The onset and withdrawal of rainy season was determined by the multi-scale moving t test method. This method is characterized by the detection of mutation points between two subsamples with equal size n, where n is the length of the subsample (n=30, 31, …, 182/183; 182/183 corresponds to half the value of length of 1 year 365/366). Theoretically, the length of subsamples in this study ranged between 1 and 182/183. However, as the onset or withdrawal of the rainy season, it will not be considered if the length of the subsample is one day or just several days when the abruption point is prominent. As a result, the length of the subsample is limited between 30 and 182/183. The determination of the mutation point can be described as (Fraedrich et al., 1997) tn,i=(x¯i2-x¯i1)n1/2(si22+si12)-1/2, where x¯i1 and x¯i2 defined as x¯i1=∑j=i-ni-1xjn;si12=∑j=i-ni-1(xj-x¯i1)2/(n-1),x¯i2=∑j=ii+n-1xjn;si22=∑j=ii+n-1(xj-x¯i2)2/(n-1), and xi is daily precipitation for Julian day i within 1 year and for one station. x¯i1 and x¯i2 are the mean values of the subsamples before and after the Julian day i, respectively.

The t value calculated above was normalized by the 0.01 test value shown in Eq. (4), which is equal to the result of Mann–Kendall test at 0.05 significance level. trn,i=t(ni)/t0.01(n), where trn,i can be taken as the threshold to detect mutations. trn,i>1.0 represents an increasing trend while trn,i<1.0 is a decreasing trend. The onset of rainy season in this study was defined as the mutation point corresponding to a maximum trn,i value. For this case, precipitation changes from a smaller to a higher value. Likewise, the withdrawal is defined as the changing point corresponding to a minimum trn,i value.

Three types of ENSO were classified based on the definition proposed by Kim et al. (2009). The years dominated by CPW, EPC and EPW are listed in Table 1.

The definition of ENSO Modoki and conventional ENSO was demonstrated. Specifically, warm (cold) episodes of ENSO Modoki, abbreviated as MEN (MLN), were defined as EMI above (below) 0.7 SD (-0.7 SD), where SD is the standard deviation (Ashok et al., 2007). EMI=SSTAA-0.5×SSTAB-0.5×SSTAC, where [SSTA]A, [SSTA]B, [SSTA]C represents the SSTA in region A (10∘ S–10∘ N, 165∘ E–140∘ W), region B (15∘ S–5∘ N, 110∘ W–70∘ W) and region C (10∘ S–20∘ N, 125∘ E–145∘ E), respectively. Likewise, the conventional EN (LN), abbreviated as CEN (CLN), was defined as SSTA above (below) 0.7 SD (-0.7 SD) in the area of 5∘ N–5∘ S, 90∘ W–140∘ W (Tedeschi et al., 2013). This definition gives an opportunity to judge the ENSO type of the rainy season rather than the whole year, which is greater than definition proposed by Trenberth (1997).

Precipitation anomaly can present the difference of precipitation between ENSO years and normal years and demonstrate the influence of ENSO regimes on precipitation more directly. Q. Zhang et al. (2013) used precipitation anomaly index to explore the effect of ENSO on precipitation in the East River basin, southern China. Q. Zhang et al. (2016) investigated the influence of ENSO and ENSO Modoki on seasonal precipitation over the Huaihe River basin by using precipitation anomaly index. Precipitation anomaly index is defined as PARSij=PRSij¯PRSNij¯-1×100%, where PARSij denotes precipitation anomaly during rainy season at ith station in jth year; PRSij¯ denotes mean daily precipitation during rainy season at ith station in jth year, and PRSNij¯ denotes mean daily precipitation during rainy season at ith station in jth normal year. The normal year refers to a year without ENSO event occurring.

The spatial variability of PARS under CPW, EPC and EPW regimes in the phase of developing and decaying years is presented in Fig. 2. The distribution of precipitation anomaly is irregular over the whole area in the developing phase of CPW. The coastal regions in southeastern China that had the largest amount of rainy-season precipitation presented the largest decreasing trend, with the precipitation anomaly reaching 30 % below average precipitation. The upper and middle reaches of the Yangtze River and the Yellow River showed decreasing precipitation, whereas the lower reaches had the opposite trend. The decaying CPW regime had a relatively regular spatial pattern. More specifically, most parts of China presented increasing precipitation during rainy season, with the largest PARS being 20 % above average precipitation. The distribution of PARS influenced by the decaying CPW is similar to that by the developing EPC, with shrinking extent of enhanced precipitation in central China for developing EPC. The distribution of PARS is similar as well in the two phases of EPC (Fig. 2, second row), with precipitation above average in northwestern China and precipitation below average in northeastern China. The difference between the two phases lies in the increasing (decreasing) precipitation in southeastern China in the developing (decaying) phase. Nonetheless, developing and decaying EPW (Fig. 2, third row) showed opposite spatial precipitation pattern. Most parts of China presented dry signals in the phase of developing EPW, which became stronger northwards, and more than 30 % below average precipitation can be identified in northern China. However, there is above average precipitation in most regions of China in the case of decaying EPW, with PARS values ranging between 0 and 30 %. In summary, the CPW decaying phase (EPC developing phase) deserves more attention than the developing (decaying) phase, since it shows more prominent wet signals. Both phases are significant for the EPW regimes, due to the obvious dry (wet) signals shown in the developing (decaying) phase.

Figures 3 and 4 present precipitation anomalies during rainy season (PARS) for warm and cold episodes of conventional ENSO and ENSO Modoki in a developing phase and a decaying phase, respectively. Precipitation increased in a band stretching from northwestern China to the coastal region in the southeast, with the largest precipitation anomaly (40 %) occurring in southeastern China under a developing CEN regime (Fig. 3a). The dry condition is more severe northwards in central China, with PARS equal to about -30 % in the northern parts of central China. W. Zhang et al. (2016) concluded that strong El Niño events are associated with summer monsoon flooding over the Yangtze River, which is consistent with our results. The distribution of rainy-season precipitation for developing El Niño is also in agreement with the research by Zhang et al. (2011). Northern China had an opposite PARS pattern for developing El Niño Modoki, in comparison to developing CEN (Fig. 3b). Nonetheless, the two phases showed similar precipitation distribution, with reduced precipitation in central China (approximately -10 %) and enhanced precipitation in southern China. Typically, developing CEN demonstrated more obvious wet or dry signals compared to MEN. Moreover, the wet and dry condition for developing CEN is the most serious among all ENSO and ENSO Modoki regimes in both developing and decaying phases, with the largest precipitation anomaly reaching 50 % above average precipitation amount and lowest 30 % below. This means that developing CEN should be paid urgent attention to for flooding and drought monitoring. The spatial pattern of PARS for developing CLN presented similar signals with developing MEN, with a shorter wet precipitation band in northern China for developing CLN (Fig. 3c). The increased precipitation was shifted westwards for the developing MLN, compared to cold episodes of conventional ENSO (Fig. 3d). ENSO and ENSO Modoki regimes in the developing phase presented various distribution of precipitation anomalies. Wet or dry signals are more easily shown for the warm episodes of conventional ENSO, in comparison to the other three regimes. Similar patterns of PARS for developing CLN and MEN should be further studied.

Decaying ENSO and ENSO Modoki years showed different features of PARS (Fig. 4). Most parts of China presented increasing precipitation for decaying CEN, with more than 30 % above average precipitation identified in northern China (Fig. 4a). The decaying phase of MEN (Fig. 4b) presented shrinking extent of enhanced precipitation, which was condensed in the central parts of China, ranging between 0 and 10 %. The result is consistent with conclusions from Feng et al. (2011), who found obvious rainfall anomalies in southern China for decaying El Niño and no prominent rainfall variations in the corresponding phase of El Niño Modoki. In terms of the cold episodes of ENSO (Fig. 4c), approximately 95 % of China showed dry signals, and the condition was more serious eastwards, being 30 % below average precipitation amount. We can see that the spatial pattern of PARS for the decaying CLN is opposite to that of CEN. Decaying MLN (Fig. 4d) showed a larger extent of enhanced precipitation in a band stretching from western China to parts of the Yellow River basin, in comparison to CLN. In conclusion, the decaying phases of conventional ENSO showed more obvious wet or dry signals compared to ENSO Modoki, with most parts of China displaying increasing (decreasing) precipitation for the CEN (CLN).

This study analyzed spatial patterns of precipitation under different ENSO regimes, since ENSO is the leading driver of precipitation anomaly in China (Xiao et al., 2015). Xu et al. (2016) revealed that increasing autumn precipitation in southern China is due to the combined ENSO and Indian Ocean Dipole (IOD) events. Other researchers also concluded that IOD and ENSO have mutual impact on precipitation anomalies in China (Weng et al., 2011; Liu et al., 2009; Wu et al., 2012). Moreover, Pacific Decadal Oscillation, subtropical high, also influences the distribution of Chinese precipitation (Chan, 2005; Wang et al., 2008; Chang et al., 2000; Niu and Li, 2008; Ouyang et al., 2014). As a result, the spatial patterns of PARS under ENSO regimes may be determined not only by ENSO but also by the combination of various drivers, which ought to be studied further.

Figure 5 presents the composites of 850 mbar vector wind for three types of ENSO. There is a strengthening of westerly and southwesterly wind in the decaying year of CPW (Fig. 5b), which brings more moisture to China, compared to developing CPW (Fig. 5a). This may explain the enhanced precipitation in decaying CPW (Fig. 2b). The difference between developing and decaying EPC (Fig. 5c–d) lies in the shift of anti-cyclonic flow in the western part of North Pacific (WNP). The eastward anti-cyclone for the decaying EPC weakened the transportation of moisture in eastern China and caused reduced precipitation (Fig. 2d). The decaying EPW (Fig. 5f) experienced stronger western and southwestern wind but weakened anti-cyclone compared to the developing phase (Fig. 5e). The WNP anti-cyclone could bring plentiful moisture to China, so weakened anti-cyclonic flow will cause reduced precipitation (Feng et al., 2011). However, most parts of China presented wetter signals in the phase of decaying EPW in comparison to developing EPW. Therefore, it can be pointed out that the India monsoon plays a more significant role in the formation of rainy-season precipitation during EPW phases compared to the atmospheric circulation.

Figure 6 shows the underlying causes of different performance of conventional ENSO and ENSO Modoki in the developing phase by analysis of the 850 hp wind. Compared to developing CEN, developing MEN experienced reduced precipitation in western China and generally enhanced precipitation in eastern parts under the combined influence of stronger monsoon and weakened anti-cyclone (Fig. 6a–b). Stronger anti-cyclonic flow in the phase of developing La Niña Modoki (Fig. 6d) may cause the enhanced precipitation in western parts of China compared to conventional La Niña regime in developing years (Fig. 6c).

The wind composites of warm and cold episodes of decaying ENSO and ENSO Modoki are presented in Fig. 7. Compared to decaying CEN, the wet signal of precipitation is weaker in the decaying year of MEN, which may be attributed to the weakened anti-cyclonic flow in WNP and western winds for the decaying MEN. The difference of wind composites between decaying CLN and MLN indicates similar configuration, with stronger westerly wind and anti-cyclone causing enhanced precipitation for decaying MLN.

In summary, westerly winds seem to play more significant role in the phase of CPW and EPW, while developing La Niña and La Niña Modoki are dominated by the anti-cyclone. The spatial pattern of PARS is the reflection of combined influence of westerly winds and anti-cyclonic flow for the EPC and decaying ENSO and ENSO Modoki regimes.

It can be seen that the spatial pattern of precipitation during the rainy season in China is dominated by westerly winds from India and anti-cyclone in WNP, which is equivalent to the results by Dai and Wigley (2000), Feng and Li (2011), and Wu et al. (2003). Generally, stronger western and southwestern winds are related to increasing precipitation. It is in agreement with the research of Zhang et al. (1996) and Wang et al. (2000), who pointed out that southeastern and southwestern winds could substantially enhance the moisture transportation to China. Wu et al. (2003) also found that East Asian monsoon is positively related to precipitation variations, which is consistent with our result. Likewise, the westward and stronger anti-cyclone is related to enhanced PARS. Wu et al. (2003) reported that the anomalous low-level anti-cyclone is determined by large-scale equatorial heating anomalies and local air–sea interactions. Westerlies and anti-cyclone are of dominant importance for the ENSO-induced precipitation during the rainy season. However, cyclonic flow may have larger influence on Chinese precipitation under certain circumstances. For example, the autumn drought in southwest China in 2009 was determined by a strong cyclone in WNP for ENSO Modoki (W. Zhang et al., 2013b). Feng et al. (2011) also revealed that the WNP circulation is cyclonic in winter and then becomes weak in the following spring and anti-cyclonic flow in summer for El Niño Modoki. As a consequence, WNP anti-cyclone has a larger effect on East Asian precipitation on the inter-annual or inter-decadal scale, but anti-cyclone and cyclone are both crucial for the determination of precipitation on the annual or smaller scale.