The potential of 230 Th for detection of ocean acidification impacts on pelagic carbonate production

Concentrations of dissolved 230Th in the ocean water column increase with depth due to scavenging 10 and downward particle flux. Due to the 230Th scavenging process, any change in the calcium carbonate (CaCO3) fraction of the marine particle flux due to changes in biological CaCO3 hard shell production as a consequence of progressing ocean acidification would be reflected in the dissolved 230Th activity. Our prognostic simulations with a biogeochemical ocean general circulation model using different scenarios for the reduction of CaCO3 production under ocean acidification and different greenhouse gas emission scenarios (RCPs 8.5 to 2.6) reveal the potential 15 for deep 230Th measurements to detect reduced CaCO3 production at the sea surface. The time of emergence of an acidification induced signal on dissolved 230Th is of the same order of magnitude as for alkalinity measurements. Interannual and decadal variability in other factors than a reduction in CaCO3 hard shell production may mask the ocean acidification induced signal in dissolved 230Th and make detection of the pure CaCO3 induced signal more difficult so that only really strong changes in marine CaCO3 export would be unambiguously identifiable soon. 20 Nevertheless, the impact of changes in CaCO3 export production on marine 230Th are stronger than those for changes in POC (particulate organic carbon) or clay fluxes.

Abstract. Concentrations of dissolved 230 Th in the ocean water column increase with depth due to scavenging and downward particle flux. Due to the 230 Th scavenging process, any change in the calcium carbonate (CaCO 3 ) fraction of the marine particle flux due to changes in biological CaCO 3 hard-shell production as a consequence of progressing ocean acidification would be reflected in the dissolved 230 Th activity. Our prognostic simulations with a biogeochemical ocean general circulation model using different scenarios for the reduction of CaCO 3 production under ocean acidification and different greenhouse gas emission scenarios -the Representative Concentration Pathways (RCPs) 8.5 to 2.6 -reveal the potential for deep 230 Th measurements to detect reduced CaCO 3 production at the sea surface. The time of emergence of an acidification-induced signal on dissolved 230 Th is of the same order of magnitude as for alkalinity measurements. Interannual and decadal variability in factors other than a reduction in CaCO 3 hard-shell production may mask the ocean-acidification-induced signal in dissolved 230 Th and make detection of the pure CaCO 3 -induced signal more difficult so that only really strong changes in marine CaCO 3 export would be unambiguously identifiable soon. Nevertheless, the impacts of changes in CaCO 3 export production on marine 230 Th are stronger than those for changes in POC (particulate organic carbon) or clay fluxes.

Introduction
Progressing ocean acidification is a fact. It can be directly seen from measurements at Eulerian time series stations (e.g. Bates, 2007;Dore et al., 2009;Santana-Casiano et al., 2007) and also at larger depth from high-quality deep hydrography data (Olafsson et al., 2009). Depending on the emission scenario for CO 2 , the decrease in ocean pH and the decrease in carbonate saturation will continue and become more pressing during this century (e.g. Orr et al., 2005;Steinacher et al., 2009;Bopp et al., 2013). Temporal and regional patterns of changes in pH and carbonate saturation are relatively straightforwardly to project by Earth system models including marine inorganic carbon chemistry formulations (e.g. Bopp et al., 2013). They also can be monitored through long-term high-quality measurements of the inorganic carbon system. However, monitoring impacts of ocean acidification on biological processes remains challenging. While for some organisms, especially corals (Kleypas et al., 1999), detrimental effects due to decreasing pH and carbonate saturation have been determined, the various physiological impacts of ocean acidification on specific organisms and ecosystem functioning are still under investigation (e.g. Iglesias-Rodriguez et al., 2008;Kroeker et al., 2013;Meyer and Riebesell, 2015;Riebesell et al., 2007). At present, a series of possible pH dependencies of governing marine carbon cycle parameters (such as elemental stoichiometric ratios and nutrient uptake kinetics under biological particle production) are discussed. So far, the potential decrease in calcification due to the lowering of carbonate satu-Published by Copernicus Publications on behalf of the European Geosciences Union. ration under high pCO 2 is among the key changes which may be expected to occur, especially when it comes to organisms building aragonite shells (the metastable form of CaCO 3 , calcite, has a lower solubility than aragonite) (Raven et al., 2005). Two important questions emerge: (1) if changes in biological calcification would indeed occur in the ocean during the coming years, how can they be detected and monitored by observational frameworks (through which methodology and through which variable)? (2) In which oceanic region could one observe early signals of these changes unambiguously at the earliest possible stage?
Development of such early warning systems is vital in order to check the validity of parameterisations of pHdependent processes in ocean models and to take appropriate mitigation/adaptation measures to diminish the consequences of potential considerable ecosystem changes due to ocean acidification. Such changes could affect the marine food chain. In a global modelling study, Ilyina et al. (2009) quantified the detection thresholds for changes in alkalinity due to changes in a series of possible formulations for the reduction of calcification (with the term "calcification" we mean here the production of hard CaCO 3 parts by marine biota) with pH decrease. The logic behind the approach of Ilyina et al. (2009) is as follows: if ocean acidification leads to a decrease in calcification, more CO 2− 3 ions would be retained in the surface water and not become incorporated into CaCO 3 shell material. Changes in CO 2− 3 ion concentration would induce a change in ocean total alkalinity, which could eventually be measured against an earlier baseline. According to that study, the tropical Pacific (with high CaCO 3 production rates) would be the region for detecting such alkalinity changes first, as the anticipated change in CaCO 3 production would be largest there in absolute terms. For intermediate dependencies of CaCO 3 production on pH/carbonate saturation, a reduction in biological calcification could unequivocally only be diagnosed from approximately the year 2035 on. In the Arctic Ocean, where pH changes are expected to be most pronounced, large-scale changes in calcification would be detectable even later on the basis of alkalinity measurements due to the overall lower biological production rates. Likewise, detection may be additionally complicated due to signals from natural seasonal and interannual variations in surface total alkalinity (Carter et al., 2016). There is thus a need for the development of novel detection methods. Heinze et al. (2006) investigated the impact of varying rain ratios CaCO 3 : C org on the distribution of the radionuclides thorium ( 230 Th) and protactinium ( 231 Pa), as well as beryllium ( 10 Be). As the rain ratio describes the average amount of carbon atoms incorporated into CaCO 3 shell material relative to the amount of carbon atoms incorporated into organic matter by plankton, a reduction in calcification would lead to a rain ratio decrease. Indeed, especially for 230 Th, due to its affinity for scavenging to CaCO 3 particles, a considerable increase in the concentration of dissolved 230 Th with depth and time was obtained in a sensitivity experiment with strong rain Figure 1. Schematic illustration of the equilibration process between the dissolved and particle-attached phases of 230 Th and the increasing concentrations downward in the water column. ratio reduction. In this paper, we explore the option of using radionuclides for diagnosing changes in calcification and respectively reduced downward marine particle flux in more detail.

The concept -radionuclides and particles
We focus here on 230 Th, a long-lived radionuclide (half-life 7.5 × 10 4 yr) and a highly particle reactive metal. 230 Th is produced in the uranium decay series. As uranium has a very long residence time in the ocean and quasi-homogenous concentrations in seawater, the present natural marine 230 Th source is constant everywhere in the ocean water column and thus ideally suited for using 230 Th as a tracer in modelling studies. 230 Th is strongly particle reactive and is removed from the water column quickly through scavenging by the vertical particle flux in the ocean (for a summary, please see Henderson et al., 1999). The majority of 230 Th is removed from the water column and will not re-enter the bottom waters, e.g. through sediment pore water diffusion (though resuspension may potentially cause some re-release). In spite of its strong particle reactivity, concentrations of dissolved 230 Th show horizontal as well as vertical gradients in the water column which are induced by the oceanic current field and differential particle fluxes as well as particle concentrations (e.g. Henderson et al., 1999;Yu et al., 1996). The distribution of particle-bound 230 Th follows the particle concentrations and particle fluxes. Both the dissolved 230 Th and the particle-bound 230 Th show nutrient-type vertical profiles with low values at the surface and increasing concentrations with depth. Dissolved 230 Th increases with depth as particles continuously carry 230 Th downwards and new equilibria between the dissolved phase and the particle-attached phase establish as illustrated in Fig. 1. The equilibrium between the concentration of the dissolved phase [ 230 Th diss ] and the concentration of the particle-attached phase [ 230 Th part ] can be described in an analog way to a chemical reaction equation: The respective analog for the mass action law constant describing to which extent the reaction from left to right is carried out is then given through the k d value governing the equilibrium between the dissolved and particle-bound phases for the radionuclide: Often, k d values are formulated to account for a shift of the equilibrium towards the particle-attached phase at low particle mass concentrations, assuming that low particulate concentrations occur in parallel to low particle sizes with correspondingly high reactive surface areas when compared to large particles (e.g. Honeyman et al., 1988). Respective formulations have been used successfully in 230 Th modelling studies (Heinze et al., 2006;Henderson et al., 1999). How would then the distribution of dissolved 230 Th in the ocean reflect changes in marine calcification at the sea surface? There are indications that 230 Th is attached first of all to CaCO 3 and clay particles in the water column as preferential carrier phases (Chase et al., 2002(Chase et al., , 2003Luo and Ku, 1999;Hayes et al., 2015b). Indeed, Heinze et al. (2006) reproduced the large-scale 230 Th distribution in the world ocean with a coarse-resolution biogeochemical ocean general circulation model based on a formulation allowing 230 Th scavenging only by CaCO 3 and clay particles. Other studies have indicated that preferential carrier phases may vary regionally in the ocean (Scholten et al., 2005). However, Heinze et al. (2006) could demonstrate that rain ratio changes (CaCO 3 : C org ) in marine biological particle export production could also be recorded, if 230 Th became in addition attached to particulate organic carbon (POC). A weakening of CaCO 3 particle production would result in a less efficient 230 Th scavenging as less particles (in terms of mass) would be available to carry 230 Th to larger depths and finally the sediment. Changes in the strength of CaCO 3 production and the respective downward particle flux are reflected increasingly better in the 230 Th distribution with increasing depth (see Heinze et al., 2006) due to two reasons. First, CaCO 3 particles get less well degraded as POC (which is remineralised mostly in the upper 1000 m of the water column) and thus reach larger depth; this is also corroborated from sediment trap measurements (e.g. Honjo, 1996). Second, due to the scavenging of 230 Th by particles, the vertical downward particle flux, and the equilibration between dissolved and particle-bound phases, 230 Th is continuously transferred from shallower layers to larger depths. Thus, temporal changes in 230 Th scavenging in the upper ocean become enlarged in the deep 230 Th distribution as in a kind of "magnifying glass" (see Fig. 1). We investigate here whether this feature can be exploited for an early detection method of large-scale reductions in calcification and correspondingly reduced rates in CaCO 3 particle export and CaCO 3 particle fluxes through the water column.

Model description
In this study, we use the Hamburg ocean carbon cycle circulation model, HAMOCC (Maier-Reimer, 1993), in its annually averaged version (time step of 1 year; Heinze and Maier-Reimer, 1999;Heinze et al., 2009Heinze et al., , 2016 with a horizontal resolution of 3.5 • × 3.5 • . This coarse-resolution model is computationally very efficient and useful when multiple integrations are needed, such as for the testing and adjusting of scavenging codes. An advantage of this fast model is that it includes a fully equilibrated early diagenesis module (10 layers) under each grid point and thus can account for alkalinity changes induced by dissolution of CaCO 3 from the seafloor. The model version employed here corresponds to the version used in Heinze et al. (2009Heinze et al. ( , 2016 with a slightly updated scavenging module of Heinze et al. (2006) with respect to the formulation of the equilibrium coefficient governing the distribution of 230 Th between the dissolved and particle-attached phase (see below). For details, please see those publications. The model uses a fixed ocean velocity field (and thus provides no dynamical computation of the ocean currents; velocities, temperature, salinity, and ice cover are rather read from an input file). The model includes an atmospheric compartment ("slab atmosphere") which allows for prognostic computation of the atmospheric CO 2 concentration as well as meridional atmospheric CO 2 transport. We describe here only briefly features of specific relevance for this study. The water column is structured into 11 layers (centred at 25, 75, 150, 250, 450, 700, 1000, 2000, 3000, 4000, and 5000 m). The bioturbated top sediment zone of the early diagenesis module is divided into 10 layers which are separated by interfaces at 0, 0.3, 0.6, 1.1, 1.6, 2.1, 3.1, 4.1, 5.1, 7.55, and 10 cm "downcore". We make the simplifying assumption that no pore water reactions take place below 10 cm depth in the sediment (see, e.g. Smith and Rabouille, 2002;Boudreau, 1997). The biogeochemical model includes the processes of air-sea gas exchange, biogenic particle export production out of the ocean surface layer, particle flux through the water column and particle degradation by dissolution as well as remineralisation, transport of dissolved substances with the ocean currents, deposition of particulate constituents on the ocean floor, pore water chemistry and diffusion, advection of solid sediment weight fractions (organic carbon, organic phosphorus, CaCO 3 , opal, and clay), bioturbation, and sediment accumulation (export out of the sediment mixed layer). The model predicts the following tracer concentrations in the atmosphere, the ocean water column, and in the sediments. Atmospheric tracers include the concentrations of CO 2 (carbon dioxide) and O 2 . In the water column, concentrations of DIC (dissolved inorganic carbon), POC (particulate organic carbon), POP (particulate organic phosphorus), DOC (dissolved organic carbon), CaCO 3 (calcium carbonate or particulate inorganic carbon), dissolved oxygen O 2 , dissolved PO 3− are calculated. In the sediment pore waters, the same dissolved substances as in the water column, as well as solid sediment constituents such as clay, CaCO 3 , opal, and organic carbon, are simulated. The inorganic carbon chemistry is computed following Dickson et al. (2007). In the advection scheme and for the other chemical reactions, DIC and total alkalinity (TAlk) are used as "master tracers" form which derived quantities such as the CO 2− 3 concentration and the pH value are computed through a Newton-Raphson algorithm. In the annually averaged model as employed in this study, only export production of biogenic particles is modelled (and no explicit phytoplankton and zooplankton concentrations). Particle production takes place in the model surface layer representing the euphotic zone. Phosphate serves as the biolimiting nutrient. POC and opal export productions are simulated following Michaelis-Menten kinetics for nutrient uptake (e.g. Sarmiento and Gruber, 2006) (where the phytoplankton concentration is replaced by the phosphate concentration as ecosystem processes as such are not explicitly modelled): where P POC and P opal are the POC and opal export production rates (mol L −1 yr −1 ), Red(C : P) is the Redfield ratio C : P, V POC max and V opal max are the maximum uptake rate of phosphate and silicic acid from the water column (yr −1 ), and K POC s as well as K opal s are the respective half-saturation constants. V POC max , V opal max , K POC s , and K opal s are simulated as a function of sea surface temperature as described by Heinze et al. (2003). POP production follows POC production with a constant stoichiometry here. The export production of CaCO 3 is coupled to the local production ratio P opal / P POC . It starts to increase gradually (parameter R; see below) if P opal / P POC sinks below a threshold value S opal , i.e. when not enough silicic acid is available in the ocean surface layer to fuel full diatom growth: Parameter R is the maximum possible rain ratio C(CaCO 3 ) : C(POC), a is the CaCO 3 saturation-dependent factor to account for an ocean acidification impact (see  . Assumed sensitivities of CaCO 3 export production in response to changes in calcite saturation ( ). The rain ratio coefficient is used as factor "a" applied to the right-hand side of the equation for CaCO 3 production (see Eq. 1). For the control run, a was fixed at each grid point according to the prevailing pre-industrial carbonate saturation (following the moderate sensitivity). Blue: moderate sensitivity. Green: linear sensitivity. Red: extreme sensitivity. The shape of the curves would look similar at each grid point; only the pre-industrial crossover point would be different.
threshold value of P opal / P POC for gradual onset of CaCO 3 production. Particle fluxes and particle degradation are simulated through balance equations for sinking particulate matter as in Heinze et al. (2009Heinze et al. ( , 2016, taking the saturation state for CaCO 3 and biogenic silica into account. Scavenging of 230 Th is simulated according the reversible first-order scavenging reaction Heinze et al., 2006): c part is the concentration of particle-bound 230 Th, c diss the concentration of dissolved 230 Th, and c EQ part the equilibrium concentration of particle-bound 230 Th. The first-order rate constant K (in yr −1 ) is set here to 10 4 yr −1 , thus assuming a quasi-instantaneous equilibration. Concentration of suspended particulate material is represented by M. For the partitioning coefficient k d between the dissolved and particleattached phases of 230 Th, we follow the formulation of Honeyman et al. (1988), which accounts implicitly for the changing reactive surfaces of particles with particle size log 10 k first guess d = A + B · log 10 M, where M is the particle concentration (here taken in mg particles per litre) and A as well as B are tuneable parameters. In addition to 230 Th, we carry also 231 Pa as well as 10 Be in our model (see Heinze et al., 2006) but focus here on 230 Th only. For each of these radionuclides, the preferential carrier phase can be selected in a dedicated switchboard. For 230 Th scavenging, we used here CaCO 3 , POC, and clay as carrier phases. With introducing a particle-specific scavenging following Hayes et al. (2015b), the final formulation for the scavenging equilibrium coefficient then becomes where C and F are the weighting coefficients and fractions of total local particle mass. The A, B, and C values are included in Table 1. Atlantic and Pacific cross sections with the k d values for the control simulation without CO 2 emissions are given in Fig. S1 (in the Supplement).
The clay flux is computed according to the modern dust deposition from Mahowald et al. (1999) assuming that respective clay particles are chemically quasi-inert in seawater. Particle-bound 230 Th phases never get to zero, even if all biogenic particles are degraded, because at each grid point there is -at least a tiny -dust flux consisting of inert clay. Therefore, particle concentrations as such never go to zero (a minimum concentration could be prescribed, but this was not necessary in our case). Because the Atlantic dust deposition may be overestimated , we assumed for the 230 Th scavenging only 25 % of the respective clay flux strength in order to avoid too strong scavenging in Atlantic surface waters (see also our sensitivity experiment assuming 100 % in comparison below). We allow biological production under sea ice scaled with the local sea ice thickness.

Control run, scenario experiments, and sensitivity experiments
The model was spun up restarting from an earlier integration and computed into quasi-equilibrium including the sediment over 40 000 years. The equilibrium coefficient k d was determined through a visual fit to observations of dissolved 230 Th taken from the literature (Bacon and Anderson, 1982;Bacon et al., 1989;Chase et al., 2002;Cochran et al., 1987Cochran et al., , 1995Colley et al., 1995;Guo et al., 1995;Huh and Beasley, 1987;Moore, 1981;Moran et al., 1995Moran et al., , 1997Nozaki and Horibe, 1983;Roy-Barman et al., 1996;Scholten et al., 1995;Vanderloeff and Berger, 1993;Vogler et al., 1998) and combined with the data as given in the GEOTRACES intermediate data product version 3 (Mawji et al., 2015;Hayes et al., 2015a, b;Deng et al., 2014). Important global bulk numbers and the parameters for the partitioning coefficient k d are listed in Table 1. Meridional sections of dissolved 230 Th concentrations for the Atlantic and Pacific oceans are given in Fig. 3. The vertical distribution clearly shows the increase of concentrations with depth due to the downward transfer of 230 Th with the marine particle flux. A comparison of simulated and observed dissolved 230 Th values in the water column along a meridional Atlantic cross section (Fig. 4) indicates that the model dissolved 230 Th is in fairly good agreement with observations, except for a strong overestimation in some bottom water locations. This could be due to deficiencies in the flow field and corresponding problems or due to the lack of explicit simulations of hydrothermal vents and sediment resus- Table 1. Summary of control run results (run P0; see Table 2).

Key variable as simulated
Unit Control run value Atmospheric CO 2 mixing ratio (ppm) 277.6 POC export production (pre-industrial) pension. We carried out a respective sensitivity experiment (see discussion further below). A scatter plot of modelled and observed 230 Th values and the locations of stations with observed data are shown in Fig. 5. The standard-run export production rates for POC and CaCO 3 are given as maps in Fig. 6.
An overview concerning all model simulations is given in Table 2. We carried out model projections under four different scenarios for changes in calcification (including control runs with constant calcification but rising atmospheric CO 2 ) and four different scenarios of the future development of the atmospheric CO 2 concentration (C8.5-2.6, L8.5-2.6, M8.5-2.6, and E8.5-2.6; see Table 2). All scenarios were restarted from the same previously performed spin-up model run reflecting a pre-industrial biogeochemical state of the ocean. Throughout the experiments, the ocean circulation field was not changed. The model was integrated during the calendar years 1200-2300 spanning a period of 1100 years. In the model spin-up simulation, the atmospheric CO 2 concentration was a prognostic variable. For the computations from 1700 onwards, we prescribed the atmospheric CO 2 concentration according to the Representative Concentration Pathways (RCPs) including their extension to the year 2300 (van Vuuren et al., 2011) used in CMIP5 (Coupled Model Intercomparison Project phase 5). Our simulations follow scenarios RCP2.6, RCP4.5, RCP6.0, and RCP8.5 designed to span a range of radiative forcing between 2.6 and 8.5 W m −2 by the year 2100. The respective atmospheric CO 2 concentrations are shown in Fig. S2. For the decrease in calcification with decreasing pH as well as carbonate saturation, functional relationships between CaCO 3 export production and carbonate saturation were chosen which correspond to the respective experiments in Ilyina et al. (2009). For each RCP, we carried out a control simulation with constant calcification, a moderate decrease of CaCO 3 production with decreasing saturation, a linear dependency simulation with respectively stronger decrease in CaCO 3 production), and an extreme scenario (Fig. 2). To date, no clear bulk formulation for the dependency of CaCO 3 export on CaCO 3 saturation exists. Therefore, the scenarios carried out here are only sen-    sitivity experiments and not overall exhaustive simulations in order to reproduce the entire range of possible changes in ocean biogeochemistry. As an indication for the robustness of our results, a suite of sensitivity experiments was carried out. In a first set of sensitivity experiments (P1-P5, Table 2), we investigated how much the pre-industrial control simulation depends on selected parameter choices. In P1, parameter A for k first guess d was slightly increased from 5.8 to 6.0 in order to test the dependency of the result from small changes in the still not very precisely known k d value for 230 Th. Given the effect of release of material from hydrothermal vents and resuspension of material from the sediments, we carried out experiment P2 with a strong increase in scavenging in the lowermost wet model grid cells (directly over the ocean floor). To this end, parameter A for k first guess d was increased by 0.5 units in these grid boxes. In our general reference run (P0), we applied scavenging to clay, as if the clay flux would only be 25 % of its prescribed value. Therefore, we also added the P3 run, where 100 % clay flux was assumed also for the 230 Th scavenging. Finally, in experiments P4 and P5, we switched off the scavenging to POC and CaCO 3 , respectively, in order to see the importance of these two biogenic particle species on the 230 Th distribution.
In a second set of sensitivity experiments, we tested how sensitive the future projection of dissolved 230 Th is in view of factors other than changes in CaCO 3 production (S1-S5; see Table 2). Because we use an annually averaged model without seasonal cycle and also employed fixed annual mean velocity field, we have to use approximations in order to see the effect of a change in the velocity field or other changes involving interannual/decadal variability. We use here the time series of the Atlantic meridional overturning from the fully fledged Earth system model NorESM for a historic rampup and subsequent RCP8.5 forcing (Tjiputra et al., 2016; see Fig. 2 therein). In experiment S1, we scaled the threedimensional velocity field (and also the convective mixing representation) of our simplified model with this time series, leading to an overall reduction in circulation strength. The formulation for this scaling is  RCP2.6 Extreme Calcification reduction S1 RCP8.5 Moderate As M8.5 but overturning reduced with rising CO 2 S2 RCP8.5 Moderate As M8.5 but V max for nutrient uptake reduced with rising CO 2 S3 RCP8.5 Moderate As M8.5 but particle sinking velocity reduced with rising CO 2 S4 RCP8.5 Moderate As M8.5 but changes as in S1M8.5, S2M8.5, S3M8.5 together S5 RCP8.5 Extreme As E8.5 but changes as in S1M8.5, S2M8.5, S3M8.5 together with Y being a scalar variable (such as the velocity components in the x, y, and, z directions, the convective adjustment, or other specific biogeochemical parameters), Y scaled the value as updated for the sensitivity experiments, Y reference the original value used in the reference run, and Overturning the relative change in Atlantic meridional overturning between a specific year after the year 1850 and the year 1850 itself ( Overturning would be zero until 1850). Using this scaling, we also investigated separately from a circulation change the consequences of variations in the maximum nutrient uptake velocity V max in our simple trophic model (a reduction in V max could be seen as a deterioration of growth conditions for plankton under climate change) (run S2), and in the sinking velocity of particles (corresponding to, e.g. a loss in ballast material) (run S3). In experiments S4 and S5, finally, the simultaneous effects of changes in the circulation, V max , and the particle sinking velocity were explored for the moderate (S4) and the extreme (S5) calcification scenarios. The list of potential factors with an influence on 230 Th is much longer; however, our choice of factors covers at least the physical forcing through the velocity field, surface biogeochemistry through V max , and three-dimensional biogeochemistry through the particle sinking speed.

Results and discussion
The results from the sensitivity experiments relative to the pre-industrial control run are summarised in Fig. 7. Increasing the scavenging equilibrium coefficient in general (run P1, Fig. 7a and b) or over the ocean bottom (run P2, Fig. 7c and d) corrects to a substantial degree the too high bottom water vales for dissolved 230 Th in the southern Atlantic but leads to high relative errors in the upper few hundred metres (where 230 Th concentrations are generally low). Too low upper ocean 230 Th concentrations occur also for increasing the scavenging to 100 % of the clay flux (run P3, Fig. 7e and f). Omitting, the scavenging to POC has only minor influence on the 230 Th distribution because of the relatively shallow recycling of POC (run P4, Fig. 7g and h). In contrast, cancelling the scavenging to CaCO 3 (and only retaining that to POC and clay) leads to a completely unrealistic dissolved 230 Th distri-    bution with at least a 5-fold overestimation (run P5, Fig. 7i and j; note change of colour code in Fig. 7i and j). These results demonstrate that the choice of critical parameters for our pre-industrial reference simulation (run P0) results in a reasonable dissolved 230 Th distribution. The crucial role of CaCO 3 particles for determining water column 230 Th has been confirmed.
We discuss now the results for the projections. According to the various scenarios applied, considerable changes in surface alkalinity (not shown) and corresponding changes in CaCO 3 export production evolve at high CO 2 (Fig. 8). The effect is on average larger in the Pacific than in the Atlantic due to the overall higher modelled biological production rates of CaCO 3 in the Pacific Ocean. The extreme calcification scenario would lead to vastly reduced CaCO 3 export production after the year 2100 in all RCPs except the most moderate (RCP2.6), where even a partial recovery occurs.
For dissolved 230 Th, we show here time series for depth levels 700, 2000, and 4000 m (Figs. S3, 9, and S4, respectively). According to recent intercalibration experiments, sizable discrepancies still exist between the absolute values in 230 Th measurements from different laboratories, though these measurements show a smaller scatter for replicates within a single laboratory. Intercalibration experiments reveal a standard deviation for measurements on one sample by different laboratories of about 0.07-0.08 dpm/1000 L (Anderson et al., 2012). We take 0.075 dpm/1000 L as an approximate indicative value for the detection level for changes in dissolved 230 Th concentrations in seawater (Figs. S3, S4, and 9; see orange lines therein). This may be a somewhat optimistic estimate for samples of past decades, but future de-velopments of measuring techniques could possibly reduce measurement errors and the spread across analyses from different laboratories and application of different measurement methods. The detection level is shown here relative to the modelled pre-industrial values. The time of emergence as indicated by the orange line in Figs. S3, S4, and 9 is therefore the earliest possible time of detection if pre-industrial 230 Th would be known. Figures S3, S4, and 9 show the increase of dissolved 230 Th activities with depth from 700 m (Fig. S3), over 2000 m (Fig. 9), to finally 4000 m (Fig. S4). At 700 m, the increase in 230 Th activities due to the assumed ocean acidification effect is too small to unambiguously show an effect for RCP2.6 (Fig. S3). This picture changes when going down to 2000 m and below, where an effect would be detectable within this century at least for the stronger forcing scenarios (RCP8.5 and RCP6.0) (Figs. 9 and S4). For constant CaCO 3 production, the intermediate and deep-water 230 Th activities start to rise around the year 2100 as well (see black curves in Figs. S3, S4, and 9). This effect is due to the increasing dissolution of CaCO 3 particles in the water column in parallel with downward mixing of waters that carry anthropogenic loads of dissolved organic carbon and hence subsurface and deep acidification. The effect is most important in areas where anthropogenic carbon is mixed down quickly and induces a significant shoaling of the CaCO 3 saturation level and CaCO 3 lysocline as well the carbonate compensation depth through deep-water acidification. Parts of the deep Pacific are not as much influenced by this as compared to the Atlantic within the 21st century. The control run in the figures does not represent the reference run with constant pre-industrial atmospheric pCO 2 but the run with constant  Figure 8. Temporal evolution of Atlantic (left column panels) and Pacific (right column panels) annual mean CaCO 3 export production under the different scenarios for the sensitivity of calcification under high CO 2 (unit: Gt C yr −1 ). From top to bottom for greenhouse gas scenarios RCP8.5, RCP6.0, RCP4.5, and RCP2.6.
CaCO 3 production and rising pCO 2 according to the RCP scenarios.
In order to look for most suitable regions for detection of ocean acidification effects on CaCO 3 particle fluxes using modern 230 Th data, we plotted the year of emergence for different reference years and different assumed detection levels (Fig. 10). The year of emergence is defined here as the earliest possible year of a potential detection in the case that the 230 Th signal (rate of change over time) is only influenced by a decrease in calcification and not any other processes (such as potential shifts in circulation and associated changes in biological particle production). We choose 2010 as reference year, to which the 230 Th activity at a later stage would be compared to, and 0.075 dpm/1000 L as the limit for the de-tection of a signal. For the moderate calcification scenario, the time of emergence at least at 4000 m would be comparable to the time of emergence as potentially inferable through repeat surface alkalinity measurements (Ilyina et al., 2009) ( Fig. 10a and b). For the extreme scenario, the signal would be identifiable considerably earlier. The signal emerges earliest in deep Pacific waters.
The results on the early warning indicator as shown in Fig. 10 are based on the simplifying assumption that only changes in calcification caused by ocean acidification influence the water column 230 Th concentrations under progressing acidification and climate forcing. We therefore also mimicked the effect of interannual/decadal variability in circulation and biogeochemical parameters (runs S1-S5, Ta-  ble 2). Comparing the projections under RCP8.5 and the moderate calcification scenario for the case without interannual/decadal variability (run M8.5, Table 2) with the one including variability (runs S1-S3, Table 2) for the middle of this century reveals differences in the order of the assumed detection level for changes in the velocity field and the particle sinking velocity and smaller changes for varying V max (Fig. 11a-f). The effect on modulating water column 230 Th can be quite variable for different oceanic domains. Deviations can add up or compensate for each other locally. For simultaneous changes of circulation, V max , and particle sinking velocity, the overall deviations from the run without variability amount to 2-3 times the theoretical detection limit ( Fig. 11g and h). We repeated the analysis of the time of emergence with the runs including interannual variability for the moderate (S4) and also the extreme (S5) scenario. For this analysis, we again chose 2010 as the reference year but increased the detection limit to 2.5 × 0.075 dpm/1000 L. Under these more realistic conditions, the time of emergence would be shifted to the end of the century for the moderate calcification scenario and to the middle of the century for the extreme scenario (Fig. 12).
The question arises whether surface alkalinity and deep 230 Th measurements possibly could be combined successfully in order to detect ocean acidification impacts on calcification. We could, so far, not detect any simple relation- ship between surface changes in CaCO 3 export production, surface alkalinity, and 230 Th in spite of the conceptually clear interdependencies. In order to illustrate this, we plotted changes in surface alkalinity and 3000 m 230 Th over time for the run with the moderate calcification scenario and RCP8.5 forcing between 2040 and 2010. The distributions show a somewhat consistent picture without interannual/decadal variability (run M8.5, Fig. 13a and b) with general rises in both surface alkalinity and deep 230 Th. For the case including interannual/decadal variability (run S4, Fig. 13c and d), the temporal gradients in both variables become smaller and the deep 230 Th even may reverse. This does not mean that it would be impossible to construct a combined Alk-230 Th tracer -only that such a tracer cannot be derived from this study.
As compared to the detection approach for ocean acidification impacts through total alkalinity measurements as pursued by Ilyina et al. (2009), the 230 Th method presented here results in similar time of emergence of a signal reflecting changes in biocalcification if one omits interannual/decadal variability and other possible influence factors. Probably the most important limitation of the approach here is the fixed velocity field which does not vary realistically with seasons, interannual variability, and climatic variability modes (such as the El Niño-Southern Oscillation or North Atlantic Oscillation). Also the effect of boundary scavenging (Anderson et al., 1983;Roy-Barman et al., 2009), i.e. the transport of dissolved 230 Th from areas of low particle concentrations to those with high particle concentrations (especially at ocean margins and shelf seas), is not spatially resolved in our model. Further uncertainties are associated with the choice of the k d values and particle-specific reactivity.
The extreme scenario on pH-dependent CaCO 3 production is likely to be an overestimation. According to the results from this study and Ilyina et al. (2009), a respective large change in the real world would probably have already been detected through the alkalinity signal and other methods such as remote sensing and sediment trap measurements. In addition, the reaction of CaCO 3 shell producing organisms to sinking carbonate saturation levels is not a simple function and can vary between taxa (Kroeker et al., 2013).
The detection method for ocean acidification impacts through 230 Th would work best if the pre-industrial values for dissolved oceanic 230 Th activities could reliably be re-   constructed. From core top samples and corals, one could possibly determine whether the particle-attached 230 Th activities and hence those of the dissolved fraction would have undergone any variations over the last centuries and millennia. However, the overall uncertainties associated with the formation and analysis of the paleo-record values may be too high to provide an accurate enough baseline for comparisons with modern water column measurements.

Conclusions
In this study, we investigated the potential of the particle reactive radionuclide 230 Th for detection of reduced calcification by biota due to progressing ocean acidification. The time of emergence of a dissolved 230 Th activity signal with respect to a reference year and reference activity is about the same as for detection approaches employing alkalinity if one disregards effects of interannual/decadal variability in the flow field and influence factors other than ocean-acidificationinduced changes in CaCO 3 export production. Taking into account interannual/decadal variability including ocean cir-culation changes may delay the signal emergence by several decades. Nevertheless, regular selected reoccupations of a series of deep stations in the Pacific Ocean and Southern Ocean with highest-quality 230 Th measurements would be helpful to accompany alkalinity measurements that are easier to make in order to see whether both tracers give consistent results. Surface alkalinity measurements include signals of natural variability on seasonal and multiyear scales. Likewise, changes in ocean circulation and changes in biological particle production due to processes other than ocean acidification may lead to changes in the marine 230 Th distribution (and alkalinity). As the deep ocean is less prone to effects of natural variability and the quality of observations does not change with depth, deep-ocean observations of 230 Th could be advantageous for monitoring and detecting ocean acidification effects on calcification. In any case, 230 Th represents a fascinating magnifying glass for changes in ocean surface processes seen through the deep-ocean signal. Its potential has not yet been fully exploited. Data availability. The model output data and the model code data are available from the corresponding author by request.
Competing interests. The authors declare that they have no conflict of interest.
Special issue statement. This article is part of the special issue "Progress in quantifying ocean biogeochemistry -in honour of Ernst Maier-Reimer". It is not associated with a conference.
Acknowledgements. This work was partially supported through the "European Project on Ocean Acidification" EPOCA and the project "Changes in Carbon Uptake and Emissions by Oceans in a Changing Climate" CARBOCHANGE that received funding from the European Community's Seventh Framework Programme (FP7) under grant agreement nos. 211384 and 264879, respectively. The Research Council of Norway supported this study through the project "Overturning circulation and its implications for global carbon cycle in coupled models" (ORGANIC; grant no. 239965) and the nationally coordinated project "Earth system modelling of climate variations in the Anthropocene" (EVA; grant no. 229771). This is a contribution to the Bjerknes Centre for Climate Research (Bergen, Norway). A part of the computations was carried out under project NN2980K at the Norwegian Metacenter for Computational Science (NOTUR) and its dedicated storage and archiving project NorStore (NS2980k). This is an in-kind contribution to the GEOTRACES project. We would like to thank the GEOTRACES scientists for the interesting discussions during several workshops and for the fabulous new observational data. Jerry Tjiputra kindly made available the AMOC data from NorESM. Two constructive referees helped to improve the manuscript. Christoph Heinze acknowledges sabbatical funds from the University of Bergen.
Edited by: Arne Winguth Reviewed by: two anonymous referees