Introduction

The El Niño Southern Oscillation (ENSO) is the largest source of interannual variability in the global climate system1,2. Climate system teleconnections emanating from ENSO affect the lives, economic activity, and livelihoods of people worldwide3,4. It is thus necessary to understand its internal variability and its response to external forcings, especially in the context of anthropogenic impacts on ENSO variability due to climate change5,6. While the far-afield impacts of ENSO on global climate are well understood, the influence of other regions such as the tropical Atlantic and Indian oceans on ENSO variability remains an area of active research. Our current understanding of ENSO is primarily based on instrumental and historical records of sea-surface temperatures (SSTs) and atmospheric circulation, limited to about the last 150 years. Studying the paleoclimatic history of ENSO could provide greater insight into ENSO variability and teleconnections, especially from past climates without drastic anthropogenic perturbations and with different mean states.

The mid-Holocene (MH: 6000 years Before Present or BP) was characterized by different insolation forcing and greenhouse gas concentrations compared to the pre-industrial (PI) or the present day7. Greenhouse gas concentrations were lower, with methane levels during the MH being 15% lower than the pre-industrial concentrations. Enhanced seasonal contrast in the Northern Hemisphere due to precessional changes intensified the West African Monsoon (WAM), which further led to a “Green Sahara”, i.e., the expansion of shrub vegetation across northern Africa in place of the present-day Saharan Desert8,9,10. Basin-wide multi-proxy evidence from the Pacific has indicated that the MH was also characterized by a La Niña-like mean state with lower ENSO variability (Supplementary Fig. S1). Proxy records from core ENSO regions such as the central equatorial Pacific11,12 and eastern equatorial Pacific13,14,15 indicate fewer El Niño events, cooler and drier conditions, and a dampened ENSO during the MH. Lower ENSO variability is also shown by proxy records from ENSO-teleconnected regions such as Papua New Guinea16, Malaysia17, and eastern Australia18, where changes in isotopic signatures reflecting changes in ENSO-linked precipitation are recorded.

Climate models have also consistently shown that the ENSO variability during the MH was lower than for the present day18,19,20,21,22. The Paleoclimate Modeling Intercomparison Project (PMIP) coordinates experiments for the mid-Holocene7. An ensemble of PMIP3 and PMIP4 models (henceforth referred to as PMIP3/4 model ensemble) has shown a reduction in SST over the eastern and southeastern equatorial Pacific on the order of 0.5 °C and a decrease in precipitation over the equatorial Pacific of 0.1−2 mm/day, reminiscent of La Niña conditions23. However, some proxy-model comparisons have suggested that climate models may be underestimating the magnitude of ENSO variability that occurred during the Holocene. For example, comparisons of ENSO variability as reconstructed from coral and mollusc records from the tropical Pacific and as estimated from equilibrium simulations with nine PMIP3 models24 and transient simulations with four PMIP4 models25 indicated that the climate models showed inconsistent and relatively muted changes. Another comparison of estimates of ENSO variability changes as reconstructed from central equatorial Pacific coral records and as obtained from proxy system modeling based on outputs from the Community Earth System Model (CESM1.2) also showed that simulated estimates were lower than reconstructed estimates20. Model-proxy comparisons are hindered by a lack of sufficiently long proxy records; however, the emerging signal of a model-proxy discrepancy has also been attributed to an underestimation of ENSO sensitivity to the orbital forcing in climate models25. Here, we explore the possibility that the model-proxy discrepancy may indicate a missing teleconnection that modulates ENSO, which has not been adequately captured in climate model simulations so far.

In recent years, the vegetation changes over the Sahara have emerged as an important driver of the MH climate, with several studies showing its remote effects on the global climate system through various teleconnections26,27,28,29,30. Simulations with a coupled ocean-atmosphere global climate model EC-Earth version 3.1 indicated that the Green Sahara could have modulated ENSO variability through changes in the Walker Circulation31. In contrast, most previous studies of MH ENSO have focused on the effects of changes to orbital forcing, greenhouse gas concentrations, ice sheet extents, and meltwater discharge, but neglected vegetation and land surface changes18,19,20,21,22,32. The Green Sahara does not adequately emerge in the PMIP3/4 model ensemble33, indicating that incorporating orbital and greenhouse gas forcings alone does not lead to the simulation of a potential Green Sahara— ENSO teleconnection. In light of the increasingly recognized importance of the Green Sahara, potential model underestimations of ENSO variability reduction, and a lack of multi-model studies, it is crucial to investigate the Green Sahara—ENSO teleconnection through a model intercomparison.

We hypothesize that the under-estimation of the simulated reduction in ENSO variability during the MH is due to an inadequate representation of the Green Sahara in climate models. To test this hypothesis, we study MH ENSO variability in the context of the Green Sahara using simulations from five global climate models. For each model, we analyzed outputs from two MH simulations—with and without prescribed vegetation and land surface changes over northern Africa (MHGS and MHREF, respectively)—to isolate the impact of the Green Sahara on the equatorial Pacific and Atlantic oceans. Since the Green Saharan changes are the only difference in the experimental design for the MHGS and MHREF simulations, we attribute MHGS anomalies relative to the MHREF to the Green Sahara in the MHGS simulations. Both MH simulations incorporate changes in orbital parameters and greenhouse gas concentrations following PMIP recommendations. The MHGS simulations additionally incorporate the presence of the Green Sahara through vegetation and land surface changes. These include prescribing various vegetation types over the Sahara, reduction of dust, changes to soil albedo and composition, and including mega-lakes. The MHGS simulations were carried out differently for each model, with the central idea of adequately representing the vegetation and land surface changes that occurred over northern Africa during the mid-Holocene. Regardless of the specific Green Sahara conditions, all MHGS simulations show a significantly strengthened WAM relative to the MHREF simulations, indicated through precipitation and wind strength changes (Supplementary Fig. S2). A previous quantification of the proxy model agreement indicated that the MHGS simulations provide a more realistic picture of the MH than simulations that do not account for the Saharan vegetation changes34. Here, we show that all models show a consistent reduction in ENSO variability due to the incorporation of the Green Sahara, through Atlantic-Pacific teleconnections which are also observed in the present day.

Results

Green Sahara-induced changes in the tropical Pacific

All models in this study show negative anomalies in the boreal winter SST variability (SSTSD:DJF) over regions of the equatorial Pacific due to the Green Saharan changes, estimated through MHGS anomalies relative to the MHREF simulation (Fig. 1). The negative anomalies are statistically significant at the 90% confidence level, assessed using Bartlett’s test for equal variances. For EC-Earth, UofT-CCSM4, and GISS, the negative anomalies are zonally extensive and up to 0.1 °C, 0.2 °C, and 0.4 °C in magnitude, respectively. For iCESM and HadCM3, the negative anomalies occur largely in the eastern-central region and up to 0.2 °C in magnitude. Notably, all models show reduced boreal winter SST variability in the MHGS simulations relative to the PI over nearly all the Niño index regions (Supplementary Fig. S3). The exception to this is GISS, which shows an increase of 0.02 °C over the Niño3.4 region. This small and exceptional increase in SST variability in the Niño3.4 region is not due to the Green Sahara, but instead arises because of the exceptional insolation-driven increase shown by GISS. This increase is unique to GISS, with all other models showing an orbitally driven decrease in SST variability in the Niño3.4 region. The large insolation-driven increase of about 0.4 °C is not entirely offset by the Green Sahara-driven decrease in SST variability. Thus, all models consistently show a reductive effect of the Green Sahara on equatorial Pacific SST variability, though this reduction is weaker in the case of HadCM3 (Fig. 1E) due to substantial warm bias and inadequate representation of changes in the tropical Atlantic (discussed further below).

Fig. 1: Impact of the Green Saharan changes on the tropical Pacific during boreal winter.
Fig. 1: Impact of the Green Saharan changes on the tropical Pacific during boreal winter.
Full size image

MHGS anomalies relative to MHREF during boreal winter for A EC-Earth v3.1, B iCESM v1.2, C UofT-CCSM4, D GISS-E2.1-G, and E HadCM3. Left: Anomalies in interannual SST variability (°C). Gray boxes indicate the Niño index regions Niño4, Niño3, and Niño1 + 2. Black box index indicates Niño3.4 region. Right: Anomalies in precipitation (mm/day) and surface wind strength (m/s). The arrow keys on the top right indicate wind strength of 1 m/s. Only the anomalies statistically significant at the 90% confidence level are shaded.

Our simulations also suggest that the equatorial Pacific was characterized by a La Niña-like mean state during the MH (Fig. 2). The models show negative boreal winter SST anomalies over the equatorial Pacific of up to 1.2 °C. The majority of the models (except GISS) show a cooling over the present-day Southern Pacific Convergence Zone (SPCZ), accompanied by a warming to its southwest. Zonal SST gradients, which are typically higher during La Niña events than during El Niño events, were calculated using three indices. The West Pacific Gradient increases from PI through MHREF to MHGS for four out of five models (Supplementary Table S2). The exception to this increase is in the case of GISS, which shows an increase in zonal SST gradient from PI to MHREF, but the lowest gradient occurs in the MHGS simulation. Other indices for the zonal SST gradient, such as the Trans-Niño index and the Eastern Pacific Gradient, did not show consistent trends among models. The La Niña-like mean state is also reflected in precipitation and wind anomalies. During boreal winter, the equatorial Pacific was characterized by negative precipitation anomalies of up to 4 mm/day. Surface wind strength anomalies indicate anomalous easterlies in the western and central equatorial Pacific. The Green Sahara alone induced extensive drying of up to 2 mm/day and easterly wind anomalies over the western and central equatorial Pacific (Fig. 1). The present-day SPCZ region witnessed drying, with GISS showing minimal drying and HadCM3 showing maximum values of more than 4 mm/day. Combined with the SST anomalies, the precipitation anomalies suggest a southwesterly shift in the SPCZ. The Equatorial Southern Oscillation Index, a measure of large-scale surface pressure changes which typically has positive values during La Niña events and negative values during El Niño events, increases consistently for all five models going from PI to MHREF to MHGS simulations (Supplementary Table S2). Lastly, anomalies in the mean annual zonal mass streamfunction reflect changes to the large-scale atmospheric Pacific Walker Circulation (PWC, Fig. 3). There is a consistent intensification of 2–18% in the PWC intensity in response to the orbital changes from PI to the MHREF (Fig. 3A–E). In the MHGS simulations, the PWC intensity increases by 19–47% relative to the PI (Fig. 3F–J). EC-Earth and HadCM3 show a westward expansion of 4° of the PWC in response to the orbital changes, with the other models showing no clear change (Supplementary Table S2). On the other hand, all models show a westward expansion of 1–6° in the MHGS simulations. The models do not show a consistent westward or eastward shift in the position of the PWC centre in response to the orbital changes or the incorporation of the Green Sahara. In summary, our simulations indicate three key mean state changes during the MH: a La Niña-like cooling and drying over the equatorial Pacific, a southwesterly shift of the SPCZ, and an intensification and westward expansion of the Pacific Walker Circulation.

Fig. 2: Mean state changes in the tropical Pacific during boreal winter.
Fig. 2: Mean state changes in the tropical Pacific during boreal winter.
Full size image

MHGS anomalies relative to PI during boreal winter for A EC-Earth v3.1, B iCESM v1.2, C UofT-CCSM4, D GISS-E2.1-G, and E HadCM3. Left: Anomalies in SST (°C) and sea level pressure (hPa). Right: Anomalies in precipitation (mm/day) and surface wind strength (m/s). The arrow keys on the top right indicate wind strength of 1 m/s. Only the anomalies statistically significant at the 90% confidence level are shaded.

Fig. 3: Changes to the Pacific Walker circulation.
Fig. 3: Changes to the Pacific Walker circulation.
Full size image

Anomalies in mean annual zonal mass streamfunction (in 1010 kg/s) for MHREF simulation relative to the PI (A–E), MHGS simulation relative to the PI (F–J), and MHGS simulation relative to the MHREF (K–O). Contours represent values (in 1011 kg/s) for the PI simulations (A–J) and MHREF simulations (K–O). Numerical values to the bottom left represent percentage change in intensity of the Pacific Walker Circulation relative to the PI simulation (A–J) and MHREF simulation (K–O).

Green Sahara-induced changes in the tropical Atlantic

The models in this study show several biases that limit their ability to capture mean state and interannual variability over the tropical Atlantic Ocean, especially during boreal summer. Comparisons of the PI simulations with the HadISST and NOAA OI SST observational datasets reveal substantial warm biases in the tropical Atlantic upwelling regions such as the equatorial Atlantic, southern tropical Atlantic (Benguela region), and the northwestern African coast (off Western Sahara, Mauritania, and Senegal) (Supplementary Fig. S4). With the exception of EC-Earth, which does not show a warm bias off the northwestern coast, these warm biases are present in all models in the aforementioned regions and can exceed 6 °C in magnitude in specific regions. The warm bias is particularly noteworthy in the case of HadCM3, which also shows a large westward extension of the Atlantic Cold Tongue. The models also cannot adequately represent the zonal SST gradients in the equatorial Atlantic (Supplementary Fig. S5), with only UofT-CCSM4 and GISS showing a reasonable (but muted) seasonal cycle. Interannual variability is reasonably represented only in EC-Earth and iCESM, which show peaks in SST variability over the equatorial Atlantic during boreal summer and fall (Supplementary Fig. S7). Model biases are notably larger in the tropical Atlantic compared to the tropical Pacific (Supplementary Note S1).

In spite of substantial biases, the models show several consistent effects of the Green Saharan changes on the tropical Atlantic mean state, estimated through MHGS-MHREF anomalies (Fig. 4). During the MH, the intensified WAM strengthened southwesterly winds during boreal summer by 2–3 m/s (Supplementary Fig. S2). The Green Sahara further amplifies the WAM-related southwesterly wind anomalies by 2–3 m/s, leading to substantial westerly wind anomalies over the tropical Atlantic during boreal spring and summer (Fig. 4 and Supplementary Fig. S2). The westerly wind anomalies inhibit upwelling along the western coast of Africa and lead to positive SST anomalies of up to 1–1.5 °C over the tropical Atlantic in different models (Fig. 4). In the MHGS simulations, the Green Sahara-driven positive anomalies are masked by seasonal cooling driven by insolation and greenhouse gas changes (Supplementary Fig. S6). Thus, the models do not consistently show warmer tropical Atlantic SSTs during the MH relative to the PI. However, all models consistently show the warming effect of the Green Sahara off northwestern Africa, over the north tropical Atlantic, in the equatorial Atlantic, and the Benguela region, with the strongest signature in EC-Earth and the weakest signature in HadCM3 (Fig. 4). Two models (EC-Earth and iCESM) also show a reduction in the zonal SST gradient during boreal summer from PI through MHREF to MHGS, consistent with Atlantic Niño-like conditions (Supplementary Fig. S5). The other models show no clear changes in the SST gradient between the three simulations, with the climate signal likely being overshadowed by the SST biases.

Fig. 4: Impact of the Green Saharan changes on the tropical Atlantic during boreal spring and summer.
Fig. 4: Impact of the Green Saharan changes on the tropical Atlantic during boreal spring and summer.
Full size image

MHGS anomalies relative to MHREF for A EC-Earth v3.1, B iCESM v1.2, C UofT-CCSM4, D GISS-E2.1-G, and E HadCM3. Top: Filled contours indicate SST anomalies (°C) during February-April; overlying vectors indicate surface wind strength anomalies during March-May. Bottom: Filled contours indicate SST anomalies (°C), and overlying vectors indicate surface wind strength (m/s) during June-August. Arrow keys on the top right indicate wind strength of 1 m/s. Only the anomalies statistically significant at the 90% confidence level are shaded.

Model biases preclude identifying a consistent signature in interannual variability over the equatorial Atlantic. EC-Earth and iCESM show reductions in SST variability during boreal summer and fall, while UofT-CCSM4 shows a small increase in June-July (Supplementary Fig. S7). GISS and HadCM3 do not show the greatest interannual variability during boreal summer and fall, but instead show it during boreal spring (Supplementary Fig. S7d, e), making it difficult to associate the corresponding changes in variability with Atlantic Niño changes.

Discussion

The Green Sahara’s impact on the tropical Pacific emanated from a strengthening of the West African Monsoon. The resultant changes in regional atmospheric circulation drove various changes over the tropical Atlantic, impacting tropical Pacific variability through two key mechanisms, which also govern the Atlantic-Pacific teleconnection in the present day35. First, warming over the North Atlantic during boreal spring generates low-level anticyclonic circulation over the western North Pacific during boreal summer. This generates easterly wind anomalies over the equatorial Pacific, which can be reinforced by positive feedbacks to trigger La Niña conditions during the subsequent boreal winter36. Second, a relaxation in the trade winds and trade wind-related upwelling induces Atlantic Niño conditions during boreal summer, which further leads to the development of La Niña conditions during boreal winter due to a westward expansion and intensification of the PWC37,38. Atlantic Niño events are often associated with Benguela Niño events, both of which can be triggered by equatorial Atlantic wind anomalies39.

Our results suggest that both the above-mentioned mechanisms were operative during the MH (Fig. 5). Notably, the WAM strength during the mid-Holocene was exceptional in the context of the instrumental record, with mean annual precipitation increases of over 2 mm/day over large parts of northern Africa (estimated through MHGS anomalies relative to the PI; Supplementary Fig. S2). MHGS anomalies relative to MHREF indicate that the Green Sahara alone led to mean annual precipitation increases in the order of 0.5–2 mm/day over northern Africa. Further, the Green Sahara induced warming of up to 1.5 °C over the northern tropical Atlantic during boreal spring (Fig. 4), and anticyclonic circulation over the northern Pacific with ensuing easterly wind over the equatorial Pacific during boreal summer (Fig. 2). With regards to the Atlantic Niño mechanism, the Green Sahara led to the development of westerly wind anomalies in the equatorial Atlantic during boreal summer, which induced SST anomalies in the Benguela region and the equatorial Atlantic in the order of 0.25–1.5 °C in the different models (Fig. 4). The westward expansion and intensification of the PWC (Fig. 3) reinforced easterly wind anomalies over the equatorial Pacific during boreal summer, which led to La Niña conditions during boreal winter. The La Niña-like mean state is evidenced by negative precipitation and SST anomalies and easterly wind anomalies in the equatorial Pacific, a southwestward shift in the SPCZ, and increases in the Equatorial Southern Oscillation Index (Fig. 2 and Supplementary Table S2).

Fig. 5: Impact of the Green Saharan changes.
Fig. 5: Impact of the Green Saharan changes.
Full size image

Colors are based on multi-model mean MHGS anomalies relative to the MHREF simulation for (1) mean annual precipitation (green), (2) SST (orange), (3) zonal mass streamfunction (pink), and (4) boreal winter SST variability (blue).

The relationship between ENSO mean state and variability may not be a stationary feature of the ENSO system, as paleoclimatic evidence reveals inconsistent linkages between zonal SST gradient and ENSO variability across different climate periods. For example, reduced ENSO variability coincides with stronger zonal gradients during the Last Millennium40 and the MH18, in agreement with our findings. In contrast, lower ENSO variability is associated with weaker zonal gradients during the Last Glacial Maximum41 and Pliocene42, underscoring the non-uniform nature of this relationship. These contrasts are partly sensitive to the approaches used in different studies of the same climate periods. While some proxy records imply stronger MH gradients paired with suppressed ENSO activity, the PMIP3/4 model ensemble fails to replicate this connection, highlighting sensitivities to data sources and metrics. Our results indicate that the reduction in ENSO variability was associated with a La Niña-like mean state, which emerges more distinctly through basin-wide oceanic and atmospheric circulation shifts than via zonal SST gradients alone. Previous results31 with the model EC-Earth suggested that persistent thermocline deepening in the eastern Pacific, driven by boreal summer westerly wind anomalies and winter eastward Kelvin waves, weakens the thermocline feedback. This process dampens the Bjerknes positive feedback, which is essential for El Niño development. Intriguingly, thermocline-driven ENSO suppression also explains variability reductions during the Last Glacial Maximum41, even under weaker zonal SST gradients, bridging apparent contradictions between climate states. Future work is planned to explore inter-model consistency, focusing on thermocline adjustments and the Bjerknes feedback’s sensitivity to mean-state changes.31,40,41,42.

While all models exhibit a reduction in ENSO variability upon the incorporation of Green Saharan changes, the magnitudes and spatial extents differ in the models. Differences among model responses arise from three causes: simulation protocols, climate response in models, and model biases. Firstly, the MHGS simulations analyzed here were not a part of a coordinated modeling experiment. Thus, they employ slightly different schemes for prescribing vegetation changes, though each scheme is qualitatively consistent with proxy reconstructions (see Methods and Supplementary Table S1 for further details). The vegetation changes are further treated differently within each model. For instance, albedo over northern Africa reduces from 0.3 (typical of desert regions, similar in all models) to 0.19 in GISS, whereas it reduces to 0.15 in EC-Earth, leading to differences in the surface radiative balance and simulated surface temperatures. In addition to vegetation changes, simulations from different models further incorporate different changes like dust reduction (EC-Earth, iCESM) and soil and lake cover changes (UofT-CCSM4). Secondly, the climatic response to the Green Saharan changes differs over northern Africa (in terms of WAM strength; Supplementary Fig. S2) as well as over the tropical Atlantic (Fig. 3). However, our results indicate that the key change for transmitting the regional response to the tropical Pacific is in generating westerly wind anomalies in the tropical Atlantic. These westerly wind anomalies were responsible for inhibiting upwelling along the African coast, leading to positive SST anomalies over the tropical Atlantic. These SST anomalies modulate the PWC by driving wind anomalies over the equatorial Pacific through a Gill-Matsuno response43. EC-Earth, iCESM, and UofT-CCSM4 simulate a stronger intensification of the WAM (Supplementary Fig. S2) and show greater tropical Atlantic SST anomalies (Fig. 4), along with a more reasonable simulation of equatorial Atlantic variability (Supplementary Fig. S7). This may explain why they simulate a greater reduction in ENSO variability compared to HadCM3. On the other hand, the strong reduction shown by GISS is most likely linked to its unusually high variability in the MHREF simulations. Further, these reductions are manifested in different parts of the equatorial Pacific in different models. For EC-Earth, the Green Sahara leads to the greatest reduction in SST variability in the western-central equatorial Pacific with a 31% decrease relative to the PI over the Niño4 region. For the other models, the greatest reductions in SST variability occur in the central-eastern equatorial Pacific with decreases of up to 17% relative to the PI over the Niño3 region (Supplementary Fig. S3). Thirdly, a consistent reduction in SST variability is not observed over the tropical Atlantic with respect to the Atlantic Niño changes (Supplementary Fig. S7), due to large model biases in SST and SST variability (Supplementary Figs. S4, S5, S7). We conclude that the ability of a climate model to simulate changes to ENSO variability depends not only on an adequate representation of the Green Sahara (particularly the representation of changes to low-level winds over the tropical Atlantic), but also on model SST biases in the tropical Atlantic. This conclusion is supported by studies which show that ENSO forecasts are improved by improving model skill in the Atlantic44,45.

The evolution of ENSO variability through the Holocene has received widespread attention from the paleoclimate modeling community. Previous studies have largely focused on delineating orbitally-driven changes in ENSO variability in the context of ENSO internal variability and have proposed various mechanisms to account for reductions in ENSO variability, such as changes to annual cycle or seasonality19,20,25, PWC46, strength of ENSO feedbacks22,47, and ENSO diversity32. A previous study based on 29 PMIP2/3 models identified a 5% average intensification and a 3–4° westward shift in the PWC in the MH relative to the PI and concluded that the enhancement of the Asian and North African monsoons played a greater role in modulating the PWC than the Pacific mean state46. However, we note that PMIP2/3 models substantially under-estimated the WAM strengthening and the development of the Green Sahara during the MH, even with the use of dynamic vegetation modules48,49. More recently, another study of the evolution of ENSO during the middle-late Holocene based on the CESM v1.2 model incorporated Green Saharan changes in their experimental setup50. Lower Niño3.4 variance was simulated during the time slices with Green Saharan changes (9ka and 6ka) compared to those without (3ka and 0ka), but the design of the study prevents isolating the impact of the Green Sahara at any particular time slice.

In particular, the first study to compare the MH ENSO with and without prescribed vegetation and dust changes over northern Africa showed that the Green Sahara was critical in modulating the Atlantic mean state and variability and driving a reduction in ENSO variability31. This study was based on a single model, EC-Earth v3.1 (also included in this study)31. Here, we show that the reduction in ENSO variability due to the Green Sahara is a consistent result across five climate models, and is thus a robust conclusion, which does not dependent on model physics and parameterizations31. We further demonstrate that while the specific vegetation and land surface changes incorporated to simulate the Green Sahara play modulate regional climate, it is the large-scale atmospheric changes that govern the Atlantic-Pacific teleconnection. In addition to the Atlantic Niño-mechanism shown in the previous study31, we identify the warming over the North Tropical Atlantic as an additional mechanism for triggering La Niña conditions and identify the accompanying south-westward shift in the SPCZ. Our results do not discount the role of the orbitally driven changes to seasonality or the Walker Circulation, but instead suggest that they were amplified by the Green Saharan changes. Further, our findings may have implications for explaining changes in ENSO diversity, since the two mechanisms identified here favor different impacts on different regions of the equatorial Pacific. Further work is necessary to identify the varying importance of the different mechanisms during different phases of the Green Sahara Period and to understand the coherence between simulated and reconstructed changes to ENSO diversity. Several proxy reconstructions from the tropical Pacific have indicated a La Nina-like mean state and reduced ENSO variability during the early and middle Holocene relative to the present day. For instance, SST and SST variability reconstructions based on lake sediments13,51 from Galapagos Islands and planktonic foraminifera in marine sediments14,52 near the Galapagos Islands reflect cool and dry conditions, reduced variability, and an enhanced zonal SST gradient. The planktonic foraminiferal δ18O record from core V21-3014 suggested that ENSO variance may have been reduced by up to 50% relative to the present day. These conclusions align with reduced SST53 and precipitation variance (48% reduction before 4.4 ka)54 shown by marine records from the northeastern tropical Pacific. Similarly, a fossil mollusk record from seven sites along the Peruvian coast indicated reductions of up to 55% (around 4.7 ka)55. Records are relatively sparse in the central and western equatorial Pacific compared to the eastern equatorial Pacific. Coral11 and marine sediment12 records from or near the Line Islands in the central equatorial Pacific indicated a reduction in ENSO variance (up to 77% around 4.2 ka11). In the western tropical Pacific, a coral record from Papua New Guinea16 suggests a reduction in ENSO variance of up to 85% during 7.6–7.1 ka, and speleothem17 and lake sediment18 records from teleconnected regions of Malaysian Borneo and eastern Australia, respectively, also suggest reduced ENSO variance during the middle Holocene. Proxy-model comparisons have shown that climate models tend to under-estimate the ENSO variability reduction during the early and middle Holocene20,24,25. Since the evolution of the orbital forcing through the Holocene is widely considered to be the primary driver of ENSO variance reductions, the model underestimations are viewed as underestimations of ENSO sensitivity to orbital forcing in the climate models25. In our study, the maximum reduction observed in the MHGS simulations relative to the PI simulations is around 35% and 30% in the case of UofT-CCSM4 and EC-Earth, respectively (Supplementary Fig. S3). While this remains less than the proxy estimates, the estimates from MHGS simulations show consistent decreases over the estimates from MHREF simulations, suggesting that the inclusion of northern African vegetation changes during the Holocene may partly reconcile proxy-model differences.

Another aspect of proxy-model discrepancy pertains to the timing of the maximum variance reduction. While some records indicate less ENSO variability during the early Holocene12,52, most indicate that the lowest variability occurred during the transition from the middle to late Holocene, with variability increasing during the late Holocene11,13,14,17,54,55. On the other hand, climate models tend to show a monotonic intensification of ENSO variability from the early to the late Holocene, as seen through transient simulations25 or snapshots of equilibrium simulations18,20. The observation that proxy reconstructions do not widely show such a monotonic increase in accordance with the evolution of the orbital forcing also indicates the relevance of another factor. Notably, the “Holocene ENSO Minimum,” which is considered to have occurred between 3 and 6 ka25 appears to be coincident with the decline of the Green Sahara56. We note that the WAM intensity remained much stronger than the present day for several millennia following Saharan desertification (6–5 ka), with records south of 15 N showing the termination of the Green Sahara Period between 4 and 2.5 ka56. An interesting question that arises here is: why would the Holocene ENSO Minimum coincide with the termination, and not with the peak, of the Green Sahara Period, which occurred during the early and middle Holocene (9–6 ka)? We hypothesize that during the middle Holocene, the extreme northward movement of the Intertropical Convergence Zone could have shifted the required wind anomalies too far north from the equatorial Atlantic region, leading to conditions less conducive for the development of Atlantic Niño events than compared to the onset and decline of the Green Sahara Period. This mechanism may also help explain reduced ENSO variability during the early Holocene12,52; however, identifying the impact of the Green Sahara during the early Holocene is complicated by the influence of the ice sheets and the difference in global sea levels. Further work focusing on understanding the atmospheric circulation and its effects on the Atlantic Niño variability through the Holocene is necessary to verify this. However, the coincident Holocene ENSO Minimum and Green Sahara Period termination, an emerging signal of proxy-model discrepancy regarding ENSO variability20,24,25 and the role of the Atlantic in modulating ENSO36,37,44,57 suggest that future modeling studies of Holocene ENSO should ensure adequate representation of north African vegetation and land surface changes. Our work adds to a growing body of literature that identifies the remote impacts of the MH Green Saharan changes, which do not emerge as clearly in simulations that account only for insolation and greenhouse gas changes during the Holocene26,27,28,29,30,58,59,60.

Conclusion

This study is the first multi-model intercomparison investigating the changes in ENSO resulting from vegetation changes over northern Africa during the Holocene. Results from five global climate models suggest that the Green Sahara led to a La Niña-like mean state and a reduction in ENSO variability due to an amplification of orbitally driven strengthening of the WAM. This reduction is independent of model physics or the specific vegetation and land surface changes incorporated to simulate the Green Sahara. Instead, it depends on the simulated WAM intensity and tropical Atlantic mean state, which leads to a westward expansion and intensification of the Pacific Walker Circulation. The ENSO reduction is constrained by the ability of climate models to adequately simulate tropical Atlantic SST variability. Thus, reducing model biases in the tropical Atlantic may be key to simulating the Green Sahara–ENSO teleconnection, as well as in ensuring a reasonable representation of the Atlantic-Pacific teleconnection in future climate model projections and ENSO predictions. Due to the extreme strengthening of the WAM, which is unmatched in the instrumental record, the Green Sahara serves as a paleoclimatic test case for WAM-related Atlantic-Pacific teleconnections. Given emerging evidence of various remote impacts of the Green Sahara, it is necessary to ensure its adequate representation in paleoclimatic simulations of the relevant periods. This work further reinforces the importance of vegetation and land surface feedbacks and the role of Atlantic model biases in ENSO studies.

Methods

Climate model experiments

Climate models and simulations: We used five fully coupled global climate models for this study: EC Earth version 3.161, isotope-enabled Community Earth System Model (iCESM) version 1.262, University of Toronto version of Community Climate System (UofT-CCSM463), water-isotope enabled Goddard Institute of Space Studies (GISS) model E2.1-G64 and Hadley Centre coupled model version 3 (HadCM365). Details about the model components and resolutions are provided in Supplementary Table S1. For each model, we analysed three simulations: (i) pre-industrial (PI), (ii) mid-Holocene with changes to insolation and greenhouse gas concentrations (MHREF) and (iii) mid-Holocene with changes to insolation, greenhouse gas concentrations, vegetation dust, or land cover (MHGS). For EC Earth, changes to insolation and greenhouse gas concentrations followed PMIP3 recommendations66. For all other models, PMIP4 recommendations were followed for the same7.

Incorporating the Green Sahara: The MHGS simulations were carried out differently for each model, with the central idea of adequately representing the vegetation and land surface changes due to the mid-Holocene Green Sahara. For EC Earth, the vegetation type over the Saharan region (11–33°N; 15°W–35°E) was set to shrub, and the dust amount was set to 80% of the PI dust67. For iCESM, land conditions at 11°N were extended to the Sahara and Arabian peninsula, and Saharan dust was reduced through the use of a prognostic dust module26. For UofT-CCSM4, tropical vegetation was extended northwards, and the Saharan region was replaced by a mix of shrubs, steppe, and savanna, with different distributions across 25°N. In addition to vegetation changes, soil albedo was also reduced, and the presence of five mid-Holocene megalakes was incorporated in the MHGS simulation68. For GISS, arid shrub and grassland were prescribed over the Saharan region34. For HadCM3, a 90% grass advance was prescribed over the Sahara. These vegetation changes are in accordance with proxy reconstructions9 and broadly align with PMIP4 recommendations for vegetation sensitivity experiments7. The different MHGS simulations represent different pathways of achieving a comparable strengthening of the West African Monsoon as indicated by proxy records for the mid-Holocene Green Sahara.

Climate data analysis

ENSO variability: ENSO variability is estimated through the standard deviation of boreal winter SSTs, and is considered to represent ENSO amplitude69. No band-pass filtering was carried out.

Mean state changes: The Pacific Walker Circulation (PWC) was analysed using the annual mean zonal mass streamfunction over the equatorial Pacific (5°S–5°N). The PWC intensity was estimated using the vertical mean of the annual mean zonal mass streamfunction between 120°E and 270°E. The PWC westbound was estimated using the vertical median of the westernmost longitude of zonal mass streamfunction value of 0 kg/s. The PWC center was estimated using the longitude of the maximum zonal mass streamfunction value. The Equatorial Southern Oscillation Index was used to quantify the zonal gradient in sea level pressure, defined as the difference between the mean sea level pressure over Indonesia (5°S–5°N, 90–140°E) and the eastern Pacific (5°S–5°N, 130°W–80°W).

Statistical significance: Statistical significance for mean state changes was tested using the two-sided Welch’s t test for independent samples that does not assume equal variances. For variance changes, we used Bartlett’s test for equal variances. Only results significant at the 90% confidence level are shown.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.