Centennial decline in North Sea water clarity causes strong delay in phytoplankton bloom timing

With climate warming, a widespread expectation is that events in spring, such as flowering, bird migrations, and insect bursts, will occur earlier because of increasing temperature. At high latitudes, increased ocean temperature is suggested to advance the spring phytoplankton bloom due to earlier stabilization of the water column. However, climate warming is also expected to cause browning in lakes and rivers due to increases in terrestrial greening, ultimately reducing water clarity in coastal areas where freshwater drain. In shallow areas, decreased retention of sediments on the seabed will add to this effect. Both browning and resuspension of sediments imply a reduction of the euphotic zone and Sverdrup's critical depth leading to a delay in the spring bloom, counteracting the effect of increasing temperature. Here, we provide evidence that such a transparency reduction has already taken place in both the deep and shallow areas of the North Sea during the 20th century. A sensitivity analysis using a water column model suggests that the reduced transparency might have caused up to 3 weeks delay in the spring bloom over the last century. This delay stands in contrast to the earlier bloom onset expected from global warming, thus highlighting the importance of including changing water transparency in analyses of phytoplankton phenology and primary production. This appears to be of particular relevance for coastal waters, where increased concentrations of absorbing and scattering substances (sediments, dissolved organic matter) have been suggested to lead to coastal darkening.


| 3947
OPDAL . The spring phenology in terrestrial plants is triggered directly by temperature and photoperiod (Koerner & Basler, 2010), whereas the timing of the phytoplankton spring bloom tends to be related to hydrographic properties (Lindemann & St. John, 2014;Nelson & Smith, 1991;Sverdrup, 1953). During highlatitude winters, incoming light intensity is low and vertical mixing is deep, causing phytoplankton cells to be mixed well below the photic zone, making them light limited rather than nutrient limited (Doney, 2006) although other limitations such as grazing are also involved (Behrenfeld, 2010;Behrenfeld & Boss, 2014). In spring, increasing surface light penetrates deeper in the water column, while at the same time, increasing temperatures and reduced winds stabilize the water column, driving the shoaling of the mixed layer. This eventually leads to an exponential increase in phytoplankton concentration, traditionally referred to as the spring bloom, where the vertically integrated gross primary production exceeds phytoplankton losses due to respiration, grazing, and sinking (Sverdrup, 1953). This classical view is strongly related to hydrographic properties, of which the stratification and water column light attenuation are central. More recent work modifies the simplified scheme of Sverdrup (1953) by highlighting the role of seasonally varying grazing pressure (Behrenfeld, 2010;Behrenfeld & Boss, 2014) and the difference between a "thoroughly mixed top layer" (Sverdrup, 1953) and a turbulent diffusivity rate-driven mixed layer (Franks, 2015;Huisman, Oostveen, & Weissing, 1999;Taylor & Ferrari, 2011). However, regardless of such modifications, under unaltered optical conditions, ocean warming will lead to earlier stratification and consequently to an earlier spring bloom (Behrenfeld et al., 2006;Doney, 2006). Here, we investigate to what extent this expectation might be modulated by increased light attenuation. In the North Sea and Baltic Sea, Secchi disk observations indicate increased light attenuation during the 20th century Fleming-Lehtinen & Laamanen, 2012;Sandén & Håkansson, 1996), causing a compressed euphotic zone .
Here, we have analyzed North Sea Secchi disk data in combination with chlorophyll a concentration estimates. Our results suggest that substances other than phytoplankton have been central to the reduced transparency of the North Sea. A sensitivity analysis using a water column model indicates that the transparency loss may have caused up to 3 weeks delay in spring bloom over the last century.

| MATERIAL S AND ME THODS
Central to our analysis is the relationship between Secchi disk depth (S, m) and optical properties as recently described by Lee, Shang, Du, and Wei (2018) and Lee et al. (2015), where K S , the attenuation coefficient of downwelling irradiance (m −1 ) and Γ is a coupling constant found to be 1.48 (Lee et al., 2018).
Thus, the reciprocal Secchi disk depth (S) is an optical property with unit/m, which enables the estimation of the composite attenuation, K S . Equation (1) deviates from previous Secchi disk theory (Preisendorfer, 1986) in that the beam attenuation coefficient is not part of the denominator. Both theoretical and empirical evidence for Equation (1) are found in Lee et al. (2015Lee et al. ( , 2018. For wavelengths available for photosynthesis (photosynthetically active radiation [PAR], 400-700 nm), we considered the composite light attenuation to be a quasi-inherent optical property that to a first-order approximation is: where K W , K PHY , and K NON-PHY are contributions from clear water, phytoplankton, and other substances (such as suspended particulate inorganic matter and dissolved organic matter), respectively. Given estimates for K S , K W , and K PHY, the contribution from non-phytoplankton substances, K NON-PHY can be approximated by use of Equation (2).

| Secchi disk data
The majority (~93%, n = 9,546) of the Secchi disk measurements in the North Sea (1903-1998, 51°N-61°N, 3.5°W-11°E) were compiled by Aarup (2002) and are available from ICES (https ://www. ices.dk/ocean/ proje ct/secchi). In addition, we utilized Secchi disk measurements recorded in the World Ocean Database (~4%, n = 400) (https ://www.nodc.noaa.gov/OC5/WOD/secchi-dataformat.html) and those collected by Capuzzo et al (~3%, n = 306;Capuzzo & Stephens, 2017;Capuzzo et al., 2015), available from the Cefas database (https ://doi.org/10.14466/ Cefas DataH ub.47). This gave a total of 10,252 Secchi disk measurements in the North Sea in the period 1903-1998. The ICES Secchi disk data were also used in the study by Dupont and Aksnes (2013), and we followed their approach in separating the North Sea into a shallow (bottom depth <100 m) and a deep (bottom depth >100 m) area. In addition, we corrected for the bias originating from systematic change in locations of Secchi disk measurements over time. This bias arose from the fact that certain time periods are sampled at relatively confined and/or different areas compared to other time periods. For example, all Secchi depth measurements for the shallow North Sea between 1931 and 1949 were sampled along the east coast of England, while all those recorded in the deep areas in 1903 are from two single stations in the Norwegian trench. In addition, it has been shown that the later Secchi disk observations were taken on average closer to the coasts and at shallower bottom depths . Consequently, for each of the two areas, we used a generalized additive model (GAM) to construct a time series of annual mean Secchi depth (S t ) for two hypothetical locations (one deep and one shallow) in January throughout the time period (t, 1903-1998). The locations were defined as the point of highest sampling density for the shallow (LAT loc = 54.5°N, LON loc = 7°E) and deep (LAT loc = 58°N, LON loc = 8.5°E) areas of the North Sea. After this, a linear regression model was fitted to describe annual Secchi disk depth (S t ) as a function of time (S t ~ a + m × t). See Supporting Information for a detailed methodology description.

| Phytoplankton and the effect on composite light attenuation
To estimate light attenuation from phytoplankton, we derived the PCI-dataset (Johns, 2019) with phytoplankton color indices (PCI) sampled by continuous plankton recorders (Batten, Clark, et al., 2003) and converted to chlorophyll a concentrations following the approach by McQuatters-Gollop et al. (2007). Due to the relatively low accuracy of the PCI to predict chlorophyll a concentrations (Batten, Walne, Edwards, & Groom, 2003), these data were only used when <50 stations from the ICES-dataset were available; and (c) the CellCount-dataset, which are chlorophyll a concentrations derived from individual cell counts in 1948 and 1912, and consist of >100 stations (>650 samples; Braarud, Gaarder, & Grøntved, 1953;Gran, 1915). Estimation of the overall annual mean chlorophyll a concentration (CHL t , mg/m 3 ) was done the same way as for Secchi disk depth, using a GAM model, but including also a variable for sampling methodology. Having con- The light attenuation from phytoplankton (K PHY ) was approximated from the empirical relationship between K PAR and chlorophyll a concentration according to Morel (1988); where CHL is the chlorophyll a concentration (mg/m 3 ). To represent light attenuation of pure water (K W ), we used observations from Morel

| Water column model
To estimate the effect of reduction in transparency on phytoplankton bloom dynamics, we applied a water column model of phytoplankton growth previously applied by Huisman, Thi, Karl, and Sommeijer (2006) and Urtizberea, Dupont, Rosland, and Aksnes (2013). The model was used to simulate the annual bloom timing dynamics for an area representing the deep location in the North Sea (Figure 1; Figure  , v is the cell sinking speed (m/s), κ is the vertical turbulent diffusivity (m 2 /s), and ε is the fraction of nitrate that is recycled from lost phytoplankton ( Table 1).
The specific phytoplankton growth rate is determined by the most limiting resource, such that where µ max is the maximum specific growth rate, and H N and H I are the half saturation constants for nitrate-and light-limited growth, respectively.
Light intensity in the water column (I z ) at depth (z) is described by where I 0 is the incoming light (µmol photons m −2 s −1 ) provided hourly from the Hybrid Coordinate Ocean Model (HYCOM; Bleck, 2002) for the North Sea and K (m −1 ) is the background light attenuation (K NON-PHY + K W ) as derived from Equations (1)-(3).
A simulation was performed to represent the deep location  (Table 1) to replicate the observed seasonal bloom dynamics (January-December) averaged for the same period , and in the same area. The sensitivity of bloom dynamics to alterations in the non-phytoplankton  Figure S3).

| RE SULTS
Composite light attenuation, K S , estimated by the Secchi disk depth observations according to Equation (1) is shown in Figure 1c. It is evident that K S , through all years, is greater than the sum of the calculated K W and K PHY and that this discrepancy (K NON-PHY ) has increased across the 20th century for both the shallow and deep areas of the North Sea (Figure 1d). These findings suggest that there are other drivers for the observed reduction in transparency than increased phytoplankton concentrations.  (Figure 2c). It is evident that the peak bloom day occurs later in the year as K NON-PHY increase, and that we see a shift in peak bloom by 22 days from early (day 51) to late March (day 73) between scenarios S1903 and S1998 (Figure 2b,c).

| D ISCUSS I ON
Our results suggest that the centennial reduction in water clarity in the North Sea, as observed through Secchi disk measurements, is unlikely to be explained by changes in phytoplankton concentration.
This indicates an increase in other light-attenuating substances such as suspended particulate matter and/or DOM. We estimate a centennial increase in non-phytoplankton light attenuation (K NON-PHY ) from 0.02 m −1 in 1903 to 0.10 m −1 in 1998 in the deep areas of the North Sea, which suggest a delay in peak spring bloom by 22 days according to the idealized water column model. This finding suggests that reduced water clarity shifts bloom timing in a direction opposite to that expected from increased stratification (Behrenfeld et al., 2006;Doney, 2006), which for the North Sea appears to have measurements that the PCI is a coarse and inaccurate proxy for chlorophyll a concentrations . Moreover, the  Figure S2). This adds confidence to our conclusion that factors other than phytoplankton have contributed to the reduction in water clarity in the North Sea. Although this study is not designed to derive an exact change in K NON_PHY , we believe that the direction in which K NON-PHY has changed is correct, and thus also the direction in which the spring bloom likely has shifted during the 1900s.
Several studies have concluded that water clarity in the North Sea  (Evans, Monteith, & Cooper, 2005;Monteith et al., 2007) and centennial (Kritzberg, 2017;Meyer-Jacob, Tolu, Bigler, Yang, & Bindler, 2015) time scales. In addition, climate warming is predicted to increase terrestrial vegetation coverage, causing DOC concentrations in lakes and rivers to increase also in the future (Larsen et al., 2011).
In the Baltic Sea, river runoff is expected to increase by 15% in the next century (Graham, 2004), thus suggesting a positive correlation between higher temperatures and transport of DOC to coastal waters.
In the Norwegian coastal current, evidence for such freshening, which implies coastal water darkening, has been given by Aksnes et al. (2009).
Spatial variation in light absorption is known to be strongly associated with DOM concentrations (Højerslev, Holt, & Aarup, 1996;Kowalczuk, Olszewski, Darecki, & Kaczmarek, 2005;Stedmon et al., 2000), and potential temporal increases in terrestrial DOM load will likely decrease the North Sea transparency, particularly in the deep areas, suggesting increased light attenuation and delayed spring bloom.

ACK N OWLED G EM ENTS
We thank Knut Wiik Vollset and Mikko Heino, Eyvind Aas, and Tom Andersen for providing advice and comments on the statistical modeling, Secchi disk methodology, and writing, respectively. AFO was F I G U R E 3 Predicted phytoplankton response to increased non-chlorophyll light attenuation. In this study, we present evidence suggesting a centennial increase in non-chlorophyll light-attenuating substances in the North Sea. This implies a reduction of the euphotic zone, leading to a delayed, intensified, and prolonged spring bloom. While climate warming is suggested to advance the spring bloom due to earlier shoaling of the mixed layer, it also causes browning in lakes and rivers due to increases in terrestrial greening, ultimately reducing water clarity in downstream coastal areas. These contrasting responses highlight the importance of including water transparency in analyses of phytoplankton phenology and primary production