Inter-model variability of the CMIP5 future projection of Baiu, Meiyu, and Changma precipitation

Many studies have suggested that mean precipitation associated with the East Asian Summer Monsoon (EASM) will be increased by the ongoing global warming, but its quantitative projection by climate models has large variability, with some models suggesting even decreases. We investigate the inter-model variability of projected centennial changes of the EASM separately for Baiu over Japan, Meiyu over eastern China, and Changma over Korea by using monthly-mean model outputs provided by the Coupled Model Intercomparison Project Phase 5 (CMIP5) project. Results with the Representative Concentration Pathway 4.5 and 8.5 scenarios (RCP4.5 and RCP8.5) are consolidated by normalizing with the global-mean near-surface air temperature changes. For all the three EASM land regions, inter-model differences in the mean precipitation changes are positively correlated with the southerly moisture flux changes to the south of the regions. The correlation is highest in June among the June-to-August months, whose reason may be because precipitation in early summer relies on large-scale southerly moisture transport. These changes are localized and nearly independent among the three regions where Baiu, Meiyu and Changma occur. The low-level southerly change to the south of Japan, which affects the Baiu precipitation change, is positively correlated with upper-tropospheric meridional wind to its north; it further exhibits a stationary Rossby-wave feature associated with the Silk-Road teleconnection. This study suggests that future changes in the EASM mean precipitation depend on circulation changes and more-or-less localized.


Introduction
The east Asian summer monsoon (EASM) is characterized by the rainy areas formed in southern China (Meiyu) and Japan (Baiu) in June and July and that in Korea (Changma) in late June to July (e.g., Wang and LinHo 2002). The moisture to precipitate is supplied by the flow on the western periphery of the North Pacific Subtropical High (NPSH), and the rainy areas are situated to the south of the climatological upper-tropospheric jet axis (Fig. 1). How the EASM will change under the ongoing global warming is of great interest.
Based on physical considerations and data from the Coupled Model Intercomparison Project (CMIP) Phase 3 (CMIP3) climate models, Held and Soden (2006) provided a clear explanation on the future projection of large-scale precipitation changes due to global warming as follows. Lower tropospheric specific humidity tends to follow the Clausius-Clapeyron (CC) scaling, which is to increase by about 7% for each 1-K increase in temperature. The global precipitation change is rather governed by the changes in the net downward radiation at the surface and in the Bowen ratio. From the models, global precipitation was found to increase by about 2% for each 1-K increase, which is much smaller than the lower-tropospheric humidity increase by the CC scaling. These thermodynamic constraints further provide an important consequence applicable to oceanic or well-watered land regions: while the overturning or vertical atmospheric circulation slightly decreases, horizontal 1 3 water-vapor transport increases owing to the humidity increase, so precipitation increases (decreases) over regions with relatively large (few) precipitation. This is called a "wet-gets-wetter" change.
Monsoons occur over oceanic and well-watered land regions, so they are on the "wets" to get wetter. Based on the projection by CMIP Phase 5 (CMIP5) climate-model simulations, Christensen et al. (2013) concluded that, by the end of twenty-first century, the global monsoon area is very likely to increase and the global monsoon precipitation intensity is likely to increase, while the models exhibit decreasing trends in the lower-tropospheric wind convergence.
Multi-model analysis with CMIP3 and CMIP5 models have indicated that the EASM precipitation will increase (e.g., Min et al. 2004;Kripalani et al. 2007; Kusunoki and Arakawa, 2012;Christensen et al. 2013;Li et al. 2015), while the circulation intensity change is diverse among models (Christensen et al. 2013;He and Zhou 2015). Also indicated by many models is that the mid-summer precipitation depression around Japan will be weakened, which can be interpreted as a delay of the Baiu withdrawal (e.g., Kitoh and Uchiyama 2006;Inoue and Ueda 2012). As for the location, Seo et al. (2013) showed that the multi-model ensemble-mean (MME) based on the Representative Concentration Pathway (RCP) 6.0 experiments of CMIP5 exhibited little meridional shift of EASM precipitation from the late twentieth century to the late twenty-first century. Possible future changes of the EASM are also investigated by many other studies (e.g., Li et al. 2011;Freychet et al. 2015;Chen and Bordoni 2016;Chen et al. 2020;Zhou et al. 2020;Jiang et al. 2021;Endo et al. 2021). A review is presented by Kitoh (2017).
The Baiu precipitation occurs both over land and ocean (Fig. 1ab). The region definitions used to study possible future changes in Baiu differ among studies, such as 120° E-140° E, 20° N-40° N in Kitoh and Uchiyama (2006) and 125° E-145° E, 30° N-37.5° N in Inoue and Ueda (2012). In both cases, the areal faction of ocean is much greater than that of land. Differences among the conclusions of the preceding studies are partly attributed to the region definition differences, in addition to the differences in the models used and the analysis methods. Also, some studies conducted analyses by using circulation-based monsoon indices (e.g., He and Zhou 2015, who used a 850-hPa zonal wind difference between 10°-20° N and 25°-35° N), so direct comparison with those studies centered on precipitation variability is sometimes difficult. In this study, we limit our scope predominantly to land, considering possible application studies such as on adaptation.
Large-scale (such as planetary scale or continental/basin scale) projections based on thermodynamic constraints such as the ones suggested by Held and Soden (2006) are considered to be robust, but projections on smaller scales or on dynamical aspects have fewer consensuses. It has been reported by numerous studies and the multiple generations of the Intergovernmental Panel on Climate Change (IPCC) Assessment Reports (e.g., Christensen et al. 2013) that great differences exist among models in terms of regional climate projections on precipitation and circulations.
A traditional way to utilize the differences is to assess the reliability of the projection as an MME change from intermodel consensus. Recently, however, the limitation of this approach has been recognized in a sense that MME does not indicate plausible variability in future climate changes. An alternative strategy thus emerged is called the storyline approach, which can be defined as "a physically self-consistent unfolding of past events, or of plausible future events or pathways" (Shepherd et al. 2018). A recognition underlying the storyline approach is that the state-of-the-art coupled general circulation models (CGCMs) have matured and thus their variability is worth treating as plausible, at least after excluding the ones whose performance was found to be poor.
As for the future projection of the EASM, Horinouchi et al. (2019;H19, hereinafter) investigated, by using the RCP4.5 and RCP8.5 model outputs from CMIP5, the relation between the projected changes in the Asian jet, which is the summertime upper-tropospheric subtropical jet over Eurasia to the Western Pacific, and the projected changes in precipitation over East Asia for the June to July period. They showed that the inter-model differences in the latitudinal shifts of the jet axis and the Baiu-peak latitude are correlated, so future projection of precipitation peak latitude might be improved by improving the projection of the jet shift. They suggested that the jet-precipitation relationship is a remnant from their well-established synoptic relations (e.gKodama 1993;Horinouchi 2014;Horinouchi and Hayashi 2017). They also showed that a similar jetprecipitation relationship was not found for Meiyu, and they speculated that the reason might be because the Meiyu reproduction in CMIP5 models are rather poor, at least in terms of the spatial distribution. Jet-precipitation relation in climate projections for the EASM has also been studied by Chiang et al. (2019) and Li et al. (2021) by using a single model.
Even though H19 provides useful viewpoints on future EASM projection, it is difficult to draw useful implications for those who seek information on the future change of precipitation over land, which is an Eulerian view point. In this study, we investigate the future projection of precipitation in more direct ways in terms of changes over land regions affected by Baiu, Meiyu, and Changma. As H19, we examine inter-model variabilities to seek for robust dynamical relationships. Results are shown extensively for Baiu, on which a number of novel and interesting large-scale relationships are found. Our results for Meiyu and Changma are similar to those for Baiu in a local aspect, but large-scale relationships are different. Such contrasts elucidate similarity and differences among the EASM future projections.
We use the same CMIP5 model outputs as used in H19. Investigation with the CMIP Phase 6 (CMIP6) models will be conducted near future.
The rest of the paper is organized as follows. The data used and analysis methods are introduced in Sect. 2. The results are presented for Baiu, Meiyu, and Changma in Sects. 3, 4, and 5, respectively, which highlight the role of the variability in moisture flux and possible teleconnection associated with it. Conclusions are drawn and discussed in Sect. 6. Jet-precipitation relationship is examined in Appendix. This is because we did not obtain clear-cut results, but it will still be of interest, given the study history introduced above.

Data and methods
We used monthly-mean outputs from CMIP5 CGCMs. The modelling categories used are the historical, RCP4.5, and RCP8.5 runs. As in H19, we used the following 32 models: ACCESS1-0  Kirtman et al. (2014) and Collins et al. (2014) for the summary of the models. All the CMIP5 data are used after horizontally interpolating onto a 2.5° × 2.5° grid by using the bilinear interpolation. The available pressure levels up to 100 hPa are 1000, 925, 850, 700, 600, 500, 400, 300, 250, 200, 150, 100 hPa. Since the number of the ensemble runs in our re-gridded dataset (prepared for our project stated in acknowledgement) are very limited, being only one for most models, we simply used one ensemble run for each model. The results thereby depend on the internal variability of each model. This aspect is treated to some extent by comparing RCP4.5 and RCP8.5 runs, as described in Sect. 3. There, we will show that the effect of the internal variability to our conclusions is limited. In this study, we analyze long-term means over 1979-1999 (historical runs) and 2079-2099 (RCP runs) for each of June and July. We also present results for August as well for comparison and also to cover boreal summer. Although many studies on EASM projection use seasonal mean over June to August, we treat the three months separately to crudely resolve sub-seasonal time evolutions. We shall call the means over 1979-1999 and 2079-2099 as the present climatology and the future climatology, respectively. The difference between the averages of a quantity over the two periods is called as the centennial change for each model run, which will be indicated by the symbol Δ (e.g., ΔP for mean precipitation P ). The fractional change of P in percentage is defined as ΔP(%) ≡ 100 × ΔP∕P(present) , where P(present) is P for the present climate for each model run.
In most analyses, which are based predominantly on correlations and linear regressions, we consolidate the RCP4.5 and RCP8.5 results by normalizing changes (from historical to RCP results) with the change in the globalmean near-surface air temperature for each model and scenario, which will be referred to as ΔT g . For such cases, MME means are defined as the mean over the 32 models across the two scenarios, which consists of 64 samples. According to the Student t test, the 99% significance level against the null hypothesis for the correlation coefficient r Fig. 2 The MME present climatology (historical, 1979-1999) of precipitation (color shading) and its projected change as the future (RCP4.5, 2079-2099) minus the present (contours with an interval of 0.25 mmday −1 ) for a June, b July, and c August. The red "+" marks indicate the ten 2.5° × 2.5° grid points from which the Baiu region precipitation is derived. d-f As in a-c but for RCP8.5 is |r| ≥ 0.320 ( r 2 ≥ 0.109 ) when the ensemble size n is 64, and it is |r| ≥ 0.449 ( r 2 ≥ 0.201 ) when n = 32 . The 95-% significance level is |r| ≥ 0.246 ( r 2 ≥ 0.061 ) when n = 64 and |r| ≥ 0.349 ( r 2 ≥ 0.122 ) when n = 32 . In our quantitative evaluation, we mainly use r 2 , since it is a definition of the coefficient of determination, a measure on how much the variability of one factor may be attributed to the relationship with another. Simple linear regression of a quantity y onto another quantity x is expressed by the variability of y per the standard deviation of x , x , i.e., the regression coefficient multiplied by x , or the correlation coefficient between x and y multiplied by the standard deviation of y.
Before proceeding to the results Sects. 3-5, we briefly mention the relationship between the RCP4.5 and RCP8.5 projections for same models. As seen in Fig. 3d-i, if the model is the same, ΔP Baiu (or ΔP Baiu (%) ) normalized by ΔT g are often close to each other between RCP4.5 and R C P 8 . 5 . Q u a n t i t a t i v e l y , t h e M M E o f , where the superscripts designate the scenarios, is 0.53 times its expected value when the results for the two scenarios are independent (2 times the variance of ΔP Baiu ∕ΔT g across the two scenarios); the ratio is 0.51 for ΔP Baiu (%)∕ΔT g . It suggests that the results for the two scenarios are partially independent, so the effective degree of freedom is between 64 and 32. The statistical significance shown in this figure is based on the 99% level assuming the full independence among the 64 results, but it can be conservatively interpreted as actually being around 95% (see Sect. 2). The same argument applies to other quantities such as moisture flux (ordinate of Fig. 3d-f). The closeness between the RCP4.5 and RCP8.5 results for each model gives some justification to use only one ensemble member for each model and scenario; in our analysis results, the effect of the internal variability in each model is likely suppressed to some degree. Figure 2 shows the MME precipitation and its future centennial change based on the RCP4.5 scenario for each of June, July, and August; August is after the Baiu withdrawal, but it is shown for reference. The present MME climatology of Baiu precipitation is weaker than the CMAP observational estimates (Fig. 1a, b), but the overall distribution is not unrealistic in a crude sense. The future change is positive as pointed out by many studies (Sect. 1), and it is distributed to shift the Baiu precipitation southward (H19), which is evident in June (Fig. 2a).

Relationship with moisture flux and low-level flow
We conducted a variety of correlation analyses and found that the inter-model variability of the future change of precipitation in early summer is tightly correlated with the intermodel variability of the future change of the moisture flux impinging on the region of interest, which is dominated by low-level flow variability. Here we show the results for the Baiu precipitation. Figure 3a-c shows that the inter-model variation of the centennial change of the Baiu region precipitation is positively correlated with that of the change in the northward component of the vertically integrated water-vapor flux, q v , to its south. In other words, the more the moisture supply to the region increases, the more the precipitation increases. The spatial extent of the positively and significantly correlated regions (yellow) reflects the shape of the western periphery of the NPSH for each month as shown in Fig. 1g-i. The correlation is particularly high in June, with the peak r 2 value around 0.6 ( Fig. 3a), indicating a tight correlation as shown in Fig. 3d and g for the box A (127.5° E-137.5° E, 25° N-30° N); this box was chosen to focus on the maximum correlation in June. The MME centennial change of the Baiu region precipitation ( ΔP Baiu (%)∕ΔT g ) is around 5%/K (Table 1). It varies widely among models at the standard deviation of 5%/K (Table 1), spreading between the maximum around 20%/K and the minimum around −10 %/K (Fig. 3d). As for ΔP Baiu ∕ΔT g , it spreads between 1.5 mm/day/K and −0.5 mm/day/K (Fig. 3g). The inter-model spread of the Baiu region precipitation change percentage for July is similar to that for June, but it is smaller for August ( Fig. 3d-f).
If Fig. 3a-c is drawn by using the eastward component of the vertically integrated water-vapor flux, q u , instead of q v , we obtain similar results around the Baiu region but with smaller r 2 values (not shown). The upstream extension of the region of positive significant correlation is also weaker. The extension is somewhat deflected toward west compared with the q v case, which is natural because q u is associated with the moisture supply from the west. These results suggest that we can evaluate the effect of the moisture flux variability mainly with its northward component. Note that a somewhat similar result was obtained by Ito et al. (2020), who investigated inter-model correlations in future centennial changes in seasonal mean climatology by using CMIP5 models and showed that the mean precipitation changes in the June-July-August period over Japan and surrounding ocean is positively correlated with the southwesterly wind changes over the same region.
In the correlation maps of Fig. 3a-c, the peak values in July ( Fig. 3b; r 2 peaking at around 0.4) and August ( Fig. 3c; r 2 peaking at around 0.2) are smaller than in June (Fig. 3a).

Fig. 3
Result of the inter-model correlation analysis across scenarios (RCP4.5 and RCP8.5) between the centennial changes in the Baiu region precipitation ΔP Baiu (a single scalar) and the centennial changes in the vertically integrated northward moisture flux Δq v (ac: at each 2.5° grid point; d-i: areal mean). a-c Squared correlation coefficient r 2 (contours with an interval of 0.1) and their 99% significance (color-shading for where r 2 ≥ 0.109 ; yellow: r>0, and cyan: r < 0 ) between the future changes (1979-1999 to 2079-2099) in the Baiu region precipitation ΔP Baiu normalized by the global-mean near-surface air temperature change ΔT g ( ΔP Baiu ∕ΔT g ) and the future change in Δq v ∕ΔT g for a June, b July, and c August. d-f Scatter diagram between Δq v ∕ΔT g averaged over the box A (indicated in a-c) and ΔP Baiu (%)∕ΔT g for d June, e July, and f August. The models are indicated by alphabets (Sect. 2), and the scenarios are indicated by colors (red: RCP4.5, blue: RCP8.5). For each month, the MME values (over the 64 points across the two scenarios) are indicated by the black cross mark, and the correlation coefficient is shown at the bottom. g-i As in d-f but for using ΔP Baiu instead of ΔP Baiu (%) The highest correlation in June among the three months is expected, because sea-surface temperature (SST) around Japan in June has not risen to mid-summer levels, so the precipitation is more dependent on the moisture supply by the southerly flow around the NPSH, as indicated by the largest climatological discrepancy between precipitation (Fig. 1a) and evaporation (Fig. 1b).
The vertically integrated moisture flux change normalized by ΔT g is highly correlated with the 850-hPa meridional wind ( V 850 ) change (Fig. 4). This is because the water vapor is mainly distributed in the lower troposphere and because the effect of specific humidity change on moisture flux change is limited; note that specific humidity change is dominated by SST change, so its effect is reduced by the normalization by ΔT g . Figure 3 changes only slightly if created by using V 850 instead of q v (not shown). The regions of positive significant correlations shown in Fig. 3a, b (shaded in yellow) is much smaller than the extent of the NPSH. This is the case if the correlation map is created using 850-hPa geopotential height Z 850 instead of V 850 (not shown). In other words, the projected Baiu precipitation change depends on local changes in NPSH. We conducted broad investigation and found two factors associated with it, as described in what follows (Sects. 3.2, 3.3).
Since the moisture flux is a multiplication of velocity and humidity, one may question that we should also investigate the effect of humidity changes in the moisture flux changes. However, doing so is not always fruitful, because the humidity changes are affected by flow changes though advection and evaporation (dependent on wind speed). Fortunately, there is a tight inter-model correlation between the vertically integrated moisture flux changes and the low-level flow changes for the region of interest (Fig. 4). This makes interpretation simple as argued above.

Role of the Silk-Road teleconnection and Arctic variability
In Sect. 3.1, we showed the correlation between the intermodel variability in low-level southerly flow change and that in Baiu precipitation change, whose causality should be from the former to the latter. Here we show its possible teleconnection in the extratropics to affect the variability in the southerly flow. The inter-model variation of the Baiu region precipitation change has significant correlation with that of the 200-hPa Table 1 The mean (MME) and the standard deviation (denoted as σ) of the normalized precipitation changes ΔP Baiu (%)∕ΔT g , ΔP Meiyu (%)∕ΔT g , and ΔP Changma (%)∕ΔT g for each of June, July, and August. The runs with the two scenarios RCP4.5 and RCP8.5 are consolidated, so the sample number is 64 Mean (%K −1 ) 4.9 6.1 4.0 3.5 1.0 3.1 4.6 4.8 4.7 σ (%K −1 ) 5.4 6.5 5.0 5.5 8.5 7.8 6.5 6.5 6.0 Fig. 3d-f but for the 850-hPa meridional wind ( V 850 ) and the vertically integrated meridional moisture flux ( q v ) averaged in the box A meridional wind, V 200 , locally over Japan as shown in Fig. 5; the peak r 2 value is the greatest in June, being greater than 0.3 (Fig. 5a). Also, significant correlation with V 200 changes is found broadly in June and July; the distribution of the regression of ΔV 200 ∕ΔT g at each location onto ΔP Baiu ∕ΔT g exhibits wavy distribution from subtropics to midlatitude (Fig. 5a,  b), which resembles the meridional wind anomaly patterns associated with the Silk-Road teleconnection (Enomoto et al. 2003). This teleconnection occurs as stationary Rossby waves whose energy propagates downstream along the subtropical jet. Although the teleconnection induces spatial pattern of temporal variability (e.g., Kosaka et al. 2009), the stationary wave feature also exists in the meridional-wind climatology of June (Horinouchi et al. 2021; Fig. 6a), which is well reproduced in the historical MME (Fig. 7a). The wave-train-like correlation distribution for June (Fig. 5a) is present along this climatological wave, and their longitudinal phases are similar. In other words, the Baiu region precipitation tends to be increased more than the MME changes in the models in which the MME present climatological Silk-Road pattern is enhanced. Note that, for both RCP4.5 and RCP8.5, the MME changes of the mean meridional winds (Fig. 7c, e) are in quadrature to shift the climatological wavy feature (Fig. 7a) slightly to the east. These results are interpreted as follows: there is a wide inter-model variability in the longitudinal phase changes in the future meridional winds, and among them, the Baiu region precipitation increase is selectively enhanced in the models in which the upper-tropospheric southerly around Japan is enhanced. Because of the wave guide along the subtropical jet, such a change tends to have an origin in the upstream. The normalized 200-hPa meridional wind change ( ΔV 200 ∕ΔT g ) over Japan in June is more correlated with ΔV 850 ∕ΔT g in the box A (Fig. 8a) than with ΔP Baiu ∕ΔT g (Fig. 5a). The high correlation (with the r 2 peak greater than 0.4) is not surprising because the upper-tropospheric and lower-tropospheric circulation are coupled, and the meridional cross-section of horizontal wind fields tend to be tilted toward north with altitude, as explained in the next paragraph. The fact that the correlation between ΔV 200 ∕ΔT g and the box-A ΔV 850 ∕ΔT g is higher than the correlation between ΔV 200 ∕ΔT g and ΔP Baiu ∕ΔT g indicates that the latter correlation is induced indirectly via the former correlation through the enhancement of southerly moisture flux. In other words, the appearance of the Silk-Road pattern in Fig. 5 is interpreted as being caused passively by the required southerly enhancement in the downstream (over Japan) even though the teleconnection is dynamically originated in the upstream.  The vertical coupling mentioned above can be explained as the tilting of jets along quasi-stationary fronts as demonstrated by the numerical experiments by Chou et al (1990) conducted to understand dynamics of Meiyu fronts; thermally direct secondary circulations create low-level jets to the south of upper-tropospheric jets. Examples for Baiu cases in reanalyses are found in Fig. 3 of Horinouchi et al (2021) for zonal winds; see Horinouchi and Hayashi (2017) for jet meanders to affect meridional winds. This is basically synoptic features, but H19 showed that such synoptic relationship tends to remain to much longer time scales to even affect inter-model climatological variability.

Fig. 4 As in
Another interesting feature in the correlation maps for ΔP Baiu ∕ΔT g in June (Fig. 5a) or ΔV 850 ∕ΔT g in the box A in June (Fig. 8a) is the positive correlation with ΔV 200 ∕ΔT g at around the Barents Sea and the Kara Sea. The regression pattern is similar whether only the RCP4.5 runs are used or only the RCP8.5 runs are used (Fig. 8b, c), but the signal over the Barents and Kara seas are weaker in the RCP8.5 case. We speculate that this Arctic signal might be induced by changes in the distribution of the Arctic sea ice; if this is the case, its effect may not be linear with ΔT g because of the nonlinearity inherent in sea-ice cover (which cannot decrease beyond zero). However, to elucidate it would require further investigation. It is currently unclear how the Arctic changes propagate to around Japan, but one can hypothesize two candidate processes: a direct propagation over Siberia more-or-less along the great circle, and/or the wave-train connection change at around 70° E to stimulate the Silk-Road teleconnection (Ding and Wang, 2005;Kosaka et al. 2012;Horinouchi et al. 2021). Fig. 7 a, b The MME present climatology of V 200 , c, d its centennial change with the RC4.5 and e, f the RC8.5 scenarios (a, c, e: for June; b, d, f: for July). The changes c-f are shown only where they are more significant than by 99% (in terms of the sample number of 32) Fig. 8 a As in Fig. 5a, but for using V 850 ∕ΔT g averaged over the box A instead of ΔP Baiu ∕ΔT g . b, c as in a but for limiting the scenario to RCP4.5 and RCP8.5,respectively It is also noteworthy that the correlation coefficients over Japan is similar between RCP4.5 and RCP8.5 in Fig. 8b, c, but the regression coefficients are smaller in the latter, which indicates some nonlinearity with ΔT g . The nonlinearity might be because dynamical response to global warming can be saturated or limited. Again, however, it would require further studies to elucidate possible nonlinearity as for the Silk-Road teleconnection too.
The Silk-Road pattern is also present in the regression of ΔV 200 ∕ΔT g at each grid point onto ΔP Baiu ∕ΔT g in July (Fig. 5b). Thus, a similar interpretation can be made. However, it appears that a caution is needed, since the correlation is weaker than in June, and the positive correlation at around India is even more southward (centered around 80° E, 30° N), which is not on the subtropical jet in July situated around 40° N (Fig. 14b). The correlation features are more localized in August, without exhibiting a hint of large-scale teleconnection (figure not shown).

Meiyu
The present climatology of the CMIP5 MME precipitation in eastern China (here, 110° E-120° E) in June and July peaks at lower latitude than in the observations (Figs. 1a,b,  2a, b), so it appears that the CMIP5 projection for the Meiyu precipitation changes may have severe limitations. Nonetheless, we present the results of our analyses similar to those conducted for Baiu, since we believe that to document them have some importance to be compared in future studies with more advanced models. We also found it useful to investigate Meiyu and Changma (Sect. 5) to gain insight on the locality in the EASM changes. Note that the projected MME future change (contours) peaks at higher latitude than that of the present peak in June with the RCP4.5 scenario (Fig. 2a), indicating a northward shift of Meiyu in June; whereas, a contrasting result by Seo et al. (2013) was obtained from analyses of the mean over the June to August period. The northward shift is also found in the case with the RCP8.5 scenario, but it is only slightly (Fig. 2d).
The grid points to define the Meiyu region precipitation P Meiyu over 115° E-120° E and 25° N-30° N (Sect. 2) are shown in Fig. 9. Its centennial change ΔP Meiyu normalized by ΔT g has positive correlation with the northward moisture flux to the south of the region throughout summer (Fig. 9), exhibiting the highest correlation in June (Fig. 9a). This result is similar to the Baiu case (Sect. 3). The high correlation in June, again, is consistent with the fact that the EASM precipitation in early summer relies heavily on large-scale southerly moisture transport, which is evident from the apparent deficit of evaporation against precipitation over Japan and the nearby ocean to its south (Fig. 1a); the relatively low evaporation in June as compared with July and August is consistent with the fact that the SST around Japan is lower in June than in mid-summer. As in the Baiu case (Sect. 3), we obtain correlation maps similar to Fig. 9, if lower-tropospheric meridional wind is used instead of vertically integrated southerly moisture flux (not shown).
If Fig. 9a-c is drawn by using the eastward component of the vertically integrated water-vapor flux q u instead of q v , we obtain similar results around the Meiyu region (not shown). The upstream extension is somewhat deflected toward west compared with the q v case, as in the Baiu case (Sect. 3).
As shown in Table 1, the standard deviation of ΔP Meiyu (%)∕ΔT g is 5.5 to 8.5 %K −1 depending on the months. They are greater than the MME changes (1-3.5 %K −1 ), so the change is negative in some models. It also means that MME changes are not statistically significant when measured from inter-model variability.
The distribution of the regions of significant correlations have few overlap between Fig. 3a-c and Fig. 9, which indicates that the meridional flow changes associated Fig. 9 As in Fig. 3a-c, but for using ΔP Meiyu instead of ΔP Baiu with Baiu and Meiyu changes are uncorrelated. Actually, the correlation coefficient between ΔP Baiu (%)∕ΔT g and ΔP Meiyu (%)∕ΔT g for June is 0.32 (i.e., r 2 = 0.10 ). Unlike the Baiu case, the Meiyu precipitation changes do not exhibit large-scale correlation features suggesting the effect of the Silk-Road teleconnection (Fig. 10).
To summarize, as for the projected centennial changes in Meiyu, we have not found large-scale dynamical features beyond the correlation with the southerly moisture flux to the south of the region. This might be due to the poor reproduction of Meiyu in the CMIP5 models, but it might also be because the region is more away from the subtropical jet than the case of Baiu.

Changma
The climatological Changma period is late June to July (Sect. 1). In the monthly climatology, precipitation peak over the Korean Peninsula is found in July and August (Fig. 1b,   c). The CMIP5 MME precipitation crudely reproduces the peak over the Korean Peninsula in July (Fig. 2b), although it is weaker than in the observation. The August peak is obscure in the MME (Fig. 2c).
The grid points to define the Changma region precipitation P Changma over 125° E-127.5° E and 35° N-40° N (Sect. 2) are shown in Fig. 11. Its centennial change ΔP Changma has the standard deviations around 6 %K −1 for the three months when expressed in percentage and normalized by ΔT g , which is slightly greater than the MME change around 5 %K −1 ( Table 1).
The precipitation change has positive correlation with the northward moisture flux to the south of the region throughout summer (Fig. 11), exhibiting the highest correlation in June (Fig. 11a), as in the cases of Baiu (Sect. 3) and Meiyu (Sect. 4). Also, the correlation is relatively large in August (Fig. 11c). In the Changma period of July, the correlation is significant but not tight, with r 2 peaking at around 0.3. The regions with significant positive correlations (yellow shading) have little overlap between the Baiu and Changma cases for July and August   Figs. 3b,c,11b,c), while there is some overlap in June (Figs. 3a, 11a). Consistently, the correlation between ΔP Baiu (%)∕ΔT g and ΔP Changma (%)∕ΔT g are nearly zero for July and August ( |r| < 0.1).
If Fig. 11a-c is drawn by using the eastward component of the vertically integrated water-vapor flux q u instead of q v , we obtain similar results around the Changma region with smaller r 2 values (not shown). The upstream extension is weaker and somewhat deflected toward west compared with the q v case, just like the Baiu case (Sect. 3).
The correlation between ΔP Changma ∕ΔT g and ΔV 200 ∕ΔT g at each grid point has a wave-like feature in June (Fig. 12a). The wave-like feature extends zonally as in the case of Baiu (Sect. 3), but the signal is much weaker. The correlation coefficient between ΔP Baiu (%)∕ΔT g and ΔP Changma (%)∕ΔT g for June is only 0.20 ( r 2 = 0.04 ). However, the spatial distributions in Figs. 5a and 12a have some similarity, presumably because the Baiu region and the Changma regions are not far (even though). For reference, the correlation coefficient between ΔP Meiyu (%)∕ΔT g and ΔP Changma (%)∕ΔT g for June is nearly zero.
Regions with significant correlation signals are fewer for July (Fig. 12b) than June (Fig. 13a), which is even more case for August (not shown). For these months, we do not find large-scale features that appear meaningful, which is similar to the case of Meiyu (Sect. 4).

Conclusions and discussion
We have investigated inter-model variability in the CMIP5 future projections on the centennial changes in the EASM. Our investigation was made separately for Baiu (over western and eastern Japan), Meiyu (over eastern China), and Changma (over the Korean Peninsula) for each of June and July; results for August were also shown in some cases to elucidate the intra-seasonal time evolution through summer and to cover the conventional June to August period. All changes are evaluated after normalization by the globalmean near-surface air temperature change for each model run. In most analyses, thereby, the changes according to the two scenarios of RCP4.5 and RCP8.5 are consolidated.
Our findings are summarized in Fig. 13. For all the three EASM land regions, the inter-model variations of the mean precipitation changes are positively correlated with that of the southerly moisture flux change to the south of the regions. For all the cases, the correlations are highest in June among the three months. Its reason may be because precipitation in June relies on large-scale southerly moisture transport around the western periphery of the NPSH, as the local SST is relatively low, and evaporation is fewer. There is little correlation among the inter-model variations of the precipitation changes over the three regions as shown in Sects. 4 and 5. The associated changes in the northward component of the vertically integrated moisture flux (as elucidated by correlation maps) are more-or-less localized. This The inter-model differences in the moisture flux changes are dominated by those in the low-level meridional wind changes. The inter-model variability in the low-level southerly change to the south of Japan, which affects the centennial Baiu precipitation change, exhibits large-scale teleconnection. It is positively correlated with the uppertropospheric meridional wind change to its north in June and July. This upper-tropospheric signal is on the subtropical jet, and the correlation maps suggest teleconnection along the jet, which form a waveguide for stationary Rossby waves, known as the Silk-Road teleconnection. Via this teleconnection, the Baiu precipitation change in June tends to be enhanced in the models in which the climatological wave train is enhanced. Association with Arctic variability is also found. On the contrary, the centennial Meiyu precipitation change does not exhibit large-scale teleconnection. There is a hint of a teleconnection in the centennial Changma precipitation change in June, as in the Baiu case, but the signal is much weaker.
Recently, Zhou et al. (2020) conducted studies on the dynamical variability of the future projection of the EASM by using empirical-orthogonal-function analyses for the June-July-August averages. While their approach is useful to elucidate large-scale seasonal features, the present study highlights the usefulness of simple correlation analyses to elucidate local sub-seasonal features and to further investigate their mechanisms. It should also be stressed that the inter-model variability in the EASM projection can be sensitive to the region and variable settings. Our study focused on precipitation mainly over land regions, while some others treated precipitation predominantly over the ocean or focused on indices on circulations (Sect. 1).
Storylines for the future change of the EASM according to the global warming may be described like the following; Fig. 13 can serve as their summary. As is well-known, the EASM precipitation will likely increase, which is explained as a consequence of thermodynamic constraints including the overall wet-gets-wetter effect. However, there can be a wide variety of changes depending on circulation. The future precipitation change will increase, if low-level southerly winds increase to the south of the region of interest, which is especially the case in early summer. The southerly wind change can be localized, so changes in Baiu, Meiyu, and Changma may not be similar to each other. As for the case of Baiu, it can have large-scale teleconnection to the Silk-Road wave train along the waveguide associated with the subtropical jet; the Baiu precipitation is enhanced if the climatological upper-tropospheric stationary wave is enhanced, which increases upper-tropospheric southerly wind over Japan to further increases low-level southerly to its south, increasing moisture supply.
This study was conducted with the CMIP5 simulations. We plan to use CMIP6 simulations to compare with the present study and to further extend it.

Appendix: Inter-model correlations between changes in precipitation and upper-tropospheric zonal winds
As introduced in Sect. 1, jet-precipitation relation has been highlighted in recent studies. We also investigated it in this study but did not obtain clear-cut results. This is presumably because we focused on precipitation mainly over fixed and limited regions (unlike in H19, for example). Here we show some of the results here for reference. Figure 14 shows historical MME zonal winds and jet axis for each model for reference. Figure 15 shows the inter-model correlation between ΔP Baiu ∕ΔT g and ΔU 200 ∕ΔT g , where U 200 is the zonal wind at 200 hPa. As for July, the normalized precipitation change ΔP Baiu ∕ΔT g is positively correlated with the normalized zonal wind change ΔU 200 ∕ΔT g around 35°N (Fig. 15b). The squared correlation coefficient r 2 peaks around the Yellow Sea and the Korean Peninsula with values around 0.3 (or r peaking around 0.6). Thus, the correlation is significant Fig. 14 The present (historical) climatology of the MME zonal wind at 200-hPa (color shading) and the jet axis for each model (dotted lines) against the null hypothesis, but it is not tight, explaining only ~ 30% of the variability of ΔP Baiu ∕ΔT g , when ΔU 200 ∕ΔT g is treated as the independent variable. This positive correlation is consistent with H19 as follows: the MME jet axis is around 40° N in July (Fig. 14b), so the zonal wind increase at around 35° N indicates the southward shift of the jet in many cases, and that tends to increase precipitation to the south.
As for August, the Baiu region precipitation change has only weak correlation with the upper-tropospheric zonalwind change around Japan (Fig. 15c). As for June, the correlation map is quite different from July (Fig. 15a). From The negative correlation between ΔP Baiu /ΔT g and ΔU 200 /ΔT g in June between 20° N and 30° N in Fig. 15a is explained in terms of the thermal wind balance from Fig. 9; this latitudinal band is where the uniformity in the upper-tropical temperature deteriorates as can be seen in the meridional gradient of the correlation coefficient shown in Fig. 9c. Figure 16 shows the inter-model correlation map for the upper-tropospheric zonal wind changes against the precipitation change over the Meiyu region. Unlike the Baiu case in Sect. 3, little correlation is found in June (Fig. 16a). Also, there is no significant correlation at around the subtropical jet in July and August (Fig. 16b, c). This result is consistent with H19 in a sense that no significant association was found among changes in the Asian jet and Meiyu precipitation. This result is in contrast with the inter-annual co-variation in jet and Meiyu precipitation reported by Li et al. (2021) by using a single model. interpolation onto the 2.5° grid was mainly done in the S-5-2 project of the Global Environment Research Fund, supported by the Ministry of the Environment, Japan.
Author contributions All authors contributed to the study conception and design. Material preparation and analysis were performed by TH. The first draft of the manuscript was written by TH and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.
Funding Environment Research and Technology Development Fund (JPMEERF20192004 and JPMEERF20222002) of the Environmental Restoration and Conservation Agency of Japan.

Data availability
The CMIP5 data are available from the Earth System Grid Federation server https:// esgf-node. llnl. gov/ search/ cmip5/.

Conflict of interest
The authors have no relevant financial or non-financial interests to disclose.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.