Human-induced intensified seasonal cycle of sea surface temperature

Sea surface temperature (SST) strongly couples the atmosphere and ocean, and its seasonal cycle can modulate climate even without changes in the annual mean. Variations in SST seasonality affect marine heatwaves, monsoons, precipitation, and ENSO, and have important ecological impacts through changes in ocean oxygen content. Climate models consistently project an intensified SST seasonal cycle in a warmer future, with CMIP5 suggesting ~30% amplification by late 21st century relative to preindustrial levels. Proposed mechanisms include thermodynamic effects via changes in surface heat flux seasonality and oceanic processes tied to mixed layer depth (MLD) changes; warming-driven upper-ocean stratification shoals the MLD, reducing thermal inertia and amplifying SST seasonality. However, detecting whether such intensification has already emerged in observations is challenging due to sparse ocean records and internal variability. Prior detection studies have focused mainly on land. This study addresses the gap by combining observations and model simulations to assess whether SST seasonal cycle intensification has occurred over the oceans in the past four decades and to attribute the drivers.

Model projections (CMIP5/CMIP6) indicate robust future intensification of SST seasonality, with proposed drivers including altered atmospheric circulation and surface heat flux seasonality, and oceanic processes linked to enhanced stratification and shoaling MLD. Past work has emphasized the role of reduced mixed layer heat capacity under warming, and documented widespread upper-ocean stratification increases. Detection studies of seasonal cycle changes have largely targeted terrestrial temperatures, leaving oceanic emergence uncertain due to observational sparsity, especially in the Southern Ocean, and the confounding influence of internal variability. Recent related studies point to enhanced ocean seasonality affecting extreme events and biogeochemistry, but a comprehensive detection and attribution for the global ocean in the modern observational era was lacking, which this work provides.

Observations: Three SST datasets (1982–2022) were used—ERSSTv5 (2°), HadISSTv1.1 (1°), and OISSTv2 (0.25°). Pre-1982 was excluded due to sparse winter high-latitude coverage. For mixed-layer temperature and MLD (1982–2022), objective analyses used were IAP (1°, 0–2000 m, monthly), Ishii v7.3 (1°, 0–3000 m), and EN4 (1°, >5000 m). All fields were bilinearly interpolated to 2° × 2°. Model simulations: External-forcing estimate (EXT) combined five ensembles totaling 181 realizations: CMIP6 multi-model ensemble (first member per model) extended with SSP5-8.5 from 2015–2022, and large ensembles (ACCESS-ESM1-5, CanESM5, MIROC6, MPI-ESM1-2-LR). Single-forcing attribution used DAMIP simulations (1982–2020) separating greenhouse gases (GHGs), anthropogenic aerosols (AERs), stratospheric ozone (StratO3), and natural forcings (NATs) from seven models with ≥3 members. Seasonal amplitude calculation: At each grid cell, the amplitude equals the difference between climatologically determined months of maximum and minimum (months derived from 1970–2000 climatology: ERSST for SST; IAP for ocean temperature/MLD; CMIP6 MME for SHF, DO, and CO2 flux). Annual amplitudes were computed for observations and models. MLD definitions: CMIP6 mlotst uses a 0.125 kg m−3 potential density threshold from the surface density at model timestep. Observational MLD (IAP) used monthly potential density with a 0.01 kg m−3 threshold (TEOS-10 GSW). Ishii and IAP showed consistent decreasing MLD trends; EN4 trends were treated cautiously due to relaxation to climatology in data-sparse regions. Mixed layer heat budget: The budget for mixed layer temperature Tm is ∂Tm/∂t = Qnet/(Cp ρ h) − um·∇Tm + r, where terms represent thermal forcing (surface heat flux and MLD), horizontal advection, and a residual including entrainment and unresolved processes. Each term was time-integrated (1982–2022) to obtain seasonal amplitudes at the months of Tm maxima/minima. The thermal forcing term was further decomposed into contributions from changes in annual-mean SHF, SHF seasonal amplitude, annual-mean MLD, and MLD seasonal amplitude by varying one factor at a time while holding others at climatology. Due to data availability, the budget used eight CMIP6 models (CAMS-CSM1-0, CAS-ESM2-0, CanESM5, CESM2-WACCM, FGOALS-f3-L, IPSL-CM6A-LR, NESM3). Statistical significance used effective sample size corrections for autocorrelation. Analysis choices: Trends evaluated primarily over 1983–2022 (excluding 1982 due to anomalously low amplitude). Spatial pattern comparisons used pattern correlations and stippling for 95% significance (Student’s t-test).

- Detection: The global mean SST seasonal cycle amplitude increased by 0.16 ± 0.07 °C from 1983 to 2022, a 3.9 ± 1.6% rise relative to the 1983–1992 climatological mean (~4.2 °C). Intensification is strongest in northern subpolar gyres (up to ~10% regionally), with pronounced increases in the North Pacific (horseshoe pattern), western North Atlantic, and north of the ACC (35°–50°S). - Model agreement: EXT ensembles reproduce the observed temporal evolution and spatial features; 35/36 CMIP6 models show intensification (0.3% to 7.0%). Pattern correlation between observed and simulated significant trends reaches 0.85. - Attribution: DAMIP indicates GHGs account for ~2.0% of the global increase during 1983–2020 (≈55.6% of total historical enhancement). Reduced AERs add ~0.8%, especially shaping a meridional dipole in the North Pacific that enhances subpolar intensification. StratO3 contributions are negligible; NATs produce non-significant episodic effects tied to major eruptions (El Chichón 1982, Pinatubo 1991). - Mechanisms: Mixed layer heat budget shows thermal forcing term Qm dominates the global amplification (+11.2%), partly offset by the residual term (−6.0%); horizontal advection is negligible. Decreased annual-mean MLD is the primary driver of Qm, yielding a global +29.1% amplification; MLD seasonal amplitude decreases offset part of this. Regional Qm-induced intensification is large in northern subpolar gyres (e.g., +58.1% relative to 1983–1992 mean north of 30°N), with local MLD-shoaling-driven increases up to ~158.6% relative to baseline in hotspots. Increased ocean heat uptake and enhanced SHF seasonal amplitude contribute regionally (North Pacific, North Atlantic, northern flank of ACC) but are smaller than MLD effects. - Vertical structure: Intensification extends from the surface to the base of the mixed layer across latitudes; reduction in amplitude below the mixed layer is consistent with shallower MLD limiting penetration of the seasonal signal. Observational analyses (IAP, Ishii) corroborate modeled MLD shoaling and mixed-layer-wide intensification. - Future projections: Under SSP5-8.5, the global SST seasonal amplitude increases by ~10.6% by 2100 relative to 2015–2024; SSP2-4.5 and SSP3-7.0 yield ~4.8% and ~7.7% increases. Subpolar North Pacific/Atlantic (45°–60°N) see ~10.3% increase even under SSP2-4.5. - Biogeochemical implications: The seasonal cycle of surface dissolved oxygen intensifies by ~3.7% globally (1983–2022) with a strong inter-model correlation (r = 0.84) between DO and SST amplitude trends; spatial pattern correlation between SST and DO amplitude trends is 0.60. Regional DO seasonal contrast increases: subpolar North Atlantic +15.7%, subpolar North Pacific +5.5%, Southern Ocean +6.9%. Air–sea CO2 flux seasonality also intensifies but with weak correlation to SST amplitude changes (r = 0.14), implying other drivers (e.g., winds).

The study demonstrates that the SST seasonal cycle has already intensified over the last four decades and that this change is a robust, externally forced response rather than internal variability. The attribution analysis identifies anthropogenic GHG increases as the primary driver, with additional contributions from reduced anthropogenic aerosols since the 1980s. Mechanistically, upper-ocean stratification increase and consequent MLD shoaling reduce mixed-layer heat capacity, amplifying seasonal temperature swings; changes in ocean heat uptake and SHF amplitude play secondary regional roles. The intensification occurs throughout the mixed layer, confirming the central role of mixed-layer processes. These findings resolve the observational emergence question and connect SST seasonality changes to biogeochemical impacts, notably increased seasonal contrast in dissolved oxygen, which can exacerbate hypoxia risks when superimposed on long-term deoxygenation. The results have broader implications for oceanic and terrestrial extremes given the known links between seasonality and extreme event statistics.

This work provides clear detection and attribution of a human-induced intensification of the global SST seasonal cycle since 1983, quantified at 3.9% ± 1.6%, with hotspots in northern subpolar gyres and the northern flank of the ACC. Multi-ensemble simulations closely match observations in magnitude and spatial pattern, implicating increased GHGs as the dominant driver and decreased aerosols as a notable contributor. A mixed layer heat budget shows that MLD shoaling, through reduced thermal inertia, is the primary mechanism, with smaller contributions from changes in surface heat fluxes and ocean heat uptake. The intensification extends through the mixed layer and is associated with a strengthened seasonal cycle of surface dissolved oxygen, raising ecological concerns. Future research should improve observational coverage in data-sparse regions (especially the Southern Ocean), refine MLD estimates and process representations in models, investigate interactions with marine heatwaves and biogeochemical cycles, and assess regional ecosystem impacts and adaptation strategies under varying emissions scenarios.

- Observational sparsity, particularly pre-1982 and in the Southern Ocean, limits detection and contributes to discrepancies between observations and models; pattern correlations are lower in the Southern Hemisphere. - The EN4 dataset employs relaxation to climatology in data-sparse regions, potentially biasing long-term trends, leading to inconsistencies with IAP/Ishii and models. - Differences in MLD definitions/thresholds between models and observational analyses can affect quantitative comparisons and depth of intensification. - Internal low-frequency variability may contribute regionally; although ensembles minimize it globally, attribution at basin scales can still be influenced. - CMIP6 historical simulations were extended with SSP5-8.5 after 2014; scenario dependence could introduce small differences relative to actual forcings post-2014.