The statistical data collected for the southeastern coastal regions of China include data for Hainan, Guangdong, Fujian, Taiwan, Zhejiang, Shanghai, Jiangsu and Shandong (Fig. 1). When we categorized typhoon landfall locations based on latitudes, Fujian and Taiwan are recognized as one region due to their similarities in latitude and the same applies to Jiangsu and Shanghai. It is notable that prior to 0 CE the historical record of China lacks data on typhoon activity. Consequently, this study focuses on data collected over the last 2 kyr. Furthermore, data for the period 1945–2013 CE were collected from the northwestern Pacific Ocean TC records established by the Joint Typhoon Warning Center (JTWC). The statistical results were divided into three different time frames based on keyword results and database sources: (1) 0–1000 CE; (2) 1000–1910 CE; and (3) 1945–2013 CE. To plot the number of typhoons occurring as a function of time, typhoon events in any given decade were collectively plotted to create an interdecadal bar graph dating from 1000 CE to the present (Fig. 2). The number of events which occurred in any given decade relates closely to the age of historical documents and how well they have been preserved. Records relating to TC landfall between 1945 and 2013 CE are reliant on satellite-acquired data, meaning the data source is highly reliable in terms of its location and intensity. Consequently, Fig. 2 shows extreme growth in the number of recorded TCs in the latter years of the twentieth century. Moreover, Liu et al. (2017) published TC landfall data for the northwestern Pacific Ocean region between 1945 and 2013 CE, which corresponds to the results seen here. The Fig. 2 shows clearly that TC activity grew to an extraordinary extent at around 1500 CE and has persisted to the present.

To further investigate any changes in the timing of annual TC landfall, TC landfall data were collected and analyzed for three different time periods: 0–1000 CE; 1000–1910 CE; and 1945–2013. The results are shown in Fig. 5, and monthly statistics are listed in Table S3 of the Supplement. Before 1000 CE, TCs in China mostly occurred in June, July and August (Fig. 5a). However, after 1000 CE, the entire trend in arrival times shifted by 1 month, with TC landfall occurring predominantly in July, August and September (Fig. 5b). The majority of statistics after 1000 CE were collected during the Little Ice Age (LIA; 1400–1850 CE). Figure 5c shows statistics for the period 1945–2013 CE The timing of recent TCs making landfall in southeastern China is quite similar to that which occurred during the LIA period. Recent data show that TC occurrence in the entire northwestern Pacific Ocean region can last until as late as October, November and December with TCs making landfall in Vietnam, the Philippines and Thailand after September (Liu et al., 2017). It is assumed that this relates to seasonal changes in the positions of the subtropical high and ITCZ of the northwestern Pacific Ocean region. The ITCZ begins migrating north away from the Equator in March or April. It reaches its northernmost position in August, before migrating south in September (Waliser and Gautier, 1993). The question this study raises is what occurrence shifted the predominant timing of TC arrival in southeastern China from between June and August between 0 and 1000 CE to between July and September after 1000 CE. One likely explanation is that the ITCZ was at a higher latitude before 1000 CE (Rehfeld et al., 2013), resulting in earlier (June–August) TC formation.

Not all historical records gave details on where TCs struck before 1000 CE; therefore, this study focuses solely on the landfall locations of paleotyphoons between 1000 and 1910 CE. The number of typhoons that struck each province in China is shown in Fig. 6. Table S4 of the Supplement gives additional details on landfall locations. For the period 1000–1910 CE, Guangdong was struck by the most TCs. On the whole, the number of TCs making landfall increased dramatically after 1500 CE, with the number of typhoons hitting Guangdong peaking between 1660 and 1680 CE. By contrast, regions north of Fujian did not record any increase in typhoon activity during this time period. The number of typhoons striking Zhejiang and Jiangsu, however, did start to increase after 1700 CE.

Newton et al. (2006) proved that the warmest temperatures in the Indo-Pacific Warm Pool occurred during the Medieval Warm Period while the coolest temperatures occurred during the Little Ice Age. In particular, the lowest temperatures occurred around 1660–1680 CE within the period of the Maunder Minimum (1645–1715 CE). Therefore, it is thought that the sudden change in TC tracks around 1700 CE may relate to a change in temperature lows in the Northern Hemisphere and a shift in the location of the ITCZ.

Conserving historical documents has always been a difficult task. Racial conflicts, war, rebellion and inter-court feuds could all result in precious data being damaged, destroyed or lost during certain periods in history. Consequently, statistics on paleotyphoons recorded in the historical record are only semiquantitative. On the other hand, they are very useful in terms of noting the location of landfalls and the precise timing of such events. To help overcome any anomalies in the typhoon record lost to documented history and to avoid any confusion regarding the intensity of events, this study also looked at the geological record of paleotyphoons derived from lake sediments in northeastern Taiwan (Chen et al., 2012; Yang et al., 2014; Wang et al., 2013, 2014, 2015). Since the topography of northeastern Taiwan's Yilan region is quite unique, with the summer monsoon being blocked by mountains and rainfall being mainly supplied by the winter monsoon and typhoons (Chen et al., 2012), the region is very helpful for studying TC in the northwestern Pacific. In fact, large-scale river terraces have occurred due to typhoon rainfall, and this record has been preserved in the mountain areas of Yilan since 2.7 ka (Hsieh, 2017).

In order to correlate the number of paleotyphoons from historical data with the geological record of lake sediments, the Southern Oscillation Index (SOI), the intensity of paleotyphoons determined from sedimentary particle size at Taiwan's Lake Dahu and paleotyphoon signals from lagoon sediments in Kyushu, Japan (Fig. 7), are referenced and compared. Results suggest that typhoons struck Taiwan and the southeastern coastal region of China mostly during La Niña-like stages (Fig. 7a, b, c) (Chen et al., 2012). This outcome matches that mentioned by historical maritime disaster events caused by paleotyphoons in the last 1000 years in Liu et al. (2017). According to Liang and Zhang (2007), the chances of a typhoon making landfall in the southeastern coastal region of China during La Niña years is higher than that during El Niño years. If we started entering an El Niño-like stage after 1900 CE, this means the number of typhoons striking Japan in the future will very likely increase compared to what we see now. This trend in the data since 1700 CE shows a gradual increase in typhoon numbers moving north and away from Guangdong (Fig. 6). It has also been shown that the number and intensity of typhoons recorded in Taiwan's lake sediments has grown since the LIA (1400 CE) which seems to match the general trend in the recorded number of historical events pretty well (Fig. 7a and c). This period also coincided with the timing of flooding events in southern China (Fig. 7d). Park et al. (2017) investigated the records of lake sediments in the East Asia region. Their study noted that along coastal regions including Jeju Island (Korea), lakes in Yilan (Taiwan), Lake Huguangyan in Guangdong, and lakes on Hainan Island relatively drier conditions prevailed during the MWP and wetter conditions during the LIA. This may be due to an increase in rainfall caused by typhoons along the coast.

This study, therefore, finds that the northward migration of the ITCZ during the MWP caused typhoons to move north toward Japan. In contrast, typhoons moved toward southern China during the LIA due to the southward transition of the ITCZ. This seems to be a reasonable explanation and is not out of step with other regional studies (Rehfeld et al., 2013; Chen et al., 2015; Xu et al., 2016).

Donnelly and Woodruff (2007) first suggested that the number of hurricanes in the Caribbean area has been increasing over the last 4000 years. According to ancient hurricane research along the Gulf Cost and Caribbean Sea to Puerto Rico, hurricane tracks show an antiphase in time series data (McCloskey and Liu, 2012, 2013; McCloskey et al., 2013; Liu et al., 2015). During the MWP, more TCs made landfall in the Gulf Coast as the strength of the Bermuda High enhanced and the ITCZ moved northward. During the LIA, more TCs made landfall in the Caribbean Sea (McCloskey and Knowles, 2009; McCloskey and Liu, 2012, 2013; McCloskey et al., 2013). In 1650 CE, TC frequency reached a peak, and after 1850 CE TCs began to move toward Florida and Bermuda with the northward movement of the ITCZ (Baldini et al., 2016). Ancient lake sediment data from Yilan, Taiwan, reveals the period in history when paleotyphoons occurred most frequently. This timing correlates highly with the time of paleohurricanes recorded in Belize from McCloskey and Liu (2013). This suggests that the migration paths of TCs in both the northwestern Pacific Ocean region and the northwestern Atlantic Ocean region are closely related. TC activity occurred between 200 and 600 and between 1450 and 2600 yr BP in Belize, and it occurred between 200 and 500, 1300 and 1500, and 2000 and 2300 yr BP in Taiwan's lakes (Chen et al., 2012). This phenomenon indicates a close association between TC activity in the North Pacific Ocean and the North Atlantic Ocean.

Since the ITCZ and westerlies are both linked to the Hadley Cell, and the position of midlatitude storms are determined by the westerlies, which are influenced by the North Atlantic Oscillation (NAO) (Hurrell, 1995; Morley et al., 2014), we compared the NAO record with the track of TCs. In order to compare our tracks of TCs with the NAO, we created an index called TTC1 to represent the track of TCs that either move toward southern China or toward northern China (TTC1 =∑XiFi). Xi is the number of typhoons that had made landfall in a particular province, and Fi means the location factor of the landfall locality (Table 2). When the value of TTC1 is higher, it indicates a larger amount of typhoon landfalls in northern China (Fig. 8). The TTC1 can also be normalized to values between 0 and 1. Furthermore, we used digitalization to retrieve the average data of 10 years from the 2 kyr NAO index according to the results of Trouet et al. (2009) and Ortega et al. (2015). The results calculated from Trouet et al. (2009) and our TTC1 agree quite well (Fig. 8). However, our records were fragmentary before 1470 CE, and we lack the historical data from Japan. The results in Fig. 8 reveal that our normalized TTC1 corresponding to the NAOTrouet during the LIA stage, and the 3-point smoothing of the TTC1 shows a very good correlation with the NAOTrouet. This result indicates that the NAO influences the migration of the westerlies, and it may also gently affect the tracks of the TCs.

After we performed the wavelet analysis, we found that the TTC1 shows both 30–35- and 55–65-year cycles during the LIA stage (Fig. 9). This result is also consistent with the frequency of typhoon landfall over Guangdong Province of China during the period of 1470–1931 CE based on a different data source (Chan and Shi, 2000). The 60-year cycle is clearly present in the PDO and the Atlantic Multi-decadal Oscillation (AMO), with phases coherent with a planetary signal since at least 1650 to 1850 CE (Scafeta, 2012; Solheim, 2013). This implies that the PDO also affects the TTC1 cycle.