Introduction

The Asian summer monsoon (ASM) is a seasonal response of the coupled land-ocean-atmospheric system to the thermal contrast between the Eurasian landmass and the Indo-Pacific Ocean in the presence of the elevated Tibetan Plateau (TP)1. The unique topography and geographical distributions of the region support the seasonal transition of the prevailing winds2 and considerable enhancement of rainfall3, which influence more than 60% of the world’s population4. On geological timescales, the intensity of the monsoonal climate in Asia has varied considerably since the early Cretaceous [~145 million years ago (Ma)]5 with the suggested inception of the ASM in the early Miocene6 or late Oligocene7 period, ~ 25–22 Ma. However, the initiation of the ASM has been a subject of debate as many proxy observations8,9,10,11 indicated the presence of a monsoonal climate over China and Myanmar in the Eocene ( ~56–34 Ma). Recent studies have re-evaluated many previously identified basins exhibiting monsoon conditions in the Miocene, pushing them back into the Oligocene due to advanced dating techniques. Consequently, the timing of the monsoon initiation and our understanding of its driving factors, particularly in warm climates of the past and its anticipated future response, present a great challenge to the geoscience community.

The collision between the Indian and Eurasian tectonic plates is key to the complexity of the modern ASM initiation. Previous numerical modeling studies examined the relationship between ASM initiation and late Eocene tectonic events, such as the retreat of the Paratethys Sea12,13, the uplift of the TP14,15, and the rise of the Himalayas16. Although large uncertainties exist in the timing of these events, including the complexity of TP uplift17 into its current configuration18,19, these factors altered the dynamic-thermodynamic conditions of the Asian climate and shaped the favorable ocean-atmospheric circulations for the modern ASM20). On the other hand, the role of elevated atmospheric CO2 concentration in the Eocene epoch was advocated for intensifying the ASM, while the effect of TP was found to be minimal on the ASM rainfall21,22. The early Eocene (~56–48 Ma) climate was about 9°-15°C warmer than the preindustrial period23 and atmospheric CO2 levels were estimated to be between 1200 and 2500 parts per million by volume (ppmv)24,25. The origin of ASM in the Eocene was also supported by sedimentary 26 and paleobotanical8 evidence across China. These large uncertainties in the origin of ASM and its strength during the Eocene prompted further investigations into the early Eocene ASM using state-of-the-art climate model experiments and comprehensive datasets.

Earlier numerical paleoclimate modeling attempts12,14 were mostly based on a single atmosphere-only model with prescribed sea surface temperature (SST) as the lower boundary conditions. Kitoh27 showed that coupled ocean-atmosphere models simulate a stronger East Asian monsoon due to mountain uplift than its atmosphere-only counterpart. As air-sea interaction plays a crucial role in ASM28, ocean-atmosphere coupled climate models are required for realistic ASM simulations. Moreover, the sensitivity experiments in multiple climate model frameworks with consistent boundary conditions can effectively reduce the model uncertainties relative to a single numerical model. A previous community effort with four climate models, known as the Eocene Model Intercomparison Project (EoMIP)29 examined the hydroclimate changes due to high atmospheric CO2 concentration and other boundary conditions in the early Eocene epoch and demonstrated robust paleoenvironmental conditions with lower intermodel uncertainty compared to earlier single-model-based studies. However, it is worth noting that the EoMIP models employed a non-uniform experimental framework. Participating models utilized varied paleogeographic settings and atmospheric CO2 levels, limiting the potential for a direct intermodel comparison29.

The recent Deep Time Model Intercomparison Project (DeepMIP)30 provides an opportunity to explore early Eocene paleoclimatic conditions. This paleoclimate modeling framework uses a standard set of boundary conditions and experimental design in a group of Coupled Model Intercomparison Project (CMIP) Phase 3-6 class models31. Using the DeepMIP-Eocene ensemble simulations, Williams et al.32 suggested an increase in West African monsoon rainfall associated with a southward displacement of the Atlantic intertropical convergence zone (ITCZ) at higher CO2 levels in the early Eocene. Seasonal northward progression of the ITCZ over the ASM region is intimately associated with the rainfall variation in South Asia33, therefore, any change in the seasonal position of the ITCZ during the early Eocene is likely to affect the ASM rainfall variability. The present northward migration of the ITCZ over the ASM region is constrained by the geographical land-ocean distribution and TP elevation, which had a distinctly different orientation during the early Eocene (Fig. S1). Considering the projected end-of-century atmospheric pCO2 levels ranging from 393 to 1135 ppmv for the lowest and highest emission scenarios34, and CMIP projections indicating an amplification of ASM rainfall by the end of the 21st century due to the increase in atmospheric greenhouse gas (GHG) increase35,36, examining paleoclimate evidence with elevated atmospheric CO2 levels can offer an alternative perspective for understanding future changes in the ASM37,38.

The ASM is a convectively coupled system, where winds exhibit a Matsuno-Gill response to heating associated with monsoon rainfall39. The seasonality of the ASM rainfall is regulated by the northward migration of the ITCZ3,40,41. The continental ITCZ around 25°N is sustained during boreal summer by a repeated genesis and northward propagation of the maximum cloud band from its oceanic location around 5°S in the intra-seasonal time scale42,43,44. In this study, we examine whether the ASM had begun in the early Eocene warm climate, where the rainfall seasonality and extent of northward excursion of the ITCZ were likely to be different relative to present climate in response to the change in land-ocean contrast, orography, and seasonal cycle of SST distributions. We use five DeepMIP-Eocene simulations and compare the results relative to the preindustrial control (piControl) simulations representing the present climate. The changes in the early Eocene climate can largely be categorized by their underlying drivers - non-CO2 (e.g., paleogeography, continental ice-sheet, orbital configuration) conditions with preindustrial atmospheric CO2 level (~280 ppm, 1xCO2) and enhanced CO2 forcing (3xCO2), which may provide insight into future climate scenarios with enhanced GHG concentrations.

In the context of future ASM changes45,46, despite the uncertainty in the DeepMIP models and paleoclimate proxy records21, the early Eocene monsoon simulations in warm and high GHG conditions are likely to provide a baseline for estimating possible changes in ASM in the next century. While low-resolution climate models used in paleoclimate modeling exhibit biases in simulations47, the incorporation of a multimodel ensemble mean (MMM) helps mitigate individual model biases48,49, contributing to reduced uncertainty in paleoclimate simulations. As demonstrated by Williams et al32, the MMM approach employed in the DeepMIP-Eocene proves superior to a single-model strategy in enhancing model reliability and skills. In this study, we use the five model-based MMM approach following an examination of the individual piControl simulations over the tropics and the greater ASM domain (Fig. S2, see Methods section for details). This study aims to establish a reliable early Eocene ensemble simulation, fostering increased confidence in understanding the characteristics of the Asian wet season during the early Eocene compared to the present ASM. Analyzing changes in ASM seasonal rainfall distribution, its seasonal cycle, and associated circulation alterations within the DeepMIP ensemble simulation can advance our comprehension of ASM dynamics, its initiation, and evolution from a paleoclimate perspective.

Results

Northward migration of ITCZ

The ASM is characterized by enhanced rainfall activity over the Indian subcontinent and southeast Asia from late May to September-early October with a strong low-level ( ~ 850 hPa) westerly jet, known as the low-level westerly jet (LLJ), over the Arabian Sea50. These features are linked to the seasonal northward migration of ITCZ over the ASM region. Thus, we first examine the mean annual precipitation and 850-hPa zonal wind cycles over the ASM domain (60°E-95°E) as a function of latitude in Fig. 1. Over this sector, the observed and piControl MMM rainband migrates from its mean position of ~ 5°S in southern summer to about 25°N in boreal summer and remains north of 15°N throughout the season51. This seasonal rainfall change is also accompanied by the onset of strong low-level (850-hPa) westerlies over the ASM region. Individual piControl simulations are also capable of producing the observed seasonal northward migration of the rainband and the zonal wind transition (Fig. S3). In contrast, both early Eocene simulations produce modest seasonal movements of the ITCZ in the boreal summer and the rainfall maxima are confined within 10° latitudes. Maximum convective activity occurs around the equatorial region throughout the year, while the simulated early Eocene rainband appears to be stronger than piControl.

Fig. 1: Seasonal transition of rainband and zonal wind over South Asia.
figure 1

Time-latitude sections of the monthly climatological rainfall (shaded, mm/day) and 850-hPa zonal wind (contour, m/s) averaged over 60∘E-95∘E in a observation (GPCP rainfall and NCEP reanalysis II winds), MMM of b piControl, and early Eocene c 1xCO2, d 3xCO2 simulations. The kinetic energy per unit mass (m2/s2) of 850-hPa climatological winds averaged over 50∘E-65∘E, 5∘N-15∘N are shown in red curves with a scale on the right y-axis.

Full size image

Similar to the rainfall change, the seasonal onset of 850-hPa westerlies over the northern latitudes is confined to the equatorial region, and a persistent low-level easterly regime is noted to the north of 15°N in both early Eocene simulations (Fig. 1c, d). This marginal northward extension of the rainband and the low-level zonal wind are indicative of a weaker ASM in the early Eocene MMM simulations. The strength of ASM is estimated by the mean kinetic energy of the LLJ averaged over 50°E-65°E, 5°N-15°N52. The early Eocene simulations produce notably weaker, about one-quarter of the piControl kinetic energy. It is even weaker in the 3xCO2 early Eocene simulation, perhaps due to stronger seasonal rainfall activity to the south of 10°N.

Seasonal mean state

In Fig. 2, the boreal summer (June-September, JJAS) mean rainfall and 850-hPa winds in the MMM simulations are evaluated relative to the observations and piControl. The observed seasonal rainfall enhancement over the ASM region and the rainfall maxima along the west coast of India, the eastern shoreline of the Bay of Bengal, and over the islands of the Philippines are largely captured in piControl (Fig. 2b). The land rainfall is slightly weaker than observed but this underestimation does vary across the models. Each model successfully represents most of the spatial elements of the ASM rainfall distribution (Fig. S4). A realistic simulation of these salient features is a prerequisite for climate models to produce the ASM. We note that piControl slightly overestimates the ASM rainfall (domain-averaged value is ~0.6 mm/day greater than observed), especially over the ocean and along the foothills of the Himalayas, while it has a dry bias over the Indian landmass and head Bay of Bengal regions. The LLJ is well simulated in piControl with the observed maxima occurring between 10°N and 15°N over the Arabian Sea. Both observed and piControl simulated westerlies weaken over southeast Asia, shifting to easterlies over the west-north Pacific region.

Fig. 2: Boreal summer mean rainfall and low-level winds.
figure 2

JJAS mean climatological rainfall (shaded, mm/day) and 850-hPa winds (vectors, m/s) in a observation (GPCP rainfall and NCEP reanalysis II winds), MMM of b piControl, and early Eocene c 1xCO2, d 3xCO2 simulations. The domain-averaged rainfall values are indicated at the top-right corner of each panel. The forward-rotated rainfall differences in e 1xCO2, and f 3xCO2 simulations are shown relative to piControl. The missing dataset locations due to paleo-rotation are shaded in gray. The significant rainfall difference at a 95% confidence level is marked with stippling.

Full size image

In early Eocene simulations (Fig. 2c, d), boreal summer rainfall change is confined around the equatorial region with maximum rainfall activity over the Oceans, consistent with the limited northward extension of ITCZ in Fig. 1c, d. The westerlies between 10°N and 15°N are considerably weaker; only a narrow band of south-westerlies is evident over the Tethys Ocean between the Indian subcontinent and the Asian landmass. These moisture-laden winds are mechanically uplifted due to orographic forcing from the Gangdese mountain at the southern edge of the Asian landmass53, resulting in heavy rainfall on its windward side and less rainfall over southeast Asia54. Heavy rainfall is also noted in the west of the Indian subcontinent due to its position across the monsoon flow. The enhanced atmospheric CO2 concentration causes intensification of the rainfall, particularly over the ocean (Fig. 2d), although spatial distributions of rainfall and low-level wind distributions remain similar in both the early Eocene simulations. It is indicative of topographical control on the seasonal migration of the ITCZ, while the radiative forcing from enhanced atmospheric CO2 levels plays a secondary role in rainfall intensification, in agreement with the findings of Farnsworth et al5. The enhanced rainfall amount under higher atmospheric CO2 levels, albeit having a similar wind response, suggests the simulated rainfall is predominately driven by the thermodynamic response.

To estimate the rainfall change in the early Eocene simulations, the difference in rainfall relative to piControl is shown in Fig. 2e, f. The changes in geographical locations in the early Eocene are accounted for by rotating the early Eocene simulated datasets forward in time to where it is in the piControl (based on the rotations suggested in Herold et al.55). A strong dry condition is noted over South Asia, which is further intensified in the 3xCO2 simulation. The arid condition over Asian land is significant at a 95% confidence level based on a paired t-test and also consistent with the paleobotanical evidence from China7. It should be noted that the lack of rainfall over South Asia in the early Eocene simulations is much stronger than the dry bias in the piControl simulations. Consistent with Zhang et al.56, the South Asian aridity (Fig. 2e, f) during the Eocene is simulated across the models and it is essentially due to the restricted northward excursion of the ITCZ up to only ~ 10°N. Thus, even under a high GHG scenario, the seasonal northward migration of the ITCZ is constrained without TP-Himalayan orography. In contrast, wetter conditions are simulated over western India, consistent with proxy observations in Shukla et al.57. As this part of the Indian subcontinent was located across the moisture flow in the early Eocene, regional wet conditions are expected.

The JJAS mean surface temperature (Tsfc) difference between early Eocene simulations and piControl is shown in Fig. 3. Surface warming is evident in 1xCO2 simulation with a mean global surface temperature difference of ~ 4°C, while 3xCO2 simulation produces about 11°C warmer Tsfc relative to piControl. The tropical mean SST shows an increase of approximately 1.75°C in the 1xCO2 simulation and 7.5 °C in the 3xCO2 simulation when compared to the piControl (Fig. S5). Notably, the latter value closely corresponds to the earlier estimates of the tropical mean annual SST range during the early Eocene epoch, which was in the range of 30.7°C-37.6°C58. It is worth mentioning that piControl simulates slightly colder SST relative to observations from the recent period (1979-2018, see Fig. S5) with a tropical (30°S–30°N) mean SST difference of ~ 1. 5°C. A part of this difference is likely due to the observed warming trend36 in recent decades and piControl represents a climate with lower atmospheric GHG levels (280 ppmv). However, this temperature discrepancy is smaller than the SST difference between the 3xCO2 simulation and piControl. The maximum Tsfc warming is noted over the regions where continental ice sheets are located in the present-day (e.g., Antarctica). The regional topographic changes likely result in adiabatic warming due to reduced orographic height in the early Eocene simulations. Additionally, ocean-atmospheric circulation and cloud cover change in the tropics lead to enhanced net incoming shortwave radiation absorption (Fig. S6), thereby partially contributing to Tsfc warming. Previous studies59 also suggested that a weaker Antarctic circumpolar current due to a narrower deep-water gateway in the Southern Ocean contributed to global Tsfc warming before the Eocene-Oligocene boundary.

Fig. 3: Boreal summer surface temperature change relative to piControl.
figure 3

JJAS mean surface temperature (Tsfc, in ∘C) difference in MMM a 1xCO2, and b 3xCO2 simulations relative to piControl. The early Eocene Tsfc is rotated forward to align with piControl. The global mean Tsfc difference relative to piControl during JJAS is indicated at the top-right corner of each panel. The missing dataset locations due to paleo-rotation are shaded in gray. Stippling indicates the regions where all the models agreed on the sign of the MMM difference.

Full size image

Annual rainfall cycle

The warm surface temperature and Asian arid conditions in the early Eocene simulations may influence the seasonal rainfall and the length of the rainy season relative to the present ASM regions. To examine the seasonal rainfall evolution over the Asian land, we compute the annual rainfall cycle over three subregions of ASM (boxes shown in Fig. S1): the Indian summer monsoon (ISM) in the subcontinent (70°E-90°E; 5°N-30°N); the southeast Asian summer monsoon (SESM, 90°E–110°E; 7.5°N–30°N), and the east Asian summer monsoon (EASM, 110°E–140°E; 10°N–40°N) land regions only (Fig. 4). The uncertainty in the simulated annual rainfall cycles is addressed by the intermodel rainfall standard deviation (shown in shading around the MMM annual rainfall cycle). As the Indian subcontinent was not part of the Asian mainland but located in the equatorial region in the early Eocene, the simulated subcontinental rainfall cycle in the early Eocene is also computed over 60°E-80°E, 15°S-10°N.

Fig. 4: Annual cycle of rainfall over ASM subregions.
figure 4

Annual cycle of monthly climatological rainfall (mm/day) averaged over a Indian subcontinent (70∘E-90∘E, 5∘N-30∘N), b Southeast Asia (90∘E–110∘E; 7.5∘N–30∘N), and c East Asian (110∘E–140∘E; 10∘N–40∘N) land regions in GPCP observations (OBS), MMM of piControl, and early Eocene 1xCO2, and 3xCO2 simulations (curves). For early Eocene simulations, another domain (60∘E-80∘E, 15∘S-10∘N) is also considered in a as the Indian subcontinent was away from the Asian landmass, and the corresponding rainfall cycles are shown in dashed curves. The intermodel 2 standard deviations (2σ) are shown as shading around the curves. The subregions are marked with boxes in Fig. S1.

Full size image

The simulated annual rainfall cycle over the Indian subcontinental landmass in the early Eocene is considerably weaker than piControl (Fig. 4a). The rainy season is shorter and its onset is shifted by about 2 months. However, the Indian subcontinent during the early Eocene appears as a large equatorial island surrounded by the Ocean. This continental shift has a profound impact on the timing of rainfall peaks. In piControl, the rainfall activity peaks in JJAS, while in the early Eocene simulations, the peak rainy season shifts to the austral summer months (Nov–Feb). Consequently, the early Eocene simulations produce less rainfall over Indian landmass in the northern summer, which is offset by enhanced rainfall in the other seasons. This pattern is reminiscent of the present climate over the Indo-Pacific Maritime Continent or the Indian Oceanic ITCZ (Fig. S7). The simulated rainfall amount over the Indian subcontinent is aligned with the mean annual rainfall estimates from fossil records from northwest India57. An increase in atmospheric CO2 levels only produces a marginally stronger annual rainfall cycle.

A substantial change is also noted in the early Eocene simulated SESM rainfall. The amount of rainfall in the peak rainy season is considerably lower than piControl in both early Eocene simulations (Fig. 4b). A limited northward shift of the ITCZ over the ASM region, as shown in Fig. 1, is likely to cause a weaker rainfall cycle over the SESM region. The peak of the rainy season is also shifted later by about 2 months, suggesting the delayed onset of the wet season. This change is common and robust in the DeepMIP models, indicated by small intermodel variability. While a large seasonality with an amplitude of ~ 8 mm/day is a characteristic feature of the present-day SESM, a weak seasonal cycle with about half of the present amplitude during the early Eocene barely qualifies to be called a ‘monsoon’ system. This substantial weakening of the monsoon rainfall in the early Eocene simulations is likely to be associated with reduced diabatic heating due to the lower-elevated TP60 and other complex regional paleogeographic impacts, e.g., the height of the Zargros/Iranian Plateau5 and East African highlands61, while CO2 increase has a trivial role in the SESM rainfall variability.

On the contrary, the rainfall change over the EASM region is less prominent in the early Eocene simulations (Fig. 4c). Although the peak wet season is delayed by about 2 months, the region still receives at least ~ 60% of the piControl rainfall in the early Eocene simulations. This reduction in rainfall is marginally compensated in the 3xCO2 experiment. As the interaction between the midlatitude frontal system and the monsoon trough governs the EASM rainfall4, the weakening of the western North Pacific subtropical high (Fig. 2) is likely to reduce the moist southerly flow62, leading to a decrease in seasonal rainfall. Additionally, the absence of elevated TP may weaken the EASM rainfall63, as it affects other subregions of ASM. The absence of the TP and its role in the seasonal displacement of the tropical rain belt over the ASM region are explored in the next subsections.

Thermal impact of the Tibetan Plateau

The present seasonal transition from dry to wet conditions64 in Asia, as shown in the annual rainfall cycle, is associated with the change in large-scale atmospheric heating3 and circulation65. The non-adiabatic heating over the TP acts as the heat pump and drives the ASM circulation66,67. During boreal summer, the upper tropospheric temperature (UTT, averaged air-temperature between 200 and 500 hPa) is observed to be enhanced around 30°N (Fig. 5a), and the warm upper tropospheric condition is concurrent with ASM onset and withdrawal68. This seasonal upper-troposheric warming sets up a meridional UTT gradient on the south side of the TP (Fig. 6a), which regulates the ASM onset60,69.

Fig. 5: Seasonal transition of upper tropospheric temperature and zonal winds over South Asia.
figure 5

Time-latitude sections of the monthly climatological upper tropospheric temperature (UTT, averaged between 200 and 500 hPa, shaded, ∘C) and 200-hPa zonal wind (contour, m/s) averaged over 70∘E-100∘E in a observation (NCEP reanalysis II), MMM of b piControl, and early Eocene c 1xCO2, d 3xCO2 simulations.

Full size image

The diabatic heating over the TP in the boreal summer induces an upper tropospheric continental anticyclone5,67 that supports the poleward retreat of upper-level (200 hPa) subtropical westerly jets (Fig. 5a). Retreat of the westerlies in early summer is replaced by the northward migration of the easterlies. By the end of September, the continental upper tropospheric anticyclone weakens and westerlies return to their winter position after ASM withdrawal.

The piControl realistically simulates the seasonal UTT enhancement and the retreat of upper-level westerlies (Fig. 5b). In contrast, the warmer UTT is evident around the equatorial region in the early Eocene simulations (Fig. 5c, d), indicating a weaker meridional UTT gradient between the equatorial region and 30°N in the boreal summer (Fig. 6a). Unlike piControl, the seasonal retreat of westerlies is not prominent in the early Eocene simulations and northward migration of the upper-level easterlies is limited to 10° latitudes only. The weaker easterly wind shear in the early Eocene simulations during JJAS (Fig. 6b) aligns consistently with these results, underscoring the pivotal role of the TP in regulating the monsoonal climate in Asia.

Fig. 6: Meridional distribution of JJAS upper tropospheric temperature and vertical zonal wind shear over South Asia.
figure 6

Seasonal (JJAS) mean meridional a upper tropospheric (averaged between 200 and 500 hPa) temperature (UTT) (in K), and b vertical zonal wind shear (zonal wind difference between 200 and 850 hPa, in m/s) distributions in NCEP reanalysis II (OBS), MMM of piControl, and early Eocene 1xCO2, 3xCO2 simulations (curves). The intermodel 2 standard deviation (2σ) range is shown as shading around the curves. All datasets are averaged over 60∘E-100∘E.

Full size image

Mean meridional circulation

The asymmetric off-equatorial monsoon heat source to the south of the TP shifts the mean meridional atmospheric circulation70 with an ascending motion of ~ 20°N and descending motion to the south of the equator and ~ 45°N (Fig. 7a). The ascending branch of this cell is aligned with the position of the maximum rainfall zone (ITCZ) as shown in Fig. 2. In contrast to the equatorial position of the ascending branch over rest of the globe, its largest northward displacement over south Asia is associated with the location of the monsoon trough71. A southward vertical tilt in the maximum vertical motion is apparent in observations, indicative of stronger lower-level convergence at the northern flank of the ITCZ convection.

Fig. 7: Seasonal mean meridional circulation.
figure 7

Vertical-latitude section of JJAS mean climatological vertical p-wind (shaded, Pa/s) and meridional wind (vector) over the ASM region in a observation (NCEP reanalysis II), and MMM of b piControl, c 1xCO2, d 3xCO2 simulations. Both the meridional and vertical p-wind fields are averaged over 60∘E-100∘E. Vertical wind is scaled up by a factor of 100 and the corresponding vectors are multiplied by –1 to represent upward motion by the upward vectors. The vector scale indicates the meridional wind with a magnitude of 5 m/s or, equivalently, a vertical wind of magnitude 0.05 Pa/s.

Full size image

The strong ascending motion ~ 20°N is largely captured in piControl with realistic descending branches to the northern subtropics and ~ 30°S. However, the observed vertical tilting of the vertical velocity axis is not well reproduced in piControl (Fig. 7b). In both early Eocene simulations (Fig. 7c,d), the ascending branch of vertical motion is located around the equatorial region. As the off-equatorial heating is absent in the early Eocene simulations and the ITCZ convection does not migrate to the northern latitudes as in piControl, the ascending branch of the meridional circulation remains near the equatorial region. A stronger and narrower equatorial ascending branch in the early Eocene simulations is accompanied by strong subsidence around  30°N. The enhanced subsidence over the South Asian landmass is indicative of suppressed convection, consistent with the arid condition over the region during the early Eocene, as shown in Fig. 2.

Discussion

We examine the change in the Asian monsoonal climate in the early Eocene epoch ( ~ 56–48 Ma) using DeepMIP-Eocene simulations30 to better understand deep-time paleoclimate modeling uncertainty. The DeepMIP multimodel piControl and early Eocene simulations provide a consistent experimental framework for studying the evolution of ASM, which can be compared with proxy-based reconstructions. These paleoclimate simulations consider the changes in paleogeography, vegetation, and continental ice sheets using a standard set of boundary conditions across the models for the early Eocene when atmospheric CO2 levels were at least three times higher than in the preindustrial period25. The multimodel ensemble analysis indicates a significant dry condition over South Asia in the early Eocene (Fig. 2), promoted by the convectively suppressed large-scale circulation (Fig. 7). The topographic blocking of Gangdese mountain (a mountain range of ~ 1500m elevation) to weaker summertime cross-equatorial flow at the southern edge of the Asian mainland could also contribute to South Asian arid conditions. These led to significantly weaker seasonal rainfall over the region relative to the present ASM climate, consistent with proxy estimates7.

The boreal summer maximum rainfall activity was confined to the equatorial oceanic region as the location of ITCZ had a reduced seasonal northward migration in the early Eocene. The warmer tropical SST in the early Eocene simulations may contribute to the enhanced rainfall around the equator. While the early Eocene period had a considerable global Tsfc warming, attributed to factors like the absence of continental ice sheets, alterations in surface elevation across Asia, and a more constrained Southern Ocean gateway, these conditions were insufficient to induce a seasonal displacement of the tropical rainbelt from its preferred oceanic location. This invokes the importance of diabatic heating over the TP in regulating the monsoon flow through the generation of positive potential vorticity on the southern flank of the TP in the present climate66 and driving the northward excursion of the ITCZ. The lower-level quasi-stationary cyclonic vorticity induces strong LLJ, while the upper-level anticyclonic vorticity supports the retreat of the subtropical westerly jets poleward67. The ellipsoidal shape of the TP also determines the position of the westerly jets and moisture advection over the EASM region72. Additionally, a large mean easterly wind shear setup by the upper-level monsoon anticyclone is critical for the northward propagation of the ITCZ on the intra-seasonal timescales73.

The absence of a TP-like elevated heat source in the early Eocene results in a weaker north-south UTT gradient (Fig. 6a). This is associated with a considerably weaker upper-level anticyclone and reduced mean easterly shear (Fig. 6b), impeding the feedback necessary for the northward movement of the ITCZ. As a result, the northward excursion of the rainband was limited in the early Eocene, contributing to the aridification of South Asia. Due to a weak seasonal rainfall cycle and inadequate wind amplitude, the early Eocene wet season over South Asia does not distinctly qualify as a ‘monsoon’ system. It appears that the initiation of the South Asian summer monsoon is likely dependent on a specific degree of upliftment of the Tibetan Plateau. While previous studies have emphasized the importance of the uplift of eastern Tibet15 and the retreat of the Paratethys Sea13 for the ASM initiation, some studies have reported the presence of monsoons in China during the Eocene8,21,74. The disparity between our findings and previous studies is ascribed to the substantial uncertainty inherent in paleoclimatic observations and earlier modeling studies. Nevertheless, our results still reveal a clear shift in wet and dry seasonality with distinct regional characteristics during the early Eocene over South Asia. It is essential to note that while the wet-dry season rainfall ratio may not serve as a reliable proxy for the monsoon circulation75, the altered seasonality patterns persist as a noteworthy feature in our analysis. Moreover, the interannual variability of a slightly weaker EASM during the early Eocene may also contribute to the seasonal transition from dry to wet conditions in the annual rainfall cycle over the eastern part of the continent.

Due to the equatorial position of the Indian subcontinent in the early Eocene, the region used to receive high rainfall in the early-late months of the year with prominent wet and dry seasons57. The JJA total rainfall amount in western India ( ~ 83 cm) in the 3xCO2 simulation matches with the proxy observation at Gurha 72 m ( ~ 93.08 cm). However, the equatorial Indian monsoon system was not part of the Asian wet season, and its seasonality and dynamics were likely to be regulated by the insolation, land-ocean contrast, and the surrounding oceanic processes. In contrast, the annual rainfall cycle over South Asia was markedly weaker by a magnitude of less than half of the present SESM. As the off-equatorial elevated heat source was absent and northward migration of ITCZ was limited to south of 15°N, it is perhaps not surprising that the rainfall activity was weaker with a peak rainy season delayed by about 2 months relative to the present climate. Seasonal increase of UTT to the southern flank of TP determines the onset of ASM68. In the absence of this large-scale upper tropospheric thermal factor in the early Eocene, the role of regional land-ocean contrast and insolation maxima become fundamental to drive the rainy season, which exhibits different phase lag due to local convective anomalies and causes delay at the beginning of the rainy season. Compared to South Asia, the rainfall changes in the EASM region in the early Eocene are less prominent, where the monsoon system was likely to be significantly established by the late-middle Eocene ( ~ 40 Ma)76. With an equatorward retreat of the northern edge of the monsoon, the Meiyu/Baiu frontal system can partially compensate for seasonal rainfall over East Asia.

The impact of the high GHG forcing (3xCO2) on the ITCZ and rainfall during the early Eocene is significantly different than what is expected in the present geophysical configuration. An ensemble of CMIP6 model projections indicates that the ISM is going to intensify at ~ 3%/K with the monsoon rainfall over land expected to expand westward77. The westward expansion of the ISM is facilitated by a westward expansion of the Indo-Pacific warm pool and a westward movement of the oceanic ITCZ. However, the SESM was almost non-existent during the early Eocene (Figs. 4b and 2f). This is consistent with the westward expansion of the warm pool, showing little sensitivity to the GHG forcing (Fig. 3b) under the modified coupled climate mean state.

Our analysis suggests that a significant weakening of the Asian wet season in the early Eocene is primarily influenced by the changes in paleogeography, vegetation, and continental ice-sheet boundary conditions, while enhanced atmospheric CO2 levels amplify the rainfall variability. The dominance of non-CO2 boundary conditions on the early Eocene monsoon dynamics is in line with the findings of Farnsworth et al.5, who studied the evolution of the East Asian monsoon system in HadCM3 for the last 150 Ma. However, the 3xCO2 experiments produce a higher rainfall amount over the ASM domain with an increase of ~ 11% (4%) relative to piControl (1xCO2 simulation). This outcome is expected, given the approximately 11°C warming observed in the 3xCO2 experiments, which corresponds to the lower limit estimated by Inglis et al.23. This aligns with the fact that 3xCO2 simulations represent the lower range of atmospheric CO2 levels in the early Eocene24. In the warmer ‘high CO2’ climate, rainfall enhancement is noted along the ITCZ, and the annual rainfall cycle is slightly stronger than the 1xCO2 simulation.

Despite a set of accurately reconstructed boundary conditions used in the DeepMIP-Eocene models, uncertainties persist in the paleo-reconstructions55, particularly related to the elevation of the Gangdese, the uplift of TP17, and the timing of the India-Asia collision78. Additional uncertainties may arise from globally homogeneous soil parameters derived from preindustrial conditions, as well as non-CO2 GHGs, orbital configuration, and aerosols. As the monsoon inception, strength, and variability are highly dependent on these factors, future studies employing sensitivity analyses with tectonically constrained paleogeography could provide deeper insights into the complex ASM system. This aspect will be addressed in our forthcoming research.

Models, data, and methodology

Eight modeling groups contributed to DeepMIP-Eocene paleoclimate simulations. In addition to a piControl simulation similar to the CMIP6 standard79, a set of early Eocene experiments was conducted with standardized atmospheric CO2 concentrations and other boundary conditions, such as common paleogeography, vegetation, and river routing from the reconstruction of Herold et al.55; solar constant, orbital configurations, non-CO2 GHG concentration, globally homogeneous soil properties from the preindustrial period. The early Eocene experiments were run without continental ice sheets and with atmospheric CO2 concentration at preindustrial value (280 ppmv, 1x) and various CO2 levels, for instance, at 1x, 3x, 6x, or 9x the preindustrial value for the sensitivity experiments31. The details of the utilized climate models and the experimental setup used in this study are described in refs. 30,31, which are summarized in Table 1. For equilibrium and model spin-up, each simulation was run for a few thousand years (see Table 1 for the exact simulation length) and only the last 100 years are analyzed in this study.

Table 1 The DeepMIP climate model configurations for the early Eocene simulations used in this study
Full size table

Out of the nine participating climate models, we consider five models based on the availability of the piControl and sensitivity experiments (1xCO2 and 3xCO2). To examine the CO2 and non-CO2 forcing in the simulations, the 1xCO2 simulation and at least one higher CO2 level early Eocene simulation are required to compare with the piControl. Three models (INM-CM4.8, IPSL-CM5A2, and NorESM1_F) are excluded as no 1xCO2 simulation is available and HadCM3BL is also excluded due to the use of its higher resolution version. Consequently, we analyze the most common 3xCO2, along with 1xCO2 and piControl simulations, from the remaining five models in this study.

The zonal, meridional, and vertical winds and air-temperature in piControl are evaluated against monthly National Center Environmental Prediction -National Center for Atmospheric Research (NCEP-NCAR) reanalysis II80 dataset, which is available globally on a 2.5° latitude × 2.5° longitude grid from January 1979–2018. The simulated rainfall in piControl is verified using the monthly Global Precipitation Climatology Project (GPCP), version 2.3 dataset on a 2.5° spatial resolution for the period of 1979–2018. We also use global monthly Extended Reconstructed SST (ERSST)81, version 5, at 2° longitude × 2° latitude resolution from 1979-2018. To mitigate potential artifacts stemming from data regridding, these observed/reanalysis datasets are primarily selected for their resolutions close to the coarse horizontal resolution of the DeepMIP models.

The model outputs are interpolated onto a common reanalysis or observation grid. The evaluation of the piControl mean climate during the boreal summer (JJAS) season is conducted using the Taylor diagram, as illustrated in Fig. S2. The spatial pattern correlation coefficient, the normalized (by the observed) spatial standard deviations of the simulations, and the root-mean-square error (RMSE) between the observation and simulation are indicated by a single point on the Taylor diagram82. Generally, the models reasonably simulate the seasonal mean climate in the tropics as well as over the ASM region, with a pattern correlation range between 0.8 and 0.99, and a normalized standard deviation of about 1 for most of the variables. Only COSMOS and GFDL have slightly lower correlations ( ~ 0.6–0.7) for rainfall and 500-hPa vertical velocity. It may be noted that this evaluation may not be entirely rational as piControl and the verification datasets from the recent period represent different climate regimes, especially due to changes in the atmospheric CO2 concentrations since the preindustrial period. However, piControl and the observations are from the same climate with present topography, continental ice sheets, and other boundary conditions relative to the early Eocene climate. Therefore, the inconsistency in atmospheric GHG concentration may be neglected relative to model biases32,83.

The assessment reveals that the performance of all five models is remarkably consistent, exhibiting a relatively low intermodel spread. It directs us to use the five models-based MMM approach. The MMM outperforms most models in simulating the mean boreal summer climate with a higher pattern correlation, and lower RMSE in the Taylor diagram (Fig. S2). This robust performance gives us confidence in utilizing the MMM strategy for a comprehensive examination of the ASM characteristics in both piControl and early Eocene simulations.