For each sample, chironomid remains were extracted from 1–5 g of freeze-dried sediment. The preparation procedure followed the standard techniques described in Brooks et al. (2007). The sediments were deflocculated in warm 10 % KOH for about 15 min, and then sieved with 212 and 90 µm mesh sieves. Head capsules were hand picked from the sieve residues under a stereomicroscope at ×20–40 magnification, and mounted on slides, ventral side up, in Hydromatrix beneath a 6 mm coverslip. Chironomid head capsules were identified to the highest possible taxonomic resolution under a compound microscope at ×100–400 magnification with reference to Wiederholm (1983), Rieradevall and Brooks (2001), Brooks et al. (2007), Walker (2007) and the chironomid collections housed at the Natural History Museum, London.

The modern calibration set from around Gonghai Lake obtained by Wang et al. (2016) was used to identify temperature-sensitive chironomid taxa in the region. The data set comprises 44 water bodies in northern China (Fig. 3a), samples from only 30 of which contained sufficient chironomid head capsules for analysis.

Mean annual temperature (TANN), mean summer (from June to August) temperature (summer temp), and the mean temperatures for June (June temp), July (July temp) and August (August temp) were interpolated from meteorological data from 2001 to 2011 (Dataset of monthly values of climate data from Chinese surface station, 2017; Zhao et al., unpublished data). It should be noted that the surface of Gonghai Lake freezes in winter which disrupts the linear relationship between water temperature and air temperature. Moreover, the winter season is not the growing season of chironomids (Armitage et al., 1995), and therefore the mean temperature of the winter months was not included in the numerical analysis.

A large amount of detailed documentary evidence is available for China. This material documents a wide range of human activities and it provides a valuable reference for the present study. Information pertaining to wars was obtained from the Tabulation of Wars in Ancient China, an appendix of the Military History of China, which was summarized by the Editorial Committee of Chinese Military History (1985); it has been widely utilized in previous research (Zhang et al., 2005, 2015). Only ancient wars which occurred within the current territory of Shanxi Province were counted in the present study. In addition, fluctuations in population size are a major component of human societal evolution and therefore population information was also collated and used to characterize social change. Data documenting fluctuations in the population size of Shanxi Province were obtained from Lu and Teng (2006).

Only taxa which were present in at least two samples with an abundance of > 2 % were selected for analysis. A chironomid percentage diagram was plotted using Tilia 2.0.2 (Grimm, 2004). Zonation of the chironomid assemblages was accomplished using stratigraphically constrained cluster analysis (CONISS) in Tilia 2.0.2 (Grimm, 2004). Both redundancy analysis (RDA) and detrended correspondence analysis (DCA) were performed using R 3.2.1 (R Core Team, 2014) to explore the relationship between modern chironomid taxa and temperature variables, and to analyse the distribution characteristics of fossil assemblages, respectively. In addition, Pearson correlation and Granger causality analysis were performed to explore the relationship between climate change and the occurrence of wars.

Air temperature is widely assumed to play a key role in controlling the abundance and composition of chironomid taxa in freshwater bodies (e.g. Walker, 2001; Brooks, 2003; Walker and Cwynar, 2006). RDA of the chironomid taxa and temperature variables shows that TANN tends to be more significant in influencing the chironomid assemblages than summer temp, June temp, July temp and August temp (Fig. 3b). This result also passed the Monte Carlo permutation test (p = 0.001) even though the explanatory ability is relatively low (Fig. 3b). The taxa are plotted in Fig. 3c according to the taxon scores in the RDA of chironomid assemblages and TANN. Taxa on the left side of the plot currently prefer a warmer environment in the Gonghai Lake region because they are distributed close to the positive axis of TANN in Fig. 3b; conversely, those taxa on the right side of the plot prefer a colder environment.

The following criteria were used to identify temperature-sensitive species: (1) those located at the ends of Fig. 3c and (2) those species previously reported as warm or cold stenotherms. On the left side of the diagram, Polypedilum nubifer-type, Dicrotendipes nervosus-type and Tanytarsus mendax-type have been previously reported as warm stenotherms (Watson et al., 2010; Brooks and Heiri, 2013), and were defined as thermophilous taxa here. Procladius choreus-type and Microchironomus were eliminated because their high scores on the positive axis may be because in the Gonghai Lake region they are indicators of deep water (Wang et al., 2016). On the right side of the diagram, Hydrobaenus conformis-type, Psectrocladius sordidellus-type and Chironomini first instar have been widely regarded as cold stenotherms (Cranston et al., 1983; Brodin, 1986; Brooks and Heiri, 2013), and were defined as cold-water taxa here. Chironomus gonghai-type was included given that it was located at the end of the diagram and tends to live in cold environments (see Fig. 5 in Wang et al., 2016).

We identified 44 major taxa within 25 genera and 4 subfamilies (Tanypodinae, Chironomini, Tanytarsini and Orthocladiinae), and three chironomid assemblage zones were recognized (Fig. 4). Of the chironomid head capsules, 95.7 % were identified to genus or species morphotype. Due to poor preservation, the remaining 4.3 % were only identified to subfamily level; this was especially applicable to the head capsules of the tribe Tanypodinae because the key identification segments of fragmented subfossils were often covered by other material. The concentration of chironomid head capsules appeared to follow variations in the organic matter content of the samples. The concentration was high before 1500 cal yr BP and then decreased to very low values until the present (Fig. 4). The chironomid assemblage zones are described below.

Zone 1 (ca. 4000–2700 cal yr BP). This zone is dominated by Cladotanytarsus mancus-type, Procladius and Stictochironomus. Many Tanytarsini taxa, including Tanytarsus “no spur”, Tanytarsus mendax-type, Tanytarsus lugens-type and Tanytarsus glabrescens-type, are present at a low abundance.

Zone 2 (ca. 2700–1270 cal yr BP). This zone is characterized by the rapid decrease in the abundance of Cladotanytarsus mancus-type and by the sudden appearance of Parakiefferiella bathophila-type. In addition, there is an increasing representation of Paratanytarsus, Hydrobaenus conformis-type and Psectrocladius sordidellus-type.

Zone 3 (ca. 1270 cal yr BP–present). This zone is characterized by a significant increase in Cladotanytarsus mancus-type and a decrease in Parakiefferiella bathophila-type. Hydrobaenus conformis-type remains at a relatively high level throughout the zone. There are large fluctuations in the representation of most of the taxa and therefore the zone is divided into the following subzones.

Subzone 3a (ca. 1270–1040 cal yr BP). This subzone is characterized by an abrupt increase of Cladotanytarsus mancus-type and decrease of Parakiefferiella bathophila-type.

Subzone 3b (ca. 1040–970 cal yr BP). This subzone, which only consists of two samples, is dominated by Propsilocerus jacuticus-type, Chironomus gonghai-type, Chironomini larvae and Procladius.

Subzone 3c (ca. 970–570 cal yr BP). Although they are very poorly represented in the previous subzone, Cladotanytarsus mancus-type, Parakiefferiella bathophila-type and Hydrobaenus conformis-type became dominant in this subzone.

Subzone 3d (ca. 570–270 cal yr BP). In this subzone, Psectrocladius sordidellus-type increases abruptly and reaches its maximum abundance, and Hydrobaenus conformis-type is highly abundant throughout.

Subzone 3e (ca. 270 cal yr BP–present). The dominant taxon in this subzone is Paratanytarsus penicillatus-type. Both Cladotanytarsus mancus-type and Glyptotendipes severini-type increase slightly, whereas Hydrobaenus conformis-type and Psectrocladius sordidellus-type decrease significantly.

Based on the definition of warm- and cold-preferring taxa given above, their totals were calculated to reconstruct temperature changes during the past 4000 years (Fig. 4). The results indicate an overall trend of decreasing temperature; furthermore, fluctuations in the abundance of cold-preference taxa indicate that the temperature was high in zone 1, decreased sharply around 2700 cal yr BP but remained relatively high in zone 2 and fluctuated significantly and reached a minimum in zone 3. It should be noted that the abundance of warm-preferring taxa is much less than that of cold-preferring taxa, and the former were often absent during the past 2700 years (Fig. 4). To avoid the potential limitations of presence/absence data, the changes in abundance of cold-preferring taxa (which provide more detailed information about temperature variations on a centennial timescale) were primarily used to investigate temperature changes.

We calculated a total of 418 wars from 718 BC to AD 1911. Given that the resolution of the Gonghai Lake samples ranges from 50 to 100 years, the incidences of wars were summed to produce a 50-year resolution. The record of chironomid-inferred temperature variability (Fig. 5a) and the pollen-based precipitation reconstruction for Gonghai Lake (Fig. 5b; Chen et al., 2015) were compared with the cumulative frequency of these events (Fig. 5c). The distribution of wars reveals that they occurred more frequently when temperature and precipitation decreased abruptly, and they also lasted for a relatively long time (Fig. 5c). For example, these events were the most severe during the Little Ice Age (LIA), which lasted for nearly 350 years, when both the temperature and precipitation decreased significantly. The results of Pearson correlation and Granger causality analysis show that the change in abundance of the cold-preferring taxa are significantly related to the incidence of wars (r=-0.189 in Table 1, p<0.01 in Table 2).

Only 19 records of population size in Shanxi Province from 340 BC onwards were mentioned in Lu and Teng (2006), and they are used in the present study. These data are evenly distributed within each dynasty (Fig. 5d). Although the population size fluctuated significantly, an overall increasing trend is evident, together with frequent population collapses following intervals with a significant number of wars.

Relevant physical, chemical and climatic variables were all included in the investigation of the relationships between chironomid assemblages and environmental parameters in the Gonghai Lake region (Wang et al., 2016). Although previous analyses indicated that the fossil chironomids mainly responded to changes in precipitation through water depth since the last deglaciation (Wang et al., 2016), the existence of certain typical stenothermic taxa provides high potential for extracting a temperature signal. To further verify whether the stenothermic taxa (based on the published literature) also have a thermal significance in the Gonghai Lake region, the temperature variables (TANN, summer temp, June temp, July temp and August temp) were used as the only variables to constrain the changes in the abundance of the taxa in the calibration set. The results of RDA of modern chironomid assemblages and temperature variables (Fig. 3b), as well as the Monte Carlo permutation test, demonstrate that TANN was a significant environmental variable influencing the modern chironomid taxa. In addition, TANN has a higher score on the first axes than the other variables in Fig. 3b; furthermore, TANN was the only variable selected in the interactive-forward-selection (p = 0.026). This result has rarely been observed in the published literature, although it has been noted that chironomids often respond significantly to mean July or summer temperature (e.g. Brooks and Birks, 2001; Self et al., 2011; Samartin et al., 2017). Our observed correlation between modern chironomid assemblages and TANN provides a valuable reference for extracting temperature signals from the fossil chironomid assemblages of Gonghai Lake. For example, Chironomus gonghai-type is ranked at the end of the RDA of the modern assemblage data and TANN, indicating that it is a cold-temperature indicator in the Gonghai Lake region. Moreover, this taxon was abundant during the Younger Dryas, clearly indicating that it prefers a cold environment. However, Chironomus is reported as a temperate indicator in chironomid records from Scotland and northern Russia (e.g. Brooks et al., 2007, 2012b; Nazarova et al., 2015). The reason for these contradictory findings may be that Chironomus gonghai-type is a new species, or that Chironomus has different preferences in the Gonghai Lake region. These observations indicate that it is necessary to improve the taxonomic resolution of chironomid identification and to establish more precisely the environmental preferences of chironomid taxa from local training sets to enhance the reliability of paleotemperature reconstructions.

Temperature variability in the Gonghai Lake region during the past 4000 years is revealed by changes in the abundance of the warm-preferring and cold-preferring chironomid taxa (Fig. 4). As described in Sect. 5.3, the variations in the abundance of the cold-preferring taxa were primarily used to investigate the temperature changes. An explanation for the more resolved temperature signal carried by the cold-preferring taxa may be that they were easily able to become dominant in Gonghai Lake and respond quickly to temperature fluctuations due to the lake's relatively high elevation (1860 m a.s.l.) and the decreasing trend of late Holocene temperature.

In addition to the warm- and cold-preferring taxa, there are other chironomid taxa in the Gonghai Lake record which could also be regarded as temperature indicators (although to a significantly lesser extent), and it is worth investigating whether they exhibit a similar trend of temperature change to the warm- and cold-preferring taxa. Details of the faunistics and inferred environmental change for each of the three intervals of the record are given below.

4000–2700 cal yr BP. During this interval, the temperate-preferring taxon Cladotanytarsus mancus-type (Brooks, 2006) is dominant. Thus, we infer that the temperature was relatively high during this interval. Stictochironomus and Procladius were abundant in this stage as well as in the mid-Holocene (Wang et al., 2016), and this may indicate a relatively warm environment in this stage. This is similar to a record from Norway which showed that Stictochironomus and Procladius indicate a relatively warm environment (Brooks and Birks, 2000).

2700–1270 cal yr BP. The abundance of the previously dominant temperate-preferring Cladotanytarsus mancus-type decreased abruptly and it was replaced by Parakiefferiella bathophila-type which is also a temperate-preferring taxon (Brooks, 2000; Brooks and Birks, 2000). This shift in the representation of the dominant temperate-preferring taxa probably occurred in the context of cold conditions, because the cold stenotherm Hydrobaenus conformis-type (Cranston et al., 1983) appears for the first time. In addition, another cold indicator, Psectrocladius sordidellus-type (Brooks and Heiri, 2013), also started to increase, marking the beginning of the 2700 cal yr BP cold event. However, the abundance of Paratanytarsus penicillatus-type, which is not usually indicative of cool temperatures, has also increased since 2700 cal yr BP, simultaneously with Psectrocladius sordidellus-type. This curious combination of chironomid changes also occurred in a sediment record from Gerzensee, Switzerland (Brooks and Heiri, 2013). Overall, we infer that temperature began to decrease during this second stage.

1270 cal yr BP–present. The cold-preferring taxa, including Hydrobaenus conformis-type and Psectrocladius sordidellus-type, and cool-preferring taxa Paratanytarsus penicillatus-type, are dominant in this stage, while the relatively temperate-preferring taxa, including Cladotanytarsus mancus-type, Parakiefferiella bathophila-type and Procladius, exhibit low abundances. Thus, we conclude that temperatures reached a minimum. Several climatic events can be recognized; for example, chironomid subzones 3a, 3c and 3e correspond to the Sui-Tang Warm Period (STWP), the Medieval Warm Period (MWP) and the modern warm period, respectively; in addition, subzones 3b and 3d correspond to the cold periods of the Five Dynasties and Ten Kingdoms in China and the LIA, respectively.

The foregoing analysis indicates that the temperature variability inferred from typical chironomid temperature indicators is in accordance with that inferred from most of the other taxa in the Gonghai Lake sediments, which supports our reconstruction.

In addition, a recent study indicated that the organic matter content of the Gonghai Lake sediments was dominated by the authigenic fraction during the past 4000 years (Chen et al., 2018). This suggests that most of the organic matter is of within-lake origin (Birks and Birks, 2006) and thus that variations in its content probably reflect past regional temperature changes. The variation of the organic content of the Gonghai Lake sediments (Fig. 4; Chen et al., 2018) is consistent with the decreasing trend of chironomid-inferred temperature, validating the reliability of our temperature reconstruction.

As mentioned previously, climate model simulation results indicate that TANN in China was higher in the late Holocene than in the mid-Holocene (Jiang et al., 2012). In addition, even the global TANN indicates a warming trend from the early Holocene onwards, due to the retreating ice sheets and rising atmospheric greenhouse gas concentrations (Liu et al., 2014), in contradiction to the cooling trend inferred from various proxy records for 30–90∘ N (Marcott et al., 2013). Our qualitative reconstruction of TANN in northern China suggests that the warming trend estimated for the late Holocene by the simulation results is not convincing.

To validate our chironomid-inferred temperature record (Fig. 6a), all the Holocene temperature reconstructions for China were collected. However, as mentioned in the Introduction, many of the records are problematic in that they have large dating uncertainty, low resolution or are environmentally ambiguous; for these reasons, they were excluded. Only two unambiguous and high-resolution temperature reconstructions were finally chosen for further comparison due to their precise high-quality dating which was the most important selection criterion used in this study. The first record is based on stalagmite layer thickness at Shihua Cave, close to Gonghai Lake (Tan et al., 2003; Fig. 6b); the second is based on historical documents pertaining to winter temperature changes in eastern China (Ge et al., 2003; Fig. 6c). The three (Fig. 6a, b, c) records exhibit a consistent pattern of temperature change on both millennial and shorter scales: cold intervals of 1650–1350, 1150–950 and 650–300 cal yr BP (LIA); and warm intervals of 1350–1150 cal yr BP (STWP) and 950–650 cal yr BP (MWP). In addition, an integrated temperature record for the whole of China, produced by combining multiple paleoclimate proxy records from ice cores, tree rings, lake sediments and historical documents (Fig. 6d, Yang et al., 2002), was compared with the chironomid-inferred temperature record from Gonghai Lake. Both records show the same pattern of warm and cold intervals during the past 2000 years: for example, the cold intervals of 1650–1350 and 1150–950 cal yr BP, and the LIA, STWP, MWP and modern warm periods.

In addition to the consistency of the records described above, the trend of generally decreasing temperature during the past 4000 years is also evident in several other recent proxy-based reconstructions: for example, the U37K′ record from the sediments of Gahai and Qinghai lakes in the northeastern Tibetan Plateau (He et al., 2013; Wang et al., 2015), a novel microbial lipid record from Dajiuhu in central China (Huang et al., 2013), percentages of thermophilous trees in Huguangyan Maar Lake in southern China (Wang et al., 2007) and an integrated temperature reconstruction for 30–90∘ N (Fig. 6e; Marcott et al., 2013). The similarity of these proxy-based temperature reconstructions to a record of total solar irradiance (Fig. 6f; Steinhilber et al., 2009) and the similar decreasing trend of the various reconstructions and solar insolation (Fig. 6g; Berger and Loutre, 1991) suggest that solar irradiance and insolation are important external drivers of temperature variability during the late Holocene at centennial and millennial scales, respectively.

The foregoing analysis demonstrates that our chironomid-based temperature reconstruction is reasonable and representative and that the approach can be extended to longer timescales. The success of our approach can be attributed to the following factors: (i) chironomids are sensitive to temperature changes; (ii) the precise, high-resolution chronology increases the usefulness of the temperature reconstruction; and (iii) the 60-year resolution enables the results to be compared with other high-quality temperature reconstructions and with documentary evidence. Furthermore, our results, combined with a pollen-based precipitation reconstruction from the same core, enable the identification of trends in both temperature and humidity (Fig. 5b; Chen et al., 2015). Generally, there were pronounced changes in warm and humid/cool and dry climatic patterns both on millennial and centennial scales in the Gonghai Lake region, consistent with a previous synthesis study (Tan et al., 2011b). However, this pattern is not evident during the last 300 years (Fig. 5). Given the general consistency between our temperature reconstruction and other records in this period (such as the rapid warming at the end of LIA; Fig. 6), more high-quality precipitation records are needed to further validate this warm and dry configuration.

Although past wars in China were often the consequence of social–geopolitical factors, including territorial disputes (Zhao, 2006), nomadic invasions and agricultural expansion (Di Cosmo, 2002), the impact of climate change should also be considered when analysing societal evolution (Ge, 2011). Traditionally, China was an agricultural society, the productivity of which was very low during most of its history. When temperature or precipitation decreased abruptly, or fluctuated significantly, there tended to be an increase in the incidence of natural disasters such as floods and droughts (Zhang et al., 2008), which seriously affected agricultural production. The combination of a large population and a poor grain harvest often resulted in high rice prices and famines, which generated large numbers of homeless refugees and outbreaks of plague. These factors would finally trigger wars and social unrest which acted to reduce the population size. To analyse the societal response in Shanxi Province to climate change, the occurrence of wars (Fig. 5c) and changes in population size (Fig. 5d) were summarized for comparison with the chironomid-inferred temperature record (Fig. 5a) and the pollen-based precipitation reconstruction (Fig. 5b; Chen et al., 2015) from Gonghai Lake.

Although both temperature and precipitation in the Gonghai Lake region exhibit a decreasing trend during the last 4000 years, temperature changes were not always in phase with precipitation changes. For example, four cold events can be recognized from the chironomid-inferred temperature record (Fig. 5a), which occurred during ∼ 2710–2180 cal yr BP (Spring and Autumn Period and Warring States Period), 1690–1300 cal yr BP (Period of Disunity), 1050–900 cal yr BP (Five Dynasties and Ten Kingdoms), and 650–300 cal yr BP (Ming dynasty). The reconstructed precipitation record only exhibits two dry events during this interval, during 1050–900 cal yr BP and during 650–300 cal yr BP. The societal response to such events varied during different periods. The incidence of war was especially high during 1050–900 and 650–300 cal yr BP when both temperature and precipitation were lower; it was higher at these times than during the periods of 2710–2180 and 1690–1350 cal yr BP when the decrease in temperature was more severe than that of precipitation. This relationship is confirmed by the results of Granger causality analysis (see Table 2), which show that the incidence of wars is more strongly correlated with temperature changes than with precipitation. However, this may only be a statistical artifact and the causal relationship between climate change and societal crises needs to be further tested in future research. A sharp decrease in temperature may have been an important precondition for the outbreak of war in China, but it may be insufficient as an isolated variable, and decreases in precipitation during the past 3000 years may also have been important. Moreover, the fact that historical documents in China became increasingly detailed and reliable as human society developed (Ge et al., 2010) may be an additional explanation for the observation that increases in the frequency of wars persistently coincided with decreases in temperature and precipitation. With regard to population, an increase often occurred during warm periods which would have created latent economic pressures when the crop harvest was poor following a cold period. In addition, population collapse often occurred following an increase in the frequency of wars, famines and plagues during cold periods, suggesting that population size could be indirectly influenced by climate change (Zhang et al., 2011).

The demise of the Ming dynasty provides an example of how climatic deterioration, as well as the related socioeconomic impacts, severely undermined an empire in historical China. The late Ming (390–306 cal yr BP) coincided with the Little Ice Age, when temperatures decreased significantly (Fig. 5a). During this cold period, the incidence of natural disasters such as flood and droughts was the highest in Shanxi history (Chen, 1939). Rapid cooling accompanied by large-scale desertification began in the 1620s and had a devastating effect on agricultural production (Wang et al., 2010; Yin et al., 2015). Zheng et al. (2014) noted that the total grain yield in Shanxi in the 1630s ranged from 1219.8×106 to 1951.3×106 kg, a reduction of almost 50 % compared to the yield in ∼ 1580 of (2439.1×106 kg). The population increased from 8.42 to 9.50 million during this period (Zheng et al., 2014) and it seemed that widespread famines would be unavoidable given the additional factor that governmental disaster relief did not function due to political corruption in the late Ming (Zheng et al., 2014; Xiao et al., 2015). Furthermore, the fiscal situation of the Ming was precarious since conflicts with the Jurchen people soon exhausted the treasury and the government was forced to levy higher taxes on the peasants (Huang, 1974; Gu, 1984; Wei et al., 2014). The exacerbation of the food crisis consequently triggered a prolonged peasant uprising which broke out in northern Shaanxi, spread to Shanxi, and finally overthrew the Ming Empire in 1644. The historical records at a provincial level are voluminous and the socioeconomic context was complex and further research is needed to explore the relationship between climate change and the societal response on a regional scale in China.