound interannual variabilities of the spring bloom magnitude and phenological features.

Interannual variabilities of spring bloom magnitude and deep-water composition
The spring bloom magnitude in each year was represented by the spatiotemporally averaged chlorophyll concentration. The results showed that the bloom magnitude declined from 1.88 mg m À3 in 2001 to 1.29 mg m À3 in 2013, followed by a moderate increase to $ 1.55 mg m À3 until 2016 (Fig. 3). Meanwhile, a sharp decline of the dSSW proportion from 38% to 5% occurred between 2001 and 2013 with a slight increase afterward (Fig. 3a). The mean proportion of dSSW in the NEC from 2001 to 2016 was 23.9%. A significant positive correlation was found between the spring bloom magnitude in the GoM and the proportion of dSSW in the NEC (r = 0.73; Fig. 3a), suggesting a possible link between dSSW and the spring bloom magnitude. The proportions of two slope water masses (LSW and WSW) were negatively correlated (r = À0.90; Fig. 3b), while neither was significantly correlated with the spring bloom magnitude (LSW-bloom: r = À0.32; WSW-bloom: r = À0.01).To reveal the spatial heterogeneity of dSSW's impact on the spring bloom magnitude, we estimated the correlation between the dSSW proportion in the NEC and the spring bloom magnitude in each grid across the entire GoM (Fig. 4). The results showed an overall positive correlation except for the downstream of the Penobscot Bay and the southeastern GoM (Fig. 4a). The significant correlation coefficients (p < 0.05) were mainly distributed along the path of the deep-water inflow (Fig. 4b).

Nutrient regimes in the GoM
The spatial patterns of nitrate and silicate over the top 20 m in spring are shown in Fig. 5a,b. Both nitrate and silicate concentrations were lower in the central GoM and higher nearshore. Since diatoms take up nitrate and silicate in roughly equal proportions (Townsend et al. 2006), residual silicate (silicate minus nitrate) was estimated to reveal the primary limiting nutrient in the GoM (Fig. 5c). Positive residual silicate located mostly in nearshore regions, indicating low nitrate concentration limited phytoplankton growth. In the central GoM, negative residual silicate indicated silicate limitation exceeded that of nitrate. The similar spatial distributions of negative residual silicate (Fig. 5c) and dSSW-spring bloom magnitude correlation coefficients (Fig. 4b) suggested that the spring bloom magnitude in the central GoM was associated with silicate-rich dSSW due to stronger silicate limitation.

Deep silicate flux into the GoM through the NEC
The mean deep silicate flux through the NEC from January to May was estimated using along-channel deep-water flux between 100 and 200 m across the transect S1 (see Fig. 1), and mean silicate concentration of newly entered deep waters (13.33 μM; Townsend et al. 2010). The deep silicate flux into the GoM was $ 2.7 Â 10 8 mol d À1 ($ 4.0 Â 10 10 mol silicate from January to May). Compared with depth-integrated silicate inventory between 0 and 200 m in the GoM ($ 9.9 Â 10 10 mol) estimated using the World Ocean Atlas dataset (Supporting Information Section S3), deep silicate influx through the NEC from January to May accounted for 40% of silicate inventory in the GoM.
We estimated the dSSW silicate flux across the transect S1 using mean silicate concentration of dSSW, along-channel deep-water flux, and the averaged proportion of dSSW in the NEC (23.9%) estimated above. The mean silicate concentration of dSSW was estimated following Eq. 1, and its value was $ 17.56 μM. The silicate flux associated with dSSW was $ 0.8 Â 10 8 mol d À1 , which contributed to $ 30% of total silicate flux through the NEC.

Discussion
The principal nutrient source supporting primary production in the GoM has been generally thought to be via the influx of Slope Water, and previous studies have emphasized the contributions of different Slope Water components in nutrient supply (Schlitz and Cohen 1984;Ramp et al. 1985;Townsend 1991;Thomas et al. 2003). More recent studies, however, find that the increasing importance of silicate-rich dSSW contributes to the variation of nutrient regime in the GoM (Townsend et al. 2015). As a potentially limiting nutrient in spring, silicate concentration in the GoM is influenced by both external deep-water intrusion and internal recycling (Townsend and Thomas 2002;Townsend et al. 2010;Switzer et al. 2020). The concentration of nitrate, another important nutrient for phytoplankton growth, is modulated by the composition of deep-water inflow and bottom sedimentary denitrification (Christensen et al. 1996;Townsend et al. 2015). The decrease of bottom nitrate concentration from the NEC to the central GoM indicates that denitrification is important in nitrate removal (Switzer et al. 2020). Quantitative analysis based on a 3D biogeochemical model suggests that denitrification removes $ 14 Gmols yr À1 (Zhang et al. 2019). Unlike the depletion of nitrate in the deep water, silicate at depth tends to accumulate from the NEC to the GoM (Christensen et al. 1996;Switzer et al. 2020). The increase of silicate concentration along the deep circulation pathways indicates the importance of silicate recycling in fall (Switzer et al. 2020). However, the significant correlation between the dSSW proportion and the spring bloom magnitude found in this study implies that the dSSW intrusion is more important than internal recycling in modulating silicate concentration in the GoM. A possible reason for such a discrepancy is that the dissolution of diatom cells mainly occurs after the spring blooms, which provides a large number of diatom frustules to the deep water via sinking (Townsend and Thomas 2001;Switzer et al. 2020). In addition, it should be noted that silicate supply might not be the only reason for the correlation between dSSW and the spring bloom magnitude. The distinct hydrographic feature of dSSW (cooler and fresher than the Slope Water) might also influence the spring bloom magnitude by modulating overturning and vertical mixing patterns in the GoM (Christensen and Pringle 2012;Cai et al. 2021). Thus, future work should focus more on the synergistic effects of dSSW-related nutrient and hydrographic conditions on spring bloom dynamics.
The decoupling between the spring bloom magnitude and the dSSW proportion downstream of the Penobscot Bay and the southeastern GoM implies different mechanisms regulating the spring bloom magnitude in these two regions. For the region downstream of the Penobscot Bay, freshwater input with high silicate concentration (up to 200 μM) is mixed with coastal waters and transported by Western Maine Coastal Current southwestward along the coast, supplying a significant amount of silicate for spring diatom bloom (Schoudel 1996;Anderson et al. 2008). Therefore, nitrate in this region is the primary limiting nutrient during the spring blooms. In the southeastern GoM, the surface nutrient availability can be directly influenced by the offshore oligotrophic surface water rather than deep-water nutrient (Thomas et al. 2003). Given the significant correlation between the spring bloom magnitude in the GoM and the proportion of dSSW in the NEC, it is thus crucial to understand the drivers of dSSW transport into the GoM from upstream. Previous studies have suggested that the dSSW transport from the Scotian Shelf to the GoM is modulated by wind stress (Schwing 1992;Li et al. 2014;Feng et al. 2016), baroclinic forcing , remote waves (Schwing 1992), and bathymetric variability (Greenberg et al. 1997). The study of Townsend et al. (2015) suggests that the transport of dSSW from the NEC to the Jordan Basin takes about 3 months. If our hypothesized link between the dSSW and the GoM spring bloom remains valid in the future, such a time lag makes it possible to forecast the spring bloom magnitude in the GoM by monitoring the proportion of dSSW in the NEC during winter and early spring. It is worth noting that not all the deep water in the NEC can be transported into the GoM: part of deep-water mass entering the GoM circulates counterclockwise over the Georges Basin and flows out of the GoM on the western side of the NEC (Ramp et al. 1985;Smith et al. 2001). More recent observations suggest sustained deep outflow during winter, implying the decrease in nutrient supply to the GoM (Smith et al. 2012). The contribution of inflow through the Northern Channel (between Browns Bank and Nova Scotia) to external silicate supply into the GoM was also examined here. Both model results and moored observations suggest that the eastward outflow along Browns Bank's northern flank negatively influences silicate influx into the GoM ). Moreover, the water flux into the GoM through the Northern Channel decreases with water depth, and the transport direction of silicate-rich deep water mass (depth > 100 m) is outward in winter and spring (Fig. 4b;Hannah et al. 2001). Thus, external nutrient supply into the GoM through the Northern Channel could be much lower than that through the NEC.
Our results also suggest the need to consider adding silicate as a limiting nutrient for the regional ecosystem simulations. Many Nutrient-Phytoplankton-Zooplankton-Detritus models have a sole limiting nutrient (mostly nitrogen) for simulating phytoplankton dynamics (Fennel et al. 2006;Song et al. 2011;Zhang et al. 2019). Such a simplification, however, should be well justified based on the phytoplankton community structure and the objectives of simulations. If the model focuses on the variability of bloom magnitudes controlled largely by diatoms, having silicate as a limiting nutrient becomes essential. Our analyses provide a strong support for a serious consideration of including silicate as a limiting nutrient, and this is particularly important if the focus is on the diatom bloom.