2.1 Total organic carbon, XRF scanning, ¹³⁷Cs, and PCB analyses

Total organic carbon (TOC) was calculated by the subtraction of total inorganic carbon (TIC) from total carbon (TC) values. TC was analyzed by means of an EA1110 CHN (CE-instruments). TIC was determined by means of a TIC module connected with a Multi EA 2000 CS (Analytik, Jena) elemental analyzer, involving the acidic removal of carbonates from sediment samples and analysis of the CO₂ released in a carrier gas stream (Leipe et al., 2011). Polychlorinated biphenyls (PCBs), ¹³⁷Cs, and XRF data were used to elaborate the age model of core MSM51-2/20 following the approach developed in Moros et al. (2017). The activity of ¹³⁷Cs radionuclide was measured by gamma spectrometry with a Canberra BE3830 broad energy germanium detector (Moros et al., 2017). The analysis of PCBs was performed with gas chromatography coupled to mass spectrometry (Thermo Fisher Scientific Trace DSQ GC-MS) (Schulz-Bull et al., 1995; Moros et al., 2017). The sum content of the most abundant 23 PCB congeners was used here. An ITRAX X-ray fluorescence (XRF) scanner (Cox Analytical Systems) equipped with a Cr tube operated at 30 kV and 30 mA (exposure time of 15 s) was used to estimate the elemental composition of core MSM51-2/20 sediments with a 200 µm resolution. Mn, Fe (expressed as count per second, cps), and the incoherence ∕ coherence ratio (inc ∕ coh; e.g., Chawchai et al., 2015) were used here to help correlate sediment cores EMB1215/7 (Moros et al., 2017) and MSM51-2/20 (this study) together.

2.2 Age model

The age model of core MSM51-2/20 (42 cm; Fig. 2; Table S2) was obtained by correlating time markers defined in sediment core EMB1215/7 (39 cm), retrieved from about the same location (Fig. 1), and published in Moros et al. (2017). Briefly, using an event stratigraphy approach Moros et al. (2017) have assigned dates to the sediment using (1) the early increase in Hg and ²¹⁰Pb_unsupp. that occurred around 1900 CE; (2) the early increase in PCBs in 1935 CE; (3) the beginning and maximum of atom weapons tests in 1953 and 1963, respectively, as recorded by ²⁴¹Am; (4) the ¹³⁷Cs peak related to the Chernobyl nuclear accident in 1986; and (5) the sedimentary chemical signature (Mn, Fe, Hg) related to two major inflows of oxygenated North Sea water into the central Baltic Sea (Major Baltic Sea Inflows; Mohrholz et al., 2015) in 1994 and 2003. We assigned an age of 2016 CE for the core top, i.e., the year of the core recovery, and an extrapolated age of 1860 CE to the core bottom. Estimated error bars for an age model based on an event stratigraphy are ca. 5–10 and 1–2 years for the time markers, respectively, before and after 1960 CE (Kaiser et al., 2018). Linear sedimentation rates were assumed between the time markers. This dating method is the most adapted in marine environments, where the sediment composition can vary between homogenous (low organic carbon content) and laminated (high organic carbon content) over relatively short depth intervals, as is the case for the Baltic Sea and the Black Sea (Moros et al., 2017; Kaiser et al., 2018). The mean sedimentation rate of core MSM51-2/20 increases constantly from ca. 0.2 cm yr⁻¹ before the 1900s to ca. 0.7 cm yr⁻¹ toward the present day, in agreement with published data (Moros et al., 2017). In the upper 1–2 cm of the core, the sedimentation rate drops to 0.2 cm yr⁻¹. This is an artifact related to the evaporation of the water from the fluff layer (>95 % of water) resulting in sediment compaction before core sampling. The mean sedimentation rate of core MSM51-2/20 is thus ca. 0.4 cm yr⁻¹. This means that each sample represents an average of ca. 2–3 years.

Figure 2Age models of short sediment cores EMB1215/7 (grey curves) and MSM51-2/20 (black curves) based on records of pollutants (mercury, Hg; sum of polychlorinated biphenyls, PCBs), radionuclides (unsupported ²¹⁰lead, ²¹⁰Pb_unsupp.; ²⁴¹americium, ²⁴¹Am; ¹³⁷caesium, ¹³⁷Cs), and XRF-scanning elements (Mn in counts per second and the inc ∕ coh ratio). The dashed lines represent core EMB1215/7 depths dated by event stratigraphy (year CE) as published in Moros et al. (2017) (except for the PCBs record) and their equivalent depths in core MSM51-2/20. The data are listed in Table S2.

Download

2.3 Lipid analysis

Sediment trap material (n=18) and sediments (n=118) were extracted using accelerated solvent extraction (Dionex ASE 350) with DCM ∕ MeOH (9 : 1, v : v). After extraction, squalane was added as an internal standard. The apolar lipid fractions were obtained by column chromatography (SiO₂) using hexane as eluent. They were analyzed by gas chromatography–mass spectrometry (GC-MS) using an Agilent Technologies 7890B GC system coupled to a 5977B Mass Selective Detector equipped with a HP-5ms capillary column (30 m × 0.25 mm × 0.25 µm). The oven temperature was programmed from 40 to 320 ^∘C at 8 ^∘C min⁻¹ followed by a 15 min isotherm. GC-MS data were collected in total ion current (TIC) (m∕z 50–600). The apolar fractions were also analyzed using a multichannel TraceUltra gas chromatograph (Thermo Fischer Scientific) equipped with a DB-5ms capillary column (30 m × 0.25 mm × 0.25 µm) and a flame ionization detector (FID). The temperature program was identical to that used for analysis by GC-MS. Peak identification was made by comparison of retention indexes using the two methods. For quantification, the GC-FID response of each lipid was normalized to that of the internal standard and the amount of sediment extracted. The detection limit was estimated to ca. 5 ng g⁻¹. Because 6Me-C_17 : 0 and 7Me-C_17 : 0 monomethyl alkanes elute very close to each other on a DB-5ms capillary column, their total sum (6+7Me-C_17 : 0) was considered preferentially. However, the contents of 6Me-C_17 : 0 and 7Me-C_17 : 0 monomethyl alkanes are also provided. In order to correct for potential effects of lipid preservation, degradation, and/or dilution by terrestrial inputs, the data were normalized to total organic carbon (TOC) in both core MSM51-2/20 and core POS435/10 (TOC data published in Häusler et al., 2017), and the results are expressed as µg gTOC⁻¹. The data are listed in Tables S3, S4, S5, and S7.

2.4 Monitoring data

The phytoplankton data were gathered according to the guidelines for the Baltic Monitoring Program of HELCOM (HELCOM, 1988), which has been only slightly modified over the 38 years of the monitoring program (for regular updates, see HELCOM, 2020). For representative sampling of the euphotic zone, the upper 10 m was considered, in most cases by pooling discrete samples from 1, 2.5, 5, 7.5, and 10 m depth to one integrated sample. Acetic Lugol solution was used as a fixative. The principles of the quantitative and qualitative microscopic analysis of phytoplankton are described by Utermöhl (1958) and Olenina et al. (2006). The species list created by the HELCOM Phytoplankton Expert Group forms the basis of unified species identification and biomass calculation in the Baltic Monitoring Program. The HELCOM monitoring data, including the IOW data, are kept in the ICES database and are freely available (http://www.ices.dk, last access: 1 October 2019). They were downloaded and restructured for our purposes. A total of 530 observations from 14 stations located in the EGB (Fig. 1; Table S1) were used here. The data for the years 2010–2011 are listed in Table S3, while the annual mean and summer (July–August) biomass data for the period 1983–2016 are listed in Table S4.

2.5 Frequency of cyanobacteria accumulations (FCA)

Quantitative estimates of the presence and extent of cyanobacteria blooms by satellite sensors are affected by the frequent cloud cover. In order to normalize the detections of cyanobacteria accumulations to the amount of available cloud-free imagery, Kahru and co-workers (Kahru et al., 2007; Kahru and Elmgren, 2014) introduced the frequency of cyanobacteria accumulations (FCA) index, which normalizes the number of cyanobacteria detections to the number of unobstructed views of a satellite pixel. In 2010 the satellite observations were pooled from the MODIS-Aqua and MODIS-Terra sensors. FCA can be calculated over variable periods. For the characterization of the annual bloom we used a period of 2 months (July–August), and for the characterization of the seasonal dynamics we used 5 d intervals. FCA values for each 1 km² pixel were averaged for the whole EGB region (Fig. 1; Table S3). The FCA data for the years 2010–2011 are listed in Table S3 and for the period 1987–2016 are listed in Table S4. The latter data have been published recently in Kahru et al. (2018).

2.6 Statistics

For the monitoring time interval (1983–2016), the biomass, FCA, and biomarker data (Table S4) were processed without pre-treatment. For comparison with the ca. 2-year resolved 6+7Me-C_17 : 0 record, the annually resolved HadISST1, AMO, and NAO data (Table S5) were smoothed with a two-point adjacent averaging, resampled with a 2-year step, normalized (min-max normalization), and linearly detrended before computing the correlation coefficient using the PAST v3.25 software (Hammer et al., 2001). Considering the different datasets and uncertainties related to the age model of core MSM51-2/20, the correlation analyses were restricted to the period 1870–2006.

4.1 Diazotrophic cyanobacterial blooms in the central Baltic Sea in 2010–2011

The highest fluxes of 6+7Me-C_17 : 0 in July–August 2010 reflect the main cyanobacterial blooming period that is characteristic of the central Baltic Sea as observed in sediment trap material (Fig. 3). The occurrence of 6+7Me-C_17 : 0 fluxes correspond to an abundant to highly abundant presence of both Aphanizomenon sp. and Nodularia spumigena. While Aphanizomenon sp. produces only relatively small amounts of 7Me-C_17 : 0, both 6Me-C_17 : 0 and 7Me-C_17 : 0 are produced in relatively high amounts by Nodularia spumigena (Bauersachs et al., 2017). However, 6Me-C_17 : 0 and 7Me-C_17 : 0 were present in May 2010, despite Nodularia spumigena not being observed. Aphanizomenon sp. is very likely the source of 7Me-C_17 : 0, but the presence of 6Me-C_17 : 0 is somewhat puzzling. The absence of 6+7Me-C_17 : 0 in December 2010 and January 2011 despite Aphanizomenon sp. being observed very likely results from both its low relative amount (mainly only “present”) and its relatively low production of monomethyl alkanes (Bauersachs et al., 2017). The fluxes of 6+7Me-C_17 : 0 are not correlated to the relative amounts of Aphanizomenon sp. (r=0.19; p=0.66; n=8) but are slightly positively correlated to Nodularia spumigena (r=0.53; p=0.28; n=6). Both the biomass data and FCA data suggest that the main phase of the cyanobacterial bloom occurred in July, with a maximum extent of ca. 70 000 km² estimated on the 12 July (Hansson and Öberg, 2010), and decreased in August. Relatively high abundance of cyanobacteria in sediment trap material as late as early September and early October may be related to a delayed signal between the water surface and the sediment trap related to the lower sinking velocities after the main phase of the bloom. Despite this possible delay, the 6+7Me-C_17 : 0 content of the sediment potentially semiquantitatively reflects the summer bloom of Nodularia spumigena, which is usually the most abundant diazotrophic cyanobacterial strain in central Baltic Sea summer blooms (Kononen, 1992; Finni et al., 2001; Wasmund et al., 2018).

4.2 Diazotrophic cyanobacterial blooms in the central Baltic Sea between 1983 and 2016

In order to further constrain 6+7Me-C_17 : 0 as potential cyanobacterial biomarker for the Baltic Sea, the 6+7Me-C_17 : 0 contents of the dated core MSM51-2/20 were compared to both July–August monitoring and satellite-derived FCA data (Fig. 4). Monitoring and satellite (FCA) data are not easy to compare on a long-term trend for two main reasons. First, monitoring data are punctual in space and time, while the satellite-based data used here are integrated on a basin-wide scale. Second, the satellite data used here mostly reflect the surface and near-surface accumulations and not the organisms thriving deeper in the water column (Kahru and Elmgren, 2014), while monitoring data integrate the upper 10 m of the water column. However, when removing the exceptionally high biomass of Nodularia spumigena in 1992 (2149 mg m⁻³; Table S4), the FCA index is significantly correlated to the biomass of Nodularia spumigena (r=0.61; p<0.01; n=27) but not to the biomass of Aphanizomenon sp. (r=0.01; p=0.97; n=27). This can be expected as Nodularia spumigena concentrates at the water surface (Eigemann et al., 2018), whereas Aphanizomenon spp. may accumulate in subsurface water layers (Hajdu et al., 2007). The 6+7Me-C_17 : 0 content is not significantly positively correlated to the FCA index (r=0.08; p=0.71; n=22), the biomass of Nodularia spumigena (r=0.10; p=0.62; n=26), or the biomass of Aphanizomenon sp. ($r=-\mathrm{0.36}$; p=0.07; n=26). This is very likely because each sediment sample integrates 2–3 years and because the age uncertainties of the biomarker-based record are large (± 1–2 years) compared to satellite and monitoring observations. However, considering this age uncertainty, it seems that 6+7Me-C_17 : 0 contents are relatively high (low) when both Nodularia spumigena biomass and the FCA index are high (low), e.g., between the early 1990s and the mid-2000s. Therefore, we suggest that 6+7Me-C_17 : 0 is potentially semiquantitatively reflecting fluctuations in cyanobacterial biomass and accumulations in the EGB, as already indicated by the sediment trap data.

4.3 Diazotrophic cyanobacterial blooms in the central Baltic Sea over the last 140 years

Based on a compilation of published historical records, Finni et al. (2001) could provide a discontinuous record of cyanobacterial open-sea blooms before the beginning of Baltic Sea monitoring programs. In the early 20th century, cyanobacterial accumulations were rare, which may be explained by relatively cool to moderate summers. Blooms occurred abundantly in successive summers between 1924 and 1930. Favorable weather conditions with several warm summers promoted strong blooms in the 1970s (Kononen and Niemi, 1984). These observations are in agreement with the 6+7Me-C_17 : 0 record, which suggests a quasi-absence of diazotrophic cyanobacteria (and more specifically Nodularia spumigena) in the central Baltic Sea between 1860 and 1910, in the late 1940s, the late 1950s, and the mid-1960s but a relatively high abundance in the 1920s–1930s, in the mid-1950s, in the early and late 1960s, and in the early 1970s (Fig. 5a). While it has been suggested that the massive anthropogenic loading of nutrients to the Baltic Sea increased cyanobacterial productivity (Finni et al., 2001; Poutanen and Nikkilä, 2001; Stal et al., 2003; Vahtera et al., 2007), it is interesting to note here that no significant increase in cyanobacteria occurred in the 1950s, when nutrient inputs, eutrophication, and deoxygenation of the Baltic Sea became explicit (Fig. 5b; Zillén et al., 2008; Gustafsson et al., 2012; Savchuk et al., 2012; Carstensen et al., 2014). However, the small increase (ca. 16 %) of riverine phosphorus input to the Baltic Sea may have partly triggered the early increase in cyanobacteria abundance in the 1920s (Fig. 5b). Our data provide the first, continuous record of cyanobacterial activity in the central Baltic Sea at a sub-decadal resolution beyond instrumental data and allows for analyzing its decadal to multi-decadal variability over the last 140 years.

A positive correlation exists between fluctuations in cyanobacterial biomass in the central Baltic Sea, as reconstructed with the 6+7Me-C_17 : 0 proxy and Baltic Sea summer (July–August) sea surface temperature at a decadal to multi-decadal timescale (r=0.25; p<0.05; n=70; Table S6), i.e., higher (lower) cyanobacterial blooms during years with higher (lower) temperature (Fig. 5a–b). Although temperature has been suggested as a key factor that triggers the formation of a bloom (Sellner, 1997), temperature does not affect the euphotic depth (below which cyanobacteria are unable to perform net photosynthesis) or the critical depth (below which the depth-integrated cyanobacterial primary productivity is zero) of the cyanobacterial community (Stal et al., 2003). Instead, changes in solar irradiance reaching the surface and in the angle of the sun affect the critical depth. Indeed, years with more sunshine in July–August tend to have more cyanobacterial accumulation (Kahru et al., 1994). Temperature is, however, an important contributory factor since it causes a stabilization of the water column and decreases the mixing depth, thereby increasing the light irradiance available for the cyanobacteria (Stal et al., 2003). Recently, Kniebusch et al. (2019) and Börgel et al. (2018) found that the Atlantic Multi-decadal Oscillation (AMO; Fig. 5c; Knight et al., 2006) causes changes in Baltic Sea surface temperature. The AMO, which is defined as the alternation of cold and warm phases in North Atlantic surface temperature with a period of 60–90 years, is positively correlated to the 6+7Me-C_17 : 0 (r=0.30; p<0.05; n=70; Table S6). Therefore, the AMO is likely a factor determining the variability of cyanobacterial biomass in the central Baltic Sea at a multi-decadal timescale. The North Atlantic Oscillation (NAO; Hurrell et al., 2003) is another large-scale climate mode that dominates winter climate variability in Europe. At a multi-annual to decadal timescale, the NAO determines, to a large extent, ice cover, surface wind, freshwater runoff, and temperature in the Baltic region (Hänninen et al., 2000; Omstedt and Chen, 2001; Kauker and Meier, 2003). Using computer simulations, Janssen et al. (2004) proposed a cause-and-effect chain between the NAO and cyanobacterial blooms. Positive NAO leads to low ice cover; high wind stress, triggering a deepening of the surface mixed-layer; and an increase in excess dissolved inorganic phosphorus, favoring late summer cyanobacterial blooms during the following summer. However, the 6+7Me-C_17 : 0 record is not correlated to the NAO index over the last 140 years (r=0.10; p=0.43; n=70; Table S6). There is also no correlation between the 6+7Me-C_17 : 0 and the annual riverine P input records ($r=-\mathrm{0.09}$; p=0.47; n=70; Table S6). Therefore, our data suggest that fluctuations in the cyanobacterial biomass in the central Baltic Sea at a decadal to multi-decadal timescale are at least partly related to sea surface temperature changes, which are ultimately triggered by the AMO. However, it has to be considered that the response of cyanobacterial blooms to environmental variables, such as nutrient loading and temperature, are likely nonlinear, which may explain the relatively low correlations. Indeed, a very recent study shows that cyanobacterial blooms are highly correlated to environmental variables at a decadal timescale when considering a set of biogeochemical variables related to the amount of phosphorus and hypoxia in bottom layers, as well as surface water temperature (Kahru et al., 2020).

4.4 Diazotrophic cyanobacterial blooms in the Bothnian Sea during the Holocene

The 6+7Me-C_17 : 0 contents in the Bothnian Sea during the Holocene thermal maximum (HTM; 4500–8000 cal. yr BP sensu lato; Borzenkova et al., 2015) were up to 100-fold higher than in the modern central Baltic Sea (Figs. 4 and 5a). This suggests a high abundance of diazotrophic cyanobacteria (at least regarding Nodularia spumigena) in a region of the Baltic Sea where the diazotrophic cyanobacterial biomass is ca. 4- to 5-fold lower than in the central Baltic Sea in the present day (Wasmund et al., 2018). An abundant presence of cyanobacteria in the central and northern Baltic Sea during the HTM has already been shown in previous studies (Bianchi et al., 2000; Funkey et al., 2014; Jilbert et al., 2015; Sollai et al., 2017). This pattern is very likely related to three main factors. First, the volume of the Baltic Sea was much larger than today due to glacial overdeepening and early Holocene eustatic sea-level rise, resulting in a stronger influence of North Sea water and, very likely, a pronounced water column stratification (Jilbert et al., 2015; Häusler et al., 2017). Second, temperature and insolation were higher than today during the HTM (Seppä et al., 2005, 2009; Borzenkova et al., 2015). Third, intensely hypoxic to anoxic conditions during the HTM may have stimulated the sediment-bound phosphorus to be released into the water column (Vahtera et al., 2007; Jilbert and Slomp, 2013; Jilbert et al., 2015; Funkey et al., 2014). These conditions very likely provided an ideal environment for diazotrophic cyanobacteria to thrive. In turn, frequent cyanobacterial blooms may have contributed to the formation of hypoxic to anoxic deep waters (Jilbert et al., 2015; Häusler et al., 2017), as is the case for the central Baltic Sea in the present day (Zillén and Conley, 2010). The 6+7Me-C_17 : 0 record suggests a fluctuating abundance of diazotrophic cyanobacteria between 0 and 3000 years before present. This pattern is absent in a previous study using echinenone and zeaxanthin pigments to reconstruct cyanobacterial blooms in a sediment core from the Bothnian Sea (Jilbert et al., 2015). The non-specificity to diazotrophic cyanobacteria (Bianchi, 2007) and the diagenetic instability (Britton et al., 2008) of these organic biomarkers may explain the discrepancies between both records. The variability in diazotrophic cyanobacterial blooms in the Bothnian Sea over the last 7000 years seems to be related to mean annual temperature changes as recorded in lake sediments from southern and central Sweden (Fig. 6b; Seppä et al., 2009). Periods with relatively high temperatures co-occur with a higher presence of cyanobacterial lipids in the sediments, which may be related to the enhancement of cyanobacteria blooms by higher temperatures (Stal et al., 2003). However, as suggested by Jilbert et al. (2015), the shoaling of the Åland Sea sills (Fig. 1) and the resulting decline in P availability may have been a major factor explaining the decline in cyanobacterial blooms from ca. 6300 cal. yr BP. The present results indicate that 6+7Me-C_17 : 0 is a relevant proxy for diazotrophic cyanobacterial blooms in the Baltic Sea for the whole duration of the Holocene.