Assessing Global Warming Induced Changes in Summer Rainfall Variability over Eastern China Using the Latest Hadley Centre Climate Model HadGEM3-GC2

Eastern China encompasses a large area from the subtropics to the midlatitudes. Regional precipitation here is influenced by both local processes, such as land-atmosphere interactions and human activities, and teleconnections from oceanic forcing and upstream disturbances. In the summer, the rainfall distribution is characterized by meridionally banded structures (Zhu and Wang, 2002; Ma, 2007; Ye and Lu, 2012; Yang et al., 2016). Alternation and pattern shifts in the rainfall distribution often cause regional droughts and floods with severe societal and economic consequences to this highly populated region with concentrated economic development (Zhai et al., 2005; Piao et al., 2010;Ren et al., 2011).

Two major patterns of summer rainfall anomalies over eastern China have been reported based on recent observations: the so-called dipole and tripole patterns (Lau, 1992; Weng et al., 1999; Huang et al., 2006; Ding et al., 2008; Han and Zhang, 2009; Ye and Lu, 2012; He et al., 2016). These patterns have been found at both interannual and interdecadal time scales (Ma and Fu, 2003; Ma and Shao, 2006; Huang et al., 2011; Ding et al., 2013). The dipole pattern is characterized by opposite signs of rainfall anomalies between the north and south of the Yangtze River. The tripole pattern is characterized by a sandwich-like distribution, with rainfall anomalies over the Yangtze River valley varying out-of-phase with North and South China.

Although these rainfall distributions are regional, their driving mechanisms may involve remote factors, including tropical sea surface temperature anomaly (SSTA) forcing (Hsu and Lin, 2007; Liu et al., 2008; Wang et al., 2009; Chen and Huang, 2012; Chen and Zhou, 2014; Stephan et al., 2017; Ying et al., 2017), mid-high latitude teleconnections (Hsu and Lin, 2007; He et al., 2016), snow-cover change at midlatitudes, e.g., the Tibet Plateau (Ding et al., 2009), Arctic sea-ice variation (He et al., 2017), and so on. On interannual time scales, the evolution of ENSO events has notable influences on these rainfall modes (He et al., 2016; Jin et al., 2016). It can affect the summer rainfall in eastern China through air-sea interaction during the same season, and also from the previous winter through local thermodynamic feedback over the northwestern Pacific (Wang et al., 2000) or storing the signals through other basins, e.g., the Indian Ocean (Xie et al., 2009, 2016). Apart from the ocean's direct forcing, several atmospheric teleconnections are also linked to the climate variability over eastern China. For example, (Ding and Wang, 2005) showed that a circumglobal teleconnection wave pattern that propagates mainly along the westerly jet stream could affect rainfall over China via a high/low pressure anomaly center sitting over North China. (Nitta, 1987) and (Huang, 1992) proposed that the Pacific-Japan/East Asia-Pacific wave pattern could affect eastern China by meridionally organized anomalous anticyclonic and cyclonic circulation along the East Asia coast. Similar wave trains (Si and Ding, 2016; Wu et al., 2016) and SST (Yang et al., 2016, 2017) patterns have also been reported to have modulated the rainfall variability in China on interdecadal time scales. These wave patterns and SSTAs might not be independent but closely linked with each other (Ding et al., 2011). It is the combined influences of the local and remote factors that dictate rainfall anomalies over eastern China on various time scales.

It is well known that anthropogenic global warming has a strong impact on the global hydrological cycle (e.g., Allen and Ingram, 2002; Wu et al., 2010, 2013), regional climate modes such as the Walker circulation, ENSO and the South Asian monsoon system (Chadwick et al., 2012; Cai et al., 2014; Wu et al., 2015), and consequently teleconnections and regional rainfall patterns (Lee et al., 2014; Yang et al., 2017). Increased greenhouse gas emissions combined with land-use changes, air pollution and rapid urbanization have already contributed to intensifying regional drought (Wang et al., 2016; Zhang et al., 2017) and precipitation extremes (Xiao et al., 2016). Apart from changes in mean precipitation and extremes, any positional shift or polarity changes in seasonal precipitation distribution will impact on local ecological systems and water resources. Some previous studies also show that anthropogenic forcing could increase the occurrence of certain rainfall patterns while suppressing some others (Shi et al., 2009; Sun and Ding, 2010).

Unlike most previous studies concentrating on long-term trends, this paper focuses on potential changes in summer rainfall modes over eastern China on interannual and interdecadal time scales, using the latest version of the global coupled climate model of the Met Office, HadGEM3-GC2. The main questions are: (2) What are the leading modes and their driving mechanisms in this particular model? (2) How would these modes change under the anthropogenic greenhouse gas forcing? These questions are addressed by comparing an idealized CO₂ forcing scenario with a pre-industrial control simulation.

The paper is organized as follows: Details of the model data and analysis methods are given in section 2. The simulated internal variability of rainfall patterns and their corresponding mechanisms are presented in section 3. Section 4 describes the changes in rainfall patterns in a CO₂ scenario. A summary and further discussion are given in section 5.

Simulations using the latest version of the Met Office's global coupled climate model, HadGEM3-GC2, are investigated in this study. The model has a vertical resolution of 85 levels in the atmosphere and 75 levels in the ocean. A horizontal resolution of N216 (∼60 km in the midlatitudes) is used for the atmosphere and 0.25^° for the ocean. More details of the model can be found in (Williams et al., 2015). This model has good skill in representing the large-scale climate modes related with East Asian rainfall, e.g., ENSO (Williams et al., 2015), as well as some aspects of the East Asian monsoon, e.g. the intraseasonal variability (Fang et al., 2017).

Given that our focus is on potential future changes in regional summer rainfall variability, two particular model experiments are exploited. One is the model's pre-industrial control simulation, in which all external forcing factors including the atmospheric CO₂ concentration are fixed at the pre-industrial level. The other is an idealized 4×CO₂ simulation, in which all remain the same as in the pre-industrial simulation except that the atmospheric CO₂ concentration is instantly quadrupled and then held at that fixed level. Different from a gradually increased (1% per year until quadrupled) CO₂ simulation, an abrupt CO₂ simulation is more suitable for investigating "slow" climate system responses (Taylor et al., 2012). Considering that we also focus on the effect of CO₂ forcing on interdecadal time scales, which is more unambiguous in a slow response, the abruptly quadrupled CO₂ experiment is chosen here.

The control simulation has 315 years of data, while the 4×CO₂ simulation has 170 years. In order to avoid initial model adjustment, the first 20 years of the 4×CO₂ data are dropped. Summer (June, July and August) mean data are used in all the following analysis, unless otherwise explained in context.

By comparing the above two model simulations, we attempt to identify potential future changes in rainfall patterns, admittedly unrealistically. Although the pre-industrial simulation is not the same as observations and thus cannot be verified by observational data, it represents the model's internal variability with long-term simulation. Our focus is the relative change between these two experiments. Therefore, the results of this study reflect the possible influence of anthropogenically induced global warming on the internal variability of eastern China's summer rainfall. The revealed rainfall variability in terms of leading rainfall modes in this model is also beneficial for understanding the model's internal variability.

As we are only interested in the interannual and interdecadal variability, all model data are first detrended at each grid point. A 7-year Lanczos low-pass filter (Duchon, 1979) is then applied to extract the decadal and interdecadal signals. The remaining is treated as interannual signals.

Empirical Orthogonal Function (EOF) analysis is applied to identify the major spatial characteristics (leading modes) and its time variance of rainfall variability on interannual and interdecadal time scales. Before this, anomaly data at each grid are normalized by the standard deviation of unfiltered data to emphasize the relationship between each grid.

Based on the principle component (PC) of the EOF, linear regression is used to reveal the mode-related circulation and SSTAs. Composite analysis is also applied to check the possible asymmetry between the positive and negative phases of these modes, but no obvious difference is found. Thus, in this paper, only the results of linear regression are presented.

In order to obtain more detailed features of the teleconnections related to the rainfall modes, the wave activity flux proposed by (Takaya and Nakamura, 2001) is calculated. By showing the direction and source of the Rossby wave energy propagation, it is widely adopted for studying stationary quasi-geostrophic eddies on a zonally varying asymmetric basic flow (Hsu and Lin, 2007). The horizontal components of the wave activity flux under the pressure coordinate are calculated as \begin{equation} \label{eq1} w=\dfrac{1}{2|{U}|}\left( \begin{array}{c} U(\psi'^2_X-\psi'\psi'_{xx})+V(\psi'_x\psi'_y-\psi'\psi'_{xy})\\[2mm] U(\psi'_x\psi'_y-\psi'\psi'_{xy})+V(\psi'^2_y-\psi'\psi'_{yy}) \end{array} \right),\ \ (1) \end{equation} where ψ is the streamfunction and U(u,v) is the horizontal wind velocity. The overbars represent the climatology, calculated as the long-term mean during the whole period of the experiment; and the primes denote the anomalies with respect to the climatology.

Figure 1 shows the first two leading modes of eastern China summer rainfall variability in the control simulation, as revealed by EOF analysis of the unfiltered, high-pass (interannual) and low-pass (interdecadal) filtered data. Similar meridional structures of rainfall anomalies are shown on both time scales. In EOF1, most of eastern China experiences the same sign of rainfall anomalies and the coastal regions of South China vary oppositely (named as the monopole mode below). In EOF2, rainfall anomalies over the middle and lower reaches of the Yangtze River are out of phase with those in North China and coastal South China (named as the dipole mode below). The variance contributions of the monopole and dipole mode in the unfiltered data are 10.24% and 8.64%, respectively, which are well separated from each other and from the rest of the modes according to North's criteria (North et al., 1982). This result is consistent with the leading modes revealed by (Pei et al., 2015) by analyzing the historical proxy data over 400 years.

The unfiltered mode and interdecadal mode in this model share not only the same spatial pattern, but also the same variations. As shown in Figs. 1g and h, the interdecadal components of PCs in the unfiltered data (red bars) vary synchronously with the PC derived from the interdecadal data (green lines). The linearity implies that, in the internal variability, physical mechanisms associated with the two leading modes might be similar on interannual and interdecadal time scales. To make it clearer, the driving mechanisms of the two modes are examined separately in the following based on their time scales.

Figure 1.  The first two EOF modes of summer rainfall variability over eastern China simulated by the HadGEM3-GC2 pre-industrial control experiment: (a, d) from unfiltered data; (b, e) high-pass (7-year truncation) filtered data (interannual); and (c, f) low-pass filtered data (interdecadal). (g, h) Corresponding PCs, with gray for the unfiltered, green the interdecadal, and red the low-pass filtered component of the gray.

Circulation anomalies and SSTAs related to these modes are shown in Fig. 2, as obtained by linear regression onto the corresponding PCs. For EOF1, the wet anomalies over most of eastern China and the dry anomalies over the South China coast and South China Sea are associated with a large-scale anomalous anticyclonic circulation over the western North Pacific. The enhanced southerlies over eastern China bring more moisture from the ocean to over the land, favoring the excessive precipitation. Meanwhile, at midlatitudes, an anomalous cyclone over Lake Bagel enables the southerlies to penetrate further north. Thus, the combined effect from mid and low latitudes plays an important role in the increased rainfall over most of eastern China. As controlled under the anomalous anticyclonic circulation, a rainfall deficit is seen along the southeastern coast of China and the South China Sea due to anomalous moisture divergence.

Figure 2.  Linearly regressed atmospheric fields and SSTAs associated with the normalized PCs of the two interannual EOF modes shown in Fig. 1: (a, b) precipitation (shading; units: mm d^-1) and 850-hPa winds (arrows; units: m s^-1); (c, d) 200-hPa geopotential height (shading; units: m) and zonal winds (contours; units: m s^-1); (e, f) SSTA (units: K); and (g, h) 500-hPa vertical velocity field (shading; units: 10^-2 m s^-1), 200-hPa velocity potential (contours; units: 10⁵ m² s^-1) and divergent winds (arrows; units: m s^-1).

Based on previous studies (Wang et al., 2013), the anomalous anticyclonic circulation over the western North Pacific as seen in Fig. 2a can be forced by tropical SSTAs. As shown in Fig. 2e, the SSTAs related to the monopole mode resemble a La Niña-like pattern. The cold SSTAs over the central eastern Pacific from 180^° to 100^°W could generate descending Rossby waves over the western North Pacific, causing an anomalous anticyclonic circulation. It is also accompanied by a strengthened Walker circulation over the tropical Indo-Pacific (Fig. 2g). With the accelerated overturning circulation, the Indian Summer monsoon becomes stronger as more moisture is converging in this region (Fig. 2a), which is consistent with previous studies (Chen and Yen, 1994; Kumar et al., 1999). On the one hand, the enhanced Indian summer monsoon could in turn strengthen the ascending branch over the northern Indian Ocean. On the other hand, the latent heat released by the abundant rainfall could also interact with the wave activity along the westerly jet at midlatitudes, influencing the East Asian monsoon (Wu, 2002). As shown in Figs. 2c and 3a, the monopole mode is correlated with a wave train propagating globally over the midlatitudes. This mode is very similar to the EOF2 of the 200-hPa geopotential height field in this model (figure not shown), reminiscent of the circumglobal teleconnection pattern proposed by (Ding and Wang, 2005). A negative geopotential anomaly center of this wave train sitting over Lake Bagel induces anomalous southerlies over North China. Meanwhile, the downstream positive center spans from eastern China to the North Pacific but locates further south compared to the typical circumglobal teleconnection pattern (Ding and Wang, 2005). This center may enhance and broaden the anomalous anticyclonic circulation over the western Pacific generated directly by tropical SSTAs, producing stronger southerlies within eastern China and feeding moisture to this area.

Figure 3.  Linearly regressed vorticity anomalies (shading; units: 10^-6 s^-1) and wave activity fluxes (arrows; units: m-2 s-2) at (a, c) 200 hPa and (b, d) 850 hPa associated with the normalized PCs of the two interannual EOF modes shown in Fig. 1.

In contrast to the monopole mode, the dipole rainfall pattern is associated with a pair of anomalous anticyclonic and cyclonic circulations along the coast of East Asia (Fig. 2b). In this case, the wet anomalies emerge zonally from the Yangtze River to South Korea and Japan, and the dry anomalies are prominent over North China, the South China coast and South China Sea. The anomalous northerlies over North China, induced by the anomalous cyclone sitting north of its anticyclonic circulation counterpart, inhibit the moisture transported further north and lead to less rainfall there. Compared to the monopole mode, the dipole mode has a closer link with the tropics. There are significantly warm SSTAs (Fig. 2f) over the northern Indian Ocean, the Maritime Continent, the eastern tropical Pacific, and the tropical Atlantic, corresponding to the anomalous upward motion (Fig. 2h) over these regions. The entire upper troposphere over the tropical region is lifted by the latent heating (Fig. 2d). The anomalous upward motion over the Maritime Continent induces a local Hadley cell with the subsidence center located over the Philippine Sea (Fig. 2h). The subsidence center is also enhanced through the local anomalous Walker circulation related with a warmer Indian Ocean (Fig. 2f) and the wind-evaporation-SST feedback (Xie and Philander, 1994) related with the cold SSTAs southeast of it. This strong anomalous center over the Philippine Sea then triggers a Rossby wave train (Fig. 3d) propagating northward along the East Asian coast following the route of the Pacific-Japan/East Asia-Pacific teleconnection pattern (Nitta, 1987; Huang, 1992).

The above analysis indicates the important role of tropical SSTAs in forming the monopole and dipole rainfall distribution over eastern China, but how do these SSTAs form themselves? Figure 4 shows the evolution of 850-hPa horizontal winds and SSTAs from the preceding winter (December-January-February), spring (March-April-May) to the summer (June-July-August). For the monopole mode (Figs. 4a, c and e), the SSTAs correspond to a developing La Niña event. The cold SSTAs over the central and eastern Pacific are first developed from north of the equator between 180^° and 120^°W in previous seasons. During the previous winter, an anomalous anticyclonic circulation sits over the position of the Aleutian low, producing anomalous northerlies along its eastern side. When the climatological trade wind starts building in the spring, cold SSTAs form under it and then become enhanced and extend southward in the following summer. The atmosphere responds to the cold SSTAs following the Gill response (Gill, 1980), producing an anticyclonic anomaly to its northwest. The cold SSTAs and the anomalous anticyclonic circulation both become strengthened through the wind-evaporation-SST feedback, eventually forming the La Niña-type SSTAs.

Figure 4.  Linearly regressed evolution of SSTAs (shading; units: K) and horizontal winds (arrows; units: m s^-1) onto the normalized PC1 (left column) and PC2 (right column) on interannual time scales. Statistically significant (95%) wind anomalies are shown in black arrows and grey arrows are not significant.

For the dipole mode (Figs. 4b, d and f), a mature El Niño SSTA is seen from the previous winter and spring. As the anomalous westerlies over the equator become weaker and even change direction in summer, a decaying phase of El Niño appears. The weakened anomalous westerlies and the strengthened climatological easterly wind blow the water from the warm pool to the Maritime Continent, inducing warmer SSTs and enhanced convection there. During the decaying of the El Niño, the basin-wide warmer Indian Ocean forming from the previous winter to the summer, also contributes to the formation of the anomalous anticyclonic circulation over the Philippine Sea.

On interdecadal time scales, the circulation and SSTAs related to the monopole and dipole are similar to those on interannual time scales. The monopole mode (Figs. 5a, c and e) results from the combined effect of a midlatitude wave train and the western North Pacific anticyclonic circulation induced by the cold SSTAs over the central and eastern tropical Pacific. The dipole mode (Figs. 5b, d and f) is linked to a Pacific-Japan/East Asia-Pacific wave pattern correlated with the warm SSTAs over the Maritime Continent. However, there are also notable differences.

For the monopole mode, the SSTAs (Fig. 5e) resemble the structure of a typical Interdecadal Pacific Oscillation (Zhang et al., 1997). Compared with the SSTAs on interannual time scales (Fig. 2e), the warm SSTAs over North Pacific have a larger extent and the cold SSTAs over the tropics also extend further into the eastern Pacific. Moreover, the rainfall anomalies over India (Fig. 5a) are much less than those on interannual time scales and the wave train over midlatitudes (Fig. 6a) propagates more meridionally over the Eurasian continent. For the dipole mode, more influences from midlatitudes are found than on interannual time scales, particularly over the Ural Mountains (Fig. 6c). Besides, a stronger relationship between North China and Indian summer monsoon rainfall is related with the dipole pattern (Fig. 5b).

Figure 5.  Linearly regressed atmospheric fields and SSTAs associated with the normalized PCs of the two interdecadal EOF modes shown in Fig. 1: (a, b) precipitation (shading; units: mm d^-1) and 850-hPa winds (arrows; m s^-1); (c, d) 200-hPa geopotential height (shading; units: m) and zonal winds (contours; units: m s^-1); (e, f) SST (units: K).

Figure 6.  Linearly regressed vorticity anomalies (shading; units: 10^-6 s^-1) and wave activity fluxes (arrows; units: m^-2 s^-2) at (a, c) 200 hPa and (b, d) 850 hPa associated with the normalized PCs of the two interdecadal EOF modes shown in Fig. 1.

Although the rainfall patterns and their driving factors in the control simulation are similar on interannual and interdecadal time scales, their responses to anthropogenic greenhouse gas forcing are very different. In the 4×CO₂ scenario (Fig. 7), leading patterns in the control simulation still dominate in 4×CO₂ on interannual time scales (Figs. 7b and e) and in the unfiltered data (Figs. 7a and d), but not on interdecadal time scales (Figs. 7c and f). On interannual time scales, the dipole becomes the first leading mode in 4×CO₂ (Fig. 7b), while the monopole becomes the second (Fig. 7e), indicating a more dominant role and more frequent occurrence of a dipole distribution of rainfall anomalies in the future. On interdecadal time scales, both leading modes (Figs. 7c and f) show limited similarity with the ones in the control simulation (Figs. 1c and f), but with a more east-west orientation. EOF1 shows rainfall anomalies with opposite signs between the southeast and the northwest of eastern China, while EOF2 shows opposite rainfall anomalies between most of eastern China with south coastal region and Inner Mongolia.

Figure 7.  The first two EOF modes of summer rainfall variability over eastern China simulated by the HadGEM3-GC2 abrupt 4×CO₂ experiment: (a, d) from unfiltered data; (b, e) high-pass (7-year truncation) filtered data (interannual); and (c, f) low-pass filtered data (interdecadal). (g, h) Corresponding PCs, with gray for the unfiltered, green the interdecadal, and red the low-pass filtered component of the gray.

Considering that the data length of the 4×CO₂ experiment (150 years) is shorter than in the control simulation (315 years), the changed order of the two leading EOFs in 4×CO₂ may result from the sensitivity of the EOF analysis to the input data length or the internal variability instead of CO₂ forcing. To clarify this, running EOFs are performed with each 150-year length of data in the control simulation and a 10-year moving length is applied. There are 15 150-year sub-spans obtained in the control simulation. For each sub-span, the spatial correlation coefficients between its first five leading modes with the monopole (Figs. 1a-c), dipole (Figs. 1d-f), and the EOF3 (not shown) obtained in the control simulation on the corresponding time scales are calculated. The maximum correlation among them for each mode is shown in Figs. 8a and b, in which a red box means the mode in this particular sub-span most resembles the monopole, a green box represents resemblance to the dipole, and a purple one to the EOF3, in the whole control simulation. Here, we only show those correlation coefficients larger than 0.5. The corresponding variance contribution of the monopole-like (red) and dipole-like (green) leading modes for each sub-span on interannual time scales and their differences (purple) are shown in Fig. 8c. Modes that are well separated from the other modes (North et al., 1982) are filled.

Figure 8.  Spatial correlation coefficients between the EOFs of each sub-span (150 years) in the control simulation and 4×CO₂ (150 years) with the EOF1-3 of the whole control simulation (315 years) at (a) interannual and (b) interdecadal time scales. EOFs with maximum correlation with the monopole, dipole and EOF3 in the control simulation are marked as red, green and purple, respectively. The percentage variance explained by the first two leading modes on interannual time scales and their difference are shown in (c). Modes that separate well from other modes according to North's criteria (North et al., 1982) are filled.

Figure 8a shows that the monopole and dipole mode are nearly always the first leading modes on interannual time scales, despite varied percentage of variances. There are 10 in the 15 sub-spans in which the first two leading modes are in the same order as in the control simulation (Fig. 1), and eight of them are well separated according to the rule of thumb (North et al., 1982). Only two in the 15 sub-spans (starting at model year 2172 and 2182) have the dipole as the first leading mode and monopole as the second, but both modes cannot be well separated (Fig. 8c). The first two leading modes of the remaining three sub-spans, however, both resemble the dipole, and again, cannot be well separated (Fig. 8c). This indicates the consistently dominant role of the monopole mode in the control simulation. Therefore, the leading two modes that are well separated in 4×CO₂ but in inverse order to the control simulation are more likely a result of climatic change under anthropogenic CO₂ forcing.

On interdecadal time scales, the dominance of the monopole rainfall mode is even more consistent (Fig. 8b). All the 15 sub-spans show the same order of leading modes as in Fig. 1, albeit the variance contribution varies. However, as shown in Fig. 8b, Fig. 7c and Fig. 7f, both EOF1 and EOF2 on interdecadal time scales under 4×CO₂ show limited similarity with the ones in the control simulation.

In order to examine the possible mechanism for the changes in rainfall patterns under 4×CO₂, the related circulation and SSTAs are given in Figs. 9-11.

As shown in Fig. 9, on interannual time scales, the circulation anomalies associated with the leading rainfall modes under 4×CO₂ anthropogenic warming are similar to their counterparts in the control simulation (Fig. 2). Differences are found mainly in the SSTAs related to the dipole (Fig. 9e——compare to Fig. 2f), i.e., the strengthened and extended warm SSTAs over the Maritime Continent and Indian Ocean compared to the control simulation.

The analysis in section 3.2 shows that ENSO-like SSTAs in the preceding winter and its related signals over the Maritime Continent and the Indian Ocean in the summer play a critical role in the formation of the dipole rainfall pattern. To figure out whether the strengthened relationship between this mode and the aforementioned SSTAs is a response to anthropogenic forcing, the correlation coefficients between the PC of the dipole patterns (corresponding to the green boxes in Fig. 8a) and the area-averaged SSTAs over the Indian Ocean (5^°S-10^°N, 50^°-85^°E), the Maritime Continent (10^°S-5^°N, 95^°-140^°E) in the summer, and Niño3.4 region (5^°S-5^°N, 170^°-120^°W) in the previous winter, are calculated. Results for each sub-span of the control simulation (dots), the whole control simulation (triangles), and for 4×CO₂ (stars), are shown in Fig. 10. The greatly enhanced relationship between the dipole rainfall pattern and the preceding Niño3.4 SSTAs is not only seen in the comparison with the whole control simulation, but also for each sub-span. However, it is not the case that both the Indian Ocean and Maritime Continent SSTAs show same result. The strengthened relationship between the Indian Ocean and the dipole rainfall mode in 4×CO₂ is unprecedented in the control simulation, but not for the Maritime Continent (one sub-span in the control simulation shows a similar magnitude of correlation with the rainfall mode as in 4×CO₂). This indicates that the strengthened relationship between the preceding ENSO with the dipole is more likely a response to anthropogenic forcing, and operates mainly via an enhanced role of the Indian Ocean.

Figure 9.  Linearly regressed atmospheric fields and SSTAs associated with the normalized PCs of the two interannual EOF modes in the 4×CO₂ simulation shown in Fig. 7: (a, b) precipitation (shading; units: mm d^-1) and 850-hPa winds (arrows; units: m s^-1); (c, d) 200-hPa geopotential height (shading; units: m) and zonal winds (contours; units: m s^-1); (e, f) SSTA (units: K).

It has been pointed out that the Indian Ocean can act as a "capacitor" (Xie et al., 2009), extending the ENSO signals over the western North Pacific from the previous winter to the next summer. Under global warming, an enhanced "capacitor" effect of the Indian Ocean has been reported (Zheng et al., 2011). Consistent with these studies, more obviously wedge-shaped geopotential anomalies at 200 hPa over the Indo-western Pacific is seen in 4×CO₂ (Fig. 9c), indicating a warm Kelvin wave response to the change of conditions over the Indian Ocean. An enhanced Pacific-Japan/East Asia-Pacific pattern and corresponding rainfall anomalies over eastern China therefore appear.

Figure 10.  Correlation coefficients between the PCs of the interannual dipole modes and the summer SSTAs averaged over the Indian Ocean (5^°S-10^°N, 50^°-85^°E), the Maritime Continent (10^°S-5^°N, 95^°-140^°E), and the previous winter SSTAs averaged over the Niño3.4 region (5^°S-5^°N, 170^°-120^°W) during the whole control simulation (triangles), each 150 year sub-span in the control simulation (dots, corresponding to the green boxes in Fig. 8a), and the 4×CO₂ scenario (stars), respectively. Correlation coefficients statistically significant at the 95% confidence level are marked in red.

On the interdecadal time scales, the first two leading rainfall modes in 4×CO₂ are very different from those in the control simulation——not only for the rainfall anomalies over eastern China, but also for the related rainfall anomalies in other regions. In the control simulation, the regional rainfall modes over eastern China are related to a larger-scale banded structure extending from East Asia to the western North Pacific, particularly for the dipole pattern (Fig. 5b). However, under 4×CO₂, the similarities are limited (Fig. 11), and the SSTAs related to them are also very different. One notable difference is the SSTA pattern related to EOF1 (Fig. 11e). In the control simulation, it resembles the Interdecadal Pacific Oscillation pattern, with the SSTA signals most significant over the eastern Pacific (Fig. 5e). However, under 4×CO₂ forcing, the significant tropical SSTAs are more confined to the central Pacific (Fig. 11e), indicating a potential impact of a weakening Walker circulation under global warming (Vecchi and Soden, 2007). Another notable difference appears over the high-latitude North Atlantic and the Arctic. Significant correlations of the SSTAs are found over these regions, related to both leading rainfall modes (Figs. 11e and f). Meanwhile, strengthened zonally characterized wave patterns over the mid and high latitudes are also seen in both modes (Figs. 11c and d). Recently, the role of the decadal variation of SSTAs over the North Atlantic has been emphasized as an important source for the midlatitude wave train, affecting the summer climate variability in China (Si and Ding, 2016; Wu et al., 2016; Yang et al., 2017). Thus, the enhanced relationship between the leading modes on interdecadal time scales with the high-latitude circulation might result from the influences of the North Atlantic Ocean.

Figure 11.  Linearly regressed atmospheric fields and SSTAs associated with the normalized PCs of the two interdecadal EOF modes shown in Fig. 7: (a, b) precipitation (shading; units: mm d^-1) and 850-hPa winds (arrows; m s^-1); (c, d) 200-hPa geopotential height (shading; units: m) and zonal winds (contours; units: m s^-1); (e, f) SSTA (units: K).

Using the latest version of the Hadley Centre's coupled climate model, HadGEM3-GC2, this paper focuses on the potential changes in summer rainfall patterns over eastern China with two idealized experiments: a pre-industrial control simulation and an abrupt quadrupled CO₂ simulation (4×CO₂). Instead of looking at long-term trends in mean precipitation, we concentrated on modes of internal variability on interannual and interdecadal time scales. The main conclusions are:

(1) In the control simulation, the first two leading modes account for about 20% of summer rainfall variability. The spatial patterns, variance contribution and their associated circulation and SSTAs (possible driving factors) are similar on both interannual and interdecadal time scales. EOF1 is largely a monopole mode that covers most of eastern China, apart from the South China coast. EOF2 is a meridional dipole mode with a boundary around 35^°N.

(2) On interannual time scales, the monopole mode is associated with the developing phase of ENSO events, during which the tropical SSTAs emerge from an anomalous anticyclonic circulation over the North Pacific in the previous winter. The combined effect of the SST driving anomalous western Pacific anticyclonic anomaly and a midlatitude wave train leads to enhanced rainfall anomalies over eastern China. The dipole mode is associated with the decaying phase of ENSO events. Warmer SSTAs from the Maritime Continent and the Indian Ocean strengthen the anomalous western Pacific anticyclonic circulation and induce a Pacific-Japan/East Asia-Pacific pattern propagating northward along the East Asian coast. The combined circulation anomalies result in contrasting rainfall anomalies over the north and south part of eastern China. The analysis on interdecadal time scales shows similar key regions of SSTAs and atmospheric circulation, which might well be model dependent.

(3) In a 4×CO₂ world, the leading patterns of summer rainfall variability over eastern China remain unchanged on interannual time scales, but with the dipole mode becoming the first EOF. The changes on interdecadal time scales are stronger than those on interannual time scales, as the first leading mode becomes a dipole mode with a more east-west orientation. On interannual time scales, the main reason is the strengthened influence from the Indian Ocean of excessive warming and stronger relationship with the SSTAs in the previous winter over the tropical Pacific. On interdecadal time scales, there is evidence of increased influence from the high latitudes and potential impact of a weakening Walker circulation.

The leading modes of summer rainfall variability over eastern China have been pointed out as a dipole and tripole during recent years. In the model's control simulation, they may not be exactly the same as recent observations may have shown (e.g. Huang et al., 2011; Ye and Lu, 2012; He et al., 2016), as historical external forcing factors are not included and observational data cover a much shorter period of time compared with the 315 years of the model simulations. As a matter of fact, using over 400 years of historical proxy data, (Pei et al., 2015) suggested that the monopole rainfall mode might have been the dominant pattern of eastern China rainfall variability until recently, when the dipole mode became more frequent. Their findings, from a historical perspective, support our result revealed by climate models. There is a possibility that external forcing, be it anthropogenic or natural, may have played a role in adjusting the leading patterns of summer rainfall variability over eastern China.

For the tripole, which has been considered an important mode of eastern China rainfall by previous studies on the interannual time scale (Hsu and Lin, 2007), this model shows limited ability in simulating it. Although it is the EOF3 of the pre-industrial control simulation in this model (figure not shown), only a weak relationship with the SSTAs over the Maritime Continent is seen and no obvious upper-level teleconnection is found. The mechanisms are therefore not very clear or consistent with previous studies (Huang et al., 2006; He et al., 2017). Under 4×CO₂ forcing, the tripole no longer exists in the first three leading modes. Whether this is due to the model's poor performance in simulating it, or it is the response to the anthropogenic forcing, needs further investigation.

The known bias in this model regarding its poor simulation of the Indian monsoon (Williams et al., 2015), which has been thought to play an important role in maintaining or triggering the midlatitude circumglobal wave pattern (Ding and Wang, 2005), may have contributed to the difference as well. The amount and position of the latent heating released by the Indian summer monsoon could have a downstream influence on North China rainfall.

The important message from this study is the potential relative change between the 4×CO₂ scenario and the control simulation. The exact change in rainfall patterns for the future requires further studies with multi-model ensembles and even higher resolution models to resolve the complex orography, coastlines and air-sea interactions. The main point for readers to take away from this paper is the qualitative message rather than the absolute details.