2025 Volume 103 Issue 5 Pages 559-572
Climate models project stronger summer monsoons and weaker winter monsoons in a warmer future over East Asia. This is because the large-scale land–sea temperature contrast and resulting pressure differences will be larger during the boreal summer but smaller during the boreal winter. However, how the atmospheric circulation evolves during the transition between the seasons remains poorly understood. In this paper, we analyze the modelling results from the historical and shared socioeconomic pathways (SSP) 5–8.5 future scenario experiments generated as part of the sixth phase of the Coupled Model Intercomparison Project (CMIP6). We show that the timing of the sign-reversal of the land–sea sea-level pressure difference moves significantly later through the boreal autumn during the late 21st century, but that there is little difference in the sign-reversal during spring. This asymmetric seasonal response is associated with a greater land–sea difference in surface temperature increase and pressure drop in autumn than in spring. In addition, models with a greater global annual mean surface temperature change show a later autumn transition, whereas no such relationship is evident for the spring transition.
The monsoon has long been described in terms of the seasonal reversal of the surface wind system (Ramage 1971), which in turn is caused by the seasonal reversal of the continental–ocean thermal contrast. The critical importance of precipitation to agriculture, as well as the amount and quality of water resources available for industrial and domestic use, has led to much research into the interannual, interdecadal, and secular variability of monsoonal precipitation. How the global warming will affect regional and global monsoons has therefore been widely debated (Kitoh et al. 2013; Christensen et al. 2013; Kitoh 2017; Wang, B. et al. 2021).
As climate warms, the warmer air can hold more water vapor, and this increases the amount of water vapor transported from the oceans to the land. Thus, warming-related changes in large-scale atmospheric circulations affect the strength, extent and duration of regional monsoon rainfall (Kitoh et al. 2013; Seth et al. 2013; Endo and Kitoh 2014; Endo et al. 2018; Ha et al. 2020; Dai et al. 2022; Liu et al. 2023). Previous studies show that future changes in regional monsoon precipitation are influenced mainly by dynamical effects, rather than thermodynamical effects (e.g., Endo and Kitoh 2014). Among the regional monsoons, the East Asian monsoon is somewhat different when compared with the other tropical and sub-tropical monsoons (Wang and LinHo 2002; Ding and Chan 2005). Under a warming climate, the East Asian summer monsoon circulation is more closely correlated with the lower-troposphere thermal contrast between the Tibetan Plateau and the northwestern Pacific, rather than with the upper-troposphere thermal contrast, whereas the South Asian summer monsoon circulation is driven by the upper-troposphere thermal contrast between the Tibetan Plateau and the tropical Indian Ocean (Wu and Wang 2001; Sun and Ding 2011; Endo et al. 2018). The role of the enhanced latent heating over the Tibetan Plateau in strengthening the East Asian summer monsoon circulation is also discussed (He et al. 2019).
Changes in the hydrological cycle and atmospheric aerosol loading play an important role in moderating the temperature difference between land and sea (Joshi et al. 2008; Christensen et al. 2013). Global warming will cause surface temperatures to rise throughout the year, but there are geographic and seasonal variations that will result in greater temperature increases during the Northern Hemisphere winter than during the summer months.
The effects of rising temperatures on seasonality and the biome are largely robust in the observations. With respect to the changes in seasonality, a delayed start to the snow and ice season, as well as an earlier end to this season, have already been widely observed (Shaw et al. 2022). In addition, observations of the biological season have shown an earlier onset of spring and a later onset of autumn, which is consistent with the rising temperature. However, investigations into changes in the seasonality of the surface temperature depend on how the four seasons are defined. For example, Ho et al. (2021) defined the temperature between June 1 and August 31 as a reference period for the summer season. They found that the start and end dates of the summer season in Korea had advanced by 13 days and been delayed by 4 days, respectively, over recent decades (1988–2017) relative to an earlier period (1919–1948). On the other hand, the extent of advance and delay are similar (24 days and 23 days, respectively) at the end of the 21st century under the representative concentration pathway 8.5 (RCP8.5) scenario. Wang, J. et al. (2021) show that global warming results in longer summers, but shorter winter, spring and autumn seasons, when the lengths of the seasons are defined as the local temperature thresholds using the 25th and 75th percentiles over the historical period (1951–2011). Wang, J. et al. (2021) project that the onset of spring will become earlier and that of autumn will be delayed in a warming world.
There have been numerous modeling studies focusing on future changes to the monsoon circulation during their peak season, but how the atmospheric circulation evolves during the transition between the seasons remains poorly understood. Greenhouse gases increase the boreal summer land–sea thermal contrast in the Northern Hemisphere, which strengthens the Asian summer monsoon circulation (Endo et al. 2018; Jin et al. 2020). Endo et al. (2021) pointed out that the effects of continental warming become visible in the late summer precipitation over East Asia. The summer monsoon circulation is projected to strengthen over East Asia but weaken over South Asia, and these changes are associated with an enhancement of the zonal thermal contrast and a weakening of the meridional thermal contrast between the Eurasian continent and the surrounding oceans, respectively (Wu et al. 2022). The former is considered to be dominated by a fast response whereby Eurasian continental warming leads to a continental pressure decrease, and thus enhanced monsoon circulations (Li et al. 2022). The latter is thought to be dominated by a slow response related to the so-called El Niño-like changes in tropical sea surface temperatures (SSTs), which are large in the eastern Pacific and are caused by a weakening of the North Pacific subtropical high associated with the oceanic warming (Li et al. 2022). However, whether future tropical SST patterns undergo El Niño-like or La Niña-like changes remains to be determined (Vecchi and Soden 2007; Kociuba and Power 2015; Coats and Karnauskas 2018; Chung et al. 2019).
Some studies have investigated the seasonal mean differences in the changing atmospheric circulation caused by global warming. During the boreal summer season, the Eurasian continent has a low-pressure area at the surface and thus winds blow from the sea to the land, whereas during the winter season, the Eurasian continent has a high-pressure area and winds blow from the land to the sea. Seasonal changes in the mean sea-level pressure (SLP) are much greater on land than over sea. Global warming is expected to bring relatively greater warming over the Eurasian continent than over the oceans, so that the SLP difference between land and sea is expected to be larger in summer and smaller in winter, relative to the present.
Previous studies have investigated future changes in regional and local atmospheric circulation. Using the seasonal mean SLP generated by the fifth phase of the Coupled Model Intercomparison Project (CMIP5) models under the RCP2.6 and RCP8.5 scenarios, Ito et al. (2020) defined the near-surface westerly wind change index around Japan as the difference in SLP between 25°N and 45°N averaged over 120–150°E, and the near-surface southerly wind change index as the difference in the SLP between 150°E and 120°E averaged over 25–45°N. They showed that the atmospheric circulation around Japan weakens in winter and summer. Ito et al. (2025) further calculated the above wind change index using the monthly mean SLP from the CMIP6 models under the historical and the shared socioeconomic pathways (SSP) 5–8.5 scenario. They found that the spring SLP pattern appears earlier, and the autumn pattern is delayed, by the end of the 21st century relative to the present day. Dai et al. (2022) investigated future changes in the eight monsoon stages of the East Asian annual cycle based on the 850 hPa winds over [110–140°E, 20–45°N] using the self-organizing map developed by Dai et al. (2020). They found prolonged mid-summer and autumn monsoon periods, and a shortened transition in spring, based on the CMIP5 models under RCP4.5 and RCP8.5.
The timing of the transition between winter and summer atmospheric circulations, together with additional regional factors such as local land–sea distribution and orography and aerosol loading, will affect regional monsoon precipitation onset and retreat. As the land–sea thermal contrast is the main driving factor of the large-scale monsoons, we will focus on the continental-scale SLP differences between land and sea as an indicator of the large-scale monsoon circulations. In the remainder of this paper, we will investigate future changes in the timing of the transition from the winter monsoon to the summer monsoon, and vice versa, using the output from the CMIP6 models based on the historical and the SSP5-8.5 scenario experiments.
We used the monthly mean outputs from the historical simulation (years 1850–2014) and the future projection under the SSP5-8.5 scenario (years 2015–2099) from the 44 CMIP6 models (Eyring et al. 2016; see Table S1). One realization from each model was used. The monthly mean data were interpolated onto a common 2.5° × 2.5° (longitude × latitude) grid. Additionally, the daily mean SLP from six CMIP6 models was used to assess the suitability of using the monthly mean data.
2.2 Reanalysis dataWe used three global reanalysis datasets: the Japanese 55-year Reanalysis (JRA-55; Kobayashi et al. 2015), the Japanese Reanalysis for Three Quarters of a Century (JRA-3Q; Kosaka et al. 2024), and the fifth generation of the European Centre for Medium-Range Weather Forecasts (ECMWF) atmospheric reanalysis (ERA5; Hersbach et al. 2020). We used the monthly mean 850 hPa meridional winds and SLP, interpolated onto a common 2.5° × 2.5° grid, covering the periods 1958–2022 (JRA-55), 1947–2022 (JRA-3Q), and 1959–2022 (ERA5).
There have been various attempts to define an East Asian monsoon index (e.g., Li and Zeng 2002; Zhu et al. 2005; Wang et al. 2008; Li et al. 2022). Li and Zeng (2002) defined the East Asian monsoon index as the area-averaged normalized winds at 850 hPa over the domain [10–40°N, 110–140°E]. Zhu et al. (2005) used the large-scale zonal and meridional land–sea thermal contrast with a combination of the vertical shear of zonal winds over East Asia and the zonal SLP difference between 160°E and 110°E. Li et al. (2022) used the area-averaged meridional wind at 850 hPa over [20–45°N, 110–125°E] as an East Asian summer monsoon index to investigate changes to the East Asian monsoon circulation caused by global warming.
As an example of the East Asian monsoon index, Fig. 1 shows the monthly (January–December) and annual (1850–2099) distribution of the mean 850 hPa meridional winds from the 44 CMIP6 multi-model ensemble (MME) mean averaged over [20–45°N, 110–125°E], and the differences relative to the 1850–1899 average for each month. The southerly wind in eastern China reaches a maximum in July. The zero line, where the north–south winds reverse, appears around March in spring and between August and September in autumn. The zero line shows a delaying trend for the autumn season, when it changes from a southerly to a northerly wind, through the 21st century. The deviation from the 1850–1899 average (Fig. 1b) clearly shows a positive anomaly in late summer (July–September). The linear trends (1850–2099) and the differences between the 1991–2020 mean and the 2070–2099 mean are statistically significant through May through October (positive trend) and through November and December (negative trend). On the other hand, the timing of the change from northerly to southerly winds in the spring season shows no significant secular changes. Therefore, the summer monsoon in East Asia is extended as a result of the delay in its termination.
(a) Year–month distribution of monthly mean East Asian Monsoon index (EAMI) defined as the 850 hPa meridional wind (m s−1) averaged over [20–45°N, 110–125°E] calculated using the 44 CMIP6 models and the historical and SSP5-8.5 scenario experiments. Zero lines are black. (b) As for (a), but showing the deviation from the 1850–1899 average. Black dots indicate that 75 % of the models have the same sign.
The CMIP6 models indicate a period of weakening in the East Asian summer monsoon, mainly in July, during the mid to late 20th century. Figure 1b shows the negative anomalies associated with the 850 hPa southerly winds during the late 20th century in summer. Song et al. (2014) investigated the observed weakening of the East Asian summer monsoon circulation between 1958 and 2001, and showed that aerosol forcing caused the cooling over East Asia, and this reduction in the land–sea thermal contrast resulted in the weakening of the monsoon circulations. Zhu et al. (2012) suggest that the local warming around Lake Baikal and related circulation changes are a possible cause of this monsoon weakening. Shao et al. (2024) used climate model experiments to demonstrate the effect of anthropogenic aerosols on the East Asian summer monsoon changes over the periods 1950–1980 and 1980–2010.
The winter–summer reversal of large-scale monsoon circulations is caused mainly by the temperature differences between the Eurasian continent and the surrounding ocean. Here, we used the SLP difference between land and sea over the Asia–Pacific region, because the continental-scale land–sea thermal contrast is associated mainly with SLP changes over land (e.g., Webster 1987; Meehl 1994). For the purpose of discussing the transition between winter and summer, we define “spring” and “autumn” as the spring/autumn sign-reversal date when the SLP difference between land and sea turns from positive (winter) to negative (summer), and from negative (summer) to positive (winter), respectively, and then consider the future changes when the value becomes zero.
Figure 2b shows the seasonal change in the SLP difference (land minus sea) averaged over the period 1991–2020 from the three reanalysis datasets and the 44-model CMIP6 MME. “Land” and “Sea” values are averaged over the regions [25–60°N, 30–180°E] and [0–60°N, 45–180°E], respectively (Fig. 2a). These domains are slightly different to those used by He et al. (2019), where employed land and sea areas of [25–50°N, 100–180°E]; and those used by Jin et al. (2020), who employed a land area of [0–60°N, 30–180°E] and a sea area of [10°S–60°N, 30–180°E]. In this paper, we excluded the land area south of 25°N because the subtropical land regions suffer surface temperature drops due to monsoonal precipitation and low clouds when the summer monsoon is enhanced. We checked the sensitivity of the northern latitudes with respect to the land area between 50°N and 70°N, and the southern latitudes with respect to the sea area between 10°S and 0°N, and did not find any significant differences in the characteristics of sign-reversal between summer and winter.
(a) Map showing the (red) land and (black) sea regions used in the analysis. (b) Monthly mean SLP differences (hPa) between land and sea. Green, red, blue, and black lines denote ERA5, JRA-55, JRA-3Q, and CMIP6 MME, respectively.
We note that, comparing the three reanalysis datasets, JRA-55 shows greater seasonal variation than ERA5. The value for JRA-3Q is very close to that of JRA-55. The CMIP6 MME lies between the values from the JRA-55/JRA-3Q and ERA5 reanalysis datasets. The land SLP is lower than the sea SLP from June to August, whereas the land SLP is higher than the sea SLP from September to April. The reversal of the sign occurs in mid-May and late August to early September based on this definition, but the precise dates depend on how the sea and land definition areas are defined.
Figure 3 shows the interannual variations of the monthly mean SLP difference between land and sea for the three reanalysis datasets and the CMIP6 MME. The CMIP6 MME reproduces the characteristics of the reanalysis. They all show positive values in winter and negative values in summer. The transitions occur in May and in August or September, as shown in Fig. 2b.
Year–month distribution of the monthly mean SLP differences (hPa) between land and sea from (a) JRA-55, (b) JRA-3Q, (c) ERA5, and (d) CMIP6 MME. (e) Time series of July SLP differences (hPa) between land and sea for the four datasets. In (a)–(d), zero lines are black.
Looking at the secular variations of the July SLP differences over this period 1960–2020 (Fig. 3e), the JRA-55 and JRA-3Q profiles are nearly constant, while the ERA5 shows a weak positive trend. The value of the CMIP6 MME lies between the JRA-55/JRA-3Q and ERA5 reanalysis datasets as in Fig. 2b. The CMIP6 MME shows a trend since the 1980s towards a larger negative SLP difference between land and sea, but this feature is not seen in the reanalysis data. The discrepancies we have highlighted between the three reanalysis datasets in Figs. 2 and 3 imply that further clarification is required, even for this fundamental variable, although this is beyond the scope of the present study.
3.2 Future changes projected by the CMIP6 MMEIn this sub-section, we investigate the temporal evolution of the monthly mean SLP and the thermal contrast between land and sea. Figures 4a and 4c shows the 1850–2099 interannual variation of the CMIP6 MME monthly SLP and surface air temperature differences (land minus sea) over the Asian region. Figures 4b and 4d shows their deviations from the 1850–1899 average.
(a) Year-month distribution of the monthly mean SLP differences (hPa) between land and sea from the CMIP6 MME using the historical and SSP5-8.5 scenario experiments over the period 1850–2099. (b) As for (a), but showing the deviation from the 1850–1899 average. (c) and (d) As for (a) and (b), but for the monthly mean surface air temperature (T2m). In (a) and (c), zero lines are black. In (b), black dots indicate that 75 % of models have the same sign.
The CMIP6 projections indicate that, during the 21st century, summer temperatures on land will rise (relative to the oceans) and the land will become a low-pressure area, implying an enhanced Asian summer monsoon circulation. The linear trends (1850–2099) of MME SLP and surface air temperature differences (land minus sea) are statistically significant in all months. The onset of the SLP difference becoming zero in autumn (from negative to positive) will be delayed because the difference in the land–sea temperature rise is largest in August (Fig. 4d) and the SLP drop is also largest in August (Fig. 4b). On the other hand, the land–sea temperature rise difference and the SLP drop in May are relatively small. Therefore, CMIP6 projects only small changes in the timing of when the SLP difference becomes zero (from positive to negative) in spring, thus leading to the asymmetric changes between spring and autumn. We note that besides a minimum rise in April and May and a maximum rise in August, there is a minimum rise in October and a maximum rise in December. The latter is related to a larger temperature rise caused by global warming across northern Eurasia (Fig. S1).
In the latter half of the 20th century, positive anomalies are evident in the land–sea SLP difference during summer, and this is consistent with a decrease in the surface air temperature contrast (negative anomalies) and a negative East Asian Monsoon index (Fig. 1b). This mid-summer temperature drop and positive land-sea SLP difference is caused mainly by aerosol effects (Song et al. 2014; Li et al. 2016; Hegerl et al. 2019; Wang et al. 2019). Further analysis of the Detection and Attribution Model Intercomparison Project (DAMIP; Gillett et al. 2016) results would clarify the factors responsible for this late 20th century responses.
What causes these differences in the magnitude and spatial distribution of the surface air temperature rise between spring and autumn? Differences arise in the magnitude of the surface air temperature rise caused by global warming depending on the season. Over land in the Northern Hemisphere high latitudes, the winter temperature increases are greater than the summer temperature increases, whereas in the subtropical to mid-latitudes, the summer temperature rise is greater (Lee et al. 2021). Figure S1 shows the monthly mean surface air temperature changes between the present (1991–2020) and the future (2070–2099 under the SSP5-8.5 scenario). In central Eurasia, July to September are the months when the surface air temperature rises are greater than the annual average.
The changes in surface air temperatures between the present (1991–2020) and the future (2070–2099) are shown in Figs. 5a and 5b for May and September, respectively, and Fig. 5c shows the difference between them. A greater increase in the future surface air temperatures over land than over the ocean has been assessed as being virtually certain through the 21st century (Lee et al. 2021). Figure 5 shows a greater increase in September surface air temperatures over most of the Eurasian continent as compared to in May. Over South Asia, the summer monsoon related precipitation and low-level clouds are responsible for smaller surface temperature increases in September relative to those in May.
(a) Monthly mean surface air temperature changes in May between the periods 1991–2020 and 2070–2099. (b) As for (a), but for September. (c) Differences between (b) and (a).
Ito et al. (2020) analyzed the CMIP5 projections under the RCP8.5 scenario and found that the future change in seasonal mean surface air temperatures at the end of the 21st century around Japan in spring will be smaller than for the other seasons. On the other hand, seasonal mean surface temperatures observed by 15 stations in Japan during 1898 and 2022 indicated a linear increasing trend of 1.19, 1.56, 1.19, and 1.31 °C per century for winter, spring, summer and autumn, respectively (Japan Meteorological Agency 2023). This contrast in temperature rises between past and future may be partially related to the difference in aerosol forcing in the region, but should be investigated further.
3.3 Variability among modelsIn the previous sections, we analyzed the CMIP6 MME data, but there is significant uncertainty among the model projections. As warming trends are large in the 21st century, variability of the spring/autumn sign-reversal dates is calculated during the 150-year period 1850–1999. The 44-model average of interannual variability (standard deviation) during the 150-year period of each model is 0.29 month (7 days) in spring and 0.12 month (4 days) in autumn. There is also uncertainty among the various model projections, which we sought to evaluate. To do this, we defined inter-model variability as the standard deviation of the 150-year average of each of the 44 models. This was calculated to be 0.37 month (8 days) for spring and 0.24 month (6 days) for autumn. Thus, the inter-model variability is slightly greater than the interannual variability, and is larger in spring than in autumn. Consequently, we show the 10th, 30th, 50th, 70th, and 90th percentile values among models.
To investigate the variability among the models, Fig. 6 shows the time series from 1850 to 2099 of the dates when the difference in SLP between land and sea becomes zero for each model. Note that both the year-to-year variability of each model and the variability between models overlap in this figure.
(a) Time series of the month when the SLP difference between land and sea regions changes from positive to negative during spring from all 44 CMIP6 models. Bold white lines indicate the 10th, 30th, 50th, 70th, and 90th percentiles for the 44 models in each year. (b) As for (a), but showing when the SLP difference between land and sea regions changes from negative to positive during autumn.
The year-to-year variability of the winter-to-summer reversal in the land–sea thermal contrast is large, whereas the year-to-year variability of the summer-to-winter reversal is small. The differences between the models are also larger for spring than for autumn. From the data we have analyzed, it is not possible to investigate the reason why the interannual variability and inter-model variability are larger for spring than for autumn. We speculate that, from the viewpoint of seasonal variability, these differences might reflect the different mechanisms associated with the warming and cooling of the land surface. Land surface warming is the result of incoming solar radiation and longwave radiation from the atmosphere, whereas cooling is driven by the emission of longwave radiation, as well as the latent and sensible heat fluxes. The presence of clouds has a significant impact on solar radiation, and this cloud-atmosphere interaction, which is highly model-dependent due to the different parameterizations used, could be responsible for large interannual and inter-model variability. Thus, in the former case, the continental warming is highly variable, but cooling may be less variable from year to year because the radiative cooling from the ground may have a larger role during this season.
To examine the uncertainty associated with the transition timing among the models, Fig. 7 shows a histogram of the difference between the present (1991–2020) and the end of the 21st century (2070–2099) in the date when the SLP difference between land and sea becomes zero for the spring (Fig. 7a) and autumn (Fig. 7b) seasons. The model-to-model variability of the change in the transition during spring varies widely from −14 days to +12 days among the models. However, the ensemble mean results in a small change of −0.9 days (median = −1 day). On the other hand, for the change in autumn, there is a delay in the sign change date for most models. With an average of +4.5 days (median = +5 days), the end of summer is projected to be significantly delayed, by about 5 days, within 80 years.
(a) Number of models for which the SLP difference between land and sea regions changes from positive to negative during spring between the present (1991–2020 mean) and the end of the 21st century (2070–2099 mean). On the horizontal axis, positive values indicate a delay. (b) As for (a), but for when the SLP difference between land and sea regions changes from negative to positive during autumn.
We assessed the reliability of the projected future changes of the sign-reversal dates, and the asymmetry between spring and autumn, based on the monthly mean SLP data. To do this, we analyzed the daily SLP data from six selected models: CESM2, CNRM-CM6-1, IPSL-CM6A-LR, MIROC6, MPI-ESM1-2-LR, and MRI-ESM2-0. As daily data are highly variable even when the 30-year mean is calculated, we instead used the land–sea SLP difference reversal dates calculated using the pentad (5-day) mean (Figs. S2, S3). The results are qualitatively similar; i.e., the sign-reversal date in autumn is delayed, but in spring varies among the models.
We considered whether a relationship exists between the timing of the transition of the land–sea SLP difference and the global warming rate among the models. Figures 8a and 8b shows the scatter plots of the global annual mean surface air temperature difference (dTgm) between the present (1991–2020 mean) and the end of the 21st century (2070–2099 mean), versus the date when the SLP difference between land and sea changes. There is no relationship between dTgm and the timing of when the land–sea SLP difference changes from positive to negative in spring (correlation coefficient = −0.05). On the other hand, there is a significant positive correlation between dTgm and future changes in the timing of the land–sea SLP difference changes from negative to positive in autumn (correlation coefficient = +0.38, which is statistically significant). Models that incorporate greater global warming project a later date in the autumn for when the land–sea difference in SLP changes from negative to positive; i.e., the autumn sign-reversal date is delayed more when future warming increases.
(a) Global mean surface air temperature difference (dTgm) between the present (1991–2020 mean) and the end of the 21st century (2070–2099 mean), versus the date when the SLP difference between land and sea regions changes from positive to negative during spring. Units on the x axis are K and on the y axis are days. (b) As for (a), but for when the SLP difference between land and sea regions changes from negative to positive during autumn. (c) As for (a), but the horizontal axis is the SLP changes over land for the April–June period. (d) As for (b), but the horizontal axis is the SLP changes over land for the July–September period.
Future changes in the April–June average land SLP on the Eurasian continent (see Fig. 2b) are significantly positively correlated with future changes in the time when the land–sea difference in spring SLP changes from positive to negative (Fig. 8; correlation coefficient = +0.78). Figure 8d also shows that the future changes in the July–September average land SLP are significantly negatively correlated with future changes in the time when the land–sea difference in autumn SLP changes from negative to positive (correlation coefficient = −0.72). In other words, these results strongly suggest that land SLP anomalies control the alternation of the summer and winter large-scale atmospheric circulations.
We believe that the robust SLP response over land caused by the continental warming during the mid-summer, and the resultant passive radiative cooling during the late summer, cause the delay in the timing of the summer-to-winter alternation in the autumn circulation field. On the other hand, the alternation from winter to summer is affected by various forcing factors, as well as internal variability, causing large model-related uncertainties within the projections. The reasons for the latter should be investigated in future studies.
Climate models broadly project that atmospheric circulations will weaken in the tropics in a warming world. In East Asia, on the other hand, the summer monsoon circulation is projected to intensify due to an increase in the land–sea temperature and pressure differences during the summer season. This paper has examined the winter-to-summer and summer-to-winter transitions associated with the monsoon circulations and considered the potential changes to these transitions projected by the CMIP6 experiments based on the SSP5-8.5 scenario. Our result show that when the boundary between summer and winter is defined as the time when the difference in SLP between land and sea becomes zero, there is an asymmetry in the projected changes to the East Asian monsoon circulation in the future, with smaller changes in spring and larger changes in autumn, with the latter seasonal transition occurring later in the year. Our analysis of the intermodel differences demonstrates that the larger the change in global annual mean surface air temperature, the later the autumn transition; however, we found no such relationship associated with the spring transition.
The same is true for the 850 hPa meridional winds averaged over East Asia [20-45°N, 110-125°E]. Northerly winds prevail during the winter and southerly winds during summer, with the wind direction reversing in March and August–September. Late summer positive anomalies for the meridional winds are prominent in the second half of the 21st century, indicating a delay in the transition from southerly to northerly winds. On the other hand, the timing from northerly to southerly winds shows little change. Overall, the models project that the East Asian summer monsoon in the second half of the 21st century under a warming world will have a greater duration via a delay in the end of the summer monsoon season.
As for the seasonal changes in surface air temperature and SLP, the timing of when the pressure difference reaches zero (from negative to positive) will be delayed because of the large difference between the land and sea temperature rises in August and the large pressure drop over land. On the other hand, the land–sea temperature rise and pressure drop in May are relatively small, resulting in minor changes in the timing of when the pressure difference moves from positive to negative (zero), thereby generating the asymmetry in the response of the two seasons (i.e., spring and autumn).
The seasonal change in SST and its future change are small when compared with the land surface temperature; consequently, the seasonal future change in land temperature is dominant. Recent studies have attributed differences in the land–sea surface temperature rise to differences in evaporation from the ground and sea surface in a warming world. Evaporation from the sea surface is possible at any rate, but on land there is a limit to the amount of water held in the soil and available for evaporation. Therefore, the latent heat flux at the land surface is limited, by increasing the sensible heat flux. Furthermore, the rising temperature results in a decrease in relative humidity on land and has a feedback effect that strengthens the elevated temperature contrast (Byrne and O’Gorman 2016; Chadwick et al. 2016). Kamae et al. (2014) pointed out that land surface heating related to increasing CO2 levels is an important contributor to the increase in the continental land–sea thermal contrast and changes in atmospheric circulations during the summer months. Therefore, the temperature increase is greater in the late summer season when the land becomes drier.
Importance of surface moisture conditions during the monsoon transition seasons was discussed by Seth et al. (2011) with respect to future changes in the annual cycle of monsoon precipitation in the CMIP3 models. They suggest spring drying as a barrier to monsoon onset, whereas summertime wetting favors a prolonged monsoon. With the data available to us and analyzed in this paper, it was not possible to investigate why the increased warming occurs during late-summer rather than mid-summer. Previous observational and modeling studies have investigated the impact of land surface conditions on the atmosphere; i.e. drying the land surface during summer (e.g., Seneviratne et al. 2013; Vogel et al. 2017; Cook et al. 2020), although most of them assessed the seasonal mean responses. Other studies considered the positive feedback between soil moisture and the atmosphere throughout the season have raised the possibility of a lagged response (e.g., Ramarao et al. 2016).
Using the CMIP6 model results, Ito et al. (2025) investigated the seasonal changes in the SLP pattern around Japan. They found a delay in the seasonal progression from summer to autumn (July–September) that is consistent with our study. Ito et al. (2025) also noted an earlier seasonal progression from winter to spring (January–April); however, the season they used was slightly different to the period focused on in this study (around May). Our study investigates the continental-scale monsoon circulation around Eurasia, whereas the SLP changes around Japan reported by Ito et al. (2025) were influenced by changes in the cold frontal jet and storm track activity.
The timing of the sign-reversal of the large-scale land–sea sea-level pressure difference should affect the regional seasonal monsoon rainfall cycle, but the regional aspect of the onset and retreat of regional monsoon rainfall cycle would be different from the large-scale circulation changes. There are numerous studies on the progression of regional monsoon rainfall onset and retreat and their future changes (Kitoh et al. 2013; Moon and Ha 2020; Hu et al. 2024). Modulations in the timing of the monsoon circulation transition affect various sectors, including agriculture, and further research is needed into climatic changes during this period of high model uncertainty. This paper has discussed only the differences in land SLP and global warming rate with respect to inter-model differences related to the seasonal progression in spring and autumn, and further investigation on rainfall, ground wetness, atmospheric stability and others factors is needed. Different aspect between large-scale circulation changes and regional rainfall changes should be investigated. How differences in future aerosol concentrations would affect the timing of the seasonal transition is also remains open to question, particularly with respect to the SSP3-7.0 experiment, a scenario with a small future decrease in aerosol concentrations (Shiogama et al. 2023). Furthermore, the change in the spring transition was found to be close to zero in the model ensemble average, but uncertainty among the models was very large. We await with interest the forthcoming CMIP7 model results, which will have improved physical process parameterizations and reduced model bias.
The CMIP6 dataset is publicly available at https://esgf-node.llnl.gov/projects/cmip6/. The JRA-55 dataset can be accessed from https://search.diasjp.net/en/dataset/JRA55, the JRA-3Q dataset is available from https://search.diasjp.net/en/dataset/JRA3Q, and the ERA5 dataset is available from https://cds.climate.copernicus.eu.
The Supplementary Material comprises Figs. S1–S3 and Table S1.
Figure S1. Monthly mean surface air temperature changes for each month between the periods 1991–2020 and 2070–2099 based on the CMIP6 MME.
Figure S2. As in Figs. 4a and 4b, but for six selected CMIP6 models using pentad mean data. From left to right: CESM2, CNRM-CM6-1, IPSL-CM6A-LR, MIROC6, MPI-ESM1-2-LR, and MRI-ESM2-0.
Figure S3. (a) Number of models (among CESM2, CNRM-CM6-1, IPSL-CM6A-LR, MIROC6, MPI-ESM1-2-LR, and MRI-ESM2-0) for which the SLP difference between land and sea regions changes from positive to negative during spring when comparing the present period (1991–2020 mean) with the end of the 21st century (2070–2099 mean) using monthly mean data. On the horizontal axis, positive values indicate a delay. (b) As for (a), but for when the SLP difference changes from negative to positive during autumn. (c) and (d) As for (a) and (b), but using pentad mean data.
Table S1. CMIP6 models used in this study.
This work was funded by the Research on Attribution and Projection of Climatic/Global Environmental Change of the Meteorological Research Institute, by the Environment Research and Technology Development Fund (JPMEERF20242001) of the Environmental Restoration and Conservation Agency provided by Ministry of the Environment of Japan, and JSPS KAKENHI Grant JP21K13154. This work was also supported by the MEXT-Program for the Advanced Studies of Climate Change Projection (SENTAN) Grant Number JPMXD0722680734. We acknowledge the World Climate Research Programme, which, through its Working Group on Coupled Modelling, coordinated and promoted CMIP6. We thank the climate modeling groups for producing and making available their model output, the Earth System Grid Federation (ESGF) for archiving the data and providing access, and the multiple funding agencies who support CMIP6 and ESGF. We also thank Osamu Arakawa for archiving and re-gridding the CMIP data onto the 2.5° × 2.5° grid.
