Here we use simulations with a global coupled ocean-atmosphere model (CM2.1)10 in which sea surface temperature (SST) is nudged toward observations in the tropical Pacific (TP, marked in Fig. 2d; 10% of Earth surface area), but the ocean and atmosphere are otherwise fully interactive (Methods). The model is run 20 times, different only in initial conditions, over a 42.5-year period from January 1982 to June 2024. The ensemble mean represents the climate response to imposed radiative forcing and TP SST variability. A great advantage of the TP pacemaker simulation is that the results can be directly compared to observations to determine the TP contribution and underlying mechanisms2. The TP pacemaker ensemble mean tracks observed GMST very well (Fig. 1a), and the detrended time series are correlated at 0.73, significant at 99% level. A 10-member TP pacemaker ensemble with a newer climate model (CESM2)11 yields very similar results (Fig. 2c, S1), confirming that ENSO is a major driver of interannual variability in GMST.
ENSO is often tracked by Niño3.4 SST anomalies (5°S-5°N, 170°W-120°W), which peak in the three-month season of November-January. A conventional annual averaging breaks one ENSO event into two calendar years, a more natural ENSO year is defined as 12 months from July 1 to the following June 30, which captures ENSO variability better than calendar years. Indeed, the model-observation correlation is higher (r = 0.76) for detrended GMST averaged for ENSO years (Fig. S2). Record GMST anomalies were observed during 13 months of June 2023-June 2024 (Fig. S3), a time window consistent with the development of the 2023-24 El Niño. El Niño-induced warming is efficiently communicated throughout the tropical troposphere, resulting in surface warming over tropical continents and the tropical Indo-Atlantic oceans at a time delay of 2-3 months due to the ocean’s thermal inertia12.
Observed GMST jumped by 0.31 °C from 2022 to 2023, a record since 1880 when GISTEMP data started. The TP pacemaker ensemble mean simulates a jump of 0.18 °C or 58% of what was observed (Fig. 1b). Almost all the 2022 to 2023 GMST jump in the TP pacemaker ensemble mean is due to that in the tropics (30oS-30oN, covering half Earth surface area). At 0.47 °C, the observed tropical warming is larger but within the range of the TP pacemaker ensemble (Fig. 1b). Indeed, intense SST warming was observed over the tropical North Atlantic from May 2023 onward (Fig. 2a), coupled with the weakened northeast trade winds (Fig. S4). Tropical North Atlantic SST warming began to develop in April 2023 and reached 2 °C, or 2.5 standard deviations off West Africa, aided by positive feedback of ocean-atmosphere interaction13. The pre-El Niño warming of the tropical North Atlantic explains the early commencement of tropical and global warming in 2023, and why the TP pacemaker ensemble mean under-simulates the observed warming magnitude.
Extratropical contribution
The TP pacemaker simulates large variability in extratropical Northern Hemisphere (NH) temperature change, ranging from −0.60 °C to +0.35 °C (Fig. 1b), equivalent to –0.15 ~ + 0.09 °C GMST. The temperature variability averaged over the extratropical NH is not significantly correlated with ENSO (Fig. S5). The observed extratropical NH temperature change stands far above the TP pacemaker ensemble mean, by 0.33 °C, equivalent to 0.08 °C GMST. The observed surface warming in 2023 is indeed much larger than the pacemaker ensemble mean over central Asia, the North Pacific Ocean east of Japan, and mid-high latitude North America (Fig. 2a vs. 2d). Remarkably, TP pacemaker run #14 produces a temperature distribution very similar to observations (Fig. 2b vs. 2a), while increasing GMST anomaly by a whopping 0.18 °C above the TP pacemaker ensemble mean. It indicates an important role of extratropical variability as well as El Niño in making the record GMST in 2023. Extratropical Southern Hemisphere temperature variability is smaller.
Among the contributors to GMST variability, ENSO is predictable a few seasons in advance, and tropical Indo-Atlantic SST variability is of seasonal persistence. The predictability of extratropical temperature variability remains to be fully determined.
Prediction
The Niño3.4 SST anomaly turned negative in August 2024, signaling the arrival of a La Niña state in the tropical Pacific. Nevertheless, the 2024 monthly GMST anomalies matched or exceeded those of 2023 for the first eight months (Fig. S3), stirring speculations that 2024 could become yet another hottest year on record. We have removed the SST nudging over the tropical Pacific on 1 July each year and integrated the freely running model for 12 months. The annual-mean forecast is obtained by combining January-June observations and the 20-member TP pacemaker ensemble mean for July-December. The correlation of detrended annual GMST amounts to 0.90 between observations and the forecast for 1982–2023, and it remains high at 0.64 for the July-December mean. The forecast skill originates from high predictability of Niño3.4 SST (r = 0.77 for November-January) and the associated Indo-Atlantic response through tropical cross-basin interactions.
Our forecast initialized on 1 July 2024 puts the 2024 GMST 0.08 ± 0.08 °C (ensemble mean ± inter-member standard deviation) higher than 2023 (Fig. 1a), despite the transition of the Pacific from an El Niño to a La Niña state. As validation, observed GMST was 0.10 °C higher in 2024 than 2023. We recently updated the CM2.1 TP pacemaker simulation through 2024. The simulated two-year change in GMST from 2022 to 2024 ranges from 0.17 to 0.46 °C, inclusive of the observed value of 0.40 °C.
The prediction initialized in July 2024 indicates that the 2024/25 ENSO year will be cooler by 0.18 ± 0.10 °C than the record-shattering 2023/24 ENSO year (Fig. S2). The model over-predicts the 2024 La Niña with a July-December Niño3.4 of −0.73 °C as opposed to the observed −0.24 °C. This would exaggerate the predicted GMST decrease for the 2024/25 ENSO year.