Introduction

Human activities are increasingly influencing ocean temperatures, leading to changes not only in the mean state but also in the seasonality1,2,3,4,5. These changes in seasonality have a multitude of impacts on phenomena such as marine heat waves, Asian monsoons, tropical cyclones, and marine ecosystems6,7,8,9,10, with profound scientific and economic implications. Observational evidence points to an intensification of the sea surface temperature (SST) seasonal cycle in recent decades11,12. Over the 40 years, the amplitude of SST seasonal cycle has increased by approximately 1% per decade, with the northern subpolar gyres experiencing the strongest intensification, reaching up to 2.5% per decade12. This intensified SST seasonality has been linked to a shallower mixed layer depth (MLD) associated with a more stratified upper ocean2,13,14,15, although changes in air-sea heat exchange also contribute in the North Pacific and North Atlantic12,16,17. Climate models project that by the end of this century, the seasonal cycle of SST could intensify by up to 10% relative to the present-day level13,14,18.

The intensification of the SST seasonal cycle has enhanced seasonal contrasts in surface ocean oxygen solubility and air-sea carbon exchange2,12. These changes carry direct implications for processes such as primary production, ocean biomass, and ocean carbon uptake—all of which are already under pressure from rising mean SSTs19,20,21. However, biological activity in the ocean is not confined to the surface but extends throughout the euphotic zone, where sunlight penetrates and supports photosynthesis over a hundred meters. In this zone, seasonal temperature variations remain significant due to strong vertical mixing22, which directly affects biological processes at these depths.

Significant changes in the vertical structure of ocean thermodynamics and dynamics have been observed in recent decades, including increased ocean stratification23,24, reduced vertical mixing25, and altered ocean circulations26,27. These changes are likely to affect the seasonality of subsurface ocean temperature12,14. However, it remains unclear whether such changes in seasonality have occurred below the ocean surface. To address this gap, we analyze three different observational ocean temperature datasets and historical simulations from a broad set of climate models from the Coupled Model Intercomparison Project phases 6 (CMIP6). Our findings reveal a human-induced weakening in the subsurface temperature seasonal cycle across the extratropical oceans since the 1980s, which contributes to a shift of the temperature seasonal cycle to shallower depths. Using a hierarchy of coupled ocean-atmosphere models and a simple vertical diffusive model of ocean temperature, we demonstrate that this weakening in subsurface seasonality is primarily driven by increased stratification and suppressed vertical mixing, resulting from enhanced ocean heat uptake due to rising greenhouse gases.

Results

Contrasting changes in surface and subsurface seasonal cycles

To quantitatively assess seasonal changes in the upper ocean temperature since the 1980s, we first examine the monthly mean hemispherically averaged ocean temperature using the IAP observational gridded dataset (see Observations in Methods). Climatologically, the extratropical upper ocean (down to 100 m) exhibits pronounced seasonal variations, with a temperature difference of approximately 1 °C between seasonal maximum and minimum (Fig. 1a, b, contours). In the Northern Hemisphere (NH), the SST peaks around September and reaches its lowest values around March. This seasonal signal propagates downward, with the temperature seasonal cycle at 100 m lagging the SST by approximately 1 month in cold seasons and by 2–3 months in warm seasons.

Fig. 1: Contrasting seasonality changes between the surface and subsurface ocean over the past four decades.
figure 1

a, b Trends in monthly extratropical ocean temperature (shading in °C per decade) during 1981-2022 for the (a) Northern Hemisphere (NH; 15°−60°N) and (b) Southern Hemisphere (SH; 15°−60°S) from the IAP dataset. Climatological mean ocean temperatures (contour in °C) are superimposed. c Time-depth plot of anomalies in the extratropical mean (15°−60°) amplitude of the ocean temperature seasonal cycle (shading in °C per decade) from the IAP dataset, with anomalies referenced to the 1981–1990 climatological mean. The anomalies are smoothed with a 3-year running mean filter. Stippling indicates statistically significant trends at the 90% confidence level based on the Mann–Kendall test.

Full size image

Since the 1980s, contrasting seasonal changes have emerged between the near-surface and subsurface ocean in the NH (Fig. 1a, colors). During warm seasons, for example, positive temperature anomalies have developed in the upper 30 m, while negative anomalies have formed below, centered around 50 m. These anomalies are phase-locked to the background seasonal cycle, manifesting an intensification of temperature seasonality in the near-surface ocean and a weakening in the subsurface ocean. This pattern also indicates a shallower reach of the seasonal cycle over the past four decades. A similar pattern of contrasting changes in the temperature seasonal cycle is also observed in the Southern Hemisphere (SH) (Fig. 1b, colors). However, temperature variations in the SH are notably weaker than in the NH, largely due to different seasonal changes in the Southern Ocean and the SH subtropical ocean (Supplementary Fig. 1)28.

The shift towards shallower depths in temperature seasonality is further corroborated by the temporal evolution of seasonal amplitude in the extratropical oceans (Fig. 1c), estimated as the difference between the maximum and minimum temperatures of the seasonal cycle at each grid point (see Amplitude of seasonal cycle in Methods). Over the past four decades, increasing trends near the surface and decreasing trends in the subsurface gradually emerge. The near-surface intensification aligns with previous findings on the amplified SST seasonal cycle11,12, while the subsurface weakening indicates a growing decoupling of deeper layers from surface-driven seasonal variability. These contrasting vertical trends are also evident in other observational datasets, including those with shorter temporal coverage during the Argo era (Supplementary Fig. 2).

Given the many research efforts on the intensification of the SST seasonal cycle, we focus mainly on the distribution and mechanisms behind the reduction of the subsurface seasonal amplitude. During 1981-2022, a broad reduction of subsurface seasonal amplitude can be identified poleward of 15°N (Fig. 2a and Supplementary Fig. 3a). In the SH, this weakening is concentrated primarily within the subtropical band between 15°S and 35°S, with comparatively minimal changes in the Southern Ocean poleward of 35°S. Notably, the reductions in the temperature seasonal cycle are consistently confined beneath the climatological mixed layer in both hemispheres (Fig. 2a, thick blue line). The zero-crossing depth, where the anomalies reverse sign, deepens at higher latitudes, corresponding to the deeper mixed layer in these regions.

Fig. 2: Weakened subsurface seasonality in observations and simulations.
figure 2

a, b Trends in zonal-mean seasonal amplitude of ocean temperature during 1981-2022 (in °C per decade) from (a) IAP and (b) CMIP6 multi-model mean (historical: 1981-2014, SSP5-8.5: 2015–2022). Climatological seasonal amplitudes (contour in °C) are superimposed, with the thick blue line indicating the mean mixed layer depth (MLD) averaged over 1981-1990. c, d Geographic distribution of trends in the subsurface seasonal amplitude averaged across the 60 m layer below the climatological MLD during 1981–2022 from (c) IAP and (d) CMIP6 multi-model mean (shading in °C per decade). e Time series of the extratropical mean (15°−60°) subsurface seasonal amplitude from observations and CMIP6 simulations (dashed line, in % per decade). Thick solid lines indicate the linear trends, and pink shading indicates the standard deviation among the CMIP6 ensemble. Changes in IAP, Ishii, and CMIP6 multi-model mean are relative to the 1981–1990 average, while changes in JAMSTEC Argo are relative to the 2001–2005 average. f Boxplots of trends in the extratropical mean seasonal amplitude during 1981–2020 from the ensembled historical, greenhouse gases (GHGs), and anthropogenic aerosols (AERs) simulations. Box edges show the interquartile range, whiskers show the extrema, horizontal lines show the median value, and diamonds show the ensemble mean. Dots indicate trends of individual ensemble members. Stippling in a–d indicates statistically significant trends at the 90% confidence level based on the Mann–Kendall test.

Full size image

In contrast, the subsurface equatorward of 15° displays a more complex pattern of amplitude changes. Narrow bands of intensification are observed in the northern tropics, while reductions occur in the southern tropics. This intricate pattern may be attributed to the unique seasonality in the tropics, where the sun passes overhead twice a year, reducing the variance explained by the annual cycle to less than 50% in many tropical ocean regions (Supplementary Fig. 4). Moreover, tropical seasonal variations are strongly influenced by El Nino-Southern Oscillation29, which operates through mechanisms distinct from those at higher latitudes. Consequently, discrepancies in the tropics are more pronounced compared to the consistent weakening in the extratropics across multiple datasets (Fig. 2a and Supplementary Fig. 3a).

The subsurface weakening in the extratropical regions is most pronounced in the Indo-Pacific basin, whereas the Atlantic basin between 25°N and 40°N experiences an intensification below the mixed layer (Supplementary Fig. 5). As such, changes in the Indo-Pacific basin contribute more substantially to the zonal-mean pattern. To further investigate the geographic pattern of this weakened subsurface seasonality, we assess changes in seasonal amplitude averaged across the 60 m layer below the climatological MLD (see Mixed layer depth in Methods; Fig. 2c and Supplementary Fig. 3b). Negative trends are widespread over most of the extratropical subsurface ocean. In the NH, significant negative trends are evident in the North Pacific poleward of ~20°N and the North Atlantic poleward of ~40°N. In the SH, similar trends are observed between 15°S and 45°S. In contrast, the seasonal amplitude has intensified over large portions of the subsurface tropical ocean, particularly in the central and eastern tropical Pacific.

The weakening of the subsurface seasonal cycle is further supported by a time series analysis of seasonal cycle amplitude (Fig. 2e). Relative to the climatological mean during 1981–1990, subsurface weakening poleward of 15° shows a statistically significant decline of 1.1 ± 0.4% per decade in the IAP dataset (blue line) and 1.7 ± 0.5% per decade in the Ishii dataset (cyan line) over the past four decades. For the shorter period during the Argo era (2001–2022), the decline is even stronger, reaching 2.5 ± 1.1% per decade (gold line), indicating an accelerated weakening in the subsurface seasonality.

Attribution of subsurface seasonal cycle changes

To investigate the role of anthropogenic forcings in subsurface seasonality changes, we analyze CMIP6 historical and SSP585 simulations (see CMIP6 model simulations in Methods). The multi-model mean of these simulations effectively captures the primary spatial patterns of observed trends in seasonal cycle amplitude from 1981 to 2022, including positive trends extending from the surface to the bottom of the climatological mixed layer and negative trends concentrated below the mixed layer (cf. Fig. 2b to a). Notably, the reduced amplitude in the models occurs at greater depths compared to observations, corresponding to the deeper MLD in the models. This alignment indicates an important role of vertical mixing in shaping the vertical contrast of the seasonal cycle changes. The geographic agreement between the ensemble-mean and observed subsurface trends in extratropical regions is also high (cf. Fig. 2d to c). However, distinct local differences, particularly in the South Indian, South Pacific, and subtropical North Atlantic Oceans, highlight some regional discrepancies. Overall, the smoother patterns in CMIP6 simulations likely arise from the absence of internal variability in the multi-model ensemble averages.

Averaged over extratropical regions, climate models simulate a continuous weakening of the subsurface seasonal cycle by 1.1 ± 0.3 % per decade during 1981–2022 (Fig. 2e, red line), closely matching the observed trend of 1.4 ± 0.4 % per decade (mean of IAP and Ishii; Fig. 2e, blue and cyan lines). This weakening is robust across most of the 26 CMIP6 models analyzed (Fig. 2f, gray dots). Projections under the SSP585 scenario show a further continuation in the decline of subsurface seasonality (Supplementary Fig. 6), with a weakening of approximately 1.9% per decade from 2023 to 2099, suggesting a persistent trend extending beyond the 21st century.

The strong consistency in temporal evolution and spatial structure between observations and simulations leads us to conclude that the subsurface weakening in ocean temperature seasonality, and its shift to shallower depths, is predominantly driven by anthropogenic forcing. To determine the relative contributions of different anthropogenic forcings, we analyze single-forcing simulations from the Detection and Attribution Intercomparison Project, focusing on the individual impacts of two major climate forcing agents: greenhouse gases (GHGs) and anthropogenic aerosols (AERs) (see ‘CMIP6 simulations’ in Methods). The multi-model mean (Fig. 2f and Supplementary Fig. 7) shows that GHGs are the dominant factor, accounting for approximately 69% of the historical weakening in subsurface seasonality. While AERs contribute to ~23% of the weakening, substantial inter-model spread around zero prevents a definitive conclusion regarding their influence. Nevertheless, the ratio of the contributions to the subsurface weakening between GHGs and AERs is consistent with that identified in the SST seasonal cycle intensification12. This alignment underscores the dominant role of GHGs in driving seasonal changes in the upper ocean and in shifting the seasonal cycle toward shallower depths.

Mechanisms of the weakened subsurface seasonal cycle

The strong agreement between the observed and simulated subsurface weakening of the seasonal cycle motivates us to investigate the underlying mechanisms using further model experiments. To investigate the processes responsible for the GHGs-induced reduction in subsurface seasonality, we analyze the Flux-Anomaly-Forced Model Intercomparison Project (FAFMIP) ensemble (see ‘FAFMIP simulations’ in Methods). This ensemble separates the effects of surface heat, freshwater, and momentum fluxes on the ocean’s response to global warming.

Our analysis reveals that changes in surface heat flux (Fig. 3a, cyan line) predominantly drive the seasonal response to increased GHGs (Fig. 3a, black line), causing intensification near the surface and weakening at depth (see also Supplementary Fig. 8 for their zonal-mean patterns). Specifically, enhanced ocean heat uptake generates greater warming near the surface, resulting in a more stratified upper ocean—a phenomenon robust in both climate models26 and observations30. The increased stratification inhibits vertical mixing, limiting the downward heat penetration and thereby suppressing seasonal variations in the subsurface ocean. This relationship is supported by a high inter-model correlation of -0.70 (statistically significant at the 99% confidence level) between changes in subsurface seasonal amplitude and annual mean MLD, with each 1-meter reduction in MLD leading to an approximately 1.3% decrease in subsurface temperature seasonality (Supplementary Fig. 9). In contrast, contributions from wind stress and freshwater flux are much smaller and tend to offset each other (Fig. 3a, purple and orange lines). Overall, the FAFMIP results suggest that the increased ocean heat uptake is the primary driver of the weakened subsurface seasonality in the historical and GHG simulations, and observations.

Fig. 3: Seasonal amplitude changes driven by individual surface forcings.
figure 3

a Vertical profiles of extratropical mean seasonal amplitude changes (in °C) from individual FAFMIP forcing experiments, compared to pre-industrial control simulations. Results are averaged over years 41 to 70 of each simulation. b Zonal-mean changes in seasonal amplitude due to a uniform heat flux perturbation, obtained as the difference between OHEAT and OCTRL. The thick blue line denotes the mean mixed layer depth in OCTRL.

Full size image

Does the spatial pattern of surface heat flux changes matter for the upper-ocean seasonality changes? To address this question, we conduct ocean-only experiments using the Community Earth System Model 1 (CESM1). In these experiments, a spatially and temporally uniform heat flux perturbation of 1.5 W/m2 is applied to the ocean surface, while all other atmospheric forcing is prescribed as climatology (see CESM1 ocean-only experiments in Methods). The results show that even with homogeneous surface heating, the upper ocean exhibits a shallower reach of the seasonal cycle (Fig. 3b), mirroring patterns seen with increased GHG and in observations. This finding suggests that the ocean temperature seasonal response is primarily driven by adjustment in vertical mixing to the imposed heating, with limited sensitivity to the spatial distribution of surface heating. However, in these uniform heating experiments, the zero-crossing depth shows little variation with latitude, in contrast to the observations and CMIP6 simulations, where it deepens at higher latitudes with deeper mixed layers. Another discrepancy arises in the Southern Ocean. While observations and historical simulations show minimal subsurface changes below the mixed layer poleward of 40°S (Fig. 2a, b), the CESM1 experiments exhibit a clear weakening of the subsurface seasonal cycle in this region. Besides, the meridional dipole structure within the mixed layer is absent in the ocean-only experiments. These discrepancies imply the role of surface heat flux changes or wind forcing in shaping regional temperature seasonality patterns (Supplementary Fig. 8c).

Additional support from an idealized diffusive model

To further validate the role of reduced vertical mixing in subsurface weakening, we perform simulations using an idealized one-dimensional diffusive model that relies exclusively on diffusive processes for vertical energy transport (see ‘Idealized diffusive model’ in Methods). The model has a high constant diffusivity in the upper 60 m, representing the strong vertical mixing within the mixed layer, and a lower diffusivity below. Following a net-zero sinusoidal annual cycle, a consistent surface heat flux forcing is applied to all simulations. The diffusive model effectively captures the downward propagation of the temperature seasonal cycle signal in the extratropical upper ocean (Fig. 4a, contours).

Fig. 4: Seasonal amplitude changes in an idealized diffusive model.
figure 4

a Monthly ocean temperature changes (shading in °C) resulting from a 2% reduction in vertical mixing, obtained from the difference between the simulation with a diffusivity of 9.8 × 10−5 m2 s−1 and the control simulation with a diffusivity of 10 × 10−5 m2 s−1, for mixed layer depth (MLD) set at 60 m. Climatological ocean temperatures (contour in °C) from the control simulation are superimposed. b Vertical profiles of seasonal amplitude changes (lines in °C per decade) between the simulation with a diffusivity of 9.8 × 10−5 m2 s−1 and the control simulation with a diffusivity of 10 × 10−5 m2 s−1, for MLD set at 50 m, 70 m, 90 m, and 110 m. Bars represent the climatological seasonal amplitude (in °C per decade) from the simulation with a diffusivity of 10 × 10−5 m2 s−1 and a MLD of 110 m.

Full size image

We conduct perturbation experiments in which vertical diffusivity is reduced by 2% in the upper 60 m, mimicking the impacts of suppressed vertical mixing, which is based on the observed ~2% per decade increases in the upper ocean stratification in the extratropical oceans during the past four decades30. The resulting changes in seasonality (Fig. 4a, colors) align well with our earlier findings. Reduced mixing traps more energy near the surface, intensifying the seasonal cycle at these levels. Consequently, for a constant energy input at the ocean surface in this model, the seasonal cycle in the subsurface layer is weakened. This vertical contrast pattern is robust across a range of sensitivity experiments with varying MLD (Fig. 4b, lines). Importantly, as MLD increases, the zero-crossing depth – where amplitude changes shift from positive to negative – overall shifts deeper, mirroring observations and climate simulations where the zero-crossing depth is deeper at higher latitudes with greater MLD (Fig. 2a, b).

Further experiments with varying MLD but constant diffusivity within the mixed layer (see ‘Idealized diffusive model’ in Methods) also reveal a vertical contrast in seasonality changes (Supplementary Fig. 10). Collectively, these findings from the idealized diffusive model reinforce the notion that increased stratification is the primary driver of the observed weakening of subsurface ocean temperature seasonality.

Discussion

While previous studies have largely focused on the intensification of SST seasonality, our study reveals a weakening of the seasonal cycle in the subsurface ocean, i.e., there are contrasting seasonal changes between the surface and subsurface ocean. Between 1981 and 2022, we observe a weakening of approximately 1.4% per decade (mean of IAP and Ishii datasets) in seasonal amplitude below the mixed layer in extratropical regions. This weakening, along with the intensification of the seasonal cycle near the surface, results in the confinement of the temperature seasonal cycle to shallower depths—an aspect that has been largely overlooked in the existing literature. Our analysis of CMIP6 simulations demonstrates that the upper ocean seasonality changes are well captured by historical GHGs simulations, indicating an anthropogenic origin of these changes. The key mechanism behind this subsurface weakening is increased stratification due to enhanced heat uptake, which suppresses vertical mixing in the upper ocean. This suppression limits the penetration of heat into deeper ocean layers, trapping more energy and the associated seasonal variation near the surface and generating contrasting seasonal changes with depth. This mechanism is consistently supported by FAFMIP overriding experiments and CESM1 ocean-only model experiments, and strongly corroborated by an idealized diffusive model, where all vertical heat transport is diffusive.

Uncertainties remain despite the consistent weakening of subsurface seasonality in different observational datasets. For example, the Ishii dataset shows a weakening trend 1.5 times greater than that in the IAP dataset, suggesting potential discrepancies stemming from differences in data quality, temporal-spatial coverage, and gap-filling techniques31. Likewise, notable discrepancies exist among state-of-the-art climate models, highlighting the need for further investigation into how model resolution and parameterization may influence these trends.

The observed changes in the temperature seasonal cycle extend from the surface to the euphotic zone, posing stresses on marine organisms that are that are sensitive to seasonal variations, potentially impacting the overall health and productivity of marine ecosystems. Therefore, the ongoing and likely continuing changes in temperature seasonality throughout the upper ocean underscore the need for sustained ocean observations to advance our understanding of ocean temperature variability at different timescales.

Methods

Observations

To examine changes in the seasonal cycle of ocean temperature and the MLD from 1981 to 2022, we use three monthly observational datasets of ocean temperature and salinity: (1) the Institute of Atmospheric Physics (IAP) ocean temperature and salinity analysis, covering the period from 1940 to the present for the upper 2,000 m31, (2) the Ishii v7.3.1, covering the period from 1955 to 2022 for the upper 3,000 m32, and (3) the Japan Agency for Marine-Earth Science and Technology (JAMSTEC) gridded Argo data product, covering the period from 2001 to the present for the upper 2000 dbars33. All datasets are interpolated onto a 2°×2° grid using a bilinear interpolation for consistent analysis.

CMIP6 simulations

To estimate the human-induced impact on the ocean temperature seasonal cycle during 1981-2022, we use CMIP6 outputs of ocean temperature, and salinity from historical and SSP5-8.5 scenario simulations34. The outputs are taken from 26 models, selected based on data availability. Since CMIP6 historical simulations ended in 2014, the SSP5-8.5 scenario is used to extend the period to 202235. Only the first member from each model is used to ensure equal weighting in the multi-model ensemble mean analysis. All the outputs are interpolated onto a 2°×2° grid using a bilinear interpolation. The models are: ACCESS-CM2, ACCESS-ESM1-5, BCC-CSM2-MR, CAMS-CSM1-0, CanESM5, CAS-ESM2-0, CESM2, CESM2-WACCM, CMCC-ESM2, CNRM-CM6-1, EC-Earth3-CC, EC-Earth3-Veg-LR, FGOALS-f3-L, FGOALS-g3, FIO-ESM-2-0, GFDL-CM4, GISS-E2-1-G, HadGEM3-GC31-LL, IPSL-CM6A-LR, KIOST-ESM, MIROC6, MPI-ESM1-2-HR, MPI-ESM1-2-LR, MRI-ESM2-0, NESM3, and NorESM2-LM.

To assess the contributions of GHGs and AERs to the ocean temperature seasonal cycle changes, we use ocean temperature outputs from single forcing simulations from the Detection and Attribution Model Intercomparison Project (DAMIP)36. These simulations include scenarios with GHGs-only and AERs-only simulations. Due to the smaller number of available models in DAMIP compared to the CMIP6 historical ensemble, we select 9 models that provide at least three ensemble members, covering the period 1981-2020. To ensure equal weighting, only the first three members of each model are used. All the outputs are interpolated onto a 2°×2° grid using a bilinear interpolation. The models are: ACCESS-ESM1-5, CanESM5, CNRM-CM6-1, FGOALS-g3, HadGEM3-GC31-LL, IPSL-CM6A-LR, MIROC6, MRI-ESM2-0, and NorESM2-L.

Amplitude of seasonal cycle

The amplitude of the temperature seasonal cycle at each grid point is determined by calculating the difference between the monthly maximum and minimum temperature values. In specific, we obtain the months of maximum and minimum temperature based on the 1970-2000 climatology from the IAP dataset. These identified months serve as the basis for calculating the amplitude of the seasonal cycle for each year in both observations and simulations, ensuring that differences in calculated amplitudes among different datasets reflect true differences in seasonality, rather than discrepancies arising from the use of different reference months.

Mixed layer depth

The MLD is calculated for observational datasets and model simulations using the monthly potential density with a 0.03 kg m−3 criterion37. The potential density is derived from the temperature and salinity fields, computed using the TEOS-10 Gibbs Seawater (GSW) toolbox. This approach ensures consistency across datasets and provides a reliable estimate of the MLD, which is critical for interpreting vertical contrasts in seasonal amplitude trends in the upper ocean.

FAFMIP simulations

To evaluate the effect of changes in heat flux, freshwater flux, and wind stress on ocean temperature seasonality, we use ocean temperature outputs from the Flux-Anomaly-Forced Model Intercomparison Project (FAFMIP) simulations38, based on 7 climate models. The surface fluxes in these FAFMIP simulations are prescribed as the ensemble mean changes simulated by CMIP5 models under the 1pctCO2 scenario at the time of doubled CO2. We examine four specific FAFMIP experiments: FAF-heat, in which only net surface heat flux is prescribed; FAF-water, in which only net surface freshwater flux is prescribed; FAF-stress, in which only surface wind stress is prescribed; and FAF-all, in which all the above fluxes are prescribed simultaneously. Each experiment runs for 70 years, and our analysis focuses on the average of the last 30 years. The first member from each model is included in the analysis. All the outputs are interpolated onto a 2°×2° grid using a bilinear interpolation. The models are: ACCESS-ESM1-5, CanESM5, CAS-ESM2-0, MIROC6, MRI-ESM2-0, MPI-ESM1-2-HR, and MPI-ESM1-2-LR.

CESM1 ocean-only experiments

To assess the sensitivity of ocean temperature seasonality changes to the spatial pattern of net surface heat flux forcing, we use the ocean-only configuration of the CESM139. The ocean component, Parallel Ocean Program version 2 (POP2), utilizes a nominal 1° horizontal resolution grid (gx1v6) and has 60 vertical levels with thickness ranging from 10 m near the surface to 250 m near the bottom.

We integrate a control run (OCTRL) in which the ocean is forced with daily climatological surface heat fluxes from a preindustrial fully coupled simulation obtained from the National Center for Atmospheric Research Data Library. In the perturbation run (OHEAT), an additional uniform heat flux of 1.5 W/m² is applied to the ocean surface at every model time step. All other conditions are kept identical to the control run. Thus, any difference in ocean temperature between OHEAT and OCTRL can only be attributed to the oceanic adjustment to the implied uniform heating. Both simulations are integrated for 100 years, and our analysis is based on the average of the last 50 years.

Idealized diffusive model

To quantitatively assess the role of vertical mixing on upper ocean seasonality changes, we employ a one-dimensional diffusive model in which all vertical energy transport is realized as diffusive flux. The governing equation is:

$$\frac{\partial T(t,z)}{\partial t}=\frac{\partial }{\partial z}\left(\kappa \left(z\right)\frac{\partial T(t,z)}{\partial z}\right)+S\left(t\right),$$
(1)

where t and z represent time and depth, T is ocean temperature, κ is the vertical diffusivity coefficient, and S is the net surface heat flux. The first term on the right-hand side represents the contribution of vertical diffusion. The model represents the upper 150 m of the ocean with 15 evenly spaced layers and a time step of 3600 s. The surface heat flux S is applied to the first model layer following a sinusoidal annual cycle with net-zero flux:

$$-\frac{{S}_{o}}{{\rho }_{0}{c}_{p}H}\cos \left(2\pi t\right),$$
(2)

where So = 100 W m−2, ρ0 and cp are the density and heat capacity of seawater, respectively, and H is the thickness of the first layer. We emphasize that the model is purely energy diffusive and does not incorporate density in the governing equations. Thus, processes such as convection and advection are not represented in the model.

In the control experiment, the vertical diffusivity κ is set to 1.0 × 10−4m2 s−1 for the upper six layers and 1 × 10−5 m2 s−1 for the lower nine layers, mimicking the strong near-surface mixing typical of the ocean mixed layer. The perturbation experiment is performed the same way, but the κ of the upper four layers is reduced by 2% to simulate the impact of reduced vertical mixing. This reduction in κ is based on the observed ~2% per decade increases in the upper ocean stratification in the extratropical oceans during 1981-202230. Both experiments are initialized with SST = 20 °C and integrated for 10 years, with the last 5 years of outputs used for analysis. The response of temperature seasonality to decreased mixing is derived from the difference between perturbation and control simulations. To ensure that the results are robust across different MLD, we further perform multiple pairs of experiments where the MLD in the control simulations span the upper 5, 7, 9, and 11 layers, corresponding to depths of 50, 70, 90, and 110 m. In each corresponding perturbation experiment, the κ value in the upper mixed layer is reduced by 2%.

To further investigate the effects of a shallower MLD, we perform another set of experiments. In the control simulation, vertical diffusivity within the top 60 m is set to 1 × 10−4 m2 s−1. In the perturbation simulation, the same diffusivity is applied only to the top 50 m, representing a 10 m reduction in MLD.

Statistical significance test

The statistical significances of trends are examined based on the Mann–Kendall trend significance test.