the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 06 Nov 2025
Quantification of the influence of anthropogenic and natural factors on the record-high temperatures in 2023 and 2024
Abstract. The unexpectedly high global mean surface temperature (GMST) anomalies in 2023 and 2024 greatly exceeded the temperatures observed in the years directly prior. In this paper, we use a multiple linear regression energy balance model to quantify the contributions of several natural and anthropogenic factors to the GMST, including the large reduction of sulfur emissions from the shipping sector since 2020. The model is trained on 170 years of historical climate data, and allows for the attribution of warming to various natural and anthropogenic factors. The influence of anthropogenic activity on the GMST is quantified using a 160,000 member ensemble that considers the uncertainty in the magnitude of aerosol radiative forcing and the strength of climate feedbacks. We find that in response to a rise in global radiative forcing of either 0.1 W m−2 or 0.15 W m−2 due to the reduction of sulfur emissions from international shipping, the associated rise in GMST by the end of 2024 is either 0.028 °C [0.025 to 0.031 °C, 5−95 % range] or 0.043 °C [0.038 to 0.046 °C], respectively. We also show that approximately 0.092 °C of the rise in annual mean GMST from 2022 to 2023 can be attributed to a shift from La Niña to El Niño conditions, which is approximately a third of the observed 0.3 °C rise in GMST between these two years. Additional increases in the annual mean GMST in 2023 and 2024 (both relative to 2022) of 0.075 °C [0.036 to 0.096 °C] and 0.053 °C [0.019 to 0.074 °C] are attributed, respectively, to a strong positive Indian Ocean Dipole (IOD) event that began in 2023. Our study is the first to suggest a significant contribution from the IOD to the anomalously high values of GMST observed in 2023 and 2024. Anomalously high Sea Surface Temperatures (SSTs) in the North Atlantic region led to a rise in GMST of 0.070 °C [0.054 to 0.094 °C] and 0.069 °C [0.055 to 0.091 °C] in 2023 and 2024 relative to 2022, respectively. This contribution is almost 90 % lower when the short-term variability component of North Atlantic SSTs is removed, resulting in lower estimates of the GMST anomaly in 2023 and 2024 than observed. These results suggest that short-term variability in the North Atlantic SSTs may have played a significant role in influencing the GMST anomalies in both 2023 and 2024; however, it is unclear whether this variability is internally or externally forced. Increased incoming solar radiation due to the 11-year solar cycle led to an additional rise in GMST of 0.025 °C [−0.009 to 0.051 °C] and 0.029 °C [−0.008 °C to 0.056 °C] in 2023 and 2024 relative to 2022, respectively. While the 2023 and 2024 GMST anomalies can be reconstructed fairly well from a combination of natural and anthropogenic factors, uncertainties remain in the reconstruction, driven primarily by the imprecise knowledge of the radiative forcing of aerosols, and the strength of climate feedbacks.
- Preprint
(1853 KB) - Metadata XML
-
Supplement
(2488 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2025-4865', Anonymous Referee #1, 23 Jan 2026
- AC1: 'Reply on RC1', Endre Farago, 17 Mar 2026
-
RC2: 'Comment on egusphere-2025-4865', Anonymous Referee #2, 09 Feb 2026
Farago et al. use multiple linear regression to attribute the exceptionally high global mean surface temperatures observed in 2023 and 2024, combining an empirical energy-balance framework with time series of multiple external forcings and modes of internal variability. They constrain the regression by historical observations from 1850 onward based on suitable datasets. They find that the extreme warmth of 2023/24 is largely explained by the superposition of long-term anthropogenic warming with strong interannual variability, including ENSO, but the study also highlights a large contribution associated with the positive phase of the Indian Ocean Dipole (IOD) in 2023/24. The work provides a timely and clearly structured quantitative synthesis of drivers behind the recent record temperatures. I think that some aspects could benefit from clarification, but overall I think that this work is a valuable contribution to the discussion around the record-high recent temperatures and their attribution to various drivers, with one key strength of the study being the comprehensive inclusion of numerous external and internal drivers in a common quantitative framework. I recommend publication subject to minor revisions.
###
Before providing specific comments, I'd like to highlight three aspects that remain somewhat unclear to me:
(i) The large contribution of the IOD is surprising and also leaves some questions open about its (in)dependence from ENSO. If I am not mistaken, the IOD has thus far not been considered a significant contributor to past (annual) GMST anomalies - an amplitude around values typical for ENSO is astonishing. Can you provide an estimate for how much of the unforced GMST variability the index might explain before 2023, e.g., in terms of R-squared, and, more generally, provide more explanation why this large contribution in 2023/24 might be plausible, possibly including some hints at mechanism involved?
(ii) AMV is considered in two flavours, once with and once without high-frequency variability. However, with the "M" in AMV standing for multi-decadal, is it at all reasonable to consider a monthly AMV index with a strong high-frequency component?
(iii) The IMO-related aerosol effect on GMST in 2023/24 is determined to a high degree by the choice of the corresponding ERF, for which two values are assumed, based approximately on existing estimates. The EBM parameters somewhat influence the temperature response, but it's still strongly determined by the ERF. Please check related formulations to avoid the impression that the new estimates provided here might be more independent than they actually are from the "educated guesses" about the ERF.
###
Specific comments (including minor technical ones):
Title: "Factors" -> "factors"
First paragraph of introduction: When it comes to the role of internal variability in 2023/24 temperatures, you mention the Raghuraman et al. 2024 paper, but I think you should also mention the Terhaar et al. 2025 paper.
L47/48: Actually global SO2 emissions have been declining already since around 1980.
L57/58: It's a bit strange to point to the table with recent estimates of ERF_AER here without mentioning any of them explicitly, for example their range.
L62/63: Also here I'd consider it helpful to mention the EffCS explicitly.
L89: Do I understand correctly that sigma_OBS is large in the earlier and small in the later part of the historical record, and that the formulation thereby simply makes sure that the more uncertain past temperature observations are more weakly constrining the model parameters than the more certain recent observations? If so, maybe that can be mentioned explicitly.
Fig. 1: Why is a single ensemble member shown here instead of the ensemble mean? If I understand it correctly, the ensemble members considered here sample parameter uncertainties, not different realizations of internal variability (like ensemble members of a climate model in CMIP), so also the ensemble mean here should show similar variability, right? Why would that not be more suitable to show?
Eq. 3: Why is there a sum (capital sigma) sign in front of ERF_GHG(t)? Is that not redundant with the plus signs, given that there's only one ERF_GHG(t) time series, if I'm not mistaken?
L161: If you mention the -1.1W/m2 central estimate, I recommend to mention the left an right sigma values of the asymm. Gaussian, too, tom define the weighing function completely.
L207: Consider quantifying what is meant by "insensitive".
L220-221: Misleading formulation: The CMIP6 TSI time series certainly covers more than until 1978, you just use it only up until then.
L241: "Given that the net effect of the eruption of Hunga on GMST is small, we chose to use SAOD as a proxy for the impact of the Hunga volcano on GMST, while neglecting the additional radiative forcing from the injection of stratospheric water vapor."; The first part is per se not a good justification of the second part, given that a small net effect could in principle result from two large counteracting contributions. I think the assumption you make is OK, but the formulation here is misleading, so that should be clarified. Related, does the omission of the warming effect of the stratospheric water vapour not affect the "explanation gap"? Specifically, could the "Residual" in Tab. 2 be partly explained by the omission of that warming component of the HTHH eruption?
Fig. 3+4: I am wondering how coherent the positioning of different ensemble members would be within the "uncertainty ranges" denoted in the upper panels: E.g., would a member with relatively large anomalies in 2019 also exhibit relatively large anomalies in 2024, that is, does the parameter sampling largely reflect the long-term behaviour or also (or more) the amplitude of short-term fluctuations, so that ensemble members would fluctuate more around the ensemble mean? Maybe an additional figure in the supplement could clarify that?
L377-380: Also here, it would be useful to consider the Terhaar et al. (2025) paper.
L475/476: "whether this variability was influenced substantially by the IMO2020 regulations remains unclear"; consider reformulation, given that the permanent IMO regulations would not really "influence variability" (their effect could just be mistaken for variability).
L479: "During the positive phase of IOD" -> I suggest "During positive phases of IOD"
L503: It is unclear in what way / why Swapna et al. is cited here. (I'm not necessarily suggesting to remove the reference, it should only be clear what it's about / what it is stating).
Tab. 2: Is there a particular reason why "Non-IMO anthropogenic" is lumped together, instead of showing its components (GHG, AER, TSI, ...)?
L532: I would avoid the term "projections" here, making one think of future scenarios...
###
References:
Terhaar, J., Burger, F.A., Vogt, L. et al. Record sea surface temperature jump in 2023–2024 unlikely but not unexpected. Nature 639, 942–946 (2025). https://doi.org/10.1038/s41586-025-08674-z
Citation: https://doi.org/10.5194/egusphere-2025-4865-RC2 - AC1: 'Reply on RC1', Endre Farago, 17 Mar 2026
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 464 | 340 | 28 | 832 | 76 | 18 | 22 |
- HTML: 464
- PDF: 340
- XML: 28
- Total: 832
- Supplement: 76
- BibTeX: 18
- EndNote: 22
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The manuscript presents a comprehensive analysis of the global mean surface anomalies observed in 2023 and 2024. It builds on previous studies that focussed on individual drivers of these anomalies by jointly estimating the contributions of anthropogenic activity, volcanic eruptions, 11-year cycle variability, El Nino Southern oscillation, Atlantic Multidecadal variability, Pacific Decadal Oscillation and Indian Ocean Dipole, using a multi linear regression energy balance model.
My only concern relates to section 2.1 and the Appendix, which would benefit from improvements in the presentation order and reproducibility of the results, as suggested in the specific comments. After these corrections, the manuscript is suitable for publication.
Specific Comments:
87-88: the paragraph starts explaining the first term, then the time grid and then the other terms of equation 1. This sentence applies to all terms of the equation and breaks the flow of the description of each term of the equation. I would first mention the time grid and then each term.
89: equation 2 is briefly mentioned here before the explanation of equation 1 is complete.
113-115: iIt is unclear whether AAWR represents the slope of dTanth that is obtained from a linear fit to temperature anomalies shown in panel 1a or the slope is obtained from a linear fit to dTanth. In the latter case, how is dTanth retrieved? Please rephrase the sentence.
119-120: The text says to “see caption” of figure 1, while the caption of Figure 1 similarly and repeatedly refers the reader back to the text (“see text”). In neither case is it clear what specific information the reader is expected or how it provides additional clarification. Please make these references explicit in the text and summarize the relevant information clearly in the caption of figure 1. Figure 1 contains a large amount of detail, and repeatedly moving back and forth between the text and the figure makes it difficult to follow the main point.
In these same lines: 1) it is unclear whether the term “single fit” refers to all panels in Figure 1 or only to panel a; 2) include the values of λ here as well, as is done in panel a, and refer to the section where you explained how you got this value; 3) does “the single time series of ERFaer” refer specifically to panel (b)?
134: While EBM acronym is introduced at line 60, the meaning of EBM-1 is not defined. It is only explained later in the Appendix (but not even at the beginning of the Appendix) and it is not straightforward what the number 1 means. A brief explanation should be provided here.
135: It only explains the temperature of the upper layer. What about the lower layer? You should also mention here how this approach is an improvement? I think that lines 591-600 belong here.
279-280: same as comment for lines 113-115.
284-285: This sentence is difficult to follow and would benefit from being rephrased.
286-288: Panel a should either be discussed first or moved to the last position in the figure. The discussion of Figure 2 begins at line 281 with panels b and c, which made me wonder whether I had missed something, as panel a is only discussed later.
303: EffCS is briefly mentioned here and in the introduction at lines 61-, where it is stated that “Our model [...] provides an estimate of Effective Climate Sensitivity (EffCS)”. The purpose of this sentence (L303) is unclear in its current form. The sentence should be expanded to provide substantive information about the EffCS estimate, as suggested in the Introduction, or it should be removed from this section.
326-328: It is unclear how the colors should be interpreted or how the probabilities were calculated. Additionally, the choice of colormap is confusing: the reds and blues appear to represent higher and lower probabilities, respectively, while white seems to indicate the most probable value, but it is actually the opposite.
394-400: can you provide a value for the respective change in TSI that corresponds to the change in GMST?
519-525: This paragraph discusses the anthropogenic warming rate, which was already covered in Section 3.2. Its placement in the section on the Indian Ocean Dipole feels abrupt; it would be more appropriately included in the discussion in Section 3.2.
Conclusion: The conclusion presented here is weaker than the one in Section 3.4 (lines 523–525). Consider expanding this section by incorporating the points made in those lines.
Appendix: Please provide the values or a range of values that have been used throughout the appendix to allow reproducibility of the results (gamma L 655; value used in equation A8, only few values are defined at L675-677; initial value of gamma used in L705, three values of Cu and Cd at L726-729).
Supplementary, L51:Define the latitudinal and longitudinal boundaries of the four regions to ensure data reproducibility
Technical corrections:
118: remove the brackets: (gamma, Geoffroy at al., 2013)