More Frequent Abrupt Marine Environmental Changes Expected

We quantify an elevated occurrence of abrupt changes in ocean environmental conditions under human‐induced climate forcing using Earth system model output through a novel analysis method that compares the temporal evolution of the forcings applied with the development of local ocean state changes for temperature, oxygen concentration, and carbonate ion concentration. Through a multi‐centennial Earth system model experiment, we show that such an increase is not fully reversible after excess greenhouse gas emissions go back to zero. The increase in occurrence of regional abrupt changes in marine environmental conditions has not yet been accounted for adequately in climate impact analyses that usually associate ecosystem shifts large‐scale variability or extreme events. Estimates for remaining greenhouse gas emission targets need thus to be more conservative.

• There is an elevated occurrence of abrupt changes in key ocean state variables under human-induced climate forcing • The occurrence of abrupt changes in the upper ocean peaks around the maximum of the rate of change in human-induced forcing • A multi-centennial legacy is expected for abrupt shifts in environmental conditions, long after a stop of anthropogenic CO 2 emissions Supporting Information: Supporting Information may be found in the online version of this article.
"tipping points" in drivers that trigger such regime shifts are even more difficult to identify.Nevertheless, aggregated over the world, ocean regional abrupt state changes and associated regime shifts provide a substantial hazard to the Earth system and human societies (Heinze et al., 2021).Here we analyzed the development of abrupt changes in ocean warming, acidification, and deoxygenation from Earth system model simulations.Often changepoint analysis is used to identify abrupt changes in time series data.However, these changepoints cannot easily be attributed to a specific reason or forcing.Therefore, we explore an alternative analysis method here.

Approach and Method
We employ the Norwegian Earth System Model version 2 (NorESM2) in its configuration NorESM2-LM (with 1° resolution in the ocean component and ∼2° resolution in the atmosphere) as documented for the entire model system (Seland et al., 2020a) and specifically for the ocean biogeochemical component (Tjiputra et al., 2020).Please, see Text S1 in Supporting Information S1 for more information concerning the model and its performance.We analyzed two model experiments: (a) A worst case scenario under progressing climate forcing and (b) An idealized peak and decline emissions scenario with a long extension to explore the potential reversibility of abrupt change occurrences.
We first looked at a projection using the pessimistic strong anthropogenic forcing scenario SSP5-8.5 (O'Neill et al., 2016) (SSP: shared socioeconomic pathway).SSP5-8.5 covers the period 2015-2100 and was run in concentration driven mode (atmospheric CO 2 concentration prescribed).The data set was combined with the compatible CMIP6 historical simulation with NorESM2 (over the years 1850-2014).
In a second step, we checked the potential reversibility in the occurrence of abrupt changes in an extended experiment, where idealized anthropogenic CO 2 emissions are confined to a bell-shaped emission spike over 100 years followed by zero emissions onward.We employed the emission driven experiment B3 (see Figure 2 of Jones et al. (2019)) of ZECMIP, the Zero Emissions Commitment Model Intercomparison Project (Jones et al., 2019), which covers 800 years .This data set was merged (at its beginning) with the last 99 years of the corresponding spin-up simulation where the anthropogenic CO 2 emissions are set to zero and the prognostic atmospheric CO 2 level is kept relatively stable at preindustrial level (1751-1849).The total data set is thus 899 years long.The ZECMIP B3 forcing scenario is an idealized one.The formal calendar years used for ZECMIP B3 are thus for orientation only and do not correspond to actual calendar years as in SSP5-8.5.For the analysis of both the SSP5-8.5 and the extended ZECMIP runs, we used annual means when available or calculated them from monthly data sets, such as for potential temperature Θ and salinity.
We systematically analyzed the model data time series in order to identify the occurrence and extent of abrupt changes in ocean state variables in relation to anthropogenic climate forcing.For each grid box of the ocean model, we extracted time series of ocean state variables over the entire length of the respective model experiment.We established corresponding global anthropogenic forcing time series as a benchmark for what is abrupt or not.We considered changes in ocean state variables as abrupt if they occur at a faster rate than their associated forcing over a chosen time interval at a certain geographic position and depth.As forcings, we computed time series of changes in total global ocean contents of heat, dissolved oxygen (O 2 ), and dissolved inorganic carbon (DIC) that we diagnosed from the Earth system model experiments.This is a somewhat heuristic approach because some abrupt changes may be triggered by special local and transient forcing characteristics including natural variability.Hence, we cannot say for each single abrupt event that it is caused by human-made climate forcing.However, we can identify whether more abrupt changes occur relative to a hypothetical world without human-induced climate forcing and where the hot spots for those changes are.Our approach of applying the integrated oceanic inventory changes (additions to or removals of heat, O 2 , and DIC from the oceans) as a consequence of overall atmospheric forcing thus serves as a useful reference to discriminate between what is abrupt and what is not.We compared the relative change in the three ocean state variables potential temperature (Θ, for warming), the concentration of dissolved oxygen [O 2 ] (for deoxygenation), and the carbonate ion concentration  [ CO3 2− ] (for ocean acidification) with the respective forcing time series over subsequent 25-year intervals (25 full model years) and classified changes in the state variables as abrupt when they were faster than in the forcing (and thus accelerated in a non-linear sense) between the start and end-points in the respective window.The 25-year intervals were chosen because they are longer than timescales associated with interannual variability and shorter than long-term changes over several decades.We chose rather than pH as this variable gives a more direct link to the calcium carbonate saturation state and is on a linear rather than a logarithmic scale.For the comparison between forcing and state variable time series, we normalized both the forcing and the state variable time series signal to vary between 0 and 1 Writing -review & editing: C. Heinze, C. Michel, T. Torsvik, J. Schwinger, J. F. Tjiputra over the entire analysis period (and not only for discrete time windows).The overall changes in physical units of the ocean state variables were diagnosed a posteriori and were not used as a criterion for abruptness, that is, we looked at the severity of the abrupt events after having them identified.In order to avoid that noise in the time series would render a large amount of abrupt changes, we smoothed the time series with a 5-year running mean and required that the variance of the respective ocean state variable in the time window for analysis would exceed the variance level in the still only little perturbed time window 1850-1899 by a factor of 10.The 25-year window was shifted by 1-year increments forward for making the time series plots of abrupt change occurrences.The analysis was run for ocean state variable time series at each grid box of the model.As an example, Figure 1 illustrates the analysis method for the state variable [O 2 ] at depth level 250 m, and for one snapshot (the time window 2025-2049) during the first half of the 21st century.The map of abrupt changes occurring in physical units (mol m −3 ) (Figure 1a) and the map for normalized faster change in the state variable than in the forcing (in this case the change in global ocean O 2 content) (Figure 1b) shows that the changes found through the analysis in the normalized space are indeed substantial in most cases.The time series of [O 2 ] for grid points revealing an abrupt change during 2025-2049 indicate that the filtering technique indeed can identify locations with abrupt step-like behavior (Figure 1c).

Analysis for the SSP5-8.5 Scenario 1850-2100
We first have a look at the forcing time series in Figure 2. The pattern of the rising atmospheric CO 2 concentration (Figure 2a) is followed closely by that of the change in ocean content of DIC (Figure 2e).The forcing time series for ocean heat content (Figure 2c) and ocean O 2 content (Figure 2d) also follow each other quite closely for the shape of the curves, however, with opposite sign.Global mean atmospheric CO 2 and projected sea surface temperatures are given in Figures 2a and 2b.The development of abrupt changes for the climate projection with SSP5-8.5 forcing is summarized in Figures 2c-2e.Common to all three state variables considered is the fact that the number of abrupt change occurrences is initially increasing with rising atmospheric excess CO 2 and respective other forcings (Figures 2c-2e), however, differences exist between the state variables at depth levels.For potential temperature Θ the number of abrupt changes increases steadily with progressing warming at 1,000 m while they already start to stabilize higher up in the water column between 100 and 500 m depth after an initial increase there (Figure 2c).For [O 2 ], the quite striking increase in deep water abrupt changes (1,000, 2,000 m) indicates, that the changes in [O 2 ] due to ocean circulation and associated changes in biological production as well as remineralization below the surface may be more efficient drivers for generating abrupt changes than solubility effects at the sea surface (Figure 2d).Also, the air-sea gas exchange more directly affecting the surface waters can partly compensate for the ocean born [O 2 ] changes.The  [CO32 − ] distribution indicates that near surface depth levels are strongly affected when atmospheric [CO 2 ] starts to steeply rise and when also over short time a lot of anthropogenic carbon accumulates in the upper water column and cannot get mixed down quickly enough into deeper layers (Figure 2e).The rising trend in abrupt  [ CO3 2− ] occurrences at 1,000 and 2,000 m after 2050 is consistent with the progressing invasion of anthropogenic carbon into deeper layers and the emergence of ocean acidification signals in these depths, especially in the ventilation regions (Tjiputra et al., 2023).
Where are the areas of high abrupt change occurrences located regionally in the World Ocean?To answer this question, we collated the vertically and temporally accumulated abrupt changes in global maps (Figure 3).The maps for Θ (Figure 3a) and [O 2 ] (Figure 3b) reveal some similarities as both distributions are controlled by heat uptake and changes in the circulation and mixing.In general, there is an elevated level of abrupt change occurrence in the Arctic.Further hot spots for the occurrence of abrupt changes are the equatorial Atlantic and Indian Oceans probably due to upwelling fluctuations in these areas.Also, the Gulf Stream system reveals a higher level of abrupt change occurrence due to the strong large scale turbulence there and the influence of the deep water production areas with respective fluctuations in water mass production.It needs to be mentioned that the occurrence of abrupt events in our analysis does not say anything a priori about the absolute strength in oxygen changes; it only indicates that, for example, the spreading of oxygen deficient zones in the Pacific may be progressing more smoothly and gradually than [O 2 ] changes in the Atlantic, at least for the specific ocean model as employed here.The  [ CO3 2− ] map, in contrast indicates strong perturbations in the Southern Ocean (Figure 3c), a result that is in line with the consideration of the Southern Ocean to be a key region for regulating atmospheric CO 2 (Sigman & Boyle, 2000) and a highly variable region for anthropogenic CO 2 uptake (Gruber et al., 2023).
A presentation of candidate processes behind abrupt changes can be found in Text S2 in Supporting Information S1.

Analysis for the Idealized Zero Emission Commitment Scenario
We found above an increase in the number of abrupt changes with increased CO 2 concentration.In the current context of net zero emissions scenarios and limiting the global warming to +1.5°C, one may wonder: Is there a long-term legacy in the occurrence of abrupt changes in ocean state variables as compared to changes in forcing?
To answer this question, we will now examine the results for abrupt changes during a much longer time span for the model experiment using the idealized ZECMIP B3 forcing scenario.After an initial interval of 99 years without perturbation follows the 2,000 PgC emission spike (bell-shaped curve) over the subsequent 100 years, and then a 700-year long period with no CO 2 emissions (see blue line in Figure 4a).The black line in Figure 4a shows that the atmospheric CO 2 concentration decreases only gradually after the emission stop at year 1950 in this idealized scenario experiment.The mean global sea surface temperature (Figure 4b) increases steadily, interrupted only for a transient cooling between 1950 and 2100 caused by a partial recovery of the Atlantic Meridional Overturning Circulation after a slowing down following the peak of the emission phase (for a discussion of this phenomenon, please see other studies Schwinger, Asaadi, Goris, et al. (2022), Schwinger, Asaadi, Steinert, et al. (2022) focusing on this aspect).The forcing curves for global ocean heat content, global ocean content of dissolved O 2 , and of DIC indicate that even toward the end of the multi-centennial model experiment the system has not yet arrived at a new steady state (black dashed lines in Figures 4c-4e).The ocean is still gradually gaining heat and DIC while losing oxygen.For analyzing the occurrence of abrupt changes in ocean state variables we employed the same method as before for the SSP5-8.5 experiment.For computing the variance of the quasi-unperturbed system, the period 1775-1824 was chosen.In general, the occurrence of abrupt changes (number of grid points encountering abrupt changes in state variables in subsequent 25-year intervals) shows a strong increase with the onset of rising forcing (Figures 4c-4e).
For heat and oxygen (Figures 4c and 4d), the occurrence of abrupt changes remains at an elevated level until the end of the simulation, especially for deeper depth levels (1,000, 2,000 m).For the carbonate ion concentration  [ CO3 2− ] (Figure 4e), the abrupt change occurrences decrease over time quasi-monotonically after the CO 2 emission stop.This effect can be explained by the progressive dilution effect for water carrying anthropogenic carbon signatures during the gradual mixing of waters from the upper ocean into the vast realm of the deep sea.The regional distribution of abrupt changes (Figures 3d-3f) follows qualitatively a similar pattern as for the SSP5-8.5 experiment (Figures 3a-3c), with the exception of the equatorial Pacific, where the ZECMIP B3 experiment reveals a stronger occurrence of abrupt changes.The latter can be due to the faster increase in atmospheric climate forcing in the idealized ZECMIP B3 scenario as opposed to SSP5-8.5 so that deep water production and deep mixing are somewhat less effective in the former in relation to upper ocean processes.The ZECMIP B3 analysis reveals that there is a long legacy of abrupt ocean changes even centuries after a potential CO 2 emission stop.Over time, the occurrences of abrupt changes in Θ, [O 2 ], and (and thus carbonate saturation state) slowly invade the sub-surface and deep ocean layers (see Figure S1 in Supporting Information S1).

Assessing Modifications of the Analysis Method
As an alternative to counting the number of grid points showing abrupt events, one can consider a volume weighted assessment of the abrupt changes in order to see what fraction of the ocean is affected by abrupt changes, see Figures S2 and S3 in Supporting Information S1 for the SSP5-8.5 and ZECMIP B3 analyses respectively.We have also used a method based on linear trends to detect abrupt changes and it gave relatively similar results to Figure 4. We also tested how sensitive the analysis method is to the smoothing interval, the length of the time window for abrupt changes detection, and the filtering with respect to the naturally occurring variance.For all parameters the analysis result does not change radically qualitatively but to some degree quantitatively.Examples for sensitivity tests with respect to unperturbed variance level for noise filtering and to the smoothing interval are given in Figures S4 and S5 in Supporting Information S1.If one requires a high level of unperturbed variance exceedance in the state variable time series, the number of abrupt changes occurrences goes down as expected.On the other hand, if the smoothing interval is extended, this results in an increase in abrupt change occurrences due to the lower variance in the also smoothed unperturbed part of the time series.Finally, one has to be careful in the interpretation of Figure 3.Some abrupt changes (i.e., with faster change than the forcing) may last longer than one 25-year period and thus in some cases are counted more than once.

Conclusions
We analyzed a multi-decadal projection with a strong anthropogenic climate forcing (SSP5-8.5)and a multi-centennial idealized simulation with a short initial CO 2 spike (bell shaped curve, ZECMIP B3) for occurrence of abrupt changes.In contrast to many studies focusing on change point analysis, we could link the elevated occurrence of abrupt changes in key ocean state variables for warming, oxygen changes, and acidification to the overall human-induced ocean forcing of heat content, oxygen content, and DIC content.The occurrence of abrupt changes in the upper ocean peaks around the maximum of the rate of change in human-induced forcing.A multi-centennial legacy is expected for abrupt shifts in environmental conditions, long after anthropogenic CO 2 emissions might have been stopped.The main focus in the occurrence of abrupt changes will move toward deeper layers over time.Thus, not only upper ocean ecosystems will be affected through abrupt changes in living conditions but also the intermediate and deep waters with ecosystems that are usually optimized for stable conditions in environmental factors (Gehlen et al., 2014).This can lead to a long legacy of negative impacts to oceanic ecosystems at a broad spectrum of depth levels through sudden changes in ocean state variables that come in addition to the multiple stressors (Bopp et al., 2013) already provided by gradually changing conditions.Due to the small spatial scales and long timescales of oceanic motion, the detection and monitoring of abrupt ocean state changes through observations remains a challenge.Studies have revealed that interactions and combinations of processes and forcings require a downward correction of forcing thresholds for critical changes in Earth system elements (Lade et al., 2020;Steinacher et al., 2013;Willcock et al., 2023).Our study shows that the rising occurrence of COMFORT-Our common future ocean in the Earth system-quantifying coupled cycles of carbon, oxygen, and nutrients for determining and achieving safe operating spaces with respect to tipping points) and by the Norwegian Research Council through Grant.295046 (project KeyCLIM).The contents of this article reflect only the authors' viewsthe European Commission and their executive agencies are not responsible for any use that may be made of the information it contains.The computations were performed on resources provided by Sigma2-The National Infrastructure for High Performance Computing and Data Storage in Norway through projects NN2980K, NS2980K, NN9708K, NS9708K, and NS9560K.Mats Bentsen (NORCE) helped the authors with access to some of the data sets.

Figure 1 .
Figure 1.Detailed analysis snapshot for one depth level (250 m) and one time window (2025-2049) for the dissolved oxygen concentration for the historical + SSP5-8.5 experiment.(a) Change in the raw (un-normalized) O 2 concentration between 2025 and 2049 and (b) ratio of the difference in normalized smoothed O 2 concentration and the difference in the normalized global ocean O 2 content (forcing) where abrupt changes occur.Absolute values <1 mean that no abrupt change has been detected.(c) Normalized smoothed O 2 concentration time series for the 10 largest positive changes detected (each colored line represents one grid point) superimposed on the normalized global ocean O 2 content forcing time series (black line).The vertical dashed lines highlight the time window considered here for the abrupt changes analysis.

Figure 2 .
Figure 2. Overview of the forcings applied and the number of abrupt changes occurring.(a) Atmospheric CO 2 concentration for the historical + SSP5-8.5 scenario.(b) Global average sea surface temperature.(c-e) Black dashed lines show the three different forcing used in the abrupt changes analysis: (c) global ocean heat content, (d) global ocean content of dissolved oxygen, and (e) of dissolved inorganic carbon.Blue lines show the number of grid points where abrupt changes occur during 25-year periods shifted by one year at five different depth levels for (c) potential temperature, (d) dissolved oxygen concentration, and (e) carbonate ion concentration.The color code for the depth levels is given within panel (c).

Figure 3 .
Figure 3. Regional distribution of abrupt changes occurrence.Number of abrupt changes integrated over all depths from 100 m downwards and all considered 25-year periods between 1900 and 2100 for the historical + SSP5-8.5 experiment (a-c) and between 1850 and 2649 for the ZECMIP B3 experiment (d-f) for potential temperature (Θ) (a, d), dissolved oxygen concentration (O 2 ) (b, e), and carbonate ion concentration (  CO 2− 3 ) (c, f).

Figure 4 .
Figure 4. Analysis for an idealized forcing scenario with zero emissions after 200 years.Same as Figure 2 but for the ZECMIP B3 experiment.The gray shading highlights the 100 years (1850-1949) with CO 2 emissions whose distribution is shown with the blue line (a).The color code for the depth levels is given within panel (e).