Drought is an extreme climate event common in eastern China. It has a huge impact on the environment, economy, and society. Against the background of global warming, the occurrence of extremes will increase (Ning and Bradley, 2015c,d), and the intensity, frequency, and duration of drought have shown an increasing tendency in both observations and model results (Dai, 2012;Huang et al., 2015). Given the increasing severity of its influence, it is becoming ever more urgent to understand the characteristics and mechanisms of drought, and to better forecast drought events in the future, especially in the eastern China.

Previous studies that have analyzed drought events in eastern China tended to focus on observational datasets during the 20th century warm period (Ye et al., 2013;Zhang and Zhou, 2015), revealing that the drought in this region is significant influenced by the variability of the East Asian summer monsoon (EASM). Meanwhile, many studies have shown that remote oceanic influences have a huge impact on the EASM, such as the El Niño–Southern Oscillation (ENSO) (Wang et al., 2000;Ning and Bradley, 2014,2015a;Li et al., 2015) and Indian Ocean dipole (IOD) (Ashok et al., 2004;Zhao et al., 2011). Both of these may cause a change in the SST gradient in the tropical ocean and shift the longitudinal position of the Walker circulation (Bjerknes, 1969). As a result, anomalous subsidence occurs in the East Asian monsoon region, weakening the strength of the EASM and leading to serious and persistent droughts (Chen et al., 2013;Gao et al., 2014;He and Liu, 2016). Therefore, the studies mentioned above have revealed some interesting results with respect to the period of global warming, but are they applicable over a broader historical timeframe?

Considerable effort has been put into reconstructing summer precipitation or drought indices in eastern China using historical documents (Zheng et al., 2006), stalagmites (Zhang et al., 2008;Kuo et al., 2011;Li et al., 2011;Wan, 2011), lake sediments (Liu et al., 2011), and tree rings (Cook et al., 2010). Based on instrumental and paleoclimate observations,Zheng et al. (2006) used a proxy precipitation index and found that precipitation in eastern China exhibited four dry epochs: the 500–870, 1000–1230, 1430–1530, and 1920–1990s. However, there were strong regional differences in terms of precipitation variation.Li et al. (2011) used stalagmite δ¹⁸O measurements from Furong Cave to show that the climate in Chongqing was relatively dry during the 250–1150 and 1450–1600s. However, a record from Wanxiang Cave failed to show any drought events in the 550–850s, although a decreasing trend in terms of precipitation during this period was detected (Zhang et al., 2008). The reason for such inconsistent results may derive from uncertainties in reconstructed datasets. Furthermore, these efforts concentrate on the characteristics of drought events; the climate dynamics and causes of drought historically are less clear because of the constraints imposed by the spatial resolutions of reconstructed datasets and the inability to rebuild the wind fields of these times.

Recently, an increasing number of scientists have begun to simulate and analyze the characteristics and mechanisms of drought events over longer historical periods. One such attempt indicated that solar irradiation may be the main driver for the occurrence of drought events on the decadal time scale (Peng et al., 2014). However, they also found that large volcanic eruptions play an important role in some drought events. They analyzed 18 cases of large volcanic eruptions and the results showed that summer precipitation over eastern China decreased significantly in the year of an eruption, and the year after (Peng et al., 2010). The reason may be the decrease in water vapor over the tropical oceans. Additionally,Man et al. (2014) found that anomalous northerly winds dominated the area of East Asia after large volcanic eruptions, contributing to the decrease in the northward transport of tropical water vapor.Hernandez et al. (2015) used a pre-industrial 1300-yr control run of the Community Earth System Model (CESM) to analyze the dynamics of South Asian monsoon failure on the interannual and decadal time scales. Their study showed that eastern Pacific El Niño events had a huge impact on the interannual variability of monsoon rainfall; meanwhile, maximum warming was observed in the central equatorial Pacific on the decadal time scale—similar to the patterns found during El Niño Modoki events. However, these studies of drought have focused mainly on the interannual or decadal time scale; if we want to predict the changes in drought over the next 100 years, it is important to understand the centennial-scale variability and mechanisms of drought over longer historical periods. Additionally, single-forcing sensitivity experiments should be conducted to distinguish the influences of different external forcings.

Precipitation variability plays the most important role in the occurrence of drought events (Zheng et al., 2006); and in the summer seasons in China, precipitation contributes 40% (in wet areas) and over 60% (in dry areas) of the annual precipitation amount (Lei et al., 2011). Therefore, in this study, we used summer precipitation to identify drought events during the past 1500 years over eastern China, and explore their characteristics and mechanisms on the centennial scale. In the next section, the model, the design of the experiments, and the climatic proxy data used in this study are described. Section 3 describes and discusses the time series and spatial distribution characteristics in different drought periods, how these linked to the ocean–atmosphere dynamics of the Indo-Pacific system, and the external forcing influences on the SST of the Indo-Pacific system. A summary is provided in Section 4.

CESM was developed by the National Center for Atmospheric Research in July 2010. The model consists of the Community Atmosphere Model version 4 (CAM4) (Neale et al., 2013), the Community Land Model version 4 (Lawrence et al., 2012), the Parallel Ocean Program version 2 (POP2) (Danabasoglu et al., 2012), and the Sea Ice Model version 4 (Briegleb and Light, 2007). Scientists can freely access the model code and user guide via the website (http://www.cesm.ucar.edu/ models/cesm1.0/). CESM is one of the main models cited in the Fifth Assessment Report of the IPCC (IPCC, 2013), and the performance of CESM has been verified by many researchers (Lehner et al., 2015;Ning et al., 2015;Otto-Bliesner et al., 2015;Yan et al., 2015a,b;Ning and Bradley, 2016).

Due to the long time required for integrating the model and the limited computing resources we had at our disposal, we used the low resolution version (T31_g37, which is equivalent to 3.75° × 3.75°) of CESM to carry out the simulation over the past 2000 years. Specifically, that meant CAM4 had a global range of 48 × 96 grids (latitude × longitude) and 26 vertical levels. POP2 had 116 ×100 grids (latitude × longitude) and 60 vertical levels. We used CESM to simulate a long integration over the past 2000 years, including a control (Ctrl) experiment, an all-forcings (AF) experiment, total solar irradiation (TSI) experiment, volcanic eruptions (Vol) experiment, greenhouse gases (GHGs) experiment, and land use/land cover change (LUCC) experiment (Yan et al., 2013). The Ctrl experiment, which consisted of a pre-industrial 400-yr spin-up run and a 2000-yr control simulation, was run by the fixed external forcing condition of 1850 A.D. (Wang et al., 2015). Except for the Vol experiment, which only spanned from 500 to 2000 A.D, all of the experiments were carried out from 1 to 2000 A.D., based on the 2400-yr Ctrl experiment under the initial forcing of 1850 (Table 1). The time series of the external forcing factors are shown in Fig. 1.

Fig. 1 Time series of the main individual forcings used in the model experiments for the past 2000 years: (a) total solar irradiation (TSI; W m–2); (b) volcanic aerosol mass (VAM; g m–2); (c) greenhouse gases [CO2 (ppm), N2O (ppb), and CH4 (ppb)]; and (d) land use/land cover change [percentage crops in grid cell (POG; %)].

Figure options

To validate the climate variations in the AF experiment with current observations, the observed temperature data was employed, which derived from the Reanalysis-2 dataset of the National Centers for Environmental Prediction (Kanamitsu et al., 2002) and the precipitation data derived from version 2 of the Global Precipitation Climatology Project (Adler et al., 2003). The results (figures omitted) showed a pattern correlation coefficient between the simulated and observed temperature (precipitation) of 0.95 (0.64) in eastern China, and normalized root-mean-square error of 1.31 (1.92) between them. The precipitation was overestimated over the east of the Tibetan Plateau in the simulation, as in many atmospheric general circulation models (Zhou and Li, 2002;Chen et al., 2010). Meanwhile, the annual cycles of precipitation in eastern China were compared. The observed August peak was captured in the simulation, but the precipitation in June–August was underestimated. Overall, the model was found to perform well in simulating the climate variations of eastern China.

Proxy paleoclimatic data were required in order to identify the centennial-scale drought events over eastern China, and the choice of data needed to meet the following three requirements: (1) the proxy data needed to be located in the region of eastern China; (2) the temporal resolution of the proxy paleoclimatic data had to be able to reflect events lasting 50–100 years (Liu et al., 2014); and (3) each proxy record had to be relatively accurate dated in order to reveal short-duration events. Five records are listed in Table 2. The first is the dry-wet index (DWI) of the rainy season (May-September) over eastern China during the past 1500 years (Zheng et al., 2006). This record is also derived from historical documentary evidence and has a 10-yr temporal resolution. Meanwhile, the stalagmites δ¹⁸O records were selected from Furong Cave (Li et al., 2011) and Shihua Cave (Wan, 2011), representing the summer rainfall. The lake sediment record of Gonghai Lake was chosen because of its high sample resolution (Liu et al., 2011). Finally, a 1000-yr time series of annual precipitation derived from documentary evidence in Korea was chosen, which may reflect the strength of the EASM (Kim and Choi, 1987).

The rainy season (May–September) precipitation variation over eastern China (25°–40°N, 105°–123°E) during the past 1500 years from the CESM results is shown in Fig. 2a. The modeled data were processed with a 10-yr running mean to remove the interannual variability signal and maintain the longer timescale variability. The DWI, with a 10-yr temporal resolution (Zheng et al., 2006), is shown in Fig. 2b. The DWI series results indicate that there were many decadal-scale drought and flood events during the past 1500 years, and it seems that centennial-scale drought events were concentrated in the period 501–1600.Zheng et al. (2006) found that the three driest 100-yr periods of the DWI, for the whole of eastern China, were the 1120–1210, 1430–1520, and 620–710s, and their standard deviations relative to the mean value of the time series were –0.91, –0.69, and –0.53, respectively. Two of these periods (620 –710 and 1430–1520) appeared in the time series from Furong Cave (Fig. 2c), but no dryness over eastern China could be found for the 1120–1210.

We also compared the stalagmite δ¹⁸O records from Heshang Cave (Hu et al., 2008), Huangye Cave (Tan et al., 2011), and Wangxiang Cave (Zhang et al., 2008). The first two (Heshang and Huangye caves) suggest dry conditions for the periods 620–710 and 1430–1516; and whilst the record from Wangxiang Cave does not show a negative precipitation anomaly for the period 622–735, but there is a significant precipitation reduction trend. Clearly, there are differences among the different climate proxy data, and these differences may be caused by their inherent uncertainties, which requires further research. Next, we found that the proxy data from Gonghai Lake (Fig. 2d), the stalagmite δ¹⁸O record from Shihua Cave (Fig. 2e), and the drought index from Korea (Fig. 2f) suggest a relatively dry period during 1430 and 1520; however, drought events for the period from 1120 to 1210 were not evident. The reason for this discrepancy was deserving of further investigation. Therefore, we also checked the reconstruction of six regional dry/wet series during the last 1000 years in eastern China (Zhang et al., 1997), and four of them showed dry conditions for the period 1420–1516. The other two—located in the region of Shanxi and Henan provinces [approximately (33°–38°N, 110°–114°E)]—did not show an obvious decrease in precipitation. Overall, these dry/wet series suggest relatively dry conditions over eastern China in the period 1420–1516.

In order to validate the rainy-season precipitation in the historical period simulated by the model, we compared the modeled data with the DWI. We found that the modeled data and DWI had a correlation coefficient of 0.15 for the period 501–1600 (95% confidence level; 160 degrees of freedom), and therefore concluded that the modeled results were a good match in this period. The model was able to simulate the increasing precipitation trend during 1700–1920, reflecting the same trend as that found in the proxy data. However, the precipitation variability did not match well between the model and proxy data, and the correlation between them was not statistically significant after 1700. One possible reason for this might be the influence of the TSI and GHG forcings we used (Fig. 1). After applying a 10-yr running mean, the correlation coefficient over the past 300 years between the simulated precipitation and TSI (GHGs) was 0.38 (0.52) (39 and 26 degrees of freedom, respectively; 95% confidence level). The TSI forcing used was that constructed by Shapiro et al. (2011), and it has the largest variation amplitude.Shapiro et al. (2011) found that there is intrinsic uncertainty in reconstructed data of the past 300 years. The difference in the reconstructions allows the error originating from the uncertainties in the proxy data to be estimated (20%–50% in the solar forcing value, depending on the year). These uncertainties may influence the model response and cause the change in simulated precipitation variability. The process involved might be a change in TSI and GHGs leading to a change in land–sea thermal contrast (Man et al., 2012) and SST gradients across the tropical Pacific Ocean (Liu et al., 2013), ultimately influencing the rainfall in eastern China. However, the specific mechanism needs further exploration. Another reason might be the uncertainties in both the model and proxy data, as well as the difficulties involved in the dating of proxy data. These factors may lead to deviation between simulated and reconstructed results.

Using the DWI,Zheng et al. (2006) found that the anomalies of 100-yr dry periods were less than –0.5 in terms of the standard deviation relative to the mean of the series. In this study, simulated summer precipitation rate was slightly less than in the observed data. Therefore, we used the 100-yr anomalies that were less than –0.4 in terms of the standard deviations relative to the mean of the past 1500 years as the 100-yr dry periods. In the modeled data we found that the 100-yr dry periods of eastern China were 622–735, 1420–1516, 1250–1350, and 1720–1820. Two of these periods (622–735 and 1420–1516) have, through the research above, been proven to have existed, and their anomalies were –0.47 and –0.4 in terms of their standard deviations relative to the mean of the past 1500 years. The two other dry periods simulated by CESM might have been caused by the uncertainties in the simulation. It is difficult to achieve consistency between the simulated data and the reconstructions during the whole period. Another reason was that we used the volcanic forcing constructed by Gao et al. (2008), and they found that their model produced more cooling than reconstructions because IVI2 (Ice-core Volcanic Index 2) may overestimate the forcing of large volcanic eruptions owing to the linear assumption they made or the uncertainties in the model and reconstructions. Meanwhile, the TSI forcing that we used, constructed by Shapiro et al. (2011), has the largest variation amplitude in the solar irradiation forcing. The time series of TSI features minimum values during the periods 1250–1350 and 1720–1820. In our simulation, we also found that the surface temperature over the Northern Hemisphere was lower than in the reconstructions during these periods (Wang and Liu, 2014), which may have led to the reduction in precipitation. Therefore, we chose the periods 622–735 (drought period 1, D1) and 1420–1516 (drought period 2, D2) as centennial-scale drought events over eastern China during the past 1500 years.

Fig. 2 Time series of the modeled data and climatic proxy data: (a) May–September precipitation anomalies (P; mm day–1) from results of the AF experiment, relative to the past 1500 years, (b) dry–wet index (DWI) from the proxy data of Zheng et al. (2006), (c) δ18O record (‰, VPDB) from Furong Cave (Li et al., 2011), (d) environmental magnetic record (S–300 mT; %) from Gonghai Lake (Liu et al., 2011); (e) δ18O record (‰, VPDB) from Shihua Cave (Wan, 2011), and (f) the drought index (DI) from Korea (Kim and Choi, 1987). VPDB: Vienna Pee Dee Belemnite.

Figure options

From the anomalies of mean precipitation (Figs. 3a and 3b) in the rainy season, deficient precipitation mainly occurred over eastern China in both D1 and D2, but their dry conditions seem to have differed slightly. In D1 (Fig. 3a), the drought center occurred in northern China and the Yangtze River valley. In southern China, the precipitation rate was more than usual. Essentially, the precipitation anomalies exhibited a meridional dipole pattern over eastern China. Similarly, the reconstruction of Zheng et al. (2006) showed dry conditions in northern China and the area of the Yangtze-Huai River, but wet conditions in the south of the Yangtze River, during the period 620–730. The simulation results of Yan et al. (2015c) also showed drought in the central and northern part of eastern China but flooding in the southern part during the Sui-Tang Dynasty (650–700), which is similar to our results. However, in D2 (Fig. 3b), drought was widespread across almost the whole region of eastern China, but the intensity was seemingly weaker than the drought center in D1 (Fig. 3a). This pattern is also similar to the reconstruction of Zheng et al. (2006). Furthermore,Peng et al. (2014) found dry conditions for nearly the whole region of eastern China from CCSM2 (Community Climate System Model version 2) results for the period 1466–1491 (the same pattern as in Fig. 3b).

The anomalies of May–September mean 850-hPa winds in D1 and D2 are also shown in Fig. 3. Both periods were characterized by a weakening of the southerly winds over eastern China, which may indicate a weakened EASM, possibly resulting in the decreased precipitation in this region. Indeed, these results are consistent with the simulation reported by Yan et al. (2015c) for the Sui-Tang Dynasty (650–700), and the CCSM2 results of Peng et al. (2014) for the period 1466–1491. Additionally, whilst it is known that precipitation in eastern China is closely related to the EASM, might this also be the case on the centennial timescale? To answer this, we chose the index of EASM variability defined by Sun et al. (2002), which includes zonal and meridional land-sea thermal differences, to pinpoint the relationship between precipitation in eastern China and the EASM. The index contains 80% of the zonal thermal difference, which means the surface temperature difference between eastern China (the land of 27°–35°N and east of 105°E) and the northwestern Pacific (15°–30°N, 120°–150°E); and 20% of the meridional thermal difference, which means the surface temperature difference between South China (the land south of 27°N and east of 105°E) and the South China Sea (5°–18°N, 105°–120°E). After applying a 31-yr running mean, the correlation coefficient between precipitation in eastern China and the index was 0.28 (61 degrees of freedom; 95% confidence level), which means that the strength of the EASM may significantly influence the precipitation in eastern China on the centennial timescale.

Fig. 3 Anomalies of May–September mean precipitation (colored shading; mm day–1) and 850-hPa winds (vectors; m s–1) in the periods (a) D1 (622–735) and (b) D2 (1420–1516). The red box in (a, b) represents the boundary of eastern China. Areas with statistical significance at the 95% confidence level are dotted. The abbreviation AF refers to the all-forcings experiment.

Figure options

The vertical velocity and water vapor transport are very important to monsoon rainfall. In D1, anomalous upward motion (Fig. 4a) dominated to the east of 180°, while anomalous subsidence occurred between 120°E and 180° in the equatorial Pacific, indicating a shift in the Walker circulation. We checked the Walker index (Diaz and Bradley, 2005), defined by the 500-hPa vertical velocity anomaly difference between the equatorial eastern Pacific (5°S–5°N, 160°–120°W) and the equatorial western Pacific (5°S–5°N, 120°–160°E), and found that the Walker index anomaly was –0.2 in D1 (relative to the past 1500 years), meaning there was weakened Walker circulation in the tropical Pacific Ocean. This may have led to the decreased southeasterly transport of water vapor from the western Pacific (Fig. 4a) and reduction in moisture fluxes, resulting in drought conditions over eastern China (Ummenhofer et al., 2013;Hernandez et al., 2015). There was anomalous upward motion in the western tropical Indian Ocean in particular (Fig. 4a), and subsidence in the eastern tropical Indian Ocean. This may have resulted in the anomalous cyclone in the North Indian Ocean and, ultimately, a decrease in the southwesterly transport of water vapor from the Indian Ocean to eastern China (Fig. 4a). At the same time, upward motion was apparent between 15° and 25°N in eastern China, meaning an enhancement in convective activity. This may have led to excessive precipitation in southeastern China. Downward motion gathered north of 25°N, resulting in deficient precipitation over the Yangtze River valley and northern China.

Next, we analyzed the anomalies of vertical velocity and vertically integrated water vapor transport in D2. As Fig. 4b shows, there was anomalous upward motion between 120° and 160°E in the equatorial Pacific, and subsidence seems to have dominated the regions of the tropical Indian Ocean. This may have caused the tropical westerly anomalies from 60° to 150°E (Fig. 4b), blocking the low-level cross-equatorial flow, and reducing the southwesterly transport of water vapor from the Indian monsoon and the southeasterly transport from the western Pacific, which ultimately decreased the precipitation over eastern China. Additionally, subsidence was apparent over the whole region of eastern China, reducing the northward transport of moisture and resulting in deficient rainfall in eastern China (Fig. 4b).

Fig. 4 Anomalies of May–September mean 1000–300-hPa vertical velocity (colored shading; 10–4 hPa s–1) and vertically integrated water vapor transport (vectors; kg m–1 s–1) in the periods (a) D1 (622–735) and (b) D2 (1420–1516). Areas with statistical significance at the 95% confidence level are dotted. The abbreviation AF refers to the all-forcings experiment.

Figure options

Previous studies have shown the impact of the Indo-Pacific Ocean on precipitation over eastern China, but these results are mainly based on the interannual or decadal time scale. In order to find the reason for the weakened water vapor transport and anomalous vertical velocity leading to drought events over eastern China during the two centennial-scale periods in this study, we considered the relationship between precipitation in eastern China and the SST of the Indo-Pacific Ocean on the centennial timescale. The results (figures omitted) show that regions of the Indo-Pacific Ocean correlated well with precipitation in eastern, especially the eastern Indian Ocean and the western Pacific. This means that the occurrence of drought events in eastern China may be characterized by a cold SST mode in the Indo-Pacific Ocean.

In D1 (Fig. 5a), cooling occurred over the whole area. In the tropical Pacific Ocean, regions of maximum cooling were observed over the western Pacific, while there was a small cooling amplitude in the eastern equatorial Pacific. This featured a weakened zonal SST gradient in the tropical Pacific Ocean, which is the same as an El Niño event (Liu et al., 2013). It is known that the Pacific-East Asia teleconnection pattern is confined to the lower troposphere. When warm events occur in the eastern Pacific, local convection is enhanced. Subsidence occurs in the western Pacific, suppressing convective heating, inducing a Rossby wave and resulting in an anomalous Philippine Sea anticyclone. This wind anomaly causes wet conditions stretching from southern China northeastward to the east of Japan (Wang et al., 2000). In this study, in D1, regions with a small cooling amplitude were observed in the Philippine Sea (Fig. 5a), and subsidence appeared in the west of the Philippine Sea (Fig. 4a). Such anomalies will suppress the southwesterly transport of water vapor from the Indian Ocean to eastern China. Furthermore, a positive IOD SST gradient was observed (Fig. 5a), which can be linked to ENSO events through a shift in the Walker circulation to the west (Hernandez et al., 2015). This will strengthen the upward motion in southern China and downward motion over the Yangtze River valley and northern China (Fig. 4a), causing the wet conditions in southern China and the deficient precipitation over the Yangtze River valley and northern China.

As Fig. 5b shows, cooling also occurred over the whole area, but we did not find a cooling center in the tropical Pacific Ocean, and the zonal SST gradient did not change. Furthermore, the maximum cooling occurred in the tropical Indian Ocean and the whole of the North Indian Ocean. This may have resulted in anomalous subsidence in the tropical Indian Ocean, enhancing the tropical westerlies from 60° to 150°E (Fig. 4b), and reducing the southwesterly transport of water vapor from the Indian monsoon and the southeasterly transport of water vapor from the western Pacific.

Fig. 5 Anomalies of May–September mean SST (°C) in (a) D1 (622–735) and (b) D2 (1420–1516). Areas with statistical significance at the 95% confidence level are dotted. The abbreviation AF refers to the all-forcings experiment.

Figure options

Previous studies have shown that external forcings such as solar activities and volcanic eruptions may be the primary driver behind the occurrence of drought events (Man et al., 2014;Peng et al., 2014). From the forcings in Fig. 1, the solar irradiation reduced considerably in D1, and there were no large volcanic eruptions; whereas, in D2, solar activity also reduced substantially, and volcanic eruptions were more frequent. Furthermore, in both D1 and D2, GHGs and LUCC showed no obvious change relative to the past 1500 years. This is what could be found from the time series of the forcings, but further proof was needed.

For the purpose, we used single forcing experiments to uncover the external forced mode of SST on the centennial timescale. As shown in Fig. 6a, we used D2 as the period of the lowest radiative forcing to subtract the period of maximum radiative forcing in the AF experiment, to reduce the influence of the internal variability in the climate system. There was no significant SST gradient in the equatorial Pacific, and the main cooling occurred in the North Indian Ocean, which is similar to the results shown in Fig. 5b. The first empirical orthogonal function (EOF) mode of SST is shown in Fig. 6b, and the explained variance reached 83%. The pattern is similar to that in Fig. 6a, and the spatial correlation coefficient between them was calculated as 0.82. This could represent the main mode of SST on the centennial timescale in the AF experiment. As shown in Fig. 6c, we also used the difference between the period of the lowest solar irradiation (D1) and that of the highest in the TSI experiment. The results showed that the regions of main cooling were in the western equatorial Pacific, but there was a small cooling amplitude in the eastern equatorial Pacific. The EOF1 mode of SST in the TSI experiment (Fig. 6d) showed a similar pattern to that in Fig. 6c, and the spatial correlation coefficient between them was 0.91. In the same way, the results from the Vol experiment (Figs. 6e and 6f) showed an almost constant amplitude in the equatorial Pacific, similar to the results in Fig. 5b. We then computed the spatial correlation coefficient of the EOF1 mode between the AF (Fig. 6b) and TSI (Fig. 6d) simulations, and the result was 0.86; whereas, the coefficient between AF and Vol (Fig. 6f) was 0.82. We also calculated the correlation coefficients between the SST of the Indo-Pacific Ocean (30°S–50°N, 40°E–70°W) and the time series of TSI (Fig. 1a) and volcanic aerosol mass (Fig. 1b) during the past 1500 years. The results were 0.82 and 0.37 (after applying a 31-yr running mean, the degrees of freedom were 38 and 77, respectively, at the 99% confidence level). Similarly, the correlation coefficients between the first principal component (PC1) of SST in the AF experiment (Fig. 6b) and the PC1 of SST in the TSI experiment (Fig. 6d) and Vol experiment (Fig. 6f) were 0.77 and 0.31, respectively, over the past 1500 years (after applying a 31-yr running mean, the degrees of freedom were 36 and 56, respectively, at the 99% confidence level). That means, TSI and volcanic eruptions have a substantial influence on the SST of the Indo-Pacific Ocean on the centennial timescale. Furthermore, in D1, the pattern of SST may have been caused by the persistently low solar irradiation; whereas, in D2, it may have been also influenced by frequent volcanic eruptions. Additionally, we checked the results in the GHGs, LUCC, and Ctrl experiments (figures omitted), and we did not see a similar pattern to that in the AF experiment. Therefore, GHGs, LUCC, and internal variability, may have no significant impact on the SST of the Indo-Pacific Ocean on the centennial timescale.

Fig. 6 Spatial patterns of SST (°C) on the centennial timescale in different experiments: (a) D2 minus the period of maximum radiative forcing (780–880) in the AF (all forcings) experiment; (b) the leading EOF mode of AF, based on a 31-yr running mean. Panels (c, d) and (e, f) are same as (a) and (b), respectively, but for the results in the TSI (total solar irradiance) and Vol (volcanic eruptions) sensitivity experiments. The maximum radiative forcing periods in (c) and (e) are 780–880 and 750–850, respectively.

Figure options

Based on CESM model simulations, we explored the characteristics and causes of centennial-scale drought events over eastern China during the past 1500 years. The main conclusions are:

(1) Centennial-scale drought events over eastern China occurred during the periods of 622–735 (D1) and 1420–1516 (D2) over the past 1500 years, which is comparable with climate proxy data. In D1, the drought center occurred in northern China and the Yangtze River valley, but in southern China the precipitation rate was much higher than usual. In D2, decreased precipitation occurred across almost the whole region of eastern China.

(2) The direct cause of these two centennial-scale drought periods was a weakened EASM, and the specific process was closely linked to the air–sea interaction of the Indo-Pacific Ocean. In D1, regions of maximum cooling were observed over the western Pacific, which may have led to anomalous subsidence, a weakening of the Walker circulation, and reduced northward water vapor transport. Furthermore, upward motion occurred over southern China, strengthening the convective activity and increasing the precipitation. In D2, due to the decrease in the SST of the North Indian Ocean, subsidence dominated the North Indian Ocean, blocking the low-level cross-equatorial flow, enhancing the tropical westerly anomalies, and reducing northward moisture transport. Additionally, descending motion appeared in eastern China, which subsequently decreased the precipitation over the whole region of eastern China.

(3) The anomalous cooling of the SST of the Indo-Pacific Ocean may have been caused by the persistently low solar irradiation in D1; whereas, in D2, the anomalous cooling of the SST may have been also influenced by frequent volcanic eruptions.

Due to the uncertainties related to the forcings and internal variability in model simulations, the precipitation variability might not be consistent with climate proxy data in some periods. This, together with quantifying the contributions of external forcings to drought events and understanding the mechanisms of climate change, are important topics for future work.