Introduction

Anthropogenic global warming, driven primarily by emissions of greenhouse gases, has progressed at a broadly steady rate of around 0.2 °C/decade since at least 19701. Recently, however, several studies have also documented a minor step-up in the rate of global mean surface temperature anomaly (GSTA) increase around 19902,3,4, a continued rise in the global energy imbalance5, and an acceleration in the accumulation of ocean heat content6. Factors suggested to have driven this increase include continued greenhouse gas buildup from anthropogenic emissions, and loss of cooling from anthropogenic aerosols after sulfur emissions cleanup7, notably in China and in the global shipping sector.

Despite this apparent increase in warming rates and a transition of the Equatorial Pacific into an ENSO-positive state, the record surface temperature anomalies recorded through 2023 came as a surprise. All major temperature series show 2023 as the warmest year on record. The margin by which the record was set, around 0.15 °C, was also unusual, but not unprecedented for strong El Niño years. Notably, however, several ocean basins saw unprecedented surface temperatures through much of the year, including the Equatorial and North Pacific, the North Atlantic, and the Southern Ocean8,9.

A core question is whether this strong anomaly is consistent with internal variability10 and known decadal scale regional climate forcing, or if it indicates a rapid change in the climate system – or our influence on it4,11. The cleanup of shipping emissions has been implicated12, as has the 2021 Hunga Tonga volcano13, and a possible aerosol-related unmasking of a stronger-than-expected climate sensitivity. The possibility however remains that the 2023 GSTA record was simply a combination of ongoing anthropogenic influences, together with a sea surface temperature pattern that is within the ranges of observed interannual and decadal variability.

Here, we investigate this question using a recently introduced method that can isolate the contribution to the monthly or annual GSTA from the pattern of sea surface temperatures (SSTs)3,14. SST anomalies in different geographical locations are known to have varying influences on global temperatures, the so-called pattern effect15. Many studies have shown that this influence can be consistently quantified in Earth System Models, e.g. using simulations that independently perturb SSTs in multiple locations (Green’s functions)16,17,18. Our approach is to utilize a Green’s function (based on NCAR CESM1; see “Methods” section) to quantify the component of interannual (or monthly) GSTA variability that arises from the different ocean basins using time series of observed surface temperatures, i.e. the SST pattern after the pattern of global warming has been removed with gridpoint-by-gridpoint detrending14. The method makes no assumptions about the causes of SST trends or variability. It therefore lets us compare the SST pattern-induced corrections for a given year to previous observed years, regardless of whether that pattern is forced or natural, and therefore determine if the net influence is unprecedented or still within the bounds of previous observations. Note that the results will depend on the detailed response of the CESM1 model to localized SST perturbations, however, other studies have shown that the general spatial pattern and atmospheric dynamical features of such Green’s functions are broadly similar between models16,18,19,20.

Results

A common question is whether the 2023 Global Surface Temperature Anomaly (GSTA) was record-breaking by a record margin. In Fig. 1, we test this by calculating the margin for each record-breaking year, for four major temperature reconstructions (HadCRUT5, NOAA, Berkeley Earth, and GISTEMP). Most GSTA records since 1970 were set by margins of around 0.05 °C, while in El Niño years, margins of up to 0.15 °C have been recorded. Results are very similar for the four reconstructions. For 2023, the margins were indeed record high (0.17 °C) in two series (HadCRUT5 and BEST), while for the others it was on par with other recent El Niño years. This is consistent with expectations for a situation of strong but steady global warming, modulated by (ENSO-dominated) internal variability. From the observations over the last 50 years, there is therefore no immediate cause to claim that record margins are changing with time, which is, in turn, consistent with a broadly steady rate of global surface warming.

Fig. 1: Surface temperature anomalies and record margins since 1970.
figure 1

a Annual mean global surface temperature anomalies (GSTA), for four data series, relative to 1850–1900 (1880–1900 for GISTEMP). b Annual mean surface temperature record margins (increase over the previously highest recorded value). Dashed lines indicate El Niño dominated years (1998, 2016, 2023).

Full size image

We now investigate the overall SST pattern influence on 2023 global mean temperatures. Figure 2a shows the annual and monthly mean, pattern corrected, GSTA for the last 10 years, with the monthly SST-induced correction factor highlighted as red for anomalously warm (red) or cool (blue) months.

Fig. 2: Recent global mean surface temperature anomaly evolution in HadCRUT5.
figure 2

a Annual means, and corrections from the SST pattern quantified via a Green’s Function. 2023 and 2016 stand out as having notably strong positive SST pattern corrections. b–d Similar, for previous strong SST correction years (1952, 1969, 1998). e Monthly correction factors, for 1950–2023 (box-and-whisker) and the selected strong correction years. The box shows the 5–95% range, and the whiskers show the max/min range. f Annual correction factors, for CMIP6 (box-and-whisker), and HadCRUT5 (1950–2023). Strong correction years are shown in colours.

Full size image

After corrections, 2023 is still the warmest year on record. Now, however, it falls in line with recent long-term warming trends2,3. As expected for an El Niño dominated year, 2023 had a stronger-than-usual warm influence from sea surface temperatures throughout most of the year, following on from a series of relatively cool months influenced by multi-year La Nina conditions. However, there was a similar 12-month period of strong SST pattern-induced corrections around the last major El Niño episode, in 2015/2016. Taking the observational record (HadCRUT521; using 1950–2023 to ensure near-global coverage of ocean temperatures), we find that the monthly corrections through 2023 are well within the historical spread (Fig. 2e). In fact, when we select the four other years with strongest SST-induced corrections (1952, 1969, 1998, 2016), their distributions of monthly corrections, as well as their overall annual corrections, are very similar to 2023. The same is true for global mean corrections (Fig. 2f). We do find that, for HadCRUT5, 2023 is the year on record with the strongest SST-induced GSTA corrections, but with a small margin relative to the years mentioned. Compared to an ensemble of CMIP6 models, the 2023 correction is in the upper 5th percentile, but not exceptional. Note that for some models, we use multiple ensemble members to further sample the effects of internal variability. The overall conclusions are not influenced by this.

A natural question is whether the result is dependent on the temperature reconstruction of HadCRUT5. To test this, we have performed similar analyses for three other series (NOAA, Berkeley Earth, and GISTEMP). While there are minor differences in the ranking of years, all four series identify the same “strongest correction” years. All support the conclusion that 2023 had a strong contribution to its temperature anomaly from the SST pattern, but not an unprecedented one. See Supplementary Figs. 1–3.

To put the 2023 temperatures in a wider context, Fig. 3 shows an update of the analysis of the recent rate of global surface warming presented in Samset et al. 3. After Green’s function-based filtering, we find warming rates (in HadCRUT5) for the recent 10 (2014–2023), 20 (2004–2023) and 50 (1974–2023) years of 0.29, 0.27, 0.19 °C/decade, respectively.

Fig. 3: 50 years of global surface warming.
figure 3

All panels show global, annual mean surface temperature anomalies from the HadCRUT5 data series, raw (red) and SST influence filtered via a model-derived transfer function (black). The upper and lower insets show, respectively, the full data series since 1850, and the latest 10 years. Anomalies are taken relative to 1850–1900. The dashed black line shows the 50-year trend (0.19 °C/decade). Update from Samset et al.3.

Full size image

Finally, we investigate where the SST corrections come from, geographically, and whether the 2023 pattern of SST corrections is also consistent with what has previously been observed. Figure 4 compares 2023 to the four examples, strong correction, and years. The maps show the product of Green’s function and the observed SST pattern, such that the total correction (globally or regionally) is the sum of all relevant grid boxes.

Fig. 4: Ocean basin contributions to the 2023 temperature anomaly.
figure 4

SST pattern-induced corrections to the annual mean global surface temperature anomaly, for a 2023 and c the four other strongest correction years in HadCRUT5 (1950–2023). Maps show the product of Green’s function and the observed SST pattern, such that the total correction (globally or regionally) is the sum of all relevant grid boxes. Note the unit (milli°C). Box-and-whiskers b show the mean, 5–95% and max/min ranges for five selected ocean basins, coloured dots show the strong correction years. Approximate ocean basin definitions are indicated by dashed boxes in the 2023 map.

Full size image

Overall, we find very similar patterns between strong correction years, but with notable differences in some ocean basins. All five years were clearly influenced by El Niño conditions in the Equatorial Pacific, and have anomalously warm Tropical North Atlantic conditions. Elsewhere, the influence varies between years. Separating the global corrections by major ocean basins, we find that 2023 was consistent with both observed and modelled variability (Fig. 2b). The most notable outlier is in the Southern Ocean, however here the observational datasets have incomplete coverage for some of the selected years so we place less weight on this result. The North Atlantic also made very strong contributions to GSTA in 2023. For the Tropical North Atlantic (0–30 N) it was the strongest on record, but further north it was still in line with previous observations.

Discussion

The very strong deviation of the 2023 GSTA from recent global warming trends is therefore broadly attributable to the warm state of the various ocean basins. Independently, however, none of the basins had conditions that were markedly anomalous, or outside the range simulated for the historical era (1850–2014) by recent Earth System Models. What was special about 2023 was rather that multiple ocean basins had warm anomalies at the same time. Hence, it is possible that the 2023 record temperature, rather than heralding an acceleration in surface warming, was consistent with a combination of steady, anthropogenic global warming and ocean surface temperature variability on interannual and decadal scales. Previous years with comparable anomalies, and SST patterns, include 1953, 1969, 1998 and 2016, in all four observational series studied.

We note, however, that our method does not identify the underlying reasons behind the 2023 SST pattern, or for those in earlier years. A shift in global warming induced e.g. by an upwelling of previously stored deeper water temperatures, or factors such as the global energy imbalance, aerosol cleanup or changes in cloudiness anomalously affecting some ocean basins, would also produce SST-induced corrections in our analysis, and could therefore still have contributed to 2023 temperatures. The transition to El Niño conditions after a multi-year La Nina event is also rare in the observed climate history, making 2023 special also for this reason. Other studies have recently investigated the role of pacific temperatures for the 2023 GSTA, reaching seemingly differing conclusions11,22. Our study indicates a marked role of El Niño through the latter part of 2023, consistent with Forster et al.22.

Further investigations into the causes of the recent very strong surface warming are clearly warranted, notably including analyses of upper ocean heat uptake in various basins, and the regional influences of aerosol emissions changes. 2023 set record temperatures and had sea-surface temperatures in most ocean basins that were unprecedented in the observed records. However, 2023 was not necessarily a harbinger of rapid near-term surface warming, beyond the high global warming rates already known to result from anthropogenic emissions of greenhouse gases and aerosols.

Methods

The methods used in this study build on an approach pioneered and utilized in two previous publications3,14. We here explain the main elements and refer to the previous studies for further documentation and validation.

Datasets

The primary dataset used in this study is the HadCRUT5 gridded dataset of global historical surface temperature anomalies, version 5.0.2.021. We used monthly data for the period January 1850–December 2023.

For consistency checking, we also used three other gridded surface temperature data products, on monthly resolution, from their start dates and through December 2023: The GISS Surface Temperature Analysis version 4 (GISTEMP v4)23,24, NOAA Global Surface Temperature Dataset (NOAAGlobalTemp) Version 5.1 (updated July 2023), and Berkeley Earth (March 2024 update)25.

Trend removal

The pattern of monthly internal temperature variability is isolated from the long-term influence of anthropogenic warming via a boxcar smoothing with a 10-year window, applied at each gridpoint of the input dataset, relative to an 1850–1900 baseline. This removes the underlying global mean temperature increase, geographical patterns of global warming, and any seasonal differences. Near the endpoints, where there is insufficient data for the 10-year mean, we ensure consistent weighting by mirroring the data points. This was shown in Samset et al. 14 to have a negligible impact on the overall results, including trend estimates, even for the end years of the series. Note, however, that our chosen method will not remove decadal scale variability in regional temperature patterns.

Green’s functions

We use a Green’s Function (GF) calculated with the CESM1.2.1-CAM5.3 Earth System Model, monthly resolved and taking 40-year simulations with fixed sea-surface temperatures as input18,26. A GF, in our usage, relates an idealized increase in sea surface temperature at a given location, to resulting influences on radiation, clouds, water vapour and, ultimately, global mean surface temperature, and allows us to calculate the modulation of global mean surface temperatures resulting from a given pattern of SST variability. It is based on simulations where the SST was individually perturbed in 74 (partially overlapping) ocean patches of 80° longitude and 40° latitude. We use 2-metre surface air temperature to quantify the modulations.

GSTA modulations are calculated by multiplying Green’s Function for that month with the detrended SST pattern from observations (HadCRUT5 or other series). The total modulation is the sum of the contributions from all ocean-dominated grid points. For this calculation, monthly temperature fields from the observational series have been regridded to the GF resolution (2.5° latitude, 1.9° longitude).

We note that the CESM1-derived GF used here has been documented to have very similar response patterns to GFs produced with other models16,18,19,20. The applicability of equilibrium simulations to capture monthly and interannual variability has been investigated for previous publications3,14. While the full atmospheric response will not be realized for monthly varying SST patterns, the rapid timescales of Rossby wave train propagation and other dynamical features mean that the key features of a response do have time to develop27,28.

Ocean basin definitions

For the ocean basin mean corrections presented in Fig. 2, we use longitude/latitude boxes shown in Table 1.

Table 1 Ocean basin definitions
Full size table

Climate model data analysis

Output from climate model simulations was analysed identically to the surface data series, including filtering using Green’s Function. All simulations used were provided for the ScenarioMIP29 CMIP6 Endorsed MIP30, and made available to the community through the Earth System Grid Federation (ESGF). We made use of 250-year transient simulations using the CMIP6 historical (1850–2014) and Shared Socioeconomic Pathway (SSP) (2015-2100) emission datasets. Two SSPs were used (SSP1-2.6, SSP5-8.5), to extend the historical data through 2027. (The main analysis runs through 2022, but we use subsequent data to avoid the need for mirroring the endpoints). The results of Fig. 2 are insensitive to the choice of SSP (SSP5-8.5, SSP3-7.0, SSP2-4.5, SSP1-2.6). Only monthly mean temperature data (Global Surface Air Temperature at 2 metres) was used.

CMIP6 models

The CMIP6 models used for the present study are shown in Table 2, along with the relevant model and dataset references, and the number of ensemble members.

Table 2 Climate model simulations used in the present study
Full size table