On 12 June 2012 (day - 10 = t-10, 10 days before CO2 manipulation), nine floating, pelagic mesocosms (Fig. 1, KOSMOS, volume ∼ 55 m3) were deployed and moored at 59∘51.5′ N, 23∘15.5′ E in the Tvärminne Storfjärden, an open archipelago area on the eastern side of the Hanko peninsula on the south-west coast of Finland (Fig. 2). The water depth at the mooring site was approximately 30 m. The bottom ends of the mesocosm bags were lowered to a depth of 17 m below the surface to enclose the plankton community with minimal disturbance to the water column. A mesh of 3 mm was attached to the top, which was submerged ∼ 0.5 m below the surface, and bottom of the bag, at 17 m deep, to exclude any large organisms or particles with patchy distribution in the water column. Initially the mesocosm bags were kept open and covered with only the 3 mm nets at the top and bottom openings for 5 days to allow for rinsing of the mesocosm bags water and free exchange of plankton (< 3 mm). On t-5, the nets were removed, sediment traps (2 m long, Fig. 1) were then attached to close the bottom of the mesocosms and the top ends of the bags were pulled up to 1.5 m above the water surface, thereby isolating the water in the mesocosms from the surrounding Baltic Sea.

To ensure a homogeneous water column in each mesocosm at the start of the experiment, the halocline present was destroyed by bubbling each mesocosm with compressed air for 3.5 min on t-5. A video profile taken in one of the mesocosms on t-4 shows the plankton community present at the beginning of the study period (Boxhammer et al., 2015a). Figure 3 indicates the experiment timeline including important manipulations. Mesocosm bags were cleaned occasionally inside and outside throughout the experiment to minimise wall growth and keep the biofilm biomass at a minimum (see Fig. 3 and Riebesell et al., 2013, for further details). An isotope tracer (15N-N2 gas) specific to the nitrogen-fixing organisms present was injected in two additions (t22 and t26) into four mesocosm bags (M3, M5, M6, M8). Further details about the addition are described in Paul et al. (2015). No dissolved inorganic or organic nutrients were added to the mesocosms in this study. At the end of the experiment, the volume of each mesocosm (0–19 m) was determined through addition of a calibrated salt solution as described by Czerny et al. (2013). Final mesocosm volumes ranged between 53.1 and 55.1 m3 with an estimated uncertainty of 2 %. Unfortunately, three mesocosms (M2, M4 and M9) were lost because of extensive and unquantifiable water exchange with the surrounding seawater due to a welding error on the mesocosm bags and were thus excluded from sampling and analyses.

CO2 treatments were achieved by equally distributing filtered (50 µm), CO2-saturated seawater into the mesocosm as described by Riebesell et al. (2013) in four separate additions (see Table 1 for details). The first addition of CO2-enriched seawater defined the beginning of the experiment and took place on t0 following sampling activities. Seawater for the additions was collected from 10 m depth by a pipe connected to the laboratory in the research station. Different amounts of CO2-saturated seawater were added to four mesocosms to set up an initial gradient in fCO2 treatments from ambient (∼ 240 µatm) up to ∼ 1650 µatm. On t15, CO2 was manipulated in the upper 7 m to counteract pronounced outgassing in the mesocosm. Two mesocosms were selected as controls with no addition of CO2-enriched seawater. Instead, unenriched filtered seawater (50 µm) was added for the initial manipulations. For the later smaller addition, the water distributor (“spider”; Riebesell et al., 2013) was pulled up and down in each mesocosm to simulate water column mixing and manipulation side effects caused by the device on t15.

CTD casts in each mesocosm and in the surrounding water were made with a hand-held self-logging CTD probe (CTD60M, Sea and Sun Technology) from 0.3 m down to ∼ 18 m (mesocosms) and to ∼ 30 m (surrounding water in archipelago from Baltic) between 13:30 and 14:30 local time (LT) daily until t31, and then every second day until t46. Temperature, pH, dissolved oxygen and PAR (photosynthetic active radiation) sensors were deployed on the CTD as well as a conductivity cell. Details on the sensors, their accuracy, precision, and corrections applied are described in Schulz and Riebesell (2013). The potentiometric CTD pH was corrected to spectrophotometric measurements (see Sect. 2.5.1). The depth of average water column light intensity in metres was calculated by averaging all water column PAR data and relating this to the depth where this intensity of PAR occurred.

A PAR sensor (LI-COR LI-192) was placed unobstructed at the end of a 2 m pole on the roof of Tvärminne Zoological Station (∼ 1 km from mesocosm mooring site) to record incoming PAR for the mesocosms. Incoming PAR was recorded from 14:43 LT on 14 June 2012 continuously as the mean of integrated 60 s intervals until the end of the experiment at 11:23 LT on 7 August 2012.

Water samples were collected regularly from each mesocosm and the surrounding water using depth-integrated water samplers (IWS, HYDRO-BIOS, Kiel). Unless otherwise reported, all samples are from the entire water column (0 to 17 m). For example, inorganic dissolved nutrient and fluorometric Chl a samples were also taken regularly for the upper water column (0 to 10 m). Full details of mesocosm sampling procedures and equipment are described in Riebesell et al. (2013) and Schulz et al. (2013). There were two intensive sampling periods where sampling took place every day (t-3 to t5, t29 to t31), otherwise most variables were sampled every second day. Table 2 presents sampled variables, including sampling frequency and respective papers which report each data set. Samples for carbonate chemistry variables and trace gas analyses were the first to be sampled and were taken from the IWS directly on board the sampling boat. Other samples (e.g. particulate matter, Chl a, phytoplankton pigments) were collected into 10 L carboys and stored in the dark. Carboys were stored at in situ temperature onshore and subsampling from these carboys was usually within 1 h and up to a maximum of 5 h after sampling. Care was taken to mix the water samples in the carboys well before taking subsamples to ensure homogeneous sampling for all parameters.

The sediment trap was emptied every second day using a manual vacuum pump system to acquire the settled material via a silicon tube reaching down to the collection cylinder of the sediment trap (Boxhammer et al., 2015b; Riebesell et al., 2013). This material was used to quantify and characterise particle sinking flux. Subsamples of the particle suspension (< 6 % in total) were taken before the material was concentrated. Particles and aggregates were allowed to settle down within 2 h at in situ temperature before separation of the supernatant. Collected particulate material was then centrifuged, while subsamples of the supernatant were filtered and analysed analogous to water column samples for particulate matter. Centrifuged material was subsequently frozen, lyophilised and ground to a fine powder of homogeneous composition. From this powder small subsamples of between 0.7 and 1.5 mg were weighed and analysed for carbon, nitrogen, phosphate and biogenic silica content as described in this paper for water column samples (see Sect. 2.5.3). Concentrations of particulate material were calculated based on total mesocosm volume (in litres). Mesocosm volume determined on t45 by salt addition in kilograms (Sect. 2.2) was converted using mean mesocosm temperature and salinity over 0–17 m between t-3 and t43 (mean temperature = 11.42 ∘C, mean salinity = 5.70) and the algorithms described by Fofonoff and Millard Jr. (1983). A more in-depth description of sampling and processing of particles collected in the sediment traps of the KOSMOS setup is presented in Boxhammer et al. (2015b).

As in previous mesocosm experiments, an fCO2 gradient was chosen for reasons as outlined in Schulz et al. (2013). Linear regression analyses were used to determine the relationship between average fCO2 and average response of the variables during each experimental phase. Outliers were detected based on Grubb's test (p < 0.05). This test was applied to all treatments by experiment phase to account for temporal development of each variable. Detected outliers were not included in the calculation of experiment phase average. Exceptions to outlier exclusion include biogenic silicate concentrations in M8 on t23 because all data were higher on this particular sampling day, and C : N in total POM on t19 in M8 because the C : N in this treatment was also markedly higher than other treatments on the following sampling day (t21). The same line of reasoning for the latter also applies to the contribution of cryptophytes to total Ch a M8 on t17 and all five outliers in contribution of euglenophytes to total Chl a detected in Phase III for the same line of reasoning as (b). All data points are included in the figures, with excluded outliers clearly marked. Linear regression analyses and outlier detection and exclusion were undertaken using R software (http://www.r-project.org/).

Conditions in the Tvärminne Storfjärden at the beginning of the experiment and during mesocosm closure were typical for the early summer season. Daily solar irradiance was at the annual peak (summer solstice) and surface water temperatures were ∼ 10 ∘C. Daily average water column temperature was highly variable over the experiment ranging from 8.0 to 8.5 ∘C at the beginning of the experiment to 16 ∘C on t16 (Fig. 4). Temperature variations as well as the first CO2 manipulation on t0 were used to define different experimental phases (Phase 0 = t-5 to t0, Phase I = t1 to t16, Phase II = t17 to t30, Phase III = t31 to t43). Warming occurred over the first 15 days and average water column temperatures peaked at 16 ∘C (Phase I). A cooling phase (Phase II) occurred until t31 (∼ 8 ∘C), followed by a second warming period (Phase III) which continued until the end of the experiment, reaching around 12 ∘C on average in the water column (Fig. 4 and 5c). The cooling in Phase II occurred around the same time as a period of lower incoming PAR between t15 and t25 (land-based PAR measurements, Fig. 6a). Surface water temperatures reached a maximum of 18 ∘C with a surface-to-depth gradient of 6 ∘C. The water column in the mesocosms remained thermally stratified throughout the study according to daily CTD profiles. Stratification strength, defined here as the potential density anomaly (σT) difference between the surface 10 m and bottom 7 m above the sediment trap in each mesocosm, was variable but lower in Phase I than in II and III. Detected changes in density over time were largely driven by changes in temperature within the mesocosms as there was only a minimal increase in salinity during the experiment probably due to evaporation (Fig. 5). Here, M8 was arbitrarily selected as representative of all mesocosms in Figs. 5 and 6. A typical daily difference in measured average water column temperature and salinity between mesocosms was 0.04 ∘C and 0.01, respectively. The increase in salinity on t45 is from addition of a calibrated salt solution for mesocosm volume determination. A notable decrease in temperature and increase in salinity in the archipelago between t15 and t31 coincided with a period of stormy weather and a change in wind direction from north-easterly to a more westerly direction, indicating a period of upwelling. During this period, there was slightly lower incoming PAR, indicating higher cloud cover (Fig. 6). The depth of average light intensity was relatively stable between 3.7 and 4.7 m inside the mesocosms and very similar between treatments over time (Fig. 6).

All mesocosms had a similar pHT of around 8.0 prior to CO2 perturbations. Initial CO2 enrichment reached target values on t4 ranging from ∼ 240 µatm in the two ambient control mesocosms up to ∼ 1650 µatm in the highest treatment, corresponding to a pHT range of ∼ 7.45 to 8.2 (Fig. 7). Aside from the CO2 addition on t15, fCO2 was allowed to vary naturally and treatments remained well separated over the entire experiment. The decrease in fCO2 over time in the high CO2 treatment mesocosms was mostly driven by outgassing rather than biological uptake as productive biomass remained relatively low in this experiment (see Sect. 3.3). The effect of outgassing is evident in the rapid increase in surface pHT in all treatment mesocosms (Fig. 8). Surrounding water pHT (0–17 m) ranged from 8.30 initially to 7.75 during the experiment. The profound pHT variability outside the mesocosms was due to upwelling of deeper, CO2-rich seawater. Within each mesocosm, CO2 manipulations over the entire depth were relatively homogeneous initially. However, a decrease in pH in the ambient control mesocosms below 5 m depth was detected from around t15 onwards, suggesting heterotrophic activity at depth involving respiration of organic matter to CO2 (Fig. 8). DIC increased in the control mesocosms due to gas exchange. This counteracted losses through uptake by the plankton community leaving the water column undersaturated in CO2 compared to the overlying atmosphere (∼ 230 µatm in control mesocosms vs. ∼ 400 µatm in atmosphere; Schernewski, 2011). Undersaturation of CO2 is typical for post-spring-bloom conditions such as those in the Tvärminne Storfjärden before the first CO2 enrichment in this study on t0.

Calcium concentration was 2.17 mmol kg-1, which was higher than calculated from a typical mean ocean salinity relationship of 1.67 mmol kg-1 (Dickson et al., 2007), because of high riverine calcium carbonate inputs in the Baltic Sea (Feistel et al., 2010). We accounted for this in the calculation of the calcium carbonate saturation state in the water (Fig. 7d). All mesocosms apart from the two ambient controls during Phase 0 and I were undersaturated with respect to aragonite (Fig. 7d), and the highest three fCO2 treatments were also undersaturated with respect to calcite (data not shown) during the entire experiment.

Out of 105 linear regressions applied to particulate and dissolved material from the water column and the accumulated sediment trap material to analyse the effect of CO2, we detected a significant correlation in 18. These are summarised in Table 3 and highlighted in the following sections. The majority of detected responses (14) indicated a positive effect of CO2, whereas only 4 indicated a negative effect of CO2.

In this study, the low number of fCO2 treatments (six) due to the exclusion of three mesocosms limited the statistical power of our conclusions. However the effect of CO2 was consistent across biogeochemical element pools with higher sustained particulate matter concentrations and lower dissolved phosphate under high CO2. This gives us confidence that the results of our study are indicative of the response of this particular plankton community in the Baltic Sea to ocean acidification.

Chl a concentrations were low but typical of a post-spring bloom period. An increase in Chl a began after t1 and signified a phase characterised by higher Chl a concentrations (∼ 2 µg L-1) until t16 (Fig. 9; Phase I: t1 to t16). Chla concentrations decreased by ∼ 0.8 µg L-1 in the mesocosms during Phase II and remained low and relatively stable in Phase III (∼ 0.9 to 1.2 µg L-1). Between 50 and 80 % of Chl a was in the upper water column (IWS samples 0–10 m, Fig. 9c). Chl a concentrations were in general lower (0.9 to 2.5 µg L-1) in the mesocosms than in the surrounding water (1.2 to 5.5 µg L-1, Fig. 9). CO2-related differences first developed during Phase II and remained stable during Phase III, with 24 % higher Chl a in the highest fCO2 treatment in Phase III (Table 3).

No dissolved inorganic or organic nutrients were added to the mesocosms in this study, and nutrient concentrations remained relatively stable with low inorganic nitrogen concentrations throughout the entire experiment. There was low inorganic nitrogen (∼ 50 nmol L-1 nitrate and ∼ 200 nmol L-1 ammonium) relative to phosphate (∼ 150 nmol L-1) in all mesocosms at the start of the study period compared to the canonical Redfield nutrient stoichiometry (Fig. 10, C : N : P = 106 : 16 : 1; Redfield, 1958). These concentrations are within the natural range for this region in a post-spring/early summer bloom phase (Fig. 10). Fixed nitrogen availability primarily limited the development of phytoplankton biomass in this system. This is common in the Baltic Sea following the spring bloom (Matthäus et al., 1999). Temporal dynamics between phosphate and nitrate showed decoupling. Nitrate concentrations increased from ∼ 20 up to ∼ 80 nmol L-1 from t1 until the end of the experiment (t43), whereas phosphate concentrations were slightly more dynamic, decreasing in Phase I and increasing in Phases II and III (Fig. 11). Around t30, differences in phosphate concentrations between fCO2 treatments became visible with a significant negative relationship between fCO2 and phosphate concentration in Phase III (Table 3). For further details and discussion on phosphorus pool sizes, uptake rates and cycling, see Nausch et al. (2015).

Ammonium concentrations decreased from between ∼ 170 and ∼ 280 nmol L-1 on t-3 to between 40 and 150 nmol L-1 on t39, with a small increase until t43 in all mesocosms (Fig. 10c). Samples for NH4+ concentration were lost on t27 and t29 for all mesocosms. The strongest decrease occurred during Phase I and concentrations remained relatively stable in Phase II and III. No significantfCO2 effect was detected during any experimental phase above the variability in the data. Inside the mesocosms, dissolved silicate concentrations decreased minimally from around 6.2 µmol L-1 on t-1 to between 5.5 and 5.8 µmol L-1 at the end of the initial productive Phase I on t16 (Fig. 10d). Thereafter, dissolved silicate remained relatively constant until the end of the experiment. No significant effect of fCO2 on dissolved silicate concentrations was detected in any phase.

DOC concentrations ranged between 410 and 420 µmol L-1 on t2 and increased by ∼ 30 µmol L-1 up to between 440 and 450 µmol L-1 on t43 (Fig. 11a). In Phase III, DOC positively correlated with fCO2 (Table 3). There was no statistically significant correlation of fCO2 with DON or DOP concentrations in any experimental phase. No clear temporal trends were distinguished in DOP concentrations, although DON decreased during Phase I (Fig. 11). Where data points are missing, DON could not be corrected for NH4+ concentrations; hence, they are excluded from the data set.

Particulate C, N and P concentrations were higher in Phase I than in Phase II and III (Fig. 12), as also observed for Chl a (Fig. 9a). The importance of small particles was even more pronounced in Phase III, where up to ∼ 90 % of total particulate organic matter was attributed to the fraction TPC< 10 in the four mesocosms sampled for this size fraction (M3, M5, M6, M8; Fig. 12). In Phase III, there was a significant positive correlation between fCO2 and average total TPC, PON and TPP (Table 3).

C : N and C : P ratios in POMtot (Fig. 13) were above the Redfield ratio (C : N : Ptot= 106 : 16 : 1) during the productive phase, peaked at the beginning of Phase I (C : Ntot= 7–8.5, C : Ptot= 110–160) then decreased and became stable during Phase II (C : Ntot= 5.8–7.0, C : Ptot= 80–140). Differences betweenfCO2 treatments were first observed in Phase III with higher C : Ntot in the highest fCO2 treatment (Table 3). No significant effect of fCO2 on N : P or C : P was detected in any experiment phase or in any size fraction.

BSi decreased from around 1.0 µmol L-1 at the beginning to ∼ 0.3 µmol L-1 at the end of the experiment (Fig. 12). During Phase II, there was a statistically significant correlation of BSi with fCO2; however, this was absent in Phases I and III (Table 3).

The contribution to Chl a by different phytoplankton groups varied over time, although the temporal trends in all mesocosms appeared remarkably similar (Fig. 14). Results from CHEMTAX analyses of the phytoplankton community present indicate that cryptophytes and chlorophytes had the highest contribution to total Chl a during Phase I and Phase II/III, respectively. The total abundances of cryptophytes decreased from t-3 to t17 in all mesocosms, succeeded by a brief euglenophyte peak around t15, with chlorophytes being the dominant contributor to Chl a from t17 on (Fig. 14). Total abundances of cyanobacteria, probably non-diazotrophic Synechococcus, were highest during both Phase II and III. Diatoms made up a relatively small proportion of the plankton assemblage and contributed to less than 10 % of Chl a in Phases I and II and between 10 and 25 % in Phase III. Other key groups detected included dinoflagellates and prasinophytes; however, they made up minor proportions (below 15 % of total Chl a) of the plankton community throughout the entire experiment (dinoflagellate data not shown).

We analysed the relationship between fCO2 and the contribution of phytoplankton groups to Chl a by linear regression for each experimental phase (Table 4). These analyses indicated small differences in plankton community composition between CO2 treatments. There was a significant negative correlation between CO2 and total diatom contribution to Chl a in Phase III. In Phase III, fCO2 was also negatively correlated to the contribution of cryptophytes to Chl a and a significant positive effect on the contribution of prasinophytes to Chl a.

Linear regression of the absolute concentrations of a number of phytoplankton pigments in the size fraction < 2 µm indicated primarily a positive correlation to fCO2 during Phase I (i.e. Chl a, violaxanthin, neoxanthin), although a statistically significant effect was not detected in all pigments (Table 5). In Phase III, where the highest Chl a concentrations were in the size fraction < 2 µm, mass balance calculations indicated more than 100 % of total Chl a in this size range, which is not physically possible. These unbalanced Chl a measurements are the result of measurement uncertainties at such low absolute concentrations, particularly in the > 20 µm size fraction and of mass balance calculations between three independent filtrations. As the increase and decline in Chl a < 2 µm and 2–20 µm fractions, respectively, are supported by flow cytometry data for picoeukaryote and nanoeukaryote abundances, we still consider the observed temporal variations to be robust. A positive correlation between picoeukaryote abundance and CO2 treatment was also already detected in Phase I (Crawfurd et al., 2015). Absolute concentrations of Chl a, Chl b, prasinoxanthin, violaxanthin and neoxanthin in the total fraction had a statistically significant positive correlation withfCO2 during Phase III (see Table 5). Fucoxanthin concentrations (key pigment in diatoms but also present in dinoflagellates) andfCO2 were also positively correlated in the fraction > 20 µm during Phase III. Size fractionation of HPLC pigment analyses indicated a higher proportion of Chla in all treatments in biomass < 2 µm during Phases II and III (Fig. 15).

The amount of material collected in the sediment traps in each phase reflected biomass (here POM and Chl a) build-up from the water column. We calculated that > 84 % of total carbon sinking into the sediment trap was collected during Phases I and II and less than 16 % during Phase III (Fig. 16). This corresponds to average accumulation rates (± SD) of 0.303 ± 0.011, 0.203 ± 0.033 and 0.094 ± 0.029 µmol C L-1 day-1 across all mesocosms in Phases I, II and III, respectively. No significant CO2 trends were detected during any phase with regard to the total amount of C, N, P and BSi in the sediment trap material.

Phase I (t1 to t16) was characterised by the highest sustained Chl a and particulate matter concentrations in the water column. Relatively high light availability, particularly between t6 and t15 (Fig. 6a), accompanied by increasing water column temperatures likely supported autotrophic growth. However, no increase in particulate matter pool size was observed in any treatment during this productive phase. Instead carbon was diverted into the sinking particle flux and DOC pool (Fig. 11) with a net daily accumulation of DOC of between 10 and 15 % of the total TPC pool between t3 and t13. As inorganic nitrogen availability was very low, we assume this is due to carbon overconsumption (Toggweiler, 1993). Thus, organic matter turnover in the system appeared to be high during this period, although overall phytoplankton biomass production was limited by low inorganic nitrogen availability.

Although phytoplankton carbon fixation is expected to be stimulated by increased CO2 availability (Hein and Sand-Jensen, 1997; Losh et al., 2012; Riebesell et al., 2007), previous CO2 enrichment experiments using natural plankton assemblages under various conditions of nutrient repletion in different regions have shown no consistent response of primary production to elevated CO2 (Engel et al., 2005; Hopkins et al., 2010; Hopkinson et al., 2010; Nielsen et al., 2011; Riebesell et al., 2007; G. K. Schulz, personal communication, 2015; Yoshimura et al., 2013). During high organic matter turnover in Phase I, we detected no statistically significant differences in bulk organic matter concentrations or elemental stoichiometry between CO2 treatments. No effect of CO2 treatment could be detected in the most abundant, and presumably most productive, phytoplankton size class (2–20 µm, Fig. 15). Instead, detected differences between fCO2 treatments in particulate matter in Phase I were mostly confined to pigment concentrations in the smallest size fraction (< 2 µm). Here, pigment concentrations were generally higher in the highest CO2 treatment (Table 5). This is in line with flow cytometry counts which revealed a positive effect of CO2 on the abundance of picoeukaryotes (Crawfurd et al., 2015.) and is in agreement with studies in the Arctic (Brussaard et al., 2013), the subarctic North Pacific (Endo et al., 2013), and North Atlantic Ocean (Newbold et al., 2012) but contrasts the results from Richier et al. (2014) from shelf seas in the north-east Atlantic Ocean. The positive influence of CO2 on phytoplankton pigment concentrations was also detected in the largest size fraction (> 20 µm) in Phase I; however, this size class made up only a small portion of total Chl a (< 10 % Fig. 15, size fractionated pigment analyses). Thus, small CO2-driven differences in plankton community structure in the smallest and largest phytoplankton were not relevant for biogeochemical element cycling in this plankton assemblage during this productive phase.

The distinct changes in the phytoplankton communities in the mesocosms coincided with the decrease in temperature during the upwelling even in the archipelago in Phase II (t17 to t30). Temperature decreases of greater than 10 ∘C in surface water, as observed in this study, have been reported for upwelling events during periods of thermal stratification (Lehmann and Myrberg, 2008) with considerable influence on the ecosystem productivity (Nômmann et al., 1991). Here we assume that the combination of higher grazing pressure, lower PAR and cooler temperatures likely slowed down phytoplankton productivity and contributed to decreased phytoplankton biomass, observed here as a decrease in Chl a, during this period (Fig. 9).

An increase in TPCtot : Chl a from ∼ 10 µmol µg-1 on t17 to over 15 µmol µg-1 on t29 indicates that carbon was being shifted from autotrophic to heterotrophic organisms, assuming that the Chl a content of the autotrophs remained constant. CTD profiles showed a decrease in pHT below 10 m in both control mesocosms (Fig. 8) at the same time as surface Chl a (0 to 10 m) decreased between t18 and t30. This pH decrease (i.e. CO2 increase) could indicate a possible change in the equilibrium between dominance of autotrophic (CO2 uptake) and heterotrophic (CO2 release) processes during a phase of strong cooling in the lower water column. Higher organic material availability seemed to stimulate bacterial activity up until t23 (Hornick et al., 2015). Furthermore, higher zooplankton abundances after t17 (Lischka et al., 2015), as well as a peak in abundance of a potential mixotroph around t17 (Euglenophycaea), also likely contributed to higher organic matter remineralisation and CO2 release. Hence Phase II is defined by increased heterotrophy and organic matter remineralisation. Carbon was primarily channelled into sinking material flux and higher trophic levels rather than accumulating in the DOC pool, mediated by increased zooplankton grazing pressure on primary producers.

Differences between CO2 treatments in the dissolved and particulate matter pools developed during the Chl a decrease and apparent increase in net heterotrophy in Phase II. In addition, size-fractionated pigment analyses indicated a shift in phytoplankton community size to smaller organisms with up to ∼ 90 % of Chl a in phytoplankton < 2 µm at the end of Phase II. This was not caused by a remarkable gain in Chl a in the smaller size class but instead due to Chl a loss in the larger size class, which we think was driven by high grazing pressure from abundant zooplankton at this time (Lischka et al., 2015). This removal of larger phytoplankton unmasked the underlying positive CO2 response of picoplankton that was already present since Phase I but now became clearly visible. In other words, a positive CO2 effect on picophytoplankton seemed to be present throughout the entire experiment. However, their ecological and biogeochemical relevance within the plankton community was too small initially, so that the CO2 effect was not detectable in the other bulk biogeochemical element pools.

Interestingly, measured carbon fixation rates did not show any fertilising effect of CO2 (Spilling et al., 2015), whereas both respiration (Spilling et al., 2015) and bacterial production rates between t14 and t23 (Hornick et al., 2015; Nausch et al., 2015) were lower at higher CO2. This suggests slower net particulate matter loss rather than increased production under ocean acidification (see Hornick et al., 2015, and Spilling et al., 2015) in this issue for more on this topic).

While temperature increased again during Phase III, there did not seem to be any recovery of phytoplankton biomass to the same level as in Phase I. In Phase II autotrophic growth was apparently dampened so severely that it could not recover within the duration of this study and was likely strongly controlled by high zooplankton grazing pressure. There was very little change in the amount or stoichiometry of the particulate or dissolved matter pools, suggesting that production and loss of particulate matter in the water column were either very low or relatively well balanced in Phase III. Only a small amount of TPC (∼ 1 µmol L-1, ∼ 16 % of total suspended TPC) was collected in the sediment traps, implying low particulate matter sinking flux strength in this phase. The positive (picoplankton-mediated) effect of CO2 on particulate and dissolved pools unmasked in Phase II was sustained throughout Phase III in Chl a, TPC, PON, TPP and DIP. Thus, in this study, higher autotrophic biomass was sustained under elevated CO2 in this plankton community during the post-bloom phase and had a significant influence on biogeochemical pool sizes.

Variations in water column particulate matter concentrations did not translate into statistically significant differences in the amount of accumulated sediment trap material between CO2 treatments. This may be because the response of CO2 was the strongest in phytoplankton < 2 µm, which taxonomically were likely to be chlorophytes and prasinophytes (Fig. 14b and f, Table 4). The unicellular organisms are, however, too small to sink as individual cells. Instead picoplankton contribute indirectly to carbon export through secondary processing of sinking picoplankton material (Richardson and Jackson, 2007). The positive effect of CO2 on particulate matter pools was reflected positively in the DOC pool, suggesting that a higher proportion of freshly produced organic matter was directed into the microbial food web, rather than being exported during the period of low organic matter turnover in Phase III. A similar channelling of carbon and the positive CO2 response in the DOC pool was observed during nutrient-depleted conditions in an Arctic CO2 enrichment mesocosm study (Engel et al., 2013). Here, this could be a consequence of continued reduced organic matter remineralisation at elevated CO2 (Spilling et al., 2015), as hypothesised for Phase II (see also Sect. 4.2), although unfortunately no respiration data for Phase III are available.

Based on our results, we hypothesise that, under future ocean acidification the Baltic Sea in low nitrogen, summer periods may shift towards a system where more organic matter is retained for longer time periods in the upper water column but may not result in increased particulate matter sinking flux.

Although a significant, but small, response to CO2 was detected in a number of particulate and dissolved matter pools, in numerous others no significant effect of CO2 was detected in any phase (e.g. DON and DOP concentration, N : P and C : P in POM). The muted response of the plankton community and biogeochemistry to elevated CO2 observed in this experiment might be linked to higher tolerance or resilience of the plankton community. The Baltic Sea is a highly dynamic system with much larger annual temperature, light period, inorganic nutrient, pH, and salinity fluctuations than in many other major water bodies and the open ocean. Thus the community present in this study may have considerable physiological plasticity through exposure to large natural diurnal and annual fluctuations in carbonate chemistry speciation and pH (see also Joint et al., 2011, and Nielsen et al., 2011). Low nitrogen availability in this study may have dampened underlying trends particularly in larger phytoplankton size classes. In past CO2 enrichment experiments, nutrient addition amplified the existing effect of CO2 between treatments (for example Schulz et al., 2013). This is one of few plankton community experiments where nutrient concentrations were very low initially and concentrations and nutrient ratios were not manipulated. Such conditions are representative of a steady-state stratified water column present in many ecosystems for most of the year.