Using the outputs of 33 coupled models that participated in phase 5 of the Coupled Model Intercomparison Project (CMIP5), the changes of the western North Pacific subtropical high (WNPSH) in the 2050–99 period under representative concentration pathway 4.5 and 8.5 (RCP4.5 and RCP8.5) scenarios relative to the 1950–99 period are analyzed. Under both scenarios, the projected changes in the WNPSH intensity are approximately zero in the multimodel ensemble mean (MME), and large intermodel spread is seen. About half of the models project an enhanced WNPSH and about half of the models project a weakened WNPSH under both scenarios. As revealed by both diagnostic studies and numerical simulations, the projected change in the WNPSH intensity is dominated by the change in the zonal sea surface temperature (SST) gradient between the tropical Indian Ocean (TIO) and the tropical western Pacific (TWP). A stronger (weaker) warming in the TIO is in favor of an enhanced (weakened) WNPSH, and a weaker (stronger) warming over the TWP is also in favor of an enhanced (weakened) WNPSH. The projected change of the WNPSH modulates the climate change over eastern China. Under both RCP4.5 and RCP8.5 scenarios, all of the models with a significantly increased (decreased) WNPSH intensity are associated with a significant increase in the precipitation over the northern (southern) part of eastern China and an enhanced (weakened) southerly wind.
The western North Pacific subtropical high (WNPSH) in the lower troposphere is a crucial system to the summer climate in East Asia. The winds associated with the WNPSH transport water vapor from the tropical oceans into East Asia via the low-level southerly wind on its western edge (Zhou and Yu 2005). The major rain belt in East Asia is located on the northwestern edge of the WNPSH, where the southerly wind encounters the cold air mass. From late spring to summer every year, the northward jumps of the WNPSH ridge correspond to the northward jumps of the major rain belt in East Asia (Tao and Chen 1987).
The variability of the East Asian climate in summer is regulated by the WNPSH on both interannual and interdecadal time scales. The interannual variability of the circulation over the western North Pacific (WNP) is the largest among the subtropical Northern Hemisphere (Lu 2001; Sui et al. 2007; Wu and Zhou 2008; Chung et al. 2011), which greatly affects the summer precipitation anomaly over East Asia. The WNPSH connects El Niño–Southern Oscillation (ENSO) signal with the climate in East Asia (Tao and Zhang 1998; Chang et al. 2000). The Yangtze–Huai River valley along 30°N over East Asia usually suffers excessive precipitation during the decay phase of the El Niño events when the WNPSH is anomalously strong and displaced southward (Tao and Zhang 1998; Zhang and Tao 2003; Wu et al. 2009). The interdecadal change of the WNPSH also modulated the late 1970s’ decadal climate regime shift in East Asia (Zhou et al. 2009a; Xie et al. 2010b; Huang et al. 2010).
The dynamical and physical processes that affect the interannual variation of the WNPSH are well understood. Although El Niño events usually decay in boreal spring, their impact on the summer WNPSH can be maintained by the air–sea interaction in the Indo-Pacific warm pool. The warm sea surface temperature (SST) anomaly in the tropical Indian Ocean (TIO) increases the local rainfall and triggers a warm Kelvin wave emanating into the tropical western Pacific (TWP), with low pressure over the TIO and the equatorial western Pacific. The combined effects of the pressure gradient and friction within the Ekman layer drive an anomalous divergent motion and anticyclonic circulation over the WNP. The above mechanism is called Kelvin-wave-induced Ekman divergence, which explains the impact of TIO SST on the WNPSH (Yang et al. 2007; Li et al. 2008; Xie et al. 2009; Wu et al. 2009; Kosaka et al. 2013). The cold SST anomaly over the equatorial Pacific is also in favor of an enhanced WNPSH via stimulating anticyclonic Rossby wave to its northwest (Wang et al. 2013; Xiang et al. 2013). Therefore, it is recognized that the increased zonal SST gradient between a warm TIO and a cold tropical Pacific Ocean enhances the WNPSH (Terao and Kubota 2005; Chen et al. 2012; Cao et al. 2013). The WNPSH can also be enhanced by the descending branch of the enhanced local Hadley circulation induced by the warm SST anomaly over the Maritime Continent (Sui et al. 2007; Wu and Zhou 2008; Chung et al. 2011) or enhanced by anticyclonic Rossby wave generated by local cold SST anomaly over the WNP (Wang et al. 2000; Wu et al. 2010; Wang et al. 2013).
In comparison to the interannual variability, how the WNPSH responds to global warming is not well understood. Climate model projections show a weakened Walker circulation and Hadley circulation under global warming (Lu et al. 2007; Vecchi and Soden 2007; Gastineau et al. 2009; Bony et al. 2013; Ma and Xie 2013). The projected change of the large-scale rainfall pattern exhibits a wet-get-wetter (Held and Soden 2006) or a warmer-get-wetter pattern (Xie et al. 2010a) over tropical oceans. The low-level southerly wind associated with the East Asian summer monsoon may become stronger, as a result of the enhanced low-level thermal contrast between East Asia and the WNP (Chen et al. 2011; Sun and Ding 2011; Jiang and Tian 2013). A stronger interannual variability is projected for the East Asian summer rainfall (Lu and Fu 2010). In comparison to these studies, less effort has been devoted to the changes of the WNPSH (Li et al. 2012). In this study, we aim to answer the following questions: 1) How would the intensity of the WNPSH change under global warming scenarios? 2) What is the driving mechanism? and 3) How would the projected change of WNPSH affect East Asian climate?
This remainder of the paper is organized as follows. Section 2 describes the models, data, and methods. A brief model evaluation is done in section 3. Section 4 documents the projected change of the WNPSH, its connection with the tropical warming pattern, and its association with the precipitation in eastern China. Section 5 summarizes the major findings.
2. Model, data, and method
The 33 coupled models employed in this study are from phase 5 of the Climate Model Intercomparison Project (CMIP5 Taylor et al. 2012), which are used in the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. The information about the 33 models is listed in Table 1. For each model, the outputs of historical, RCP4.5, and RCP8.5 experiments are selected for analysis. The representative concentration pathway (RCP) RCP4.5 is a radiative forcing pathway to 4.5 W m−2 (equivalent to 650 ppm CO2 concentration) by 2100 without overshoot, and the RCP8.5 is a rising radiative forcing pathway leading to 8.5 W m−2 (equivalent to 1370 ppm CO2 concentration) by 2100 (van Vuuren et al. 2011). Since very few models contain multiple realizations in the RCP4.5 and RCP8.5 experiments, only the first realization (r1i1p1) of each model is adopted in most parts of this paper. To examine the impact of the natural variability of the climate system, multiple realizations from three models (IPSL-CM5A-LR, MIROC5, and MPI-ESM-LR) are analyzed.
Given the uncertainty of observation (Collins et al. 2013; Sperber et al. 2013; Jourdain et al. 2013), multiple observational and reanalysis data (hereafter referred to as observation) are employed to evaluate the models. The rainfall data used in this study include the Global Precipitation Climatology Project (GPCP) version 2 precipitation data (Adler et al. 2003) and the Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP) precipitation data (Xie and Arkin 1997). The reanalysis data used in this study include the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis (NCEP-1; Kalnay et al. 1996), the NCEP–Department of Energy (DOE) Atmospheric Model Intercomparison Project II (AMIP-II) reanalysis (NCEP-2; Kanamitsu et al. 2002), the 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40; Uppala et al. 2005), the ECMWF Interim Re-Analysis (ERA-Interim; Dee et al. 2011), and the Japanese 25-year Reanalysis Project (JRA-25; Onogi et al. 2007) datasets.
We focus on the boreal summer season [June–August (JJA)]. All the model data are interpolated onto the same 2.5° × 2.5° grid as the NCEP-2 data before analysis. Since the geopotential height systematically increases along with global warming according to the hydrostatic equation (Yang and Sun 2003; He et al. 2013), it is hard to distinguish whether an increase of the geopotential height in a certain region is associated with an enhanced anticyclonic circulation or not. Therefore, we mainly focus on the wind field instead of geopotential height field. To quantitatively evaluate the WNPSH intensity, a WNPSH index is defined as the difference of the zonal wind at 850 hPa between 25°–35°N, 120°–150°E and 10°–20°N, 130°–150°E. This kind of index denotes the anticyclonic wind shear over the WNP and has been widely used in monsoon diagnostic and modeling studies (Wang et al. 2008; Xie et al. 2009; Zhou et al. 2009b; Huang et al. 2010).
To evaluate the model performances on the mean state and the interannual WNPSH–SST relationship, we compare the 1980–99 period of the historical run with the observation, because of the available time scope of the observation. The precipitation, 850-hPa wind, and geopotential height are examined in the mean state evaluation. In the evaluation of the interannual WNPSH–SST relationship, an 8-yr high-pass Fourier filter is applied on the WNPSH index and SST to obtain the interannual variation.
For the projection of climate change, we focus on the difference between the 2050–99 period in the RCP4.5–RCP8.5 experiment and the 1950–99 period in the historical experiment. The multimodel ensemble mean (MME) is calculated as the algebraic average of the 33 models. This MME technique is helpful to suppress model drift and internal variability (Gupta et al. 2013). We use “intermodel consistency” to evaluate the robustness of the MME projected changes. For a scalar field (e.g., zonal wind), intermodel consistency is defined as the percentage of the individual models that project the same sign of change as the MME. For a vector field (e.g., wind), the intermodel consistency is defined as the maximum of the intermodel consistencies of the zonal wind and the meridional wind. Scatter diagrams are also shown for the individual models, to diagnose the relationship between the changes in the WNPSH index and the changes in other variables. Student’s t test is performed to determine the significance of the intermodel relationship in the scatter diagrams. The 95% and 99% confidence levels for a correlation of 33 models are ±0.34 and ±0.44, respectively.
To investigate the intermodel relationship between the changes in the WNPSH and the SST warming pattern, each model is treated as a sample and intermodel regression is done by regressing the mean state changes in SST onto the changes in the WNPSH index. Similar regression analysis is done for the precipitation and tropospheric temperature fields, to diagnose the air–sea relationship and the possible forcing mechanisms. Following previous studies (Xie et al. 2009), the tropospheric temperature is examined to identify the tropical waves (Kelvin wave and Rossby wave). The tropospheric temperature is calculated as the difference of the geopotential height between 200 and 850 hPa multiplied by a constant, according to the hydrostatic equation.
To reveal which part of the regressed SST pattern is responsible for the intermodel spread of WNPSH changes, numerical experiments are done using the Community Atmospheric Model, version 4 (CAM4; Neale et al. 2010). CAM4 is one of the best models in simulating the leading interannual mode of the WNPSH (He and Zhou 2014). The CAM4 is run with a finite-volume dynamic core, at a resolution of approximately 1.9° × 2.5° in the horizontal and 26 hybrid σ–p levels in the vertical. Control run is done for 20 years forced by observational monthly SST climatology. Several sensitivity experiments are performed, forced by modified SST. The modified SST is obtained by adding SST anomaly to the observational SST climatology over a specific region. Each sensitivity experiment contains 20 ensemble members, which are initialized at each 1 May in the 20-yr control run. Each ensemble member is run for 4 months until 31 August, and the outputs of JJA are analyzed. Student’s t test is performed on the composite of the 20 ensemble members to determine the significant forced signals against atmospheric internal noise.
3. Model evaluation on the mean state and the interannual WNPSH–SST relationship
The climatological circulation over the WNP in summer is characterized by a low-level WNPSH, the ridgeline of which extends to the southeast coast of China. Abundant rainfall is seen on the periphery of the WNPSH, covering the western coast of the Philippines, eastern China, and Japan (Fig. 1a). The climatological precipitation and low-level circulation is reasonably reproduced by the MME of the CMIP5 models, including the anticyclonic winds and the abundant rainfall on the periphery of the WNPSH (Fig. 1b). The major bias in the circulation field is the northward displaced ridgeline, as evidenced in the wind and eddy geopotential height fields (eddy geopotential height is the deviation of geopotential height from its zonal mean). The northward displacement of the WNPSH ridgeline is also a common bias in CMIP3 models (Inoue and Ueda 2009).
The difference of the climatological winds between the MME and the NCEP-2 (vectors in Fig. 1c) is characterized by a cyclonic anomaly over southern WNP (10°–30°N) and an anticyclonic anomaly over northern WNP (30°–45°N), consistent with the northward displaced ridgeline in the MME. This mean state wind bias is consistent among the five reanalysis datasets (figure not shown). Excessive (deficient) rainfall is seen over the southern (northern) periphery of the WNPSH when compared to GPCP data (shading in Fig. 1c) or CMAP data (contours in Fig. 1c), consistent with the previous findings that coupled models generally underestimate the mei-yu–baiu rain belt (Sperber et al. 2013). A large discrepancy between GPCP and CMAP data is seen around the Philippines. The climatological precipitation rate of the MME is greater than the GPCP data but smaller than the CMAP data over the South China Sea (SCS), and approximately equivalent to the GPCP data but smaller than the CMAP data off the eastern coast of the Philippines. The uncertainty among precipitation datasets has been noted by many previous studies (e.g., Collins et al. 2013; Sperber et al. 2013; Jourdain et al. 2013).
Since the interannual variability of the WNPSH is regulated by the tropical SST, the simulated correlation of the WNPSH with the SST at an interannual time scale is evaluated against the observation. In the observation, the most prominent feature of the correlation pattern is the positive correlation between the WNPSH and the TIO SST, while negative correlation between the WNPSH and the WNP SST is also seen (Fig. 1d). These WNPSH–SST relationships on the interannual time scale can be captured by the MME, but with weaker magnitude (Fig. 1e).
The above evaluation shows the northward displaced ridgeline is the major bias in the mean state simulation, while the positive correlation between WNPSH and TIO SST is the most prominent feature on the interannual time scale. To quantitatively evaluate the model performance, a scatter diagram for the five reanalysis datasets and the 33 models is shown in Fig. 1f. The abscissa of Fig. 1f is the latitude of the climatological ridgeline (defined as the 130°–150°E averaged latitude of the interface between the trade easterlies and the midlatitude westerlies), and its ordinate is the correlation coefficient between WNPSH index and TIO SST on the interannual time scale (TIO is referred to as 10°S–10°N, 50°–100°E; see the box in Fig. 1d).
Among the five reanalysis datasets (the red numbers 1, 2, 3, 4, and 5 in Fig. 1f), the climatological locations of the ridgeline range from 24.7°N (NCEP-1 and NCEP-2) to 25.5°N (JRA-25), with small observational uncertainty. The location of the WNPSH ridge is 26.9°N in the MME, which is displaced northward compared with the observation. The locations of the ridgeline range from 21.9°N (FGOALS-g2.0) to 31.7°N [CESM1 (BGC) and CSIRO Mk3.6.0] and the northward displacement of the ridgeline is seen in 25 of the 33 models. The correlation coefficient between the WNPSH index and the TIO SST is approximately 0.64 for all the reanalysis datasets. This positive correlation is captured but underestimated by the MME (0.24) and most of the individual models.
4. Projected changes of the WNPSH
The projected change of JJA 850-hPa wind under RCP4.5 scenario by the MME is shown in Fig. 2a, with the intermodel consistency among the models. Since the region with low model consensus is also useful for indicating insensitivity to climate change (Power et al. 2012), the regions where the projected change is smaller than the intermodel standard deviation of projected changes are marked with red crosses. An anomalous southerly wind is seen in eastern China and the intermodel consistency is larger than 70% over North China, indicating an enhanced monsoon circulation, which is consistent with previous studies (Sun and Ding 2011; Jiang and Tian 2013). Although westerly wind seems to be enhanced off the southern coast of Japan, the intermodel consistency is less than 70% over the WNP, and the MME projected change is smaller than the intermodel standard deviation of the projected changes by the individual models. These suggest a large spread among the model results. The trade easterlies along the equatorial Pacific are weakened, which is consistent with previous findings that the Walker circulation will weaken because of global warming (e.g., Vecchi and Soden 2007), but this weakening is evident in less than 70% of the models.
Under the RCP4.5 scenario, the changes of the WNPSH index are closely related to the changes of the 500-hPa vertical velocity and the precipitation among the models, in a manner that increased (decreased) WNPSH index corresponds to enhanced (weakened) descending motion and decreased (increased) precipitation over the WNP (Figs. 2b,c). The correlation coefficients with the WNPSH index are 0.67 and −0.49 for the local vertical velocity and precipitation, respectively, which are statistically significant at the 99% confidence level based on t test. As the WNPSH index based on the zonal wind is physically consistent with the vertical velocity and the precipitation, it is referred to as WNPSH intensity hereafter. The projected changes of WNPSH intensity under the RCP4.5 scenario by the individual models are between a decrease of 1.31 m s−1 (CMCC-CM) and an increase of 1.69 m s−1 (HadGEM2-AO; Figs. 2b,c, abscissa).
Under the RCP8.5 scenario, the projected change of wind over the WNP shares a similar spatial pattern with the RCP4.5 scenario but with stronger amplitude (Fig. 2d). The intermodel consistency is still lower than 70%, and the MME projected change is still smaller than the intermodel standard deviation over the WNP. Under the RCP8.5 scenario, the projected changes of WNPSH intensity are between a decrease of 1.73 m s−1 (CMCC-CM) and an increase of 2.30 m s−1 (MIROC5) for the individual models (Figs. 2e,f, abscissa). This range is larger than that of the RCP4.5 scenario, since stronger external forcing is imposed on the models. The MME projected change in WNPSH intensity remains approximately zero (Figs. 2e,f). The MME projects unchanged vertical velocity at 500 hPa (Fig. 2e) and an increase of precipitation over WNP (Fig. 2f), indicating that this increase of precipitation is caused by the increased water vapor content in the troposphere (Bony et al. 2004; Held and Soden 2006). The correlation coefficient between the projected changes in the vertical velocity (precipitation) and the changes in WNPSH intensity is 0.61 (−0.55), which is statistically significant at the 99% confidence level based on Student’s t test.
The WPNSH intensity here is defined as the zonal wind shift between a northern box and a southern box. The locations of the northern and the southern boxes are unified for all the models. As shown in Fig. 1f, the mean state locations of the WNPSH ridgeline are biased in many models. Is the projected change of the WNPSH intensity related to the bias in the mean state? To address this question, we shifted the meridional location of the northern and the southern boxes for each model, so that these two boxes are located symmetrically about the ridgeline of the model. It is found that the projected changes of this modified WNPSH index are highly consistent with the WNPSH intensity index used in this study (figure not shown). Under both scenarios, the changes of the WNPSH intensities have little relationship with the latitude of the climatological ridgeline (figure not shown). These evidences suggest the biases in the mean state location of the WNPSH ridgeline have little effect on the projected changes in the WNPSH intensity.
To investigate the relationship between the RCP4.5 projected change and the RCP8.5 projected change, the projected changes of WNPSH intensities under the RCP4.5 scenario versus the RCP8.5 scenario are shown as scatter diagrams (Fig. 3a). Among the 33 models, 27 are located in the first or the third quadrant, indicating the signs of changes in the WNPSH intensity are the same under the RCP4.5 and RCP8.5 scenarios for most models. The intermodel regression equation of Fig. 3a is y = 1.03x − 0.06, indicating a slightly stronger change of the models under the RCP8.5 scenario than the RCP4.5 scenario. Among the 33 models, 20 models (60.6%) project an increase and 13 models (39.4%) project a decrease of the WNPSH intensity under the RCP4.5 scenario and 18 models (54.5%) project an increase and 15 models (45.5%) project a decrease of the WNPSH intensity under the RCP8.5 scenario. In summary, roughly half of the models project an enhanced WNPSH and half of the models project a weakened WNPSH under both scenarios.
Table 2 lists the models in which the WNPSH intensities are significantly changed at the 95% confidence level according to a t test based on the interannual variability. The models with a significant increase (decrease) of the WNPSH intensity are referred to as P-type (N-type) models. Among the 33 models, there are seven P-type models under the RCP4.5 scenario and nine P-type models under the RCP8.5 scenario. Only five models (FGOALS-g2.0, IPSL-CM5A-MR, MIROC-ESM, MIROC-ESM-CHEM, and MIROC5) belong to P type under both RCP4.5 and RCP8.5 scenarios. The P-type models are not highly consistent between the RCP4.5 and RCP8.5 scenarios, maybe because of the natural variability of the climate system. The N-type models are exactly the same under the RCP4.5 and RCP8.5 scenarios, including CMCC-CM, INM-CM4, IPSL-CM5B-LR, and MPI-ESM-LR. The P-type (N-type) models are colored in red (blue) in Figs. 2b, 2c, 2e and 2f. Apart from the P-type and N-type models, the other 22 (20) models project no significant change in the WNPSH intensity under the RCP4.5 (RCP8.5) scenario.
How large is the natural variability? Can the natural variability overwhelm the forced response in the difference between the RCP4.5 (or RCP8.5) run and the historical run? To answer this question, three models (IPSL-CM5A-LR, MIROC5, and MPI-ESM-LR) that contain multiple realizations are selected. Under both scenarios, MIROC5 belongs to P-type models, MPI-ESM-LR belongs to N-type models, and IPSL-CM5A-LR projects no significant change in the WNPSH intensity (Table 2). Figure 3b shows the projected changes in the WNPSH intensity for all possible combinations of the ensemble members. It can be seen from Fig. 3b that the natural variability is not large enough to overwhelm the forced response, since the multiple realizations of one model are clearly distinguishable from another model. It can also be inferred that the intermodel spread (e.g., Fig. 3a) cannot be explained by natural variability, but is related to the different forced responses of different models.
Why do the WNPSH intensities increase in some models but decrease in some other models under the same external forcing? Previous studies showed evidence that the SST pattern of tropical oceans affects the WNPSH at interannual and interdecadal time scales (e.g., Terao and Kubota 2005; Xie et al. 2009; Zhou et al. 2009a; Wu et al. 2010; Wang et al. 2013). To examine the relationship between the changes in WNPSH and the tropical SST pattern, we show the intermodel regression of changes in SST and 850-hPa winds onto the changes in WNPSH intensity under the RCP4.5 scenario in Fig. 4a. The regressed pattern under the RCP8.5 scenario (figure not shown) is similar to Fig. 4a.
As seen in Fig. 4a, the changes of the SST pattern associated with an enhanced WNPSH are characterized by stronger warming of SST over the TIO, the Bay of Bengal and South China Sea (BOB-SCS, 10°–20°N, 80°–120°E), the Niño-3 region (5°S–5°N, 150°–90°W), and the midlatitude Pacific and relatively weaker warming of SST over the TWP (10°S–10°N, 150°E–180°). This SST pattern resembles the interannual SST–WNPSH relationship (Fig. 1d). Associated with the zonal SST gradient between strongly warmed TIO and weakly warmed TWP, easterly wind is seen from TWP to TIO in the regressed wind field, consistent with the mechanism of low-level wind response to SST gradient (Lindzen and Nigam 1987). Although warmer SST is seen in the Niño-3 region (Fig. 4a), it may not be responsible for the enhanced WNPSH, according to previous studies (Wang et al. 2013; Xiang et al. 2013) and the numerical experiment discussed later in this study.
The intermodel regression of the changes in precipitation and tropospheric temperature onto the changes in WNPSH intensity under the RCP4.5 scenario is shown in Fig. 4b. The spatial pattern is similar to Fig. 4b for the RCP8.5 scenario (figure not shown). The enhanced WNPSH is associated with increased precipitation over the TIO, which may be forced by the warmer SST over the TIO. Negative rainfall anomalies are seen over BOB-SCS, accompanied by warmer SST (see Fig. 4a), indicating the warmer SST in this region may be forced by the atmosphere. The WNP is dominated by negative rainfall anomaly but warmer SST, suggesting the negative rainfall anomaly (and anticyclonic wind anomaly) is not locally forced but remotely forced.
The tropospheric temperature field is characterized by two prominent features associated with enhanced WNPSH (Fig. 4b). The first is a wedgelike warm anomaly centered over the TIO and pointing eastward into the TWP along the equator. This pattern suggests a Kelvin wave forced by the positive heat source over the TIO (Gill 1980; Xie et al. 2009). The second is two relatively cold anomalies over the WNP and Australia, which are symmetric about the equator and are located on the northwest and southwest side of the colder SST of TWP (see Fig. 4a). This pattern may be induced by Rossby wave response to the relatively cold SST of TWP (Gill 1980; Wang et al. 2013; Xiang et al. 2013). The above-mentioned features indicate the intermodel spread may be dominated by the forcing from TIO–TWP or both of them.
Scatter diagrams in Figs. 4c–f show the relationship between the changes in the WNPSH intensities and the changes in the SST over some key regions, for the RCP4.5 scenario (blue) and RCP8.5 scenario (red), respectively. No matter if they are under the RCP4.5 or RCP8.5 scenario, the changes in the WNPSH intensity are very weakly correlated with the changes in the tropical mean (20°S–20°N averaged) SST (Fig. 4c), the TIO SST (Fig. 4d), and the TWP SST (Fig. 4e), as evidenced by the poor linear relationship of the scatters and the correlation coefficients (Figs. 4c–e). However, the changes in WNPSH intensity and the changes in TIO–TWP SST gradient (defined as TIO SST minus TWP SST) are significantly correlated at the 95% confidence level, with a correlation coefficient of 0.53 and 0.42 for the RCP4.5 and RCP8.5 scenarios, respectively (Fig. 4f). These correlation coefficients are 0.64 and 0.57 for RCP4.5 and RCP8.5 scenarios, respectively, if the outlier model CSIRO Mk3.6.0 (“L” in Fig. 4f) is excluded, exceeding the 99% confidence level. Previous studies showed that 0.5-K SST anomaly over the TIO is enough to stimulate prominent wind anomaly over the WNP at interannual time scale (Xie et al. 2009; Wu et al. 2010). The projected changes in the TIO–TWP SST gradient range from −0.2 to 0.5 K (Fig. 4f), which are large enough to result in distinct responses of the WNPSH.
To further confirm which regions are responsible for the projected changes in the WNPSH intensity, we examined the response of CAM4 to the SST anomalies of specified regions picked from the regressed field in Fig. 4a, including the Niño-3 region, BOB-SCS, TIO, and TWP (the boxes in Fig. 4a). To obtain stronger model response, the SST anomalies picked from Fig. 4a are multiplied by a factor of 5. The details of the experimental design are described in section 2, and the responses of the precipitation and 850-hPa winds over the WNP are shown in Fig. 5.
The response of CAM4 to the warm SST anomaly of the Niño-3 region in Fig. 4a is characterized by significant westerly wind anomaly over the TWP, without anomalous anticyclonic circulation (Fig. 5a). The response of CAM4 to the warm SST anomaly of BOB-SCS is characterized by an anomalous cyclonic circulation on the southeast coast of China, accompanied by positive rainfall anomaly (Fig. 5b). These indicate that neither the Niño-3 region nor BOB-SCS are responsible for the intermodel spread of the changes in the WNPSH intensity. Based on the negative rainfall–SST relationship over the BOB-SCS (Figs. 4a,b), it can be inferred that the warmer SST over the BOB-SCS in Fig. 4a is a forced response to atmosphere.
As a response to the warm SST anomaly of TIO in Fig. 4a, the WNP is dominated by anticyclonic wind anomaly at 850 hPa, with suppressed rainfall on its southern flank (Fig. 5c). This response to TIO forcing can be explained by Kelvin-wave-induced Ekman divergence (Terao and Kubota 2005; Xie et al. 2009). As a response to the SST anomaly over the TWP, the WNP is also dominated by anticyclonic wind anomaly (Fig. 5d), which is located to the northwest of the negative SST anomaly of TWP, reminiscent of a Rossby wave response (Gill 1980; Wang et al. 2013; Xiang et al. 2013). If the warm SST anomaly of TIO and the cold SST anomaly of TWP are both picked from Fig. 4a to force CAM4, significant anticyclonic anomaly and negative rainfall anomaly are also seen over the WNP (figure not shown).
In comparison to the regressed field in Fig. 4a, the CAM4 simulated anomalous anticyclones over the WNP are all displaced southward (Figs. 5c,d). The southward displacement of this anomalous anticyclone is a common bias in the atmospheric general circulation models (AGCMs) of CMIP3 and CMIP5 (Song and Zhou 2014). Given this model bias, a modified WNPSH index is defined as the difference in the zonal wind between 20°–30°N, 100°–130°E and 5°–15°N, 100°–130°E (red boxes in Figs. 5a–d) to quantitatively evaluate the response of the WNPSH intensity in CAM4. The responses of the modified WNPSH indices for the five sensitivity experiments mentioned above are shown as a bar chart in Fig. 5e. A significant increase in the modified WNPSH index at the 95% confidence level is seen when CAM4 is forced by the SST anomalies of the TIO or the TWP or both of them. The increase of the modified WNPSH index is stronger when CAM4 is forced by both TIO and TWP, compared to those forced by TIO alone or TWP alone. An insignificant decrease of the modified WNPSH index is seen when CAM4 is forced by the SST anomaly of the Niño-3 region or BOB-SCS. These results confirm that the TIO and the TWP have both contributed to the projected WNPSH intensity, while the Niño-3 region and BOB-SCS have not.
The regressed negative SST over the TWP in Fig. 4a actually means relatively weaker warming, instead of cooling. The TWP SST gets warmer in all of the models under both RCP4.5 and RCP8.5 scenarios, and none of them show a cooling in the TWP (Fig. 4e). What is the response of the WNPSH if the SST over the TIO and the TWP both increase but at different amplitudes? To mimic the effect of nonuniform warming, CAM4 is forced by a 1-K warm SST anomaly in the TIO and a 0.5-K warm SST anomaly in the TWP. The model response (Fig. 5f) is characterized by anticyclonic anomaly and negative rainfall anomaly over the WNP, indicating the nonuniform warming with positive TIO–TWP zonal SST gradient is indeed favorable for an enhanced WNPSH.
How do the different projected changes of the WNPSH affect the monsoon rainfall over eastern China? To answer this question, composite changes in the precipitation and 850-hPa winds are shown separately for the P-type and N-type models (see Table 2 for which models are P type or N type; Fig. 6). Under the RCP4.5 scenario, increased precipitation over the northern part of eastern China and enhanced southerly wind are seen following an enhanced WNPSH (Fig. 6a), while increased precipitation over the southern part of eastern China and weakened southerly wind are seen following a weakened WNPSH (Fig. 6b). A decrease of precipitation over the southern part of eastern China is also seen in Fig. 6a, but this decrease is not consistent among the P-type models. Under the RCP8.5 scenario, the spatial patterns of the changes in precipitation and wind are similar as the RCP4.5 scenario, but with stronger magnitude (Figs. 6c,d). These results suggest the projected changes in the WNPSH intensity will modulate the rainfall pattern in eastern China.
5. Conclusions and discussion
Using 33 coupled models from the CMIP5 dataset, the future change of the WNPSH under global warming scenarios are investigated by comparing the 2050–99 climatology of the RCP4.5 and RCP8.5 runs with the 1950–99 climatology of the historical run. The results show that the change of WNPSH depends on the change of the zonal SST gradient between TIO and TWP. The main conclusions are summarized as follows.
Although the projected changes of the models under the RCP8.5 scenario are stronger than the RCP4.5 scenario, the MME projected changes in WNPSH intensity are approximately zero under both scenarios. Large projection uncertainty is seen among the CMIP5 models, since about half of the models project an increase and half of the models project a decrease in the WNPSH intensity. Based on the Student’s t test at the 95% confidence level, only seven (four) models project a significant increase (decrease) of the WNPSH intensity under the RCP4.5 scenario, and only nine (four) models project a significant increase (decrease) of the WNPSH intensity under the RCP8.5 scenario.
The projected change in the intensity of the WNPSH is regulated by the change in the TIO–TWP SST gradient, as suggested by the diagnostic study and numerical simulation. Stronger warming in the TIO and weaker warming in the TWP are in favor of enhanced WNPSH, while weaker warming in the TIO and stronger warming in the TWP are in favor of weakened WNPSH.
The projected change in the WNPSH intensity modulates the rainfall pattern in eastern China. Under both RCP4.5 and RCP8.5 scenarios, models with an increased (decreased) intensity of the WNPSH are associated with an increase in the precipitation over the northern (southern) part of eastern China, accompanied by enhanced (weakened) southerly wind.
The relationship between the changes in WNPSH and the tropical SST resembles their relationship on the interannual time scale. The mechanisms that were proposed to explain the interannual variability of the WNPSH (Terao and Kubota 2005; Xie et al. 2009; Wang et al. 2013) may also apply on the global warming projections. Stronger (weaker) warming in the TIO stimulates warm (cold) Kelvin wave, increasing (decreasing) the WNPSH intensity via wave-induced Ekman divergence (convergence). Weaker (stronger) warming in the TWP increases (decreases) the WNPSH intensity via anticyclonic (cyclonic) Rossby wave response to its northwest.
This study focuses the intensity of the WNPSH, while the possible location shift of the WNPSH has not been addressed. Figure 7a shows the ridgeline of the MME for the historical, RCP4.5, and RCP8.5 experiments. It is shown that the mean state ridgeline is projected to shift southward slightly under both RCP4.5 and RCP8.5 scenarios compared with historical experiments. For the MME, the magnitudes of the southward shift averaged within 130°–150°E are 0.29° under RCP4.5 scenario and 0.46° under RCP8.5 scenario, which are close to zero. The large spread among the individual models is indicated by the error bar (Fig. 7b). The number of models projecting a northward (southward) shift of the WNPSH ridgeline is 14 (19) for the RCP4.5 scenario and 13 (20) for the RCP8.5 scenario. In all, the location of the WNPSH ridgeline may stay unchanged or shift southward slightly, with little intermodel consensus.
The results of this study show a large intermodel spread of the projected change in the WNPSH, suggesting attention should be paid on the WNPSH in the future studies of the climate change in the WNPSH-related regions. The intermodel spread of the projected WNPSH intensity depends on the change in the TIO–TWP zonal SST gradient, but it remains unknown what has led to the different changes of the TIO–TWP SST gradient in different models, which is worthy of further study.
We wish to thank the three anonymous reviewers whose comments and suggestions greatly helped us to improve the quality of this manuscript. The numerical experiment in this manuscript is done based on the suggestion of a reviewer. We also wish to thank Dr. Bo Wu, Dr. Liwei Zou, Mr. Fengfei Song, and Ms. Lu Dong for useful discussions. We wish to express many thanks to the modeling groups joined in the CMIP5 and the Program for Climate Model Diagnosis and Intercomparison (PCMDI; http://pcmdi9.llnl.gov), which provide the model data. This work is jointly supported by National Natural Science Foundation of China under Grant 41125017 and Chinese National Program on Key Basic Research Project (2014CB953901). The first author is also supported by National Natural Science Foundation of China (41205069, 41375095).