Leading source and constraint on the systematic spread of the changes in East Asian and western North Pacific summer monsoon

With ongoing global warming, the changes in EA-WNPSM rainfall—feeding over two billion people in East Asia and the Indochina Peninsula—projected by the Coupled Model Intercomparison Project Phase 5 (CMIP5) models show remarkable and unidentified inter-model spread. Here, we reveal the leading inter-model spread of EA-WNPSM changes in 28 CMIP5 models is related to a ‘dry north–wet south’ dipole in East Asia and a wet Indochina and WNP. This spread pattern of EA-WNPSM changes is induced by the spread of sea surface temperature changes in the equatorial western Pacific, and can be further traced back to an apparent discrepancy among the state-of-the-art models in simulating the tropical Pacific rainfall. An air-sea coupling processes involved with summer background circulation contribute to this robust spread pattern of EA-WNPSM changes. We can constrain the EA-WNPSM rainfall changes based on the current-future relationship and observation that there should be more rainfall increase in North China and the Korean Peninsula and less increase in South China, the Indochina Peninsula and WNP, relative to previous multi-model ensemble projections.


Introduction
East Asian and western North Pacific summer monsoon (EA-WNPSM) is an essential part of the Asian-Australian monsoon system (Matsumoto 1997, Wang and Lin 2002, Ding and Chan 2005, Huang et al 2007, Wu et al 2012, Hsu et al 2014. Land-sea thermal contrast and Tibetan Plateau create the EA-WNPSM (Wu et al 2012), drawing abundant moisture from ocean to East Asia and Indochina Peninsula. Meanwhile, anomalous EA-WNPSM activities can induce droughts and floods, causing considerable economic losses and ecological damage (Huang et al 2007). Besides the natural fluctuations (Wang et al 2000), human influences such as the anthropogenic greenhouse-gas and aerosol emissions also play an important role in regulating the variation of EA-WNPSM from the recent decades (Menon et al 2002, Kitoh et al 2013. Under global warming, the increase in atmospheric water vapor is believed to be responsible for an intensified global hydrological cycle (Held andSoden 2006, Durack et al 2012). Constrained by the radiative budget balance, weakened atmospheric circulation slows the increase in rainfall (Held and Soden 2006). For regional rainfall change, multiple controlling mechanisms that have not been fully understood hinder climate models to provide reliable projections (Xie et al 2015). Recent studies suggest an enhanced EA-WNPSM rainfall in climate model projections (Seo et al 2013, Lee andWang 2014). However, the robust increase in EA-WNPSM rainfall (figure 1(a)) also shows remarkable inter-model spread on intensity and spatial pattern in state-of-the-art models Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence.
Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. (Christensen et al 2013, Kitoh et al 2013, Zhou et al 2018, mainly contributed by the large spread in projected circulation changes whose origins are as yet unidentified. The unclear formation of the EA-WNPSM change spread limits the development of climate models (Taylor et al 2012) and the application of emergent constraints (Boé et al 2009, Cox et al 2013, Huang and Ying 2015 to improve EA-WNPSM change projection. We reveal the leading mode of EA-WNPSM change spread is caused by a common bias in simulating the historical tropical rainfall. We also unravel the mechanism why their coupled relationship is only pronounced in boreal summer and prove it by numerical experiments.

Models and data
We used the historical and Representative Concentration Pathway (RCP) 8.5 runs from 28 Coupled Model   circulation model developed at the Max Planck Institute for Meteorology (Roeckner et al 2003), was chosen in this study for the atmospheric experiments because of its high skill in simulating the East Asian summer monsoon (EASM; Zhou et al 2009b).

Climatology and change
The 1981-2000 mean in historical runs defined the present-day climatology, the 2079-2098 mean in the RCP 8.5 runs defined the future climatology, and their difference represented the change under global warming. The multi-model ensemble (MME) was defined as the simple average of the 28 models. The summer mean was defined as the average of June-August. The variables used included the sea surface temperature (SST), precipitation, net surface short-wave radiation flux, net surface long-wave radiation flux, surface latent heat flux, surface sensible heat flux, zonal and meridional winds at 1000, 850 and 200 hPa, surface scalar wind speed and vertical pressure velocity at 500 hPa. Analyses related to surface scalar wind speed were based on 24 of the 28 models, excluding CCSM4, CESM1(BGC), NorESM1-M and NorESM1-ME, as this variable is unavailable in these four models. To remove the effect of inter-model uncertainty from the global mean temperature change, all the future changes in this study were normalized by the global mean temperature increase in respective models.

Multivariate empirical orthogonal function
Multivariate empirical orthogonal function (MV-EOF) is an extended method based on the conventional EOF (Wang 1992), which builds a combined data matrix for the selected variables before conventional EOF analysis and thus can capture the dominant spatial relationship among various variable fields. In this study, the MME value of each variable among the 28 models is first removed from each model before the inter-model MV-EOF, and then each variable is divided by the standard deviation of that variable field to analyze coherent variations of all variables.

Numerical experiments using the ECHAM5 model
To verify the formation of the EA-WNPSM circulation response and to clarify the relative contributions of seasonal background circulation and the air-sea positive feedback in section 3.2, we performed four sets of numerical experiments using the ECHAM5 model (Roeckner et al 2003), one control run (Exp_ctrl) and three sensitivity runs (Exp_annual, Exp_summer and Exp_annual_60N). The details of experiment design were listed in table 1. We ran each experiment for 31 years and extracted the mean of the last 30 years for analysis.

Historical tropical Pacific rainfall index
To quantify the extent of model simulation and observation associated with regression pattern of tropical Pacific rainfall, we defined a historical tropical Pacific rainfall index. It was calculated by projecting the deviation of historical rainfall in a model or observation from the MME simulated historical rainfall onto the regression pattern of tropical Pacific rainfall over the tropical Pacific region (20°S-20°N, 120°E-70°W), which is shown in section 3.4. The deviations of observed rainfall (GPCP and CMAP) from MME simulated rainfall were also projected onto the regressed rainfall pattern to represent the observation, as shown by the vertical red and blue solid lines.

Local kinetic energetics analyses
The development of the perturbed kinetic energy (KE) response to external forcing is dependent on the background dynamics. The KE converted from the background circulation is based on the following formulas (Kosaka and Nakamura 2006, Hu et al 2019): The first sensitivity run in a warming climate to illustrate the role of the regressed annual-mean tropical SST warming. It is forced by SST0+SST1+SST2, in which SST2 is an SST field with corresponding values to the regression pattern of annual-mean SST warming over the tropical Indo-Pacific region (20°S -25°N, 70°E-150°W) magnified ten times (figure S1 is available online at stacks.iop.org/ERL/14/ 124059/mmedia)

years
Exp_summer The second sensitivity run in a warming climate to illustrate the role of the regressed monthly tropical SST warming. It is forced by SST0+SST1+SST3, in which SST3 is the same as SST2 but for the monthly SST warming with seasonal cycle (figure S2) 31 years

Exp_annual_60N
The third sensitivity run in a warming climate to illustrate the role of the regressed annual-mean SST warming extending to high latitudes. It is forced by SST0+SST1+SST4, in which SST4 (figure S3) is similar to SST2 but extending north to 60°N (20°S-60°N, 70°E-150°W) where ¢ u and ¢ v are the anomalous 850 hPa zonal and meridional winds, u b and v b are the present-day climatological summer-mean winds at 850 hPa, CKx represents the conversion of KE from a mean confluent flow to perturbations and CKy represents the conversion from a mean horizontal shear to perturbations.

Surface energy budget decompositions
The ocean surface energy budget balance is an efficient method to investigate the formation mechanism of SST patterns (Xie et al 2010). The energy budget balance in the mixed layer ocean can be expressed as: where Δ denotes the future change, r 0 and C P are the density and specific heat of seawater, h is the depth of the mixed layer, T is the SST, r ¶D ¶ C h P T t 0 is the tendency of heat content in the mixed layer hereafter referred as to DQ , t Q L is the net surface long-wave radiation flux, Q S is the net surface short-wave radiation flux, Q E is the surface latent heat flux, Q H is the surface sensible heat flux and D o is the ocean heat transport convergence. Positive heat fluxes (Q , are defined as warming the ocean. We used a constant mixed layer (h) of 50 m in this study to make sure that the results were not sensitive to the seasonal variation of the mixed-layer depth (Dwyer et al 2012). In this study, we focused on the summer and used the difference between the DT in May and August in the calculation of ¶D ¶ .

T t
Following previous studies Xie 2008, Xie et al 2010), the latent heat flux change DQ E in equation (2) can be decomposed into three components: and D -Q E others is a residual that represents the effect of relative humidity and stability change.

The leading systematic spread of EA-WNPSM change
The leading mode of the inter-model spread in EA-WNPSM circulation changes was extracted by a MV-EOF analysis performed on a set of five circulation variables, including the changes in zonal and meridional winds at 850 hPa and 200 hPa and the vertical pressure velocity at 500 hPa, in boreal summer mean (June-August) in the EA-WNPSM region (10°-40°N, 90°-140°E) similar to . The first mode of the inter-model MV-EOF accounts for 30.3% of the total variance of EA-WNPSM. The regression patterns of EA-WNPSM changes onto the normalized first principal component (PC1; figure 1(b)) show a conspicuous meridional dipole pattern of changes in 500 hPa vertical pressure velocity ( figure 1(c)) associated with 'dry north-wet south' rainfall ( figure 1(d) figure 1(b) are equivalent to a pattern with a PC of opposite sign. To simplify the presentation, we describe EOF1 as in figures 1(c) and (d) and all patterns in the following analyses are regressed onto the associated PC1 in figure 1(b).) Previous studies have suggested that the intermodel spread of some major components of Asian monsoon change can be attributed to the uncertainty in tropical SST warming (Yang and Lau 1998, Yun et al 2010, Chen and Zhou 2015, Li et al 2017. The regressed summer-mean tropical SST changes onto PC1 appear to be significant and positive-indicating more SST warming-in the equatorial western Pacific, but negative in the eastern Indian Ocean, Maritime Continent region and South China Sea (figure 1(e)). A broader view (only for regression pattern, not for the region of MV-EOF analysis) of the regressed 850 hPa winds ( figure 1(e)) shows that the lower-level cyclonic pattern of EASM and the westerlies of WNPSM in figure 1(c) seem to be part of a Gill-type circulation response to warming SST in the equatorial western Pacific (Gill 1980).
On close inspection, the Gill-type circulation response is asymmetric, being stronger north than south of the equator. Moreover, the regressed summer-mean SST changes (figure 1(e)) are almost equal to the annual mean in the equatorial western Pacific, but stronger than the annual mean in the Asian-Pacific summer monsoon region (figures 1(f) and S4).
Meanwhile, the summer circulation (EA-WNPSM) response is much stronger than the annual mean ( figure S4). But why is the spread of EA-WNPSM and monsoon-region SST changes more pronounced in summer, and where does the annually consistent spread of equatorial western Pacific SST changes originate from?

Mechanisms of stronger circulation response in boreal summer
A local kinetic energetics analysis was performed to understand the enlarged spread of EA-WNP circulation in summer, which is an efficient method to reveal the conversion of local KE from background circulation The apparent EA-WNP circulation response to SST warming in summer can feed back to SST in the monsoon regions, which could lead to a larger zonal SST gradient in the tropical Indo-western Pacific in summer (figures 1(e) and (f)) and enlarge the EA-WNPSM response. In the sea surface energy budget analysis, the regressed tendency of summer-mean changes in mixed layer heat storage (DQ ; t figure 3(a)) are positive over the WNP and negative from the tropical eastern Indian Ocean to the South China Sea, which indicates an increased zonal SST gradient in summer relative to the annual mean. This pattern is very similar to the regressed summer variation (the departure of the summer mean from annual mean) of changes in latent heat flux induced by the surface wind speed (DQ ; EW figure 3(b)), whereas the other surface energy budgets terms contribute little ( figure S6). Clearly, the DQ EW in figure 3(b) is the result of low-level wind spread changes lead by the EA-WNPSM wind changes ( figure 1(e)) on the background EA-WNPSM circulation ( figure S5). The Indo-western Pacific westerlies of the regressed EA-WNPSM circulation change (contours in figure 3(c)) strengthen the monsoonal westerlies over the Northeast Indian Ocean and the South China Sea and weaken the prevailing easterlies over the WNP (shaded in figure 3(c)). In summary, the original SST warming in the equatorial western Pacific with the summer background circulation can arouse a remarkable air-sea positive feedback to strengthen the zonal gradient of SST warming in the EA-WNPSM region and the EA-WNPSM circulation response.

Numerical experiments
As summarized in table 1, four sets of numerical experiments using the atmospheric general circulation model ECHAM5 were performed to verify the formation of the EA-WNPSM circulation response and to clarify the relative contributions of seasonal background circulation and the air-sea positive feedback. As shown in figure 4, the experiments reproduce the Gill-type response to the equatorial western Pacific SST warming well, albeit with a somewhat southward shift of the cyclonic circulation response in the EASM region compared to that in figure 1(e), probably due to the lack of air-sea interaction in the atmospheric model (Song and Zhou 2014). In both Exp_annual and Exp_summer, the circulation response over the EA-WNPSM region is obviously stronger in summer than other seasons (figures 4(a)-(d)), and a relatively large zonal SST warming gradient in summer in Exp_summer can indeed intensify the EA-WNPSM circulation response (figure 4(c)) relative to that in Exp_annual ( figure 4(a)). The role of the background EA-WNPSM circulation can be measured by the difference between the summer-mean and annualmean response in Exp_annual (figure 4(a) minus figure 4(b)), shown in figure 4(e), and the role of the enlarged zonal SST gradient in summer can be measured by the difference between the summer response in Exp_summer and Exp_annual (figure 4(c) minus figure 4(a)), shown in figure 4(f). It can be concluded that the two mechanisms are approximately equal for the EASM response, but the role of background circulation is dominant for the WNPSM response. Although there are significant SST signals in the North Pacific (figures 1(e) and (f)), the EA-WNPSM response in the sensitivity experiment Exp_annual_60N (table 1) is very close to that in Exp_annual ( figure S7), indicating the role of the North Pacific SST (regions (30°N-60°N, 150°E-150°W ) in figures 1(e) and (f)) in the EA-WNPSM response is negligible.

Original bias in historical simulation of tropical rainfall
To further trace the source of the annually persistent western Pacific SST change spread, we analyzed the regressed annual-mean changes in the sea surface energy budget. The only term with a positive contribution to the annually persistent western Pacific SST changes is the surface net shortwave flux change ( figure 5(a); and other terms shown in figure S8). Although the location of the surface net shortwave flux change is not well consistent with that of the SST changes, it can be well explained by a mechanism proposed in Ying and Huang (2016). The relatively strong surface net shortwave flux over the central Pacific can directly lead to a warm SST deviation over the central Pacific; then this warm SST deviation can induce a low-level convergence that suppresses (enhances) the evaporative cooling and zonal cold advection in the western (eastern) Pacific because of the prevailing easterly trades over the tropical Pacific; and finally the original warm SST deviation directly induced by the shortwave flux change in the central Pacific will move westward to the western Pacific (Ying and Huang 2016). Previous studies have suggested that the sensitive convection response to surface warming in the convective region can suppress local surface warming by modulating the surface net shortwave flux (Ramanathan and Collins 1991), which is one of the most remarkable mechanisms forming the spatial pattern of global SST warming. This mechanism implies that the spread of the equatorial western Pacific SST change and the associated net shortwave flux change could originate from the spread of convective activity in the tropical Pacific among the models. This hypothesis is confirmed by the significant regressed historical rainfall representing convective activity ( figure 5(b)). In the models with less historical rainfall (around 5°S-5°N; figure 5(b)), the western Pacific convection cannot suppress surface greenhouse warming as much as the MME, and thus these models project larger surface warming in the equatorial western Pacific ( figure 1(f)).
Based on the historical rainfall in the tropical Pacific as the source of the leading inter-model spread of EA-WNPSM changes, we can narrow the uncertainty of EA-WNPSM changes in MME projection by emergent constraints (Boé et al 2009, Cox et al 2013, Huang and Ying 2015, Hall et al 2019. Emergent constraint is based on an implicit relationship (f ) between elements of current (X ) and future (Y ) climate among 28 CMIP5 models, that is e = + ( ) Y f X , in which e is a small departure from f . In this study, X is the regression pattern of annual-mean rainfall on PC1, Y is the PC1 of inter-model MV-EOF. Their relationship ( f ) was proved in figure 5(c). We define the historical tropical Pacific rainfall index to quantify X and applied it on the bias of X so that we can place this constraint on Y , the PC1 of inter-model MV-EOF (Hall et al 2019). This index is significantly correlated with PC1 (figure 5(c); inter-model correlation 0.77), consistent with the widely significant regression in figure 5(b). Relative to the MME simulation in the CMIP5 models, the rainfall should realistically be stronger in the equatorial central Pacific but weaker in the flanks (figure S9) (Mechoso et al 1995, Dai 2006, Lin 2007, which are well-known tropical Pacific rainfall biases in these state-of-the-art models. Thus, the rainfall indexes for the observations (red and blue solid lines shown figure 5(c)) are negative when the MME simulated rainfall is zero in the index definition. Their corresponding values of PC1 predicting from the relationship between PC1 and rainfall coefficients are shown as dash lines in figure 5(c). Relatively low PC1 values predicting from observed rainfall indexes suggest that the pattern of EA-WNPSM rainfall changes as in figure 1(d), 'dry north-wet south', is overestimated in the MME projection ( figure 1(a)). Thus, we can constrain the EA-WNPSM rainfall change that the rainfall increase should be raised in North China and the Korean Peninsula but lowered in South China, the Indochina Peninsula and the WNP, relative to the MME projection. However, we should also notice that the so-called observations from different sources (GPCP and CMAP) show apparent discrepancies in the tropical Pacific (figures 5(c) and S9), implying it is still crucial in the future to improve observations and understanding with respect to present-day tropical Pacific rainfall.

Conclusions
In this study, we found a systematic spread of EA-WNPSM changes under global warming projected by the CMIP5 models, and traced the systematic mode back to the inter-model discrepancy in simulating the present-day rainfall in the equatorial western Pacific. Under global warming, one model with less historical rainfall (implying weaker convection activity) in the equatorial Pacific cannot suppress the local surface warming by the cloud-shortwave negative feedback as much as the MME, leading to a greater surface warming in the equatorial western Pacific. We found that this surface warming is annually persistent, however, it can couple with summer background circulation to induce a systematic spread of EA-WNPSM changes. In boreal summer, this SST warming can force a more apparent Gill-type atmospheric response in the Indo-western Pacific than in other seasons, appearing as a 'dry north-wet south' dipole in East Asia and wet in Indochina and WNP, in which the KE converted from the summer background circulation and the enlargement of the consequent air-sea interaction are the two key processes. Considering the common double-ITCZ bias of tropical Pacific rainfall in state-of-the-art models (Mechoso et al 1995, Dai 2006, Lin 2007, we can constrain the EA-WNPSM rainfall changes that there should be more rainfall increase in North China and the Korean Peninsula and less increase in South China, the Indochina Peninsula and WNP compared to previous MME projections. This result suggests an effective way for a more reliable EA-WNPSM projection by improving our understanding and simulation of the tropical Pacific convection and rainfall.

Data availability
The data that support the finding of this study are openly available at CMIP5 http://pcmdi9.llnl.gov/ (Taylor et al 2012), GPCP precipitation (Adler et al 2003), and CMAP precipitation (Xie and Arkin 1997).