Climate mitigation averts corrosive acidication in the upper ocean

The invasion of anthropogenic carbon into the global ocean poses an existential threat to calcifying marine organisms 1–4 . Observations indicate that conditions corrosive to aragonite shells, unprecedented in the surface ocean, are already occurring in mesoscale upwelling features of the North Pacic 2,5,6 and Southern Ocean 7 , and modeling experiments indicate that large volumes of the global ocean8 including the polar ocean’s surface might become corrosive to aragonite by 2030 4,9–13 . Such changes are expected to compress important marine habitats, but the pathways by which habitat compression manifests over global scales, and their sensitivity to mitigation, remain unexplored. Using a suite of large ensemble projections from an Earth system model14,15, we assess the effectiveness of climate mitigation for averting habitat loss at the ecologically-critical horizon of the base of the ocean’s euphotic zone. We nd that without mitigation, 40-42% of this sensitive horizon experiences conditions corrosive to aragonite by 2100, with moderate mitigation this reduces to 16-19%, and with aggressive mitigation to 6-7%. Mitigation has a stronger effect on the eastern relative to western domains of the northern extratropical ocean with some of the greatest benets in the ocean’s most productive Large Marine Ecosystems, including the California Current and Gulf of Alaska. This work reveals the signicant impact that mitigation efforts compatible with the Paris Agreement target of 1.5°C could have upon preserving marine habitats that are vulnerable to ocean acidication.

In this study we use projections from a suite of Initial Condition Large Ensemble simulations [25][26][27] of an Earth system model 14,15 (ESM; Supplementary Discussion 1 and Extended Data Figs. [1][2] to assess the bene ts of climate mitigation efforts for reducing and delaying the progression of conditions corrosive to aragonite (i.e. Ω arag < 1). Pteropods vertically migrate about the seasonally-varying euphotic zone depth (EZD), feeding above the EZD during night and migrating below the EZD during day to escape visual predation 22 . We therefore evaluate Ω arag at the EZD, thereby capturing the effects of the seasonal migratory patterns of our representative organism upon its vulnerability to ocean acidi cation (Methods, Extended Data Fig. 3). Using the Large Ensemble framework, we quantify the in uence that internal climate variability has on the onset of aragonite undersaturation , the pace of transition between saturated and predominately undersaturated conditions 11 , and on the robustness of mitigation efforts for curtailing aragonite undersaturation at the EZD.

Progression of aragonite undersaturation
We nd undersaturated waters and associated habitat compression at the EZD are projected to progressively invade vast areas of the Arctic, Antarctic, subpolar gyres and equatorial upwelling regions by year 2050 (Fig. 1a). Under a high emissions scenario (RCP8.5), the affected regions expand between 2050 and 2100 to include the subtropical gyres of the North and South Paci c, the Eastern North Atlantic and the Equatorial Atlantic and Paci c. The transition period between the onset of month-long meanstate undersaturation and predominant undersaturation varies regionally from less than 5 years to more than 25 years (Fig. 1b). The transition period is signi cantly shorter in the Southern Ocean, North Atlantic and Subtropical North Paci c ( <5 years) than in the Equatorial and Subpolar North Paci c (> 25 years, Impacts of mitigation on progression of undersaturation Without mitigation efforts, undersaturated waters are projected to comprise 40-42% (range of ensemble members) of the EZD by 2100, representing potentially wide-spread habitat loss for aragonite calci ers (Fig. 2a). However, saturation horizon shoaling is limited with moderate mitigation to the blue and green regions (16-19%), and with aggressive mitigation, to the blue regions (6-7%). Moderate mitigation averts undersaturated conditions in the yellow regions, which include the subpolar and circumpolar, Arctic, equatorial, and eastern subtropical oceans. For the northern subpolar regions, the effects of mitigation are generally stronger in the east than west, re ecting the higher upper-ocean dissolved inorganic carbon (DIC) concentrations in the west (Extended Data Figs. 5,6). High DIC pre-conditions the west to more rapid undersaturation that outpaces mitigation bene ts. The anthropogenic declines in Ω arag and the sensitivity of these declines to mitigation is mediated by the magnitude of the invasion ux of C ant , and not by anthropogenically-driven changes in circulation (Extended Data Fig. 7; Supplementary Discussion 2). This is important, as it suggests that our main results are not contingent on speci cs of model sensitivity of the Atlantic Meridional Overturning Circulation (AMOC) stability but rather on the skill with which the model represents the upper ocean carbon cycle.
Mitigation also affords longer transition times in nearly all regions, but particularly in the Southern Ocean, where mitigation increases transition times from < 5 years on average to > 25 years (Fig. 2c). This reveals that in addition to reducing the spatial extent of habitat compression, aggressive mitigation also signi cantly increases the timescales available for migration 22 , adaptation 28 , and trophic adjustments to the reduction or local extinction of pteropods 29 .
Many of the regions for which mitigation averts undersaturation are home to the world's Large Marine Ecosystem (LMEs), biogeographically de ned provinces which cumulatively contribute 80% to the global sheries yield 30 and contain the highest species richness and serve as biodiversity hotspots 31 . Mitigation more effectively averts aragonite undersaturation in LMEs located on the eastern side of the extra-tropical northern Paci c and Atlantic Oceans (Fig. 3). This contrast is exempli ed in the North Paci c, where committed emissions ensure undersaturation by 2050 in the West Bering Sea and Oyashio Current whereas it is avoidable in the Gulf of Alaska and California Current.
Theoretical and empirical evidence suggest that compression of intermediate trophic levels, such as those occupied by pteropods, can result in deleterious effects at higher trophic levels 29 . The reduced abundance or absence of pteropods could have potentially large impacts on reliant species such as salmon and herring, and thus global sheries 3 . Our results indicate that the timing of such impacts could differ by many decades between the eastern and western parts of each basin, and that near-term reductions in emissions will not mitigate the risk of undersaturation-related impacts in the western parts of the basins. This work motivates expansion of ongoing efforts to monitor plankton distributions and gut contents of key species as a key aspect of sheries management for early warning of population vulnerability.

Impacts of vertical seasonal migration
The globally-integrated progression of one-month-per-year undersaturation is consistent across multiple depth horizons in the surface ocean ( Fig. 4), with ~40% of each horizon experiencing month-long undersaturation events by year 2100 without mitigation, 20% with moderate mitigation and less than 10% with aggressive mitigation. Despite the progression of onset being similar, the area impacted over the seasonal cycle is much less severe at the seasonally-varying EZD horizon (Fig. 4a-c) than at the xed horizons ( Fig. 4e-f, 4i-k). Differences in the expression of the seasonal cycle at the different horizons result in transition periods at the seasonally-varying EZD being longer (by more than 10 years on average) and more responsive to mitigation than at the 100 meters considered in previous studies 11,12 (Fig. 4l) or the annual-mean EZD (Fig. 4h).
During winter months, a shallower EZD offers a protectionary effect when the undersaturation horizon is also the shallowest in most regions (Extended Data Figs. 3b, 8). This protectionary effect of seasonal habitat movement delays the onset of predominant undersaturation therefore extending the transition period at the seasonally-varying EZD relative to the annual-mean EZD and 100-meter depth horizon (Fig.  4d, 4h and Extended Data Figs. 9-10), and consequently provides more leverage for mitigation efforts to increase transition durations. The distribution of transition times at the EZD reveals a stark contrast between the rapid transition period for RCP8.5, and successively longer adaptation timescales for RCP4.5 and then RCP2.6 ( Fig. 4d), whereas the transition period de ned at the annual-mean EZD and 100m depth for the three scenarios reveal nearly coincident peaks at less than 5 years (Fig. 4h, 4l).
Another critical difference between the progression of undersaturation at the EZD versus xed-depth horizons is the time of year in which undersaturation begins to occur (Extended Data Fig. 10). At the seasonally-varying EZD, undersaturation arrives rst in the summer months, whereas at 100 meters, undersaturation arrives rst during winter months. This distinction may have important implications for spawning cycles and other seasonally-varying ecosystem drivers 3 .
Conceptually, the length of the transition period is related to the ratio between the pace of anthropogenicdeclines in Ω arag and the amplitude of the seasonal cycle at a given static (100m) or dynamic (EZD) depth horizon (Extended Data Fig. 6d-i). The larger the amplitude of the seasonal cycle (or interannual variability in the case of Equatorial regions) in Ω arag relative to anthropogenic declines in Ω arag , the longer the transition period, i.e. more annual cycles are required for the majority of the year to fully surpass the undersaturation threshold. The amplitude of the seasonal and interannual variability in Ω arag is also related to the vertical gradient in Ω arag , with weak gradients producing weaker seasonal cycles (Extended Data Fig. 6c). Weak vertical gradients accelerate the transition, as the saturation horizon rapidly invades the euphotic zone. This mechanism gives insight into why transition periods differ: (i) between the EZD and xed horizons (Extended Data Fig. 6g-i) -the anthropogenic declines in Ω arag are similar but the amplitude of the Ω arag seasonal cycle is larger when considered at the seasonally varying depth; (ii) across space -where the vertical gradient of Ω arag varies by an order of magnitude between the poles and tropics (Extended Data Fig. 6b); and (iii) between different emissions scenarios, where mitigation reduces the rate of the anthropogenic declines in Ω arag and therefore increases the transition period ( Fig.   4; Extended Data Fig. 6g-i).

Discussion
Our ndings extend the work presented in previous studies 11,12 by: (i) explicitly considering the impacts of mitigation on the progression of undersaturation, (ii) assessing undersaturation at the ecologicallymotivated EZD rather than a xed depth horizon, (iii) considering the global ocean and assessing risk and mitigation sensitivity across LMEs and (iv) identifying the fundamental importance of both the seasonal cycle in Ω arag and EZD, as well as the vertical gradient of Ω arag , in driving variations in transition times across regions. Our chosen threshold of strict undersaturation (Ω arag < 1) sustained for at-least a month represents a conservative estimate of when ocean acidi cation will begin to adversely and meaningfully impact pteropods. As such, the onset years presented here could understate the urgency of this ecological impact of climate change. Furthermore, other powerful changes in the ocean are underway, like warming and deoxygenation, which stand to interact and potentially exacerbate the pressures of acidi cation on pteropods and other marine calci ers [32][33][34] . Additionally, the nominal 1° resolution of our ESM limits the representation of highly-localized variability (Supplementary Discussion 1), indicating that earlier localized undersaturation events may occur prior to the LME-to-biome scale undersaturation presented here.
The progression of acidi cation in the open ocean has relatively modest model uncertainty 35 , and our analysis indicates that the rate at which aragonite undersaturation occurs in the upper ocean is largely driven by the superposition of the large-scale dynamics and the invasion ux of anthropogenic carbon (Extended Data Figs. 5-7), robust features of current generation ESMs unlikely to change meaningfully with improvements in model physics and resolution 36,37 , however there are coastal and regional exceptions to model agreement 13,38 , as well as model disagreement on the degree of ampli cation in seasonality of surface Ω arag 39 . The Large Ensemble reveals that the onset and transition period of undersaturation are sensitive to natural variability uncertainty, shifting the timing by 25+ years over the Equatorial and the North Paci c. We nd the spatial extent of this habitat compression is acutely sensitive to the magnitude of human emissions (e.g. large scenario uncertainty) and thereby displays a robust signature of mitigation efforts on reducing habitat loss (Fig. 2).
Our results emphatically present the potential of ocean acidi cation to vastly compress the habitable range of ecologically-, biogeochemically-and commercially-important taxa (Fig. 1). Habitat loss for pteropods in the ocean's polar regions is unavoidable given cumulative emissions to date. However, the extent to which habitat loss may be averted or stalled over the open ocean and in many highly productive LMEs throughout the Paci c, Atlantic, Indian and Southern Ocean is contingent on the degree to which aggressive emissions reductions are enacted (Fig. 2). Thus the degree of emission mitigation needed to avert substantial pteropod habitat compression, such as RCP2.6, is consistent with the goals of maintaining compliance with internationally recognized 1.5-2°C targets 40 .

Methods
All simulations are conducted with the coupled Earth system model GFDL-ESM2M developed at the Geophysical Fluid Dynamics Laboratory 14,15 for which delity of the biogeochemical model (TOPAZ) has been documented for preindustrial 15 , historical 41 and future 42 boundary conditions. We also benchmark the performance of ESM2M's saturation state of aragonite (Ω arag ) during the contemporary period against data-based estimates. Speci cally we assess ESM2M mean-state, anthropogenic trend and seasonal cycle relative to Ω arag derived from ETHZ-OceanSoda 43 , SOM-FFN 44 , JMA-MLR 45  New to this work is the analysis of Ω arag at the time-varying base of the model's simulated Euphotic zone.
Technically, this is done through saving the monthly mean three-dimensional Ω arag and photosynthetically-active radiation (PAR) elds. We de ne the euphotic zone depth (EZD) as the modelcalculated isoluminal horizon at which monthly-averaged PAR is equal to 0.1 Watts per square meter (Extended Data Fig. 3). The EZD exhibits seasonal variations driven by surface solar ux changes and ambient chlorophyll sustaining attenuation of light.
For each ensemble member, month, and each grid-cell, monthly-mean Ω arag is then evaluated at the coincident euphotic zone depth. We de ne the onset of signi cant undersaturation as the rst year which experiences a month of mean-state undersaturation. We de ne the transition period as the time between onset year, and the rst year which experiences six months or more of mean-state undersaturation, i.e. the time it takes to transition from a permanently saturated to a predominately undersaturated mean-state. The onset and transition metrics are computed for each gridcell (re-gridded to 1°x1°). Where indicated, we consider the fractional ocean area past onset within Large Marine Ecosystems (http://lme.edc.uri.edu/), with-in latitude bands and globally.
Additionally, in order to diagnose the difference between the radiative and chemical impact of rising

Declarations
Data availability The Large Ensemble output used in this study is publicly available through Globus, with links and dataguide available at http://poseidon.princeton.edu. Figure 1 Timing where the 10th and 90th percentile ensemble members diverge by more than 25 years (Extended Data Fig. 4).