Characterization factors for ocean acidification impacts on marine biodiversity

Rising greenhouse gas emissions do not only accelerate climate change but also make the ocean more acidic. This applies above all to carbon dioxide (CO2). Lower ocean pH levels threaten marine ecosystems and especially strongly calcifying species. Impacts on marine ecosystem quality are currently underrepresented in life cycle assessments (LCAs). Here, we developed characterization factors for the life cycle impact assessment of ocean acidification. Our main contribution was developing new species sensitivity distributions (SSDs), from which we derived effect factors for different impact perspectives: Marginal, linear, and average changes for both the past and four future emission scenarios (RCP2.6, RCP4.5, RCP6.0, and RCP8.5). Based on a dataset that covered five taxa (corals, crustaceans, echinoderms, fishes, molluscs) and three climate zones, we showed significantly higher sensitivities for strongly calcifying than slightly calcifying taxa and in polar regions compared to tropical and temperate regions. Experimental duration, leading to acute, subchronic, or chronic toxicological endpoints, did not significantly affect the species sensitivities. With ocean acidification impacts still accelerating, the future‐oriented average effects are higher than the marginal or past‐oriented average effects. While our characterization factors are ready for use in LCA, we also point to opportunities for improvement in future developments.

F I G U R E 1 Simplified cause-effect chain for ocean acidification. The focus of this study is presented in boxes with a thicker outline. PAF is the potentially affected fraction of species; PDF, the potentially disappeared fraction of species considered the "evil twin of global warming" (Pelejero et al., 2010). Compared to global warming and other environmental issues, research on ocean acidification is highly neglected and can therefore be considered a top research priority (Scherer et al., 2020a).
Surface ocean pH levels have decreased from preindustrial times in 1770 to 2000 by > 0.1 (Jiang et al., 2019). Under the Representative Concentration Pathway (RCP) scenarios, the pH would further reduce by up to > 0.3 (RCP8.5) by the end of the century (Jiang et al., 2019). Considering the logarithmic nature of the pH scale, such a change in ocean acidity would imply an increase in hydrogen ions by > 100% (RCP8.5) (Jiang et al., 2019).
Ocean acidification lowers the calcium carbonate saturation state, which especially affects calcifying species . Calcifying species use carbonate ions to form shells and skeletal components out of calcium carbonate minerals. The hydrogen ions compete with calcium for the carbonate ions and, thus, reduce calcification rates. Among others, it also affects the growth rates and reproduction success of calcifying species. Ultimately, the effects on calcifying species may disrupt the food web and alter the species composition of marine ecosystems (Fabry et al., 2008). While slightly calcifying species may be better able to cope with ocean acidification, they can still suffer from severe consequences. For example, fish have calcified structures in the inner ear, which are essential for sound detection, orientation, and acceleration (Cattano et al., 2018).
Ocean acidification can therefore lead to various behavioral changes. It can also increase fish mortality (Cattano et al., 2018). The overall impacts on marine ecosystems are largely unknown ).
The impacts of anthropogenic activities on marine ecosystems and biodiversity are currently underrepresented in impact assessment methods for life cycle assessments (LCAs) (Woods et al., 2016). LCA is a tool to quantify the environmental impacts of a product or service across all life cycle stages. While ecosystem quality is one of three typical areas of protection, the focus of impact assessments in LCA has mainly been directed toward terrestrial and freshwater ecosystems (Woods et al., 2016). For ocean acidification specifically, Bach et al. (2016) investigated how greenhouse gas emissions change ocean pH levels, and Azevedo et al. (2015) studied the sensitivity of calcifying species to ocean acidification. To be useful for impact assessment in LCA, a model must link these two aspects. Moreover, representation of global biodiversity impacts requires going beyond calcifying species. Bulle et al. (2019) developed characterization factors for ocean acidification based on the species sensitivity distributions (SSDs) of Azevedo et al. (2015). The unit of PDF⋅m 2 ⋅yr/kg emit. includes an area instead of a volume, which would be more typical for aquatic impacts (e.g., Cosme & Hauschild, 2016;Hanafiah et al., 2011Hanafiah et al., , 2013Veltman et al., 2011). Besides, they reuse the fate model of climate change for ocean acidification. Presumably, this implies that their time component reflects the atmospheric lifetime of greenhouse gases, which seems inappropriate for ocean acidification. Finally, they derived only linear effect factors, which is a simplified approach when the background concentration is unknown (Hauschild & Huijbregts, 2015).
This study aims to develop and present an impact assessment method for the impacts of carbon-containing greenhouse gas emissions through ocean acidification on marine biodiversity. While covering the entire cause-effect chain, the focus lies in estimating the adverse effects of pH changes on species richness, considering both strongly and slightly calcifying taxa. The robustness of these provided characterization factors has been assessed by taking the influence of the taxon group, climate zone, and experiment duration into account. Moreover, multiple impact perspectives-marginal, linear, and average changes for both the past and four future emission scenarios (RCP2.6, RCP4.5, RCP6.0, and RCP8.5)are considered to ensure their versatility for LCA applications.

Characterization factors
Mechanistic characterization factors in LCA link the life cycle inventory to impacts on an area of protection, such as ecosystem quality, through the product of four factors: fate, exposure, effect, and damage factors (Rosenbaum, 2015) ( Figure 1). In the case of ocean acidification, the relevant inventory data are greenhouse gases emitted to the atmosphere. Bach et al. (2016) developed fate and exposure factors for carbon dioxide (CO 2 ), carbon monoxide (CO), and methane (CH 4 ). Their fate factor represents the fraction of an emitted gas ending up in the ocean and considers the distribution within the atmosphere, the conversion of CO and CH 4 to CO 2 , and the dissolution of CO 2 in the ocean. We added a time component to their fate factor (kgCO 2 ⋅yr/kg emit. , where emit. stands for a specific emitted gas, that is, CO 2 , CO, or CH 4 ), which reflects the water residence time of the coastal ocean, assumed to be 4 years (Andersson et al., 2006). Focusing only on the coastal ocean aligns the factors with those for other marine impact categories, such as marine eutrophication (Cosme & Hauschild, 2016), and captures the areas with the highest marine biodiversity (Tittensor et al., 2010). The exposure factor (mol/kgCO 2 ), which Bach et al. (2016) called fate sensitivity factor, indicates the increase in hydrogen ion (H + ) concentration due to the dissolved CO 2 , thereby exposing the species to a lower pH. Since CO 2 can enter the ocean at any location, while we focus on the coastal areas, as discussed earlier, we multiplied their exposure factor with the area fraction representing the coastal ocean (28.3 × 10 6 km 2 for the shallow-water ocean (Andersson et al., 2006) divided by 361 × 10 6 km 2 for the total ocean area without ice shelves (Cogley, 2012)).
The effect factor (PAF⋅m 3 /mol) describes the potentially affected fraction (PAF) of species due to an increase in H + concentration. Modeling the effect factor is the focus of this study and is described in detail in the following sections. Finally, the damage factor (PDF/PAF) translates PAF to the potentially disappeared fraction (PDF) of species. In line with previous studies, we assumed that half of the affected species would eventually disappear (Rosenbaum, 2015). The resulting characterization factor has the unit PDF⋅m 3 ⋅yr/kg emit . Wittmann and Pörtner (2013) provided data on species affected by ocean acidification through a meta-analysis of experimental studies. They analyzed the species responses across different partial pressure of CO 2 (pCO 2 ) ranges: 500-650, 651-850, 851-1370, 1371-2900, 2901-10000, and above 10000 µatm. The first three pCO 2 bins include values projected for 2100 under RCPs 4.5, 6.0, and 8.5, while the higher ones could represent direct injections of CO 2 into the ocean as a climate change mitigation strategy (Wittmann & Pörtner, 2013). In most studies, the control treatment was around 380 µatm. Wittmann and Pörtner (2013) categorized the species responses as negative, positive, or none based on their own interpretations or the interpretations of the authors of the experimental studies. They did not distinguish different types of effects, such as reduced calcification, growth, and reproduction. To fill some missing values, they assumed a monotonous response to pCO 2 , with negative effects extending to higher pCO 2 levels. The 152 included species cover taxa from five phyla: cnidarians (class Anthozoa, including corals), echinoderms (four classes), molluscs (three classes), arthropods (subphylum Crustacea with two classes), and chordates (class Actinopterygii, i.e., ray-finned fishes).

Species response data
We expanded the dataset of Wittmann and Pörtner (2013) with additional species response data for ray-finned fishes and polar regions, which were underrepresented (Table S1 in Supporting Information S1). The data publisher PANGAEA hosts relevant data compilations (Yang et al., 2016).
We searched studies published in 2013 or later (the meta-analysis included studies until 2012) based on the search terms "fish," "polar," "Arctic" or "Antarctic," each together with "ocean acidification." Only those papers with clear declarations on the trend of the impact-as significantly negative or positive or not significant-were added to the dataset, leaving the interpretation to the authors of the articles. If there were multiple experimental data available for a single species and the same pCO 2 bins, priority was given to the experiments with the longest duration and then to results indicating a negative effect. Studies that focused on taxa outside the scope of this research (e.g., algae) or on the effects of fluctuating CO 2 levels in the ocean rather than specific pCO 2 bins were excluded. The search resulted in 17 additional fishes (compared to initially 25 fishes), 12 additional species for polar regions (compared to initially 12 polar species), and 4 updates of existing polar (fish) species responses. Overall, the species number increased from 152 to 179.
For the analysis, we examined all species together but also categorized them into different groups. Different classifications were made based on the taxa's calcification capability (strongly calcifying and slightly calcifying), climate zone ((sub)tropical, temperature, and (sub)polar), and the experiment duration (leading to acute, subchronic, and chronic toxicological endpoints), respectively. Following Wittmann and Pörtner (2013), we categorized corals, echinoderms, and molluscs as strongly calcifying species and crustaceans and fishes as slightly calcifying species. Strongly calcifying species are more sessile, build heavier skeletons through calcification, and are less able to regulate the pH within their bodies (Wittmann & Pörtner, 2013). Three species could not be assigned to one of three climate zones (e.g., cosmopolitan) and one not to an experiment duration class (no duration given). We keep referring to duration instead of toxicological endpoints due to the different meaning of endpoints within LCA. The thresholds for categorizing the experiment duration differ among vertebrates (fishes) and invertebrates (the other four taxa). Acute effects occur after short exposure, namely < 7 days, for both vertebrates and invertebrates. Chronic effects occur after longer periods extending over one or more life cycles or sensitive periods of the species, namely ≥32 days for vertebrates and ≥21 days for invertebrates (Rosenbaum, 2015). The period between acute and chronic was classified as subchronic.

Species sensitivity distributions
Effect factors in LCA are often derived from SSDs. They represent the sensitivity of various species, expressed in PAF, to a substance, such as H + and its associated pH value, relative to a control value of the substance. We calculated the PAF per pCO 2 bin as the fraction of species with a negative response relative to the total number of species with a known response as either negative, positive, or not significant. We represented a pCO 2 bin by taking the midrange value and converted such pCO 2 values to pH values, following Azevedo et al. (2015), who derived a relationship on a global scale based on data in Table 2  To correct for sampling bias, we averaged the control pH values and PAF of each taxon at a certain pCO 2 bin weighted by the number of extant species (Grosberg et al., 2012; Table S2 in Supporting Information S1).
We fitted the pH values and their associated PAF of species into a nonlinear regression to construct the SSDs: where pH 50 is the pH value at which 50% of the species are affected, b is the slope of the SSD, and c H + is the H + concentration (mol/m 3 ). We used the equivalent right-hand formulation to get the SSDs' derivative in the right unit for marginal effect factors (see next section). We used the nls function in R (Baty et al., 2015), which stands for "nonlinear least squares," to fit the parameters pH 50 and b of the SSDs. We used the number of species responses within a pCO 2 bin as weights in the regression analysis. While in ecotoxicology the SSDs tend to express an increasing effect with increasing concentration, in the case of pH values, this trend is reversed, as the pH value equals the negative common logarithm of H + concentration (mol/L).
The point of zero effect is of special importance for deriving effect factors. This point is unfortunately unknown for ocean acidification. In the default case, we determined this point based on the SSD as fitted earlier. As a sensitivity analysis, we assumed the zero effect to be at the preindustrial level to which marine species had adapted (Sutton et al., 2016). Therefore, we forced the SSD to stay < 0.001 up to the preindustrial pH level by adding it as an additional point for the fitting and giving it a sufficiently high weight. This weight was iteratively fitted and resulted in weight values between 0.2 × 10 6 and 1.6 × 10 6 , as opposed to bin weights of at most 179 (i.e., the number of species).
We assessed the goodness of fit of the SSDs with the pseudo-R 2 value and the residual standard error. The R 2 values typically used for linear regressions are considered inadequate for nonlinear regressions (Spiess & Neumeyer, 2010). Multiple modified versions of R 2 are available for nonlinear regressions, and we chose the Cox-Snell pseudo-R 2 index as one of the most commonly used ones (Smith & McKenna, 2013). The Cox-Snell index requires the definition of a null model to which we applied the same weights based on the number of species responses within a pCO 2 bin as to the nonlinear regression model. The second statistic chosen for assessing the goodness of fit, the residual standard error, is the standard deviation of the residuals of the regression. A higher pseudo-R 2 value and a smaller residual standard error imply a better model fit.
For testing if there is a significant difference between SSDs of different categories (e.g., tropical vs. temperate), we made pairwise comparisons and added interaction terms to the two model parameters pH 50 and b: where cat is a binary variable indicating the two categories to be compared and i 1 and i 2 are additional interaction parameters to be fitted. Both the estimates of the interaction parameters (not too close to 0) and their p-values (< 0.05) were used to verify whether the difference between the SSDs was statistically significant and it was meaningful to assess the effects within such categories separately. In such cases, we derived additional effect and characterization factors. We only developed different effect but not characterization factors for climate zones because fate factors do not distinguish them, even though the fate is expected to differ (Jiang et al., 2019).

Effect factors
For deriving effect factors from the SSDs, we convert the pH value to H + concentration in mol/m 3 to link it to the exposure factor in mol/kgCO 2 and integrate it into the characterization factors ( Figure 1): Since there is not one standard way in which to derive effect factors from SSDs, we applied multiple approaches among which the users could choose, depending on the goal and scope of their study and the comparability with effect factors for other impact categories. One can distinguish marginal, average, and linear approaches (Hauschild & Huijbregts, 2015). The marginal approach represents small changes in the environmental pressure and is based on the derivative of the SSD at the current state. The average approach reflects the average of a larger change in effect by taking the difference between the current state and a zero effect or environmental target as a preferred state. The linear approach can be used F I G U R E 2 Species sensitivity distributions for all species and different categorizations. PAF is the potentially affected fraction of species. See the regression coefficients in Table 1. The underlying data are shown in Table S4 in Supporting Information S1 when the current state is unknown and then, for example, the point at which 50% of the species are affected, can be compared to a zero effect.
Instead of a preferred state, the average approach could also consider a prospective state in the future, as was done for effect factors for climate change (De Schryver et al., 2009;Hanafiah et al., 2011).
We derived seven effect factors at the global level: A marginal one at a surface ocean pH of 8.07 in 2000 (assumed as the current state), a linear one between pH 0 (here near-zero effect with < 0.001) and pH 50 , an average one between a preferred state (here, the preindustrial pH of 8.19) and the current state, and four average ones between the current state and future states in 2100 with different emission scenarios (RCP2.6, RCP4.5, RCP6.0, and RCP8.5) at pH values between 8.02 and 7.73 (Jiang et al., 2019).
Moreover, we derived effect factors for the three climate zones: tropical, temperate, and (sub)polar. For that purpose, we delineated climate zones for the ocean ( Figure S1 in Supporting Information S2). We categorized the zones based on the minimum and maximum monthly sea surface temperature (Assis et al., 2018) and the temperature-related rules from the Köppen-Geiger climate classification (Beck et al., 2018). Afterwards, we extracted the average pH values per climate zone (Table S3 in Supporting Information S1) using zonal statistics as input to the calculation of effect factors.

Species sensitivity distributions
The SSDs for all species and different categorizations have a pH 50 at which 50% of the species are affected between 7.47 (slightly calcifying species) and 7.85 (polar regions) (Figure 2). For comparison, at the end of this century, we might reach a pH between 7.73 (RCP8.5) and 8.02 (RCP2.6) (Jiang et al., 2019). All SSDs show a good fit, with Cox-Snell pseudo-R 2 indices between 0.82 and 0.98 (Table 1). The lower pseudo-R 2 indices, indicating a poorer performance, belong to the SSDs for polar regions and subchronic effects that also covered the least species. Strongly and slightly calcifying taxa show significantly different SSDs ( Table 2). Half of the species of strongly calcifying taxa like corals, echinoderms, and molluscs are already affected at a higher pH value, that is, with less pronounced ocean acidification, than slightly calcifying taxa like crustaceans and fishes. The effects occur not only earlier but also quicker, as indicated by the steeper slope (Table 1). However, there are also marked differences between the slightly calcifying taxa. Fishes have a pH 50 value within the range of the strongly calcifying taxa, although it reduced compared to the estimates by Wittmann and Pörtner (2013) ( Table 3). The SSDs also significantly differ for polar regions compared to temperate or tropical regions. Effects in polar regions occur again earlier and quicker. In contrast, effects do not significantly differ based on the experiment F I G U R E 3 Species sensitivity distributions for all species either without forcing (solid) or with forcing (dashed) of pH 0 to the preindustrial pH value. PAF is the potentially affected fraction of species. See the regression coefficients in Table 1. The underlying data are shown in Table S4 in Supporting Information S1 duration. Due to the significant differences, we developed different effect factors for climate zones and for strongly and slightly calcifying taxa. For the latter, where global fate factors are sufficient, we also developed different characterization factors.
Since the fitted pH 0 values at which nearly no species are affected (PAF < 0.001) are relatively high (8.56-9.31), but near-zero effects could also be assumed at preindustrial pH levels, we fitted additional SSDs as a sensitivity analysis where we forced pH 0 to match the preindustrial pH. This led to steeper slopes and higher residual standard errors, while the pH 50 values remained very similar ( Figure 3, Table 1).

Effect and characterization factors
Marginal effects are larger than average effects when comparing the current state to the preindustrial state (Table 4). Because the sensitivities are further accelerating (i.e., the SSD is getting steeper at lower pH values), average effects in the future are between 15% and 32% larger than the average effects in the past. The linear effects are again lower and close to the marginal effect factors, as the reference point is not the preindustrial state but a zero effect at a much higher pH value. As expected based on the pH 50 values, the effect factors are higher for strongly calcifying than for slightly calcifying taxa. The characterization factors are highest for CO 2 (Table 4) due to the higher fate factors, as it does not require any transformation of the substance within the troposphere before dissolving in the ocean .
In the sensitivity analysis with a forced pH 0 at the preindustrial level, effect factors are generally lower because the curve starts getting steeper later. Only the linear effect factors are higher than in the default case because the pH 50 is almost the same in both cases, while the pH 0 is much lower (Table 1), reducing the difference between the two. The linear effects would be even higher than the future-oriented average effect factors, as the reference point is similar, but we would not reach the point at which 50% of species are affected even under RCP8.5. Under RCP8.5, about 24% of the species would be negatively affected in 2100 (37% in the default case). Since the SSD curve is steeper than in the default case, the differences among the different types of effect factors are much higher in the sensitivity analysis, with average effects in the future about 2 and 14 times larger than the average effects in the past.

DISCUSSION
The characterization factors we developed for ocean acidification go beyond the only other attempt so far in several aspects. Bulle et al. (2019) derived only linear effect factors as opposed to considering multiple impact perspectives (marginal, linear, and average effects for both the past and the future). Given the differences among the perspectives (Table 4), having a choice can be important. Moreover, by building on the SSDs developed by Azevedo et al. (2015), their characterization factors disregard fishes. Their unit is also less typical for marine impacts by considering the area of the system affected instead of the volume. Inconsistent units among characterization factors are a general issue for impacts on ecosystem quality and do not yet allow to properly compare the impacts on marine biodiversity across impact categories. A focus on a specific part of the ocean, such as coastal waters (and surface ocean pH changes), is also a modeling choice that requires further reflection for harmonization among marine impact categories and might not be appropriate for all anthropogenic activities in the ocean.
Compared to the characterization factors for marine eutrophication by Cosme and Hauschild (2017), our characterization factors seem rather high. Their mean characterization factors amount to 7.3 × 10 1 , 3.0 × 10 2 , and 5.8 × 10 2 PDF m 3 yr/kgN for emissions to the soil, river, and directly to the marine environment. The maximum across river basins reaches up to 5.3 × 10 3 PDF m 3 yr/kgN for direct emissions to the marine environment (Table S3 in Cosme & Hauschild, 2017). They used linear effect factors, which compares to a characterization factor based on a linear effect factor for all taxa and for CO 2 of 8.41 × 10 3 PDF m 3 yr/kg emit. (Table 4). A notable difference between the methods for the two impact categories is the water residence time that feeds into the fate factors. While we assumed 4 years, Cosme et al. (2018) assumed either 2 or even only 0.25 years, depending on the coastal archetype to which a large marine ecosystem was assigned. With residence times of 0.25 years, the characterization factors become quite similar.
Our study shows that it is important to include multiple taxa across different phyla for impact assessment, as species responses can markedly differ. In our case, strongly calcifying taxa are significantly more affected by ocean acidification than slightly calcifying taxa. With a lower bioavailability of carbonate ions under ocean acidification, building shells and skeletons becomes more energetically costly, which affects not only the calcification rate but also the general fitness of species (Spalding et al., 2017). In contrast to what studies of the paleo-record suggest, fishes showed very high sensitivity to ocean acidification, which even exceeds that of all analyzed calcifying taxa in the meta-analysis by Wittmann and Pörtner (2013). They pointed to limited evidence for fishes, a bias toward coral reef fishes, and shorter study durations. We expanded the fish dataset and estimated a slightly lower pH 50 value (7.72 vs. 7.87, Table 3). With this, the sensitivity of fishes is slightly lower than that of echinoderms and molluscs but still exceeds corals. Crustaceans, as the other slightly calcifying taxon, exhibit the lowest sensitivity to ocean acidification. Azevedo et al. (2015) did not consider fishes in their study, but also found that crustaceans are less sensitive to ocean acidification than strongly calcifying taxa. Although our study covers more taxa than Azevedo et al. (2015), the taxonomic coverage could be further improved. For example, Cosme and Hauschild (2016) cover the same five taxa as we do and additionally include species from the phylum Annelida. Strongly calcifying taxa disregarded in our study are coccolithophores and foraminifera (both unicellular, eukaryotic organisms) (Monteiro et al., 2016).
The significant differences for species sensitivities to ocean acidification in different climate zones point to the value of some level of spatial differentiation. Impact assessment would require that also fate and exposure factors distinguish climate zones, whereas currently available factors represent the world as a whole . Environmental oceanic conditions, such as the temperature, influence how much CO 2 dissolves in the ocean (fate) and how many H + ions are subsequently produced (exposure). While colder temperatures enable the ocean to absorb more CO 2 from the atmosphere (promoting ocean acidification), they also reduce the production of H + ions (inhibiting ocean acidification) (Jiang et al., 2019).
The latter also influences how pCO 2 is converted to pH values, whereas we used a globally generic equation. Because the effect of temperature change on CO 2 dissolution is slower than on CO 2 chemistry speciation, the pH in polar regions is higher than in tropical regions. At the same time, the pH declines more strongly in polar regions (Jiang et al., 2019). A spatially explicit assessment would also allow to overlay the fate and exposure factors with biodiversity distributions and, through weighting with biodiversity, to get more representative global effect factors.
Although SSDs are ideally derived from longer experiments assessing chronic effects (Rosenbaum, 2015), we found that the experiment duration did not significantly influence the species sensitivity to ocean acidification. It suggests that there is no need to prioritize chronic responses for ocean acidification and, thus, allows to consider a larger sample size by including acute and subchronic responses.
In contrast, it would be valuable to consider treatment groups with lower pCO 2 , that is, higher pH values. The median control pH value was 8.04, lower than the current global average of 8.07 (Jiang et al., 2019). For only two out of 179 species, the control pH was around the preindustrial pH value with about 8.2, and these represented the maxima. None was close to the pH 0 , which was fitted to 9.03 when considering all species responses at once. The sensitivity analysis showed how important it is to have more clarity about the point of zero effect ( Figure 3, Table 4).
We used the potentially (negatively) affected fraction (PAF) of species as the basis for the SSDs, as Wittmann and Pörtner (2013) did not report the actual species responses but only classified responses as negative, none, or positive. Alternatively, dose-response curves for single species could first be fitted, from which the effect concentration affecting 50% of the individuals (EC 50 ) can be derived and used as a basis for the SSDs (Rosenbaum, 2015). Azevedo et al. (2015) followed the latter approach. Azevedo et al. (2015) further distinguished SSDs for different life stage processes, with resulting pH 50 values of 7.11 for reproduction, 7.28 for growth, and 7.35 for survival. These values are lower than the pH 50 values we found and would lead to linear effect factors smaller by a factor of 2-3. The difference might be because we considered a larger variety of processes. Separating life stage processes is uncommon in LCA, but it could provide a clearer link between the potentially affected and disappeared fractions of species (PAF vs. PDF). Future research would have to explore which processes are most relevant and how they translate PAF to PDF.
Besides laboratory experiments, mesocosm (Falkenberg et al., 2016) and natural experiments (Scherer et al., 2020b) offer potential alternatives to examine the species sensitivities to ocean acidification and could also reflect cascading effects through biotic interactions. Another potential alternative could be a modeling approach based on the ecological niche concept (Barbarossa et al., 2021). It is also important to examine the effects of ocean acidification in the context of other stressors and potential stressor interactions (Breitberg et al., 2015). This especially applies to climate change, which is driven by greenhouse gas emissions, just like ocean acidification.
Marine ecosystems provide valuable services to society, such as carbon storage, food, livelihoods, and coastal protection (Turley & Gattuso, 2012). So, mitigation of ocean acidification, which potentially affects 37% of marine species in 2100 under a "business-as-usual" scenario (RCP8.5), not yet considering any co-stressors, is in the interest of society. However, the tractability of alleviating ocean acidification was judged as very low in a Delphi assessment (Scherer et al., 2020a). Reducing greenhouse gas emissions is the key action to combat ocean acidification (Billé et al., 2013) and simultaneously tackles the root cause of climate change (Scherer et al., 2020a;Turley & Gattuso, 2012). However, the relative strength of different greenhouse gases differs between the two impact categories. While CH 4 has a higher global warming potential than CO 2 , it has a lower ocean acidification potential, as demonstrated in the lower characterization factors (Table 4). Other acidifying substances not yet considered here, such as nitrogen oxides (NO x ) and sulfur dioxide (SO 2 ), could also be relevant, but the evidence is more limited . Since the characterization factors developed here link greenhouse gas emissions to ocean acidification and marine species loss, they can help with mitigation by identifying hotspots with the highest improvement needs or choosing between alternatives with lower impacts.

CONCLUSIONS
Reducing greenhouse gas emissions is crucial to alleviate the impacts of ocean acidification on marine ecosystems and, consequently, society.
The characterization factors developed here link carbon-containing greenhouse gas emissions through ocean acidification to marine species loss.
Thereby, they can help identify hotspots among life cycle stages that require the most improvement or choose between alternatives with lower ocean acidification impacts, depending on the emissions of the specific greenhouse gases. We provide these characterization factors for different perspectives, depending on the user's need: marginal, linear, and average changes for both the past and four future emission scenarios. While the loss of overall biodiversity is most relevant to decision-making, we also distinguish the impacts on strongly (corals, echinoderms, molluscs) and slightly (crustaceans, fishes) calcifying taxa, with significantly higher losses of the former. We found significant differences in species sensitivities among climate zones, and future research could explore fate and exposure specific to climate zones. Under a "business-as-usual" scenario (RCP8.5), we estimate that about 37% of the marine species will be negatively affected by ocean acidification alone, not yet considering any co-stressors, which demonstrates the importance of assessing ocean acidification impacts in LCA.

CONFLICT OF INTEREST
The authors declare no conflict of interest.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are partly available in the supplementary material of Wittmann and Pörtner (2013) and partly in the supplementary material of this article.