Quantifying Anthropogenic Influences on Global Wave Height Trend During 1961–2020 With Focus on Polar Ocean

This study investigates the contribution of external forcings on global and regional ocean wave height change during 1961–2020. Historical significant wave height (Hs) produced for different CMIP6 external forcings and preindustrial control conditions following the Detection and Attribution Model Intercomparison Project (DAMIP) are employed. The internal variability ranges are compared with different external forcing scenario. Statistically significant linear trends in Hs computed over regional ocean basins are found to be mostly associated with anthropogenic forcings: greenhouse gas‐only (GHG) and aerosol‐only (AER) forcing. For Hs, GHG signals are robustly detected and dominant for most of the global ocean, except over North pacific and South Atlantic, where AER signals are dominant. These results are supported by multi‐model analysis for wind speed. The remarkable increase in Hs over the Arctic (22.3%) and Southern (8.2%) Ocean can be attributed to GHG induced sea‐ice depletion and larger effective fetch along with wind speed increase.


Introduction
Wind-waves are capable of affecting continuous ocean-atmosphere exchange (Cavaleri et al., 2012), atmospheric circulation and subsequently surface winds (Chen et al., 2019), the former's driver; thus introducing a feedback loop in the global climate system.There are other ways of mutual interactions with the climate system, such as: fragmentation of ice floes in the marginal ice zone (Boutin et al., 2020), deepening of the ocean mixed layer depth due to the effect of turbulence induced by wave orbital motions (Toffoli et al., 2012), ejection of sea spray in the atmosphere and air entrainment in the water column during wave breaking (Deike, 2022).Since the industrialization, Earth's climate system has undergone some drastic changes (Masson-Delmotte et al., 2021).Being a component of the global climate system, wind-waves have experienced intensification or weakening in many regions of the global ocean (Meucci et al., 2023;Timmermans et al., 2020;Young & Ribal, 2019).Understanding long-term changes in waves is crucial for a range of socio-economic and environmental applications, such as offshore and coastal activities, coastal hazards management, and renewable energy generation.
The climate response can be interpreted as a combination of an internally unforced and an externally forced variability (Hasselmann, 1993).Anthropogenic greenhouse gases and aerosols have been identified as major external forcing for the global climate change (Bindoff et al., 2013).Moreover, detection and attribution studies have shown that aerosol influences have significantly offset the global warming induced by greenhouse gases (Gillett et al., 2021;Seong et al., 2021).It is still unclear to which extent external anthropogenic forcings (i.e., greenhouse gases and aerosols), external natural forcings (solar and volcanic activities), internal climate variability, or a combination of some of these, have influenced wave climate since the pre-industrial period.• Greenhouse gases are the major contributor for H s trends over the global ocean, but aerosols dominance is seen for a few regional basins • High increase in H s over the Polar oceans is due to greenhouse gas induced sea-ice decline, fetch enlargement and wind speed increase

Supporting Information:
Supporting Information may be found in the online version of this article.
The Detection and Attribution Model Intercomparison Project (DAMIP; Gillett et al., 2016), under the framework of Coupled Model Intercomparison Project (CMIP6; Eyring et al., 2016) provides a variety of external forcing experiments including GHG (greenhouse gases) only, AER (aerosols) only, and NAT (natural) only forcings.This gives the opportunity to assess the relative importance of individual forcing factors.The unforced climate variability can be estimated from preindustrial control simulations (CTL) from CMIP6 models.In this context, comparing unforced CTL variability with forced simulation from CMIP5 (Taylor et al., 2012) framework, Dobrynin et al. (2015) could detect climate change signal in wave climate simulations as early as 2010-2020 decade over various regions of global ocean, such as North Atlantic, equatorial Pacific, and Southern Ocean.Another study by Hochet et al. (2023) estimated that the date of emergence (when climate change emerges from the noise of natural climate variability) of forced variability in satellite-based wave climate data records would be after 2050 in the North Atlantic.The role of internal variability on wave height trend was evaluated by a recent study (Casas-Prat et al., 2022).Nevertheless, relative contribution of different external forcing is yet to be addressed for long-term changes in wave climate.
Nonetheless, decadal changes of the wave climate in the high latitudes are particularly affected by rapid changes of the oceanic and atmospheric conditions in polar regions, such as surface winds, currents and sea ice.For instance, as sea ice extent decreases in the Arctic, greater extent of fetch becomes available for waves to grow, and consequently larger waves provide a mechanism to sea ice break and accelerate the melting, leading to a positive feedback loop (Thomson & Rogers, 2014).The dominant influence of greenhouse gases in long-term trend of surface winds as a result of modulating meridional atmospheric circulation is reported by K. Deng et al. (2021) using CMIP6 multi-model simulations.
Here, we take advantage of newly available global historical wave data simulated by WAVEWATCH III (WW3) using winds and sea ice forcings from DAMIP (Patra et al., 2023b) to provide the first estimate of relative contribution of GHG, AER, NAT, and CTL influence on significant wave height (H s ) changes over the period 1961-2020.We also examine the role of sea ice and winds associated with different forcing scenario in the Arctic and Antarctic region.

Model Experiments and Analysis Method
Ocean wave data are not directly available from CMIP5/CMIP6 global climate models (GCM), hence not in the framework of DAMIP.However, Patra et al. (2023b) produced historical ocean wave climate for the 1961-2020 period, using the CMIP6 model-MRI-ESM2.0(Yukimoto et al., 2019) as forcing to WW3 over the global ocean, following the approach of DAMIP.Global 6-hourly wave height (H s ) are available at horizontal resolution of 1.0°f or preindustrial control condition (CTL) and historical experiments (ALL: natural plus all anthropogenic, GHG, AER, and NAT), each consisting of five ensemble members.The control simulation, considered as time invariant, is available for 180 years.The historical simulations were conducted by changes only in the forcing of interest, while all the other forcing were kept at preindustrial level.This is the main data set analyzed here to quantify the relative contribution from greenhouse gases, aerosols, natural forcings, and internal variability.
To confirm the reproducibility of the long-term historical changes, we use other available CMIP6-based historical wave simulations, which include WW3 simulations (Meucci et al., 2023) forced with ACCESS-CM2 (Bi et al., 2020), and EC-Earth3 (Döscher et al., 2022) GCMs and the MArine Science and NUmerical Modeling (MASNUM) wavenumber spectrum wave model (MASNUM-WAM) simulations (Song et al., 2020a) forced with FIO-ESM v2.0 (Table S1 in Supporting Information S1).It should be noted here that these simulations are available only for ALL scenario (natural plus all anthropogenic) with single ensemble member, but not for separate forcings (GHG, AER, and NAT).
For wind-speed analysis, we use multimodel data sets from CMIP6 (Eyring et al., 2016) to corroborate the patterns from multimodel mean with that of MRI-ESM2.0.Although there are many climate models participating in the CMIP6, there are only three models -namely CanESM5 (Swart et al., 2019), CNRM-CM-6-1 (Voldoire et al., 2019), and IPSL-CM6A-LR (Boucher et al., 2018) -that have at least 10 ensemble members for each of the historical experiments of DAMIP used here (Table S2 in Supporting Information S1).The daily data for wind speed from these models are interpolated to the same regular grid of 1.0°resolution.To access the robustness of trend, we identify the grids where >50% of members have statistically significant trend (at 5% level) and >80% (of the members previously identified with statistically significant trend) have the same sign for trend.In this way, the uncertainty assessment includes two types of variability: inter-member and inter-annual (Casas-Prat et al., 2024;Erikson et al., 2022;Morim et al., 2019).
We assess the 60-year linear trends in the ensemble mean H s for ALL, GHG, AER and NAT forcings.To quantitatively compare these trends with the unforced internal variability for spatially averaged values, we apply a moving window of 60-year length in a 10-year interval to 180-year long CTL simulations.Then we compute H s trends from each window, and take 95% confidence interval of those trends to represent the variability associated to the unforced signal.
Although ensemble averages can substantially reduce the internal climate variations, influence of internal variability can still persist due to small number of ensemble members (Deser et al., 2012).We compare the 5-year running mean H s anomalies for each external forcing scenario during 1961-2020 with respect to the range of modeled internal variability.These are the steps followed to compute internal variability: (a) select 13 60-yearlong samples from 180 year CTL simulations, (b) calculate ensemble mean of 13 time series, (c) compute 5-year running mean values of the ensemble mean (60 data points), (d) then take twice the standard deviation of the 5year running mean time series.
In the Arctic and Antarctic Ocean, we focus on the differences between the last two decades and the first two decades of 1961-2020 for sea ice, surface wind speed and H s .We estimate ensemble mean changes in MRI-ESM2.0based simulations for different forcing scenario and carry out a t-test for statistical significance using 5 members × 20 years = 100 data for each period.Results over the polar ocean are computed only for grid points below sea ice threshold (0.25) for the whole period.The grids, where sea ice concentration is higher than 0.25, are treated as grids with missing H s .So H s trend values are present for a grid point when there is no missing H s for the entire period.

Global Wave Height Trends for Different External Forcing in CMIP6
Figure 1 shows long-term trends in annual mean H s from five-member ensemble mean as simulated by WW3_MRI-ESM2.0during 1961-2020.The historical ALL simulation (Figure 1a) shows robust strengthening over the Arctic Ocean (AO), Southern Ocean (SO) and South Atlantic (SA) Ocean.The regions with robust weakening include the North Pacific (NP), North Atlantic (NA) and some parts of North Indian Ocean.GHG results (Figure 1b) display similar patterns to ALL in many regions of global Ocean (except South Atlantic), suggesting strong GHG influence on total H s trends.AER explains increasing H s over SA and decreasing H s over NP.Specifically, statistically robust AER trends (Figure 1c) exhibit opposite nature to GHG.It is consistent with positive forcings from greenhouse gases and negative forcings from total aerosol to total anthropogenic net effective radiative forcing (ERF) in MRI-ESM2.0(Oshima et al., 2020).NAT trends (Figure 1d) are weak and not robust over the global ocean.Furthermore, seasonal mean H s trends (Figure S1 in Supporting Information S1) are able to thoroughly reproduce the above spatial pattern and sign of trends for different forcing scenario.The influence of GHG and AER on the regions shown in boxes in Figure 1a are also clearly visible in seasonal trends (Figure S1 in Supporting Information S1).
Other CMIP6-based wave height simulations (Table S1 in Supporting Information S1) exhibit fairly similar H s trends (Figures 1e-1g) as MRI-ESM2.0based results (Figure 1a) during 1961-2020, further increasing the confidence in the wave simulations used in this study.The statistically significant downward trends in NA and NP and upward trends in AO, SO, and SA are found in WW3/ACCESS-CM2, WW3/EC-Earth3, and MASNUM-WAM/FIO-ESM v2.0, although using only single ensemble member.Yet, some differences exist between the different approaches.In particular, stronger negative values are seen in Northern Hemisphere (NH) in WW3/EC-Earth3 and stronger positive values for Southern Hemisphere (SH) in MASNUM-WAM/FIO-ESM v2.0.Overall, comparison between the three wave data sets are reasonable given the existing inter-model differences (Meucci et al., 2023).We acknowledge that dynamically coupled wave model and GCM data set are not yet available extensively, expect MASNUM-WAM/FIO-ESM v2.0.Therefore, advantage of ensembles of wave coupled GCMs (to standalone wave model) could not be quantified.It is worth mentioning that partly spurious positive H s trends (figure not shown) are seen in ERA5 reanalysis (Hersbach et al., 2020) data set during 1961-2020 period, which can be associated with the discontinuities in the assimilated observations (Meucci et al., 2023).Nevertheless, satellite altimeter measurements are not available for such long period to compare 60-year trends, although reasonable comparison with MRI-ESM2.0based H s trends are reported previously (Patra et al., 2023b).
Figure 2a shows the regionally-averaged results for AO, NA, NP, SA, and SO, which are the regions having experienced the largest changes.The historical-ALL (externally forced) trends are larger than internal variability range for all the ocean basins shown in the figure.Moreover, anthropogenic forcings (GHG and AER) are found to be responsible for the total trends during 1961-2020 over the basins.Indeed, the response of NAT forcing is small everywhere.To describe the relative dominance of GHG and AER, GHG dominates over AO, SO and NA; whereas AER over NP and SA.The polar seas show exceptional increment like none other basins.The 60-year trend in ALL simulation over NA, NP and SA are within 3%, but, in SO and AO it is around 8.2% and 22.3%, respectively.The GHG, AER, and NAT induced trends are 9.1%(±1.1%),2.7%(±1.9%)1.2%(±1.6%)respectively for SO, and 40.5%(±16.8%),0.4%(±19.3%),3.5%(±15.3%)for the AO basin.The inter-member standard deviation is quite large for AER and NAT signal over the AO basin, additionally, CTL variability is also quite considerable, though lesser than ALL signal.The difference between ALL and linear sum (GHG + AER + NAT) can arise due to nonlinear interaction between the forcings (J.Deng et al., 2020;Oshima et al., 2020) or response to land use forcing (Andrews et al., 2017).Figures 2b-2f show time-series of H s anomalies for the different external forcings, as well as the range of CTL simulations.The time-series for ALL simulation exceeds the CTL range (gray lines, 5%-95%), indicating a robust change.For the AO (Figure 2b), the external forcings are beyond the range of internal fluctuation, moreover monotonic rise in GHG H s signal induces increase in historical H s .Similar findings can be seen for the SO (Figure 2f), stronger monotonic rise in GHG H s and a weaker fall in AER H s signal cause rise in total (ALL) wave height.However, for NA, NP and SA, some of the forcings are within the range of internal fluctuation, although ALL forcing is the clear exception.In agreement with higher percentage change, higher anomalies are found for polar seas.For the NA, GHG signal exceeds the CTL ranges and the dominating forcing to total H s decrease.In contrast, AER is the robust signal as well as the dominating one for NP and SA.Overall, anthropogenic signals are found to be robust and the driving factors for total changes for all the basins discussed here.

Global Trends in Wind Speed in CMIP6 and DAMIP Experiments
The ensemble mean wind speed from MRI-ESM2.0(Figure 3) shows quite consistent pattern of trends with that of H s (Figures 1a-1d).The ALL simulation generally shows negative trends over NH, including the NA and NP, with exception over the Arctic region.Moreover, significant positive trends are seen in the SH subtropics and surrounding SO regions, similar with the findings of K. Deng et al. (2021).GHG forcing exhibits high level of agreement with the ALL forcing, with negative trends over the NA and NP, and positive trends over the AO and SO.Noteworthy influence of AER on total H s trends can be seen over the NP and SA regions.In contrast, NAT signal shows weak and insignificant changes over most regions.The inference from the above analysis is that the anthropogenic forcing (GHG and AER) can be attributed to surface wind speed changes during 1961-2020.
To address the single model limitation issue, the CMIP6 multi model mean (MMM) results are compared with MRI-ESM2.0.The MMM results (Figure 3) agree with MRI-ESM2.0 on spatial distribution and sign of trends, although the MMM shows stronger positive trends than the latter.The GHG dominance is clearly evident on total wind speed trends, not to mention the AO, SO and NA.In general, the wind speed associated with AER has weakened over the analysis period, including the NP.Consistent with above analyses, weak trends are present for NAT simulations.Overall, the MMM results agree with MRI-ESM2.0for attribution of anthropogenic forcing and signs of the trends.Here, we acknowledge the underestimation of positive trends in GHG and ALL simulations from MRI-ESM2.0 with respect to MMM, which does not include MRI-ESM2.0.The multi-member ensemble mean for each of these individual climate models (Figure S2 in Supporting Information S1) show agreement in spatial distribution and sign of trends with MRI-ESM2.0,but produce higher magnitude for trends.One should note the low spatial (Table S2 in Supporting Information S1) and temporal resolution (1 day) of these models with compared to MRI-ESM2.0(1.125°× 1.11209°-1.12148°and3 hourly).

Driving Mechanisms of Wave Height Changes Over the Arctic and Antarctic Polar Oceans
Since polar seas are shown to have experienced the largest wave height changes over the 1961-2020 period, and since polar wave climate is known to be affected by changes in sea ice extent (Thomson & Rogers, 2014), we focus on the possible drivers over these regions during the months of minimum sea ice extent (September for NH and March for SH) utilizing the MRI-ESM2.0data set.With the global warming, reduction in sea ice extent increases available fetch for waves in these basins.Although MRI-ESM2.0 has bit smaller September sea ice extents than the observation by Walsh et al. (2017), the decadal-to-multidecadal variations agree well between the model simulation and the observation (Aizawa et al., 2021).Moreover, MRI-ESM2.0has been reported as one of the few CMIP6 models which are able to simulate a plausible amount of sea ice loss and simultaneously a plausible change in global mean temperature over time compared to the observational estimate (Notz & Community, 2020).Although CMIP6 models have larger intermodel spread for Antarctic Sea ice area than that in the Arctic, moderately higher confidence is reported in simulation of Antarctic climate in CMIP6 than previous generations (Roach et al., 2020) of CMIPs.To further substantiate the sea ice area in Antarctic, 1979-2014 mean sea ice area and concentration in February from MRI-ESM2.0appear to be close to observation (Roach et al., 2020).Moreover, the sea-ice edge distribution of the Antarctic simulated by MRI-ESM2.0 is consistent with observations (Yukimoto et al., 2019).MRI-ESM2.0simulates consistent decline in NH sea ice extent over 1961-2020 period.The end of both summer and winter months (March and September) show decline, but with faster decline in September in NH (Figure S3 in Supporting Information S1).Whereas in SH both March and September exhibit almost similar level of decline (Figure S3 in Supporting Information S1).September changes between 2001-2020 and 1961-1980 over the Arctic Ocean shows statistically significant decline in sea ice area fraction (Figure 4a), in line with Arctic warming.The strong decline associated with GHG forcing is the major factor for the sea ice depletion over the Arctic.In addition, weaker decline, although statistically significant can be seen in connection with AER and NAT influence, but only limited to the Pacific sector, especially in the regions of the Chukchi and East Siberian seas.
The change pattern for wind speed indicates intensification in the recent decades for north of 70°N in Pacific sector (Chukchi, East Siberian and Laptev sea).In the further north (>80°N), the significant increase cover eastern Arctic Ocean.It seems that positive changes of GHG and NAT have induced the wind speed increment for the polar region (>70°N).There might be some traces of internal variability as we use 20-year mean in the analysis.
Finally, for wave height, strong increase is experienced over the entire circle north of 70°N (Figure 4a), this being the region where sea ice edge shifted northward as a result of ice melting.The H s increases are reflected over wider areas in Barents-Kara seas, northern Greenland sea, and bit less wider areas of Chukchi, East Siberian and Laptev sea, also bit scattered over Beaufort sea and Baffin Bay.For the GHG simulation, these increases are even stronger and extend further northward.In contrast, weaker changes are present for AER and NAT simulations.In addition, larger fetch is clear in case of GHG than AER and NAT simulations, as a result of more northward sea ice retreat associated with GHG forcings.Overall, wave height increase is majorly associated with increase of the effective fetch (less ice cover) along with wind speed increase too.Besides, there is a clear signature of the sea ice variability in H s change patterns.In line with this analysis, Liu et al. (2016) reported wave height increase over the Arctic despite decrease in wind speed during 2007-2015.Sea ice retreat majorly contributes to the variability of waves in the Arctic Shelf seas (Khon et al., 2014;Wang et al., 2015).It should be noted that, in the Atlantic sector, local winds and remotely generated swell combine together to determine changes of sea state (Semedo et al., 2015).Nonetheless, positive feedback between sea-ice area decline and wind speed increase is reported by Alkama et al. (2020), which is very clear for March over the Arctic (Figure S4 in Supporting Information S1).The regions (Barents and northern Greenland sea) with strong decline in sea ice coincide with the regions with wind speed increase and H s increase.As in September, GHG emerges as the most controlling factor in March, but strong changes appears bit southward.
For the Antarctic region, the sea ice changes during March implies decline in sea ice area concentration around the Antarctic coastline, over much wider areas in west Antarctic (Figure 4b).Consistent sea ice decline persist for the case of GHG.Contrary to this, increases in sea ice extent appear for AER simulations.And, weaker signal is present for NAT simulations.Overall, GHG forcing is found to contribute dominantly to the sea ice deline in MRI-ESM2.0.
Statistically significant wind speed increment can be seen around Southern Ocean basin in ALL simulations, mostly associated with GHG induced increase.Although these wind speed increase are not continuously present around the Antarctic coastlines, strong significant wave height increase appears all over the Antarctic seas (Figure 4b), specifically over larger areas in west Antarctic seas.This pattern of increment is mostly induced by GHG forcing.Scattered negative (weak) changes are present for AER (NAT) simulation.Similar to Arctic ocean, larger fetch becomes available for waves to grow in GHG scenario, in contrast to AER and NAT.The qualitative contribution of wind speed to wave height increase is higher for Antarctic seas than the Arctic.In addition, reduction in sea ice can induce increase in surface wind in Arctic and Antarctic seas (Alkama et al., 2020;Iwasaki, 2023;Khon et al., 2014).Thus sea ice is majorly responsible for the increase wave power over the polar seas (Iwasaki, 2023).

Conclusion
We conducted DAMIP experiments using MRI-ESM2.0based simulations and identified the external forcings responsible for historical wave height changes during the 1961-2020 period.Historical-ALL simulations, based on MRI-ESM2.0and other CMIP6 models, have undergone robust strengthening over the Arctic Ocean, Southern Ocean, and South Atlantic, but weakening over the North Pacific and North Atlantic.The DAMIP experiments suggest that historical H s changes are mostly driven by anthropogenic (GHG and AER) effects alone, additionally GHG dominates over AER in most basins except NP and SA.The surface wind speed corroborates the above changes in MRI-ESM2.0and in multi model mean as well.
We further quantified the role of sea ice and surface winds to historical H s changes over the Arctic and Antarctic regions involving DAMIP experiments.The sea ice decline appears to be a more important contributor to H s increase over the Arctic, which can be further explained by GHG-induced ice depletion.Moreover, GHG instigate poleward sea ice retreat to a greater extent (longer fetch) than the other factors.The influence of surface winds is more influential in Antarctic H s increase than that of Arctic.GHG signal emerges as the controlling factor for both sea ice decline and wind intensification, which further reflect in H s rise.
Overall, the DAMIP multimodel analyses showed that anthropogenic forcings are mostly responsible for the historical changes in wave height.The amplification of wave height in the Polar seas are remarkable, which can cause more sea ice breaks, and subsequently accelerates ice melting (Boutin et al., 2020;Thomson & Rogers, 2014).We acknowledge the non-availability of any other H s data in the framework of DAMIP.The detection and attribution of H s trends to anthropogenic and natural causes using multi models and large ensembles, when available, would be more robust.Moreover the attribution results are based on five-ensemble member, it would be ideal to use at least 10 members (Casas-Prat et al., 2022) in order to be more confident in conclusions.

•
CMIP6/DAMIP simulations show that anthropogenic signals are robustly detected for the significant wave height (H s ) trends during 1961-2020

Figure 2 .
Figure 2. (a) Region (as in Figure 1a) averaged trends in H s from WW3/MRI-ESM2.0 simulations during 1961-2020: five member mean (bar) and ±standard deviation (error bar).The rightmost vertical whiskers indicate range of internal variability defined by CTL.(b-f) Time series of 5-year running mean of region averaged H s for AO, NA, NP, SA, and SO respectively during 1961-2020.The gray lines indicate 5%-95% ranges of CTL simulations.

Figure 3 .
Figure 3. Linear trend (%) expressed as percentage change per year during 1961-2020 for annual mean surface wind speed from MRI-ESM2.0simulations for ALL, GHG, AER, NAT forcing scenario (5 member mean) (a).Same analysis but using multi model mean (MMM) derived from 3 CMIP6 models having 36 ensemble member in total for ALL, and 35 for the others (b).The percentage is calculated with respect to 1961-2020 climatology.Robustness is defined at grids (stippling) considering both inter-annual and inter-member variability.

Figure 4 .
Figure 4. (a) September Arctic Ocean changes between (2001-2020) and (1961-1980) using ensemble mean values in MRI-ESM2.0based simulations.Statistical significance (stippling) is following t-test using 5 members × 20 years = 100 data for each period.Changes over the Antarctic Ocean are shown for the month of March (b).