Ten Kiel Off-Shore Mesocosms for Future Ocean Simulations (KOSMOS) were deployed by the research vessel ALKOR and moored in the Raunefjord, Norway, on 7 May 2022 (60.25° N, 5.2° E). The technical design of these seagoing mesocosms and procedures for water column manipulation are described in detail by Riebesell et al. (2013). Briefly, the 20 m long mesocosm bags were suspended in 8 m tall floating frames, and both ends were covered with a 3 mm mesh size net to exclude larger organisms during filling. The tops of the bags were then submerged 1 m below the sea surface and fully unfolded to enclose a waterbody containing the natural planktonic community. The bags were left submerged for 5 d to allow for sufficient seawater exchange to ensure similar starting conditions in all mesocosms. Next, the waterbodies within the mesocosm bags were isolated from the surroundings by attaching a 2 m-long, funnel-shaped sediment trap to the lower end of the bags and lifting the top end 1 m above the surface. The attachment of the sediment trap on 13 May marked the beginning (day 0) of the 53 d experiment. Right after closure, a 1 mm net was pulled from bottom to top to remove any heterogeneously distributed nekton (i.e., different amounts and sizes of nekton in different mesocosms). From days 1 to 3, daily sampling was conducted to monitor the initial conditions of the enclosed waters before OAE manipulation on day 6. Additionally, the mesocosms' volume was determined following Czerny et al. (2013), yielding an estimated average volume of ∼61.6±1.9 m³.

All water column manipulations in the mesocosms were achieved using of a pumped injection device equipped with polycarbonate piping of various lengths (Riebesell et al., 2013), that was lowered and raised several times in each mesocosm during manipulations to ensure homogeneous distribution throughout the entire water column.

The alkalinity manipulation was performed on day 6. As direct mineral additions, such as lime or olivine powder, come with several caveats, such as triggering secondary calcium carbonate precipitation, and thereby removing added alkalinity (Moras et al., 2022), slow dissolution kinetics (Fuhr et al., 2022), or minerals containing a variety of potentially toxic trace elements, we simulated mineral dissolution using laboratory prepared alkaline solutions and alkalinity was added using sodium hydroxide (NaOH, Merck) solutions. To simulate the use of two alkaline feedstocks: hydrated lime (Ca(OH)₂) and olivine (as forsterite, Mg₂SiO₄), NaOH additions were followed by the addition of the respective Ca²⁺ and Mg²⁺ rich solutions. The two solutions were prepared using reagent-grade calcium chloride (CaCl₂) and magnesium chloride (MgCl₂). Furthermore, to simulate the release of SiO32-, under the olivine scenario, a Si-rich solution was prepared using Na₂SiO₃ (Roth) and added in equal concentrations to all silicate-based treatments (75 µmolL-1), regardless of the targeted alkalinity increase (more details can be found in Goldenberg et al., 2024, and below). The reasoning behind this decision was to avoid colloid formation that occurs at high Si concentrations (up to 150 µmolL-1 would have to be added to the highest TA treatment to match the TA to silicate ratio of 4:1 in olivine), and to allow separation between silicate and TA effects. Ca²⁺ and Mg²⁺ were added in a 2:1 ratio to TA. All solutions were prepared by dissolving the corresponding salts in individual bottles filled with 20 L of deionized water (Milli-Q^®, 18.2 ΩM), which were then added to the respective mesocosms. Each salt was weighed in and dissolved separately. In summary, stock solutions of NaOH and CaCl₂ were added to the calcium-based treatments, and MgCl₂, NaOH and Na₂SiO₃ to the silicate-based ones (since adding silicate inevitably increases total alkalinity (Gattuso et al., 2010), we compensated the silicate-induced rise by adding HCl to the respective control and by withholding NaOH in the ΔTA 150 µmolkg-1 treatment).

The increase in silicate and TA by NaOH was confirmed by direct measurements in the mesocosms right after the additions. However, Mg⁺² and Ca⁺² concentrations were not measured, as their expected change was considered minor (only a few percent compared to the large natural background concentrations, ∼49.8 and 9.6 mmol kg⁻¹ at a salinity of 33, respectively). Concerning silicate additions, olivine dissolution would have resulted in an increase of silicate in relation to TA by 0.25, which would have been 37.5 µmolkg-1 at the lowest TA addition of 150 µmolkg-1. This is more than one order of magnitude larger than the concentrations considered limiting for diatom growth. However, to avoid confounding effects, it was decided to keep the silicate addition the same among respective mesocosms at 75 µmolkg-1, also avoiding unavoidable precipitation at even higher concentrations. Finally, only about 10 µmolkg-1 of silicate were taken up until the end of the experiment, meaning that silicate concentrations would have been non-limiting throughout the experiment in all mesocosms.

For air-sea gas exchange determination, nitrous oxide (N₂O) additions were performed on day 14, following the procedure described in Czerny et al. (2013). One liter of a saturated stock solution was prepared by bubbling 0.2 µm filtered seawater for 2 d with N₂O (Nippon Gases). The amount of stock solution to be added to each mesocosm was calculated using solubility constants by Weiss and Price (1980) and taking into consideration in situ salinities, temperatures and individual mesocosm volumes. The stock solution was then diluted with filtered seawater into 25 L carboys, and each mesocosm was spiked with one carboy.

Given the duration of the experiment and the low levels of inorganic nutrients in comparison to the surrounding coastal water, a nutrient addition was performed on day 26. Nitrate concentrations (NO3-) were targeted at 4 µmolL-1 and phosphate concentrations (PO43-) were enhanced following the N:P Redfield ratio of 16:1 (Redfield, 1934). Further detail on specific quantities can be found in Ferderer et al. (2024).

Calcium-based treatments also received a minor silicate (Si(OH)₄) addition in a N:Si ratio of 4:1 to better mimic the natural conditions, without creating a Si-enrichment scenario. Nutrient concentrations were measured the day before nutrient addition and ∼2 h after addition, for quantification and to ensure that the targeted stoichiometry was achieved. It was then noted that the stoichiometry was not even across mesocosms due to underestimated nitrate additions. A successful nitrate amendment was performed on day 28.

Following the alkalinity manipulation, sampling in random order was carried out every second day in the morning hours (08:00–12:00 CEST). Depth-integrated (0–20 m) water samples were taken from the mesocosms and the surrounding coastal water (later referred to as “Fjord”) using a 5 L integrating water sampler (IWS, HYDRO-BIOS, Kiel).

Samples were collected from the IWS in decreasing order of sensitivity to gas exchange, i.e., N₂O, followed by carbonate chemistry and inorganic nutrients. N₂O samples were drawn with a Tygon tube directly into 20 mL caramel vials in triplicates. After ensuring that the vials were bubble-free, they were crimp-sealed immediately and kept at room temperature after fixation with 10 µL of a saturated mercury chloride solution (HgCl₂). To minimize gas exchange during air transport and storage, two paraffin wax coats were applied to the crimp seals (Glatzel and Well, 2008; Kock et al., 2016). For carbonate chemistry parameters, such as pH and TA, 0.5 L of seawater were taken into air-tight glass flasks. Clean bottles were pre-rinsed with sample water immediately prior to filling (using pre-rinsed Tygon tubing). Finally, to minimize air–water gas exchange, filling was done gently from bottom to top with an overflow of ∼1.5 times the sampling volume (Dickson et al., 2007).

Nutrient subsamples were collected next into 250 mL acid-cleaned polypropylene bottles. All samples were kept in cool conditions and protected from direct sunlight until further analysis. Subsamples for NO3-, NO2-, PO43-, and Si(OH)₄ were filtered using a PES syringe filter (0.45 µm Sterivex, Merck) and analyzed spectrophotometrically following Hansen and Koroleff (1999).

CTD casts were performed with a multiparameter logging probe (CTD60M, Sea & Sun Technology) directly after the main sampling (14:00–16:00), yielding depth profiles of salinity, temperature and pH.

The 53 d experiment was divided into three distinct phases. The pre-treatment phase (days 1 to 6), from here on referred to as phase 0 for simplicity, denotes the baseline state of the system before any alkalinity manipulation. This preliminary phase was an important step to confirm that the starting conditions in all mesocosms were similar. The reaction phase after OAE manipulation (days 7 to 28), phase I, corresponds to the period in-between alkalinity manipulation and the addition of dissolved inorganic nutrients. This post-treatment and pre-fertilization phase mainly consisted of the immediate response of the system to an increase in alkalinity levels in a post-bloom scenario. The last phase after nutrient addition is referred to as phase II. This post-fertilization period (days 29 to 53) is thought to capture the mid-term indirect responses to alkalinity addition of a biologically active ecosystem state.

Throughout the experiment, two carbonate chemistry parameters (TA and pH_T) were measured and paired with corresponding daily averages of salinity and temperature (from CTD casts), as well as nutrient concentrations, to calculate the remaining carbonate system variables, such as DIC, ΩA and pCO₂. To do so, the software PyCO2SYS version 1.8.2 (Humphreys et al., 2024b) was used with chosen constants for calculations as follows: K1 and K2 for carbonic acid from Sulpis et al. (2020), KHSO₄ from Dickson (1990), KHF from Dickson and Riley (1979), [B]_T from Uppström (1974) and the universal gas constant R.

Note that given the set-up of this experiment, pH and pCO₂ are intimately correlated, with a quasi-linear relationship between proton concentration and pCO₂. Either parameter could have been used to describe carbonate chemistry, but we chose to consistently report responses with respect to pCO₂ as the one most relevant across all aspects of the manuscript.

During phase 0, conditions in all mesocosms were relatively similar, averaging 2213±3 µmolkgsw-1 for measured TA, 8.109±0.004 for pH_T, 2030±4 µmolkgsw-1 for DIC, 339±3 µatm for pCO₂, and 2.00±0.02 for ΩA (Fig. 1). After the alkalinity manipulation, on day 7, TA levels increased according to treatment, reaching a maximum of 2740.7 µmolkgsw-1, pH_T increased to a maximum of 8.767, and ΩA up to 7.66 in the two highest treatments, while pCO₂ decreased to 61 µatm. In contrast, DIC was hardly affected by the manipulation and dropped in the initial days of the experiment consistently across all mesocosms independent from the addition of alkaline material (Fig. 1).

Overall, TA remained relatively stable throughout the rest of the experiment, regardless of the phase and the treatment (calcium- or silicate- based), with a maximum variability of ∼19 µmolkg-1. ΩA remained stable throughout phase I, but showed an increase ranging from ∼0.20 to ∼0.45 units in each treatment after nutrient addition, from day 35 onwards, consistent with a decrease in DIC and an increase in NCP_DIC and Chl a (Figs. 1a, 1c, 4a, S1).

In contrast, changes in pH_T and DIC were observed throughout the experiment. After an initial drop in DIC in all mesocosms (phase 0), subsequent DIC changes were less consistent among treatments over time, with increments of up to ∼8–28 µmolkg-1 in the treatments before the nutrient addition. About a week after the nutrient addition, DIC decreased in all mesocosms again, by ∼23–51 µmolkg-1, coinciding with average Chl a increase from 0.53±0.1 µL-1 up to 3.4±0.9 µL-1 during phase II (Fig. S1).

The experiment began with initial pCO₂ below atmospheric levels, averaging 339±3 µatm across treatments (Fig. 1). During the initial phase (0), all treatments were taking up CO₂, leading to a daily increase in DIC of ∼0.1 µmolkg-1. While CO₂ ingassing (negative FCO2) did not change significantly for the control treatments, more atmospheric CO₂ was taken up with increasing ΔTA (and the corresponding lower seawater pCO₂ (Fig. 2)). In the highest treatment, FCO2 was five times lower than in the controls, meaning ingassing was 5 times higher. Interestingly, this increasing rate of CO₂ ingassing (decreasing FCO2) persisted in all treatments except for the controls, resulting in a 50 % increase towards the end of the experiment. The total cumulative net uptake of CO₂ ranged from ∼16 to ∼30 µmolCkg-1 (Fig. S4) across treatments. Notably, no statistically significant differences in FCO2 were observed between the different minerals.

The response of normalized cumulative calcification (normalized CALC) to pCO₂ was that of an optimum curve (fitting Eq. 9 from Gafar et al., 2018) for any given phase of the experiment (Fig. S7; Table 1). The calcium-based treatment with lowest TA addition (ΔTA 150 µmolkg-1) was deemed as an outlier (see Figs. S5a, S7) and thus excluded from analysis. A plateau was reached at about 250 µatm (Fig. 3a). The highest calculated CaCO₃ production potential (CCPP) was observed at a pCO₂ of about 200 µatm (Fig. S5). With up to 12 µmolkg-1 of CaCO₃ produced, cumulative calcification was highest in low-to-intermediate TA treatments and close to zero in the two highest ones (Fig. S4a).

NCP_DIC showed no significant differences between silicate- and calcium-based treatments during phase 1 (Figs. 4a, S7). The community responded to the nutrient addition, reaching a peak ∼10–15 d afterwards. NCP_DIC in the silicate-based treatments reached about 52 µmolCkg-1, whereas in the calcium-based treatments the NCP_DIC peaks were roughly half in magnitude. To better see potential differences during the bloom phase, we calculated the changes in daily NCP_DIC relative to the mean NCP_DIC values right after alkalinity manipulation to compare only phase 1 with both phases (1+2), as well as with phase 2, for which we used the mean NCP_DIC values right after nutrient addition. The cumulative changes are plotted in Figs. 4b and S7. This gives a clearer perspective on how differently treatments reacted and the delays in bloom onset, if any. A general trend shows that silicate-based treatments reached higher bloom peaks than calcium-based ones, and higher alkalinity treatments took relatively longer to start blooming, assuming that the sampling resolution was high enough to capture all real bloom peaks.

To better assess whether a mineral or treatment effect influenced the NCP_DIC responses, we proceeded to take the maximum production value during the bloom for each treatment and plotted it against the mean pCO₂ value during the bloom period (Figs. 4c, S7). An ANCOVA analysis showed a pCO₂ effect as maximum NCP_DIC increased with decreasing pCO₂/increasing alkalinity during phase 1, only a mineral effect during phase 2 and no pCO₂ effect, but a clear mineral effect on production when both phases were considered together (Table 1).

Manipulation of total alkalinity resulted in comparable treatment pairs in the calcium- and silicate-based treatments with a similar ΔTA, which remained relatively stable over time. This stability in TA suggests that the system stayed below the ΩA threshold for secondary CaCO₃ precipitation. Based on calculations from (Marion et al., 2009), the threshold for spontaneous pseudo-homogeneous carbonate formation in the presence of organic colloids and particles, but in mineral-phase-free seawater, would be 11.1 for our average salinity and temperature of 32.6 and 11 °C respectively, which was indeed not reached (Fig. 1c). As expected for a non-CO₂-equilibrated alkalinity addition, the TA manipulation induced predictable changes in carbonate chemistry speciation, i.e., increased pH and ΩA, and decreased pCO₂, with no significant change to DIC (Fig. 1). The relative stability of DIC in the days following the TA addition, consistent with expectations for this oligotrophic phase of the experiment (phase I), supports the reliability of the DIC estimates derived from corrected pH measurements.

In the control treatments, DIC declined only slightly during the first 10 d, while remaining relatively stable thereafter. During the bloom in the second phase, slight increases in pH correlated with DIC decreases, suggesting a biological origin, i.e., primary production accompanied by nutrient uptake. DIC drawdown only began in phase II and was higher than expected from nitrate drawdown: 23–49 vs. 20 µmolkg-1 (Redfield C:N of 6.625⋅N). Higher values are commonly related to carbon overconsumption, which is known to happen in nutrient limited environments and with increasing temperature (Paul et al., 2016; Taucher et al., 2012). The pronounced DIC drawdown is difficult to detect in other carbonate chemistry parameters because they were either hardly affected (e.g., TA) or only slightly affected (e.g., pH and pCO₂), with these small signals being obscured by the relatively large initial treatment differences compared to the more uniform DIC signal. Nevertheless, before drawing further conclusions on the biological or physical origin of these changes, CO₂ uptake from air-sea gas exchange needs to be considered and is further addressed in the following sections.

FCO2 started at about -2 mmol m² d⁻¹, indicating that the system was already ingassing CO₂ from the atmosphere during phase 0. This is related to the fact that the experiment was started during a post-bloom period, i.e., seawater pCO₂ was lower than atmospheric levels, and the concentrations of dissolved inorganic nutrients were low. Furthermore, ingassing rates for the control treatments were within the range of natural air-sea flux estimates from the region (Aalto et al., 2021), indicating that the mesocosm setup did not hinder gas exchange during the experiment. After alkalinity addition, daily ingassing rates increased (FCO2 decreased) by a factor of 3–5 in the highest TA treatments (Fig. 2). The rates of daily ingassing continued to increase over time, even though the gradient between the atmosphere and seawater pCO₂ would gradually decrease with atmospheric CO₂ uptake.

Since the Schmidt number and the viscosity of gases are influenced by temperature, this is a relevant factor when it comes to diffusion of gases, which in this case may have played a major role. Throughout the experiment, temperature increased from ∼8.5 to ∼15.5 °C as we moved from late spring to early summer conditions (Fig. S1). Indeed, there was a statistically significant positive correlation of temperature and transfer velocity, and a fitting equation was derived to calculate transfer velocities [cm s⁻¹] from in situ temperature [°C] (Eq. 7). While salinity is also an important factor impacting gas exchange, the relatively small changes throughout our experiment (∼0.1 units), were deemed non-significant for the fitting: 7kCO2=-2.67e-4+9.61e-5⋅Twater In contrast to a mesocosm setup, the most important factor influencing transfer velocity in open ocean settings is wind speed (ν). Though it is widely used to parameterize kCO2, it has an uncertainty of ∼20 % and is valid only for wind speeds in the range of 3 to 15 m s⁻¹ (Wanninkhof, 2014). To investigate further, we estimated the wind speed according to Wanninkhof (1992) and Wanninkhof et al. (2009) using our derived transfer velocities. At our lowest measured temperature, ν was 1.16 m s⁻¹, and, at our highest temperature, 1.92 m s⁻¹. The mean ν obtained (at mean salinity and temperature) was 1.49 m s⁻¹, differing from the mean wind speed near the area of about 3.5 m s⁻¹ (Weather Underground, 2025). This is consistent with the use of the mesocosms, which provide some shelter to the enclosed waters in contrast to the surrounding fjord water, reflecting that wind speed alone does not drive gas exchange at low wind, but rather alters water surface texture. In this regard, Eq. (7) would be more suitable than wind speed to estimate kCO2 under similar conditions, i.e., mesocosm setup and similar ranges of temperature and salinity.

Interestingly, daily rates of FCO2 reached up to -15 mmol m² d⁻¹ for the highest treatments, which equaled to ∼30 µmolCkg-1 after a period of 47 d for a volume of 61.6 m³ (20 m mixed layer depth). Considering windspeeds of ∼ 1.5 m s⁻¹ and a temperature of ∼15.5 °C, we simulated further ingassing until equilibration (95 %) of the mixed layer. We found it would take up to ∼1050 d to equilibrate. This timeframe is in line with the findings of He and Tyka (2023), who showed most locations to have an uptake efficiency plateau of 0.6–0.8 mol CO₂ per mol of alkalinity after 3–4 years. While full equilibration spans several seasonal cycles, the timing of alkalinity addition can still influence uptake dynamics. The steepest pCO₂ gradients and thus the largest fluxes occur soon after addition, with gas transfer velocity depending on temperature. Thus, in very cold waters there will likely be slower transfer velocities, not forgetting that windspeed and mixed layer depth would also modulate equilibration. Therefore, our results indicate that seasonality of natural systems should also be considered, in the timing and deployment site of OAE, as it will affect the short-term rate and efficiency of ingassing, even if long-term equilibration eventually dampens these effects.

Because of the presence of the coccolithophore E. huxleyi, and some availability of inorganic nutrients right at the start of the experiment, there was an initial burst of calcification in all mesocosms (Fig. S4a), identified by a concomitant decrease in salinity-normalized TA. This initial calcification then quickly ceased due to nutrient limitation. After the addition of nutrients in phase II, calcification increased again, particularly in the lower TA treatments and controls (Fig. S4). When then overall cumulative calcification is related to pCO₂ (Fig. 3), we find an optimum curve response as predicted in other lab studies for this CO₂ range (Gafar et al., 2018). It is important to note, that having calcification (Fig. S4a) hovering around the zero line or being negative is most likely related to the inherent uncertainty stemming from a mass balance involving four measurements with their individual uncertainties (TA, salinity, nitrate and phosphate). Nevertheless, the consistent emergence of an optimum curve suggests that, despite these uncertainties, the overarching pattern is preserved across the whole experiment and within each phase (Fig. S7, Table 1). Specifically, mild alkalinity treatments (<ΔTA 300 µmolkg-1) enhanced calcification, while it was reduced and inhibited at ΔTA 450 and 600 µmolkg-1 respectively (pCO2<100 µatm tested here). These responses fall within the range of mean sensitivity responses to pCO₂ for coccolithophores reported by Seifert et al. (2022). Intermediate pCO₂ levels seem to provide an optimal balance of bicarbonate ions and H⁺ concentration, enhancing calcification rates.

Both Krug et al. (2011) and Bach et al. (2011) hypothesized that inhibition of calcification could be the result of substrate (CO₂ and HCO3-) limitation on the one hand, and pH / H⁺ inhibition on the other. The reason for external H⁺ constituting an inhibitor is that, during coccolithophorid calcification, H⁺ is internally being generated (e.g., Gafar et al., 2019; Suffrian et al., 2011; Taylor et al., 2011) and eventually needs to be channeled out of the cell to maintain pH homeostasis (Cyronak et al., 2016). Concerning substrate limitation for both calcification and photosynthesis, it does not matter which carbon species is actually being taken up into the cell, as at decreasing seawater pCO₂, CO₂ leakage out of the cytosol (pH ∼7, Anning et al., 1996) will increase due to the concentration gradient. Up to a certain level, this could be compensated for by boosting carbon concentration mechanisms such as active CO₂ or HCO3- uptake, but this would come at increased metabolic costs (Badger and Price, 2003; Reinfelder, 2011).

Additionally, when cumulative TA-based calcification (CALC) is compared to cumulative coccolithophore abundance (based on flow-cytometric analysis), there is a linear, statistically significant relationship, suggesting that E. huxleyi was the dominant calcifier in our experiment, and showing that abundance can be some sort of measure for community calcification (Fig. S5). However, the fact that the difference between low and high TA addition in community calcification is about a factor of 10, while it is only a factor of 5 for abundance, indicates that not only cellular calcification rates were reduced at high TA additions, but also cellular CaCO₃ quotas. A finding consistent with culture studies (e.g., Bach et al., 2011; Gafar et al., 2018; Gafar and Schulz, 2018).

While the cumulative calcification calculated here encompasses overall coccolithophore bloom dynamics, the response of calcification to changes in carbonate chemistry are typically described by changes in cellular rates. It is possible to link both by calculating the amount of CaCO₃ that would be produced in a coccolithophore bloom using rates, derived from lab experiments, termed the CaCO₃ production potential (CCPP, Gafar et al., 2018). Using the rates collected by Gafar and Schulz (2018) for the coccolithophore E. huxleyi, which are dependent on carbonate chemistry, light and temperature, and our average in-situ conditions, we estimated normalized CCPP and compared it to normalized cumulative calcification (Fig. 3b). The statistically significant linear correlation suggests that the overall bloom dynamics derived from changes in measured carbonate chemistry align with the behavior at the cellular level derived independently from rates specific to the coccolithophore E. huxleyi. In summary, there is enhanced calcification and CCPP at intermediate levels of pCO₂ (∼250 µatm) following a non-CO₂-equilibrated addition of alkalinity, but these are negatively impacted when going towards lower and higher levels of pCO₂.

By accounting for the measured air-sea gas exchange, and factoring in the reductions caused by calcification, it is possible to isolate the change in DIC driven exclusively by biological activity such as photosynthesis and respiration (the NCP_DIC). The in situ NCP_DIC derived from cumulative changes in the DIC pool (Fig. 4a) encompasses autotrophic photosynthesis, decreasing DIC through the consumption of CO2/HCO3-, and increasing DIC by both autotrophic and heterotrophic respiration. An increase in photosynthetic activity can be observed after the nutrient addition during phase II, as NCP_DIC reached peak values, corroborated by the increase in Chl a (Fig. S1f). Furthermore, there appeared to be a general trend of silicate-based treatments reaching higher levels of NCP_DIC than the calcium-based ones (Fig. 4c). This may be linked to the stoichiometry of inorganic nutrient availability. While in both treatments, nitrogen uptake was ∼3.6 µM in total. In the silicate-based treatments, silicate and nitrogen were consumed in a ratio of up to ∼3:1. In contrast, in the calcium-based treatments, the uptake ratio was reversed at ∼1:3, meaning that up to ∼10 times more silicate was drawn down in the silicate-based treatments within 20 d. This difference in uptake ratios was independent of the TA level, and thus likely driven by changes in diatom physiology and community composition, as discussed by Ferderer et al. (2024). Diatoms were likely Si-limited in phase 0, but could later allocate additional resources to growth (Inomura et al., 2023), potentially also explaining the higher measured NCP rates (Marín-Samper et al., 2024a) in the silicate-based treatments and, in turn, contributing to/causing the mineral effect detected here on NCP_DIC.

It is also important to note that Ca²⁺ is already present in seawater at high background concentrations, and our additions only altered it by 0.8 %–3.1 %. In contrast, silicate was increased by several orders of magnitude, directly affecting a limiting macronutrient for diatoms. Hence, Si is the more plausible driver of the observed mineral effect on NCP_DIC. Although no direct diatom counts were available, biogenic silica (BSi) measurements (Fig. S6) provide a useful proxy to corroborate this.

Phase-specific analysis showed only the pCO₂ effect as statistically significant on NCP_DIC during phase 1 (Fig. S7). However, this effect was small compared to the mineral-type effect observed in phase 2. When considering both phases together, only the mineral-type effect prevailed, suggesting that differences according to mineral-type might have been harder to detect under nutrient-limited conditions.

In summary, it appears that while there was no negative overall effect of the pCO₂ level on NCP_DIC, there was a positive effect of the silicate amendment. This is in line with the hypothesis of the green ocean suggested by Bach et al. (2019), though the white ocean enhanced by added calcium was not observed here. One reason could be the relatively low abundances of coccolithophores at the onset of the experiment, another the fact that only increased calcium concentrations paired with increased DIC upon full CO₂ equilibration have been hypothesized to promote coccolithophorid calcification and growth (compare Bach, 2015; Bach et al., 2019). However, given the slow equilibration time on the order of years, and the fact that during this time the TA-treated waters are likely subject to substantial dilution (significantly reducing TA and hence the DIC increase upon full equilibration), a “white ocean” might not be something to expect.

Furthermore, since NCP_DIC calculations are based on relative change over time, they are robust to potential offsets in pH-derived DIC estimates in the higher TA treatments. Thus, the mineral-type effects observed in NCP_DIC would not change even if all values are skewed.

The delay in bloom formation towards lower pCO₂ observed here (Fig. 4b) was also reflected in in vitro oxygen production rates (Marín-Samper et al., 2024a) and was linked to both the mineral treatment and the TA level. These delays can be attributed to the previously reported, species-specific negative relationships between elevated pH / low pCO₂ levels and phytoplankton growth rates (Chen and Durbin, 1994; Hansen, 2002), as well as the aforementioned substrate/inhibitor concept, affecting growth rates (see Sect. 4.2). However, in terms of ecological significance, it is not clear if a phytoplankton bloom delay causes knock-on effects for higher trophic levels (e.g., mismatches with grazers). Hence, at this stage, it is difficult to draw clear conclusions or provide strong recommendations.

The main difference between the estimated in situ NCP_DIC and NCP obtained from O₂ measurements in separate incubations (see Marín-Samper et al., 2024a) is the exclusion of grazers (larger than 280 µm) in the latter since samples are filtered before incubation. Thus, comparing both approaches provides an estimate of the contribution of larger zooplankton and fish respiration to the carbon balance, calculated as (Eq. 8): 8RZ=NCP-NCPDIC, with RZ denoting cumulative zooplankton respiration (for simplicity, all larger grazers are encompassed here) and NCP the cumulative net community production derived from O₂ incubation measurements, assuming a 1:1 O2:C conversion ratio. While this assumption provides a useful first-order estimate, it represents a simplification, since true respiratory and photosynthetic quotients can vary with community composition, nutrient availability, and metabolic pathways (Robinson, 2019).

Cumulative RZ ranged from 30 to 60 µmolkg-1 and was significantly correlated to pCO₂ as well as to mineral treatment when considering the entire experiment (phases 1 + 2; Fig. 5). Interestingly, treatments with lower NCP_DIC at higher TA addition (lower pCO₂), showed higher RZ, particularly calcium-based treatments, suggesting enhanced top-down control on primary productivity. Possible explanations include direct effects of high pH at low pCO₂ on zooplankton metabolism (Hessen and Nilssen, 1983; Pedersen and Hansen, 2003), if requiring more energy along the pCO₂ gradient to maintain homeostasis under suboptimal conditions, as well as indirect effects via changes in phytoplankton community composition, influencing prey availability between the two mineral types. Furthermore, under low pCO₂ conditions, phytoplankton can exhibit lower C:N ratios (Burkhardt et al., 1999), which increases their quality as food for zooplankton, likely leading to better growth and hence respiration rates.

Phase-specific analysis revealed that gradient effects dominated during the more oligotrophic phase 1, with respiration increasing towards lower pCO₂ (Fig. S7, Table 1). However, the mineral-type effect was only statistically significant when both phases were combined, suggesting/supporting the hypothesis that shifts in community composition across phase, and their effects on prey availability, likely drove the cumulative pattern.

All this suggests that the response to carbonate chemistry perturbations might have resulted in an enhanced top-down control on primary productivity particularly in the higher TA addition treatments (which resulted in lower pCO₂). Interestingly, Goldenberg et al. (2024), who investigated early life stages of fish during the same mesocosm experiment, showed that though fish abundance did not correlate to the treatment gradient, biomass did, and greater biomass could have a greater contribution to respiration. This aligns with our finding of higher respiration under lower pCO₂ (corresponding to higher TA additions). Although no clear explanation could be found by the authors, it highlights the complex trophic interactions in pelagic ecosystems that require untangling. Finally, a recent study showed enhanced copepod grazing rates for certain OAE scenarios, although respiration rates did not seem to correlate with alkalinity addition (Bhaumik et al., 2025), again highlighting the lack of mechanistic understanding of the underlying processes.