Periodic climate cooling enhanced natural disasters and wars in China during AD 10–1900

1. Introduction

Historians commonly attribute dynastic transitions or cycles to the quality of government and class struggles (Zhang et al. 2005). However, climatic fluctuation may be a significant factor interacting with social structures in affecting the rise and fall of cultures and dynasties (Cowie 1998; Hsu 1998). When the climate worsens beyond what the available technology and economic system can compensate for, people are forced to move or starve. Some authors suggest that climate cooling has had a huge impact on the production of crops and herds in pre-industrial Europe and China (Hinsch 1998; Atwell 2002; Zhang et al. 2007a), even triggering mass southward migration of northern nomadic societies (Fang & Liu 1992; Wang 1996; Hsu 1998). This ecological and agricultural stress is likely to result in wars and social unrest, often followed by dynastic transitions (Zhang et al. 2005). A few recent studies have demonstrated that wars and social unrests in the past often were associated with cold climate phases (Zhang et al. 2005, 2007a,b). Climate cooling may have increased locust plagues through temperature-driven droughts or floods in ancient China (Stige et al. 2007; Zhang et al. 2009). Therefore, cool temperatures may directly or indirectly be detrimental to agriculture, and thus stimulate wars of rebellion within a dynasty (defined here as the ‘internal war hypothesis’; see Zhang et al. 2005, 2007a,b). Alternatively, climate fluctuations may force nomads of the north during cold periods, through affecting livestock production (Fang & Liu 1992; Wang 1996; Hsu 1998), to migrate southwards, and thereby increase the frequency of conflicts between these northern pastoral nomadic societies and the agricultural societies (defined here as the ‘southward aggression hypothesis’). However, both hypotheses have not yet been quantitatively assessed.

The purpose of this study aims to reveal the periodic cycles of war frequency in ancient China, and the associations of external aggression war or internal war with climatic or agricultural disasters. Our emphasis is on testing the internal war hypothesis and the southward aggression hypothesis. We use historical data on war frequency, drought frequency and flood frequency, all of which having been complied by Chen (1939). In addition we use a multi-proxy temperature reconstruction for the whole of China reported by Yang et al. (2002), air temperature data for the Northern Hemisphere (Mann & Jones 2003), proxy temperature data for Beijing (Tan et al. 2003), and a historical locust dataset reported by Stige et al. (2007) and extracted from Ma (1958). Finally, we use historical data of rice price variations reported by Peng (2007). Chen (1939) distinguished wars into internal wars within the Chinese dynasties (including peasant uprisings and battles among warring states) and external aggression wars where the Chinese dynasties fought aggression from external societies (predominantly southward aggressions by northern nomads). These data on war frequencies provide us with the unique opportunity to test the temperature-associated internal war hypothesis and the external southward aggression hypothesis separately. The available data of rice price, drought/flood and locust plague occurrence enable us to assess the proposed potential causal mechanisms of how temperature variations affect war frequency. Cross-correlation function (CCF) and wavelet analysis were used to conduct statistical analysis on associations between war frequency and the influential factors.

2. Material and methods

3. Results

Figure 2 summarizes our results and shows the maxima of significant (p < 0.05) or near-significant (p < 0.1) CCFs between decadal values of war frequency and possible influencing variables in phase and with time lags of one and two decades (see the electronic supplementary material, table S1). As shown in figure 2a, there are in total six paths through which temperature may affect frequencies of all wars (AW, sum of internal wars, external aggression wars and other wars), but only three paths are reasonable if we assume that wars are caused by climatic, agricultural or biological disasters. First, low temperatures (T) may directly increase AW through the path of low T → high AW. Second, low temperature may indirectly increase AW through the path of low T → high D (drought) → high L (locust) → high AW. Third, low temperature may indirectly increase AW through the path of low T → high RP (rice price) → high AW. These results suggest that climate cooling may have increased the frequency of all wars (AW) directly as well as indirectly through its associated increases of rice price level or drought and locust occurrence. The other paths (i.e. low T → high F → low AW; low T → high F→ low RP → low AW) are not reasonable; in other words, flood disasters seem not to be key factors in causing high RP and high AW.

As shown in figure 2b, there are two possible paths through which low temperature may increase the frequency of external aggression wars (EW): low T → high EW; low T → high RP → high EW. Similarly, the other paths (i.e. low T → high F or high D → low EW; low T → high F→ low RP → low EW) are not reasonable; in other words, flood or drought disasters seem not to be key factors in causing high EW. There are also two possible paths through which low temperature may increase the frequency of internal wars (IW): low T → high D → high L → high IW; low T → high RP→ high IW. These results suggest that climate cooling may have increased directly the frequency of external aggression wars. Besides, climate cooling may have increased indirectly the frequencies of both external aggression wars and internal wars through its associated increase of rice price; it may have increased indirectly the frequencies of internal wars through its associated increase of drought and locust plague. It is notable that we did not detect significant or near-significant direct effect of temperature on frequencies of internal wars.

As shown in the electronic supplementary material, table S2, using proxy data of air temperature of the Northern Hemisphere (T_N) and temperature of Beijing (T_B), we find AW shows significant out-of-phase associations with T_N of the previous decade (CCF = −0.2209, p = 0.003) and the current decade (CCF = −0.2039, p = 0.003). EW shows significant out-of-phase associations with T_N of the previous decade (CCF = −0.1273, p = 0.0494) and the current decade (CCF = −0.1422, p = 0.0326), and with T_B of previous decade (CCF = −0.1227, p = 0.0432). IW shows significant out-of-phase associations with T_N of the previous decade (CCF = −0.1242, p = 0.0482). Besides, EW shows near-significant out-of-phase associations with T_B of the current decade (CCF = −0.1015, p = 0.0828) and previous two decades (CCF = −0.1033, p = 0.0724); IW shows near-significant out-of-phase associations with T_N of the current decade (CCF = −0.1209, p = 0.06). These results support the above observations of the out-of-phase associations between AW or EW and T. Besides, these results also reveal the out-of-phase association between IW and T_N.

Wavelet analysis shows all time series of EW, IW, T, T_N, T_B, D, F and RP variables have one or two predominant and consistent periodic bands around 160 years or/and 320 years (figure 3), suggesting that these variables may interact with each other around these two periodic bands. T has one period initially around 240 years, then two bands around 160 and 320 years (the 160 year band is more predominant; figure 3i). T_N has a predominant period around the 160 year band and an obvious band of 320 years (figure 3h). T_B has a predominant period around the 320 year band and a weak band of 160 years (figure 3j). D has one period around 160 years (figure 3e). F has two periods around 160 and 320 years (the 320 year period shows somewhat steady increase; figure 3d). IW has three periods around 160, 240 and 320 years (figure 3c). EW has one period around 240 year initially, and then two periods around 320 and 160 years (figure 3b). AW has periods around 320 and 160 years (figure 3a). L has periods of 320 and 160 years (figure 3g). RP has a period of 320 years (figure 3f). There are also obvious periods around 80 or 40 year bands for some time series (e.g. AW, IW, L, RP, T_N, T_B, F). There are good overlaps among these variables around the predominant or obvious periodic bands. For example, EW has a very good match with T at the three periodic bands of 240, 160 and 320 years (figure 3b,i), supporting the significant out-of-phase associations between EW and T (figure 2b).

Overlapped coherency analysis confirms that they are indeed closely associated with each other around these two periodic bands (see figure 4; see the electronic supplementary material, figures S1–S3). EW shows predominant out-of-phase associations (all arrows point to left) with all of the three time series of temperature (T, T_N, T_B; figure 4), supporting the significant CCF estimations in the electronic supplementary material, tables S1 and S2. All T, T_N and T_B show predominant and consistent out-of-phase associations with EW around the 320 year periodic band, but there are a few short periods showing in-phase associations (all arrows point to right) around the 160 year band (figure 4). T and T_N also show predominant out-of-phase associations with AW (see the electronic supplementary material, figure S1); however, there are no predominant in-phase or out-of-phase associations between IW and all three temperature time series. T shows about equal out-of-phase and in-phase associations with IW, resulting in a non-significant CCF between them (see the electronic supplementary material, table S1 and figure S2a).

As shown in the electronic supplementary material, figures S1–S3, L shows obvious consistent in-phase associations with AW or EW around the 320 year band, with AW or IW around the 160 year band. RP shows predominant in-phase associations with AW and IW around the 320 and 80 year bands. These results suggest that low T, high RP and high L are potential factors causing high AW, EW or IW at the specific periodic bands. F and D show obvious consistent out-of-phase associations with AW and EW around the 320 year band (see the electronic supplementary material, figures S1 and S3), suggesting that high F or D are less important key factors causing high AW or EW, or owing to a time-delayed effect caused by the covarying effect of IW on EW (see §4). In general, the observed associations by using wavelet analysis (see figure 4; see also the electronic supplementary material, figure S1–S3) are in good agreement with those observed by using CCF analysis (see figure 2). It is also notable that associations between variables are not always consistent around specific periods though they are strong or obvious. These inconsistencies often occur around the 160 year band, such as the associations between IW and D (see the electronic supplementary material, figure S2c) and between EW and D (see the electronic supplementary material, figure S3c), or they may occur between 160 and 320 year bands (e.g. figure 3d, for association between EW and L). This may be caused by several factors (see §4).

4. Discussion

Our results demonstrate that periodic climate cooling may have increased the frequency of external aggression wars during AD 10–1900, supporting the southward aggression hypothesis. All three temperature time series show significant out-of-phase associations with the frequency of external aggression war (figure 4; see also the electronic supplementary material, tables S1 and S2), and these associations are predominant and consistent around the 320 year periodic band (figure 4). Because no causal links between high EW and high F or high D are established (figure 2), the negative effect of temperature on EW may be direct. External aggression wars as defined by Chen (1939) have mostly occurred between Chinese dynasties and the pastoral nomadic societies to their north. The northern part of the Chinese dynasties are more vulnerable to temperature cooling; it is estimated that a 2^oC drop of annual air temperature can shorten the growing season of grass by 40 days (Zhang et al. 2005), adversely affecting grasslands and resulting in huge losses of domestic livestock owing to forage shortage (Fang & Liu 1992). Accordingly, previous studies have shown that cooling has had a negative impact on agricultural or livestock production in ancient China (Gong et al. 1996; Zhang et al. 2005). According to Ren (2004), pastoral nomadic regions experienced serious droughts and then southward immigration owing to collapses of livestock production in cold periods. There are reports on large-scale southward immigrations of northern pastoral nomadic societies in ancient China during the cold periods (e.g. Chen 1939; Wang 1996; Ding 2007; see also the electronic supplementary material). The collapses of the agricultural dynasties of the Han, Tang, Northern Song, Southern Song and Ming are closely associated with low temperature or rapid decline of temperature, and immediately replaced by dynasties ruled by the northern pastoral nomadic societies (see the electronic supplementary material, figure S4). We did not find significant links between high EW and high F or high D (figure 2), suggesting that the collapses of the agricultural dynasties were not caused by drought and flood events. Our results do not support the view that the collapses of the Tang and the Ming dynasties are caused by drier climate owing to weaker Asian summer monsoon (Yancheva et al. 2007; Zhang et al. 2008), but support the view of Zhang & Lu (2007; see the electronic supplementary material).

Our results also suggest that periodic low temperatures may have increased the frequency of internal wars mainly indirectly through increasing drought and locust plague frequencies during AD 950–1900, supporting the internal war hypothesis. Drought, flood and locust plague have been regarded as the three major natural disasters affecting crop production in ancient China (Zhang et al. 2009), suggesting that their detrimental effects on agricultural production may trigger internal uprisings. In the cold period of AD 1840–1890, agricultural yields in China were reduced by 10 to 25 per cent compared with the relatively warm period of AD 1730–1770 (Gong et al. 1996). Our findings provide quantitative evidence about climate-driven agricultural breakdown owing to droughts and locusts increasing the frequency of internal wars. Droughts have been well recognized as the positive effect on locust plagues in China because they produce favourable wet bank or lake beach environments for locusts to lay eggs (Stige et al. 2007; Zhang et al. 2009). The positive effect of low temperature on locust plague is obviously indirect through increase of temperature-induced droughts. Because IW is positively associated with the locust plague around the 160 year periodic band, the indirect effect of temperature on IW might be dominant around this periodic band. We did not find significant direct effect of temperature on IW and T (figure 2b). However, we found a significant out-of-phase association between IW and T_N (figure 4e). This is in agreement with findings by Zhang et al. (2005). These results suggest that direct negative effect of temperature on IW may also exist. Thus, low temperature may destroy agriculture directly or/and indirectly through its induced high level of drought, flood and locust plagues. Our study also reveals that there are more droughts and floods in cold periods in China (figure 2; see also Zhang et al. 2009). This is in agreement with some previous studies that found there were fewer floods in the warm 9th to 11th centuries and more floods in the cold 14th to 17th centuries (Singer & Avery 2007). In the Little Ice Age, Europe suffered the most serious floods and hunger in its history, and both Roman and Mayan empires collapsed in cold periods (Singer & Avery 2007).

Our results demonstrate that frequencies of both internal wars and external aggression wars were strongly and positively associated with the price of rice. Rice price may be a good criterion reflecting food availability, suggesting that food shortage indicated by high rice price may have played a key role in social conflicts in ancient China. Surprisingly, the negative association between temperature and rice price only attains a near-significance level (p < 0.1), suggesting that the close positive association between frequencies of wars and rice price may be caused more by intrinsic factors of the society such as misgovernance, corruption and population growth, and that this strong association may mask the weaker associations of rice price with climatic disasters such as droughts and floods. Cool temperature causes high price of rice directly, while cool temperature also causes high floods, which cause low rice price (figure 2). These results suggest that cool temperature may cause food shortage (reflected by price of rice). Floods may benefit rice growth by increasing area of wetlands, resulting in a near-significant out-of-phase association between RP and F (figure 2).

It is very probable that cool temperature may be the driving force in causing high frequencies of meteorological, agricultural disasters and then man-made disasters (wars) in ancient China. In our previous study we found cool temperature significantly increased frequencies of drought and flood, and then locust plagues (Zhang et al. 2009). The larger the difference of temperature between Arctic and tropical equatorial regions, the greater the strength of wind, ocean waves, ocean circulation and cyclones; thus a cold climate would bring more climatic disasters than a warm climate (Singer & Avery 2007). Such effect of climate cooling works similarly on precipitation in China (see Zhang et al. 2009; see also the electronic supplementary material). Therefore, cool temperature could not only reduce agricultural and livestock production directly, but also reduce agricultural production by producing more droughts, floods and locust plagues in ancient China indirectly. The collapses of agricultural and livestock production would cause wars within or among different societies.

We found three temperature time series (Yang et al. 2002; Mann & Jones 2003; Tan et al. 2003) had one or two predominant periodic bands of around 160 and/or 320 years during last two millennia (figure 3). These periods may be related to cyclic variations of solar activity, or cyclic changes of orbit position of the Earth (Singer & Avery 2007). Friis-Christensen & Lassen (1999) reported that solar activity can contribute 75 to 85 per cent of the climatic variation of the Earth. There are 87 and 210 year cycles of solar activity based on observations of sunspots (Braun et al. 2005). The two predominant periods of temperature are relatively wide owing to smooth variation of low-frequency temperature. The initial period of T is 240 years, and then divided into 160 and 320 year band. These may be related to the 210 year solar cycle. In some periods, the 80 year periodical band of temperature is also observed in our study, which may be related to the 87 year solar cycle. Using power spectral analysis, Tan et al. (2003) find the cycles of about 206 and 325 years of Beijing temperature are significant, and these cycles could be connected to solar variation cycles of about 208 and 350 years (Lean 2002). It is notable that the periods we discovered (e.g. 40, 80, 160, 320 years) are well linked to the solar cycle as predicted by the model T = 11 × 2ⁿ (n = 2, 3, 4, 5; see Perry & Hsu 2000; Hsu 1998). The periodic associations between frequencies of external aggression wars and temperature (figure 4), or between drought/flood and temperature (Zhang et al. 2009), are predominant and consistent in phase shift, suggesting that the periodic variations of temperature may be the direct driving forces behind these variables. With the increase of the levels in figure 2, the periodic associations of temperature-driven drought/flood or locust plagues on war frequency may become less clear, probably owing to many reasons (e.g. time delay, frequency dependency; see the electronic supplementary material), which need to be investigated further in future studies.

It is generally believed that global warming is a threat to human societies in many ways (IPCC 2007). However, some countries or regions might also benefit from increasing temperatures in some ways (Nemani et al. 2003; Stige et al. 2007; Zhang et al. 2009). Our study suggests that during the last two millennia, food production in ancient China was more stable during warm periods owing to fewer agricultural disasters, resulting in fewer social conflicts. However, the present ongoing global warming may produce different effects on our industrialized societies, which have a much higher capacity for dealing with natural disasters than pre-industrial societies. Its potential consequence on our societies and ecosystems needs to be carefully monitored and investigated. Data quality is influenced by many factors, including data coverage and time–space density of reports. Thus, it is necessary to recover more data from additional sources for gaining the utmost insight from such comparisons.