Large volcanic eruptions may cause abrupt summer cooling over large parts of the globe. However, no comparable imprint has been found on the Tibetan Plateau (TP). Here, we introduce a 400-yr-long temperature-sensitive network of 17 tree-ring maximum latewood density sites from the TP that demonstrates that the effects of tropical eruptions on the TP are generally greater than those of extratropical eruptions. Moreover, we found that large tropical eruptions accompanied by subsequent El Niño events caused less summer cooling than those that occurred without El Niño association. Superposed epoch analysis (SEA) based on 27 events, including 14 tropical eruptions and 13 extratropical eruptions, shows that the summer cooling driven by extratropical eruptions is insignificant on the TP, while significant summer temperature decreases occur subsequent to tropical eruptions. Further analysis of the TP August–September temperature responses reveals a significant postvolcanic cooling only when no El Niño event occurred. However, there is no such cooling for all other situations, that is, tropical eruptions together with a subsequent El Niño event, as well as extratropical eruptions regardless of the occurrence of an El Niño event. The averaged August–September temperature deviation (Tdev) following 10 large tropical eruptions without a subsequent El Niño event is up to −0.48° ± 0.19°C (with respect to the preceding 5-yr mean), whereas the temperature deviation following 4 large tropical eruptions with an El Niño association is approximately 0.23° ± 0.16°C. These results indicate a mitigation effect of El Niño events on the TP temperature response to large tropical eruptions. The possible mechanism is that El Niño events can weaken the Indian summer monsoon with a subsequent decrease in rainfall and cooling effect, which may lead to a relatively high temperature on the TP, one of the regions affected by the Indian summer monsoon.
Volcanic eruptions are a major natural driver of climate variability (Robock 2000). Sulfate aerosols from large volcanic eruptions that are injected into the stratosphere can shield Earth’s surface from incoming solar radiation, causing regional or global surface cooling (Briffa et al. 1998; Sigl et al. 2015; Stoffel et al. 2015; Luterbacher and Pfister 2015; Büntgen et al. 2016; Schneider et al. 2017; Esper et al. 2015). One well-known case of this phenomenon is the eruption of Tambora in April 1815, which led to “the year without a summer” in 1816 in western Europe and northern North America (Stothers 1984; Oppenheimer 2003; Luterbacher and Pfister 2015; Luterbacher et al. 2016). Following large volcanic eruptions, exceptionally low tree-ring growths across large areas of the northern mid–high latitudes have been documented (Briffa et al. 1998; Sigl et al. 2015). However, in contrast to the evidences identified in the northern mid–high latitudes, no comparable imprint of some large eruptions (e.g., the Tambora eruption in 1815) can be found on the Tibetan Plateau (TP; Bräuning and Mantwill 2004; Duan and Zhang 2014; Duan et al. 2017). The cause of this subdued volcanic effect on the TP is still unclear.
The TP is the highest and largest plateau in the world and often responds more sensitively to climate change than do other areas at the same latitude (Duan et al. 2015). Surface temperature changes in this unique geographic unit have strong effects on both the climate and ecosystems of the Asian continent and can even impact much larger areas (Yao et al. 2012; Duan et al. 2015). However, the uniqueness of this region and the complexity of the interactions between climatic drivers on the TP make it difficult to fully understand the long-term mechanisms of climate change on the TP (Sato 2009). For the interannual temperature variability, in addition to the external forcing (i.e., volcanic eruptions), internal phenomenon is also an important driving factor. For example, the contribution of ENSO to global air temperature variations over the past 150 years is 10%–30% (Privalsky and Jensen 1995). The influence of ENSO on climate was also reported on the TP and in surrounding areas. One of the most important effects is that ENSO has a direct influence on the Indian monsoon (Kumar et al. 1999). Moreover, a potential link between temperature variations on the TP and ENSO was also reported (Yin et al. 2000; Wang et al. 2003). However, most previous studies performed on the TP are based on relatively short instrumental records, and the influence of the interaction between external volcanic forcings and internal ENSO variations on the interannual temperature variability over the TP has not been explored. Here, we put forward a hypothesis that ENSO influences the interannual temperature variability on the TP by impacting the strength of the Indian monsoon, which in turn mediates the response of the TP temperature to volcanic eruptions. The identification and evaluation of climatic extremes from periods before the mid-twentieth century require high-quality proxy archives (Cook et al. 2013).
Annually resolved tree-ring data are important sources for the assessment of past summer cooling following strong volcanic eruptions (D’Arrigo et al. 2009; Anchukaitis et al. 2010; Esper et al. 2013a; Sigl et al. 2015). Compared to the commonly used tree-ring width (TRW), measurements of maximum latewood density (MXD) are more sensitive to summer temperature variability and, likely, even to light availability (Stine and Huybers 2014; Esper et al. 2015; Duan et al. 2017). Moreover, the MXD is less biased by biological memory effects or growth-inertia characteristics compared to the TRW (Frank et al. 2007; Esper et al. 2013b; Büntgen et al. 2015). The growth of the tree-ring MXD is usually controlled by temperature conditions in the growing season or late summer (Briffa et al. 1988; Schweingruber et al. 1991). This is the important period for tracheid cell wall thickening. Favorable temperatures in summer are conducive to photosynthesis and contribute to latewood cell wall thickening and the production of higher latewood density. Contrarily, cool summers can result in low latewood density. This relationship has also been demonstrated in previous TP MXD studies (Bräuning and Mantwill 2004; Fan et al. 2009; Wang et al. 2010; Duan et al. 2010, 2017; Duan and Zhang 2014). Therefore, summer cooling resulting from volcanic eruptions can also lead to an MXD decrease.
Here, we estimate the magnitude of cooling after large volcanic eruptions since 1620 over the TP based on a temperature-sensitive network of 17 tree-ring MXD sites. Furthermore, we explore the potential causes of the unobvious imprints of large volcanic eruptions on the TP compared to those documented in other regions. Especially, we will identify the cooling effect driven by external volcanic forcing with consideration of the internal ENSO influence. The structure of this paper is as follows: In section 2, the materials and methods used in this study are described briefly. Section 3 presents the main results obtained from this study. Section 4 focuses on a discussion of the results and possible mechanisms. Section 5 summarizes the conclusion.
2. Materials and methods
a. Tree-ring MXD data
Tree-ring MXD data were derived from old-growth Balfour spruce (Picea likiangensis var. balfouriana) trees at 17 high-elevation sites on southeastern TP (Fig. 1). For tree-ring-based climate reconstructions, tree-ring sampling sites without any natural or human disturbance (e.g., fire, pathogen outbreaks, and grazing) need to be chosen to guarantee that the pure climatic signal can be extracted from the collected tree-ring samples. To identify these disturbances, in addition to traces left on trees (e.g., fire scarring), detailed inquiries or investigations on the associations from the local residents, forestry bureaus, or document records can also be important means of obtaining information. Moreover, the analysis of metadata (e.g., the coherence among trees and coherence among chronologies/sites) can also be an approach to identifying pathogen disturbances in forests. In this study, all sampling sites were without any evidence or record of natural or human disturbances that might have impacted tree growth. The tree-ring density of the collected samples was measured using the X-ray method according to standard practices (Schweingruber et al. 1988; Duan et al. 2010). Considering that ring dating after densitometric analysis might be problematic because some cores are unavoidably broken during the process of sugar and resin extraction and sample sectioning, we performed ring-width measurements and cross dating before densitometric analysis. The measured ring-density data were dated based on tree-ring width cross dating, and were verified with thin wood sections when inconsistencies appeared. The quality of densitometric cross dating was tested using the quality control and dating check of tree-ring measurements (named as COFECHA) program (Holmes 1983). Thus, although the sizes of the samples vary with time, the quality of cross dating is guaranteed. A total of 739 measured MXD sequences were dated to the corresponding calendar years, and the longest series reaches back to 1375 (Fig. 2). Because the number of samples decreased back in time, the effect on the chronology signal strength was evaluated using the expressed population signal (EPS; Wigley et al. 1984). The reliable period of the regional composite growth deviation (MXD_GD) series was determined considering the criterion of EPS > 0.85, calculated based on the regional MXD raw chronology using auto-regressive standardization (ARSTAN) software (Cook and Krusic 2005). The resulting reliable period starts in 1620, with the smallest sample size of 39 cores, and ends in 2014 (Fig. 2). We developed MXD_GD based on the network comprising 739 MXD measurement series from 17 sites on the TP.
The regional MXD_GD series was calculated as the MXD deviation for each year relative to the preceding 5-yr mean:
with MXDi representing the MXD value in year i and MXDi−1:i−5 corresponding to the mean values from year i − 1 to year i − 5. This calculation was applied for each year and all MXD measurement series. The final regional MXD_GD series was obtained by averaging all 739 individual MXD_GD series. The MXD_GD series is designed to examine the relative growth change, also referred to as the abrupt growth change, of the tree-ring MXD following a volcanic eruption. This approach has also been used to explore the response of other tree-ring parameters, such as tree-ring widths, to extreme climate/disturbance events (Schweingruber et al. 1990). The formula can provide quantitative evaluation of growth responses of the tree-ring MXD to volcanic eruptions. Based on this evaluation, the quantitative magnitude of temperature changes triggered by these events can be calculated. In this study, we chose a 5-yr window (i.e., relative to the preceding 5-yr mean) and tried windows from 1–4 years. We found that there is little difference, and the 5-yr window is the best for capturing the temperature signal (figure not shown).
b. Climatic data
Homogeneous monthly temperature data from 79 meteorological stations (Duan et al. 2015) at elevations above 2000 m MSL were used in this study (Fig. 1). The record length varies among stations, but each includes at least 36 years from 1979 to 2014. The gridded land surface air temperature data used in this study are from the Climatic Research Unit Global Temperature, version 4.3 (CRUTEM4.3), and have a spatial resolution of 5° by 5° starting in 1850 (Jones et al. 2012).
c. Growth–climate relationships
The growth–climate response relationships of the regional MXD_GD series were analyzed against monthly temperature data from April–September both for individual station records and at different regional scales. To maintain consistency with the calculated MXD_GD series, observational monthly temperature series were also calculated as temperature deviations (Tdev) with respect to the preceding 5-yr mean.
d. Summer cooling estimation
Linear regression was used to reconstruct the regional August–September Tdev over the period 1620–2014. The skill of the regression models was verified using split-period calibration/verification (Meko and Graybill 1995), as well as the leave-one-out and leave-five-out cross-validation methods (Michaelsen 1987). The statistics for evaluation of the regression model include Pearson’s correlation coefficient R, the reduction of error (RE), the sign test (ST), the first-difference sign test (ST1), and the coefficient of efficiency (CE; Table 1). The reconstruction uncertainty is considered as the regression-based uncertainties associated with the residuals in the verification period. The uncertainty was calculated as the root-mean-square error (RMSE).
e. Volcanic dataset
Information regarding volcanic eruption events was obtained from the global volcanic forcing (GVF) values from Sigl et al. (2015). In total, 27 relatively large eruptions (GVF < −1) that occurred in the NH mid–high latitudes and tropics since 1620 were used in this study (Table 2). The 27 events include 14 tropical eruptions and 13 extratropical eruptions. The GVF was estimated from the reconstruction of volcanic aerosol deposition since early Roman times for both polar ice sheets. The main difference between this reconstruction and previous ones is that this reconstruction performed a time-scale revision before 1257 and added new high-resolution, ice-core sulfur measurements. The NH mid- to high-latitude eruptions were claimed using sulfate deposition from the Greenland ice cores, while tropical eruptions were confirmed based on synchronous sulfate deposition on both polar ice sheets (Sigl et al. 2015).
f. El Niño events
We considered El Niño events that occurred since 1620. This information stems from Table 2 in the paper of Brönnimann et al. (2007), which was based on a number of indices and datasets, including multiple proxy records. This El Niño dataset covers the period of 1500–2003, and three subperiods Reconstruction1 (REC1) (1500–1705), Reconstruction2 (REC2) (1706–1879), and Instrument (INS) (1880–2003) were defined. El Niño events that occurred in the period of 1880–2003 were identified using an observational sea surface temperature (SST) data (ERSST, version 2, data; Brönnimann et al. 2007). Information on El Niño events in the REC1 and REC2 periods were obtained from multisource proxy reconstructions. The El Niño events in the REC2 period were confirmed based on more proxy data, and the associated information has a higher quality than that in the REC1 period. Based on the long ENSO events chronology, Brönnimann et al. (2007) found a stationary relationship between ENSO and the European climate over a long historical period. Importantly, the volcanic forcing series, ENSO reconstructions, and TP temperature reconstruction used in the study are all data independent.
Growth-response analysis showed that MXD_GD correlates significantly with the regional Tdev during August and September (Fig. 3). Further analysis considering individual station records indicates that the MXD_GD series correlates significantly (P < 0.001) with the August–September Tdev at 69 of the 79 stations (Fig. 1). The correlation coefficients between MXD_GD and the regional August–September Tdev from 79-, 69-, and 62-station means are 0.71 [effective degrees of freedom (edf) = 58, P < 0.001], 0.74 (edf = 58, P < 0.001), and 0.75 (edf = 58, P < 0.001), respectively, over the instrumental period of 1956–2014. We used the 62-station average based on the similarity of the spatial coverage to that of the MXD network (dotted red box in Fig. 1). Sensitivity tests (not shown) suggest that averaging over a larger station network would not affect the definition of the reconstruction target appreciably. The composite MXD_GD series represents the regional MXD deviation relative to the preceding 5-yr mean and is a good proxy for the August–September Tdev on the TP from 27°–36°N to 84°–104°E (r1956–2014 = 0.75, edf = 58, P < 0.0001; Fig. 4a).
Except for the year 1962, during which the proxy indicates colder conditions, the MXD_GD series matches the instrumental data quite closely (Fig. 4a). The MXD outlier in 1962 is related to the coldest March–August (Tdev = −0.71°C) during 1956–2014, while the August–September Tdev (−0.33°C) is close to the climatology. In other words, the MXD value in 1962 reflects the temperature condition in March–August but not that in August–September. These findings demonstrate that the MXD_GD series also has the potential to indicate temperature extremes during the extended growing season (i.e., March–August) in addition to temperature extremes in August–September. Based on the MXD_GD series, postvolcanic late-summer cooling after 27 large tropical and extratropical eruptions (i.e., events exceeding GVF values of −1 W m−2) back to AD 1620 (EPS > 0.85) was quantitatively reconstructed (Fig. 4b). All the associated statistics indicate a good skill of the regression model (Table 1). The r, R2, F, ST, and ST1 values are significant at P < 0.01, and the positive REs and CEs suggest that the reconstructions have encouraging performances. The statistics of Durbin–Watson (DW) demonstrate no first-order autocorrelation in the model residuals.
Superposed epoch analysis (SEA) indicates that August–September cooling is stronger after tropical eruptions compared to extratropical eruptions (Fig. 5). This difference is likely related to the significantly different GVFs, which are significantly larger for the tropical eruptions (Table 2). Moreover, the TP temperature changes are also variable among tropical eruptions (Fig. 4b). Among the four largest tropical eruptions (GVF < −10), three eruptions (1641, 1695, 1809) forced cooling of the August–September Tdev by more than 1.28°C, whereas the largest eruption of Tambora in 1815 caused only a minor/modest temperature drop in 1816/17 (−0.25°/−0.73°C; Fig. 4b). Tropical eruptions followed by an El Niño event produced qualitative, nonsignificant warming (Fig. 5e), opposing the significant cooling from tropical eruptions not followed by El Niño events (Fig. 5c) and dampening the composite cooling over all tropical eruptions (Fig. 5a).
To verify whether El Niño compensated for the postvolcanic cooling by warming the TP, we conducted a composite analysis of the gridded August–September Tdev considering the El Niño events that occurred during the observational period 1961–2003 (Fig. 6). The composite spatial August–September Tdev field reveals a positive anomaly over a large part of the TP, demonstrating that El Niño events warmed the TP and supporting the reconstructed weakening of postvolcanic cooling effects. The possible mechanism by which ENSO warms the TP is its indirect influence on the summer monsoon. Previous studies have reported that El Niño can weaken the Indian summer monsoon, with a subsequent decrease in precipitation and accompanying drought in the regions influenced by the Indian summer monsoon (Mooley and Parthasarathy 1983; Singh and Dash 2005; Kumar et al. 2006). Summer precipitation over a large part of the TP originates from the Indian summer monsoon. Additionally, most of the annual precipitation on the TP occurs in the summer, and there is a significant negative correlation between summer precipitation and summer temperature on the TP (Wang et al. 2010; Bräuning and Mantwill 2004). This means that deficient summer precipitation on the TP can induce drought and trigger high temperatures, while abundant rainfall can dampen the summer high temperature. Thus, the mechanism by which El Niño affects the TP summer temperature can be its indirect influence on the summer monsoon. The weakened Indian summer monsoon resulting from El Niño events may lead to decreased summer rainfalls and be conducive to warming summers over the TP. This supports the hypothesis that postvolcanic summer cooling on the TP can be compensated by internal phenomena (such as ENSO warm-phase events).
These results imply that a better understanding of the interannual variability of the TP late-summer temperature is needed from the perspective of both external forcing and the mediation of internal phenomena (i.e., ENSO warm phase) on the TP temperature response to external forcing. Moreover, this interaction seems to occur more often for tropical volcanic eruptions than for extratropical eruptions. Considering the eruptions with GVF < −1 over the period 1620–2014, there are 4 out of 14 tropical volcanic eruptions with a subsequent El Niño event, but only 2 out of 13 extratropical eruptions have El Niño associations (Table 2).
To understand the mitigation effect of El Niño events on the response of the TP later-summer temperature to tropical volcanic eruptions, the 1815 Tambora event could be the most typical case. The reduced temperature effect of the 1815 Tambora eruption (Fig. 4) was likely triggered by the mitigating effects from the 1816 El Niño. The finding that the summer temperature in 1816 was not exceptional is also confirmed by another independent MXD chronology-based summer temperature reconstruction from central TP (Bräuning and Mantwill 2004). The cool summers in 1816 and/or 1817 after the Tambora eruption in April 1815 have also been found in other tree-ring studies on the TP (Liang et al. 2008; Bräuning and Mantwill 2004; Wang et al. 2010; Fan et al. 2009; Li et al. 2015). However, the magnitude of summer cooling in 1816/17 resulting from the Tambora eruption in 1815 is obviously less than that in 1810, driven by an unknown tropical eruption in 1809 (Bräuning and Mantwill 2004; Wang et al. 2010; Fan et al. 2009; Li et al. 2015). This pattern is coherent with our finding, but it is in contrast to the NH records (Briffa et al. 1998; Sigl et al. 2015). In particular, the greatest magnitude of summer cooling identified in the NH record since 1620, which occurred in 1816 (Briffa et al. 1998; Sigl et al. 2015), cannot be found in almost any tree-ring-based summer temperature reconstruction on the TP (Bräuning and Mantwill 2004; Liang et al. 2008; Gou et al. 2008; Liu et al. 2009; Fan et al. 2009; Wang et al. 2010; Yang et al. 2010; Zhu et al. 2011; Li et al. 2012; Li et al. 2015; Yin et al. 2015). This result supports the hypothesis that the TP summer cooling driven by the 1815 Tambora eruption had been mitigated by another factor. These results imply that although volcanic eruptions cause cooling in general, the particular regional response can be masked by other processes that may be independent of or even related to volcanism.
Our results indicate that the effects of tropical eruptions on the TP are significantly greater than those of extratropical eruptions. Moreover, tropical eruptions followed by an El Niño event produced qualitative, nonsignificant warming, opposing significant cooling from tropical eruptions not followed by El Niño events and dampening the composite cooling from overall tropical eruptions. Based on 27 events, including 14 tropical eruptions and 13 extratropical eruptions, over the period 1620–2014, the average magnitude change driven by large tropical eruptions without a subsequent El Niño event is approximately −0.48° ± 0.19°C, while that caused by events with an El Niño association is approximately 0.23° ± 0.16°C. These results imply a mediation effect of El Niño events on the TP temperature response to tropical eruptions. A possible mechanism of the El Niño mediation of the response of the TP temperature to volcanic eruptions can be its indirect influence on the summer monsoon. Specifically, El Niño can weaken the Indian summer monsoon, in which a subsequent decrease in precipitation and accompanying drought occurs in the regions influenced by the Indian summer monsoon. Summer precipitation over a large part of the TP, where most of the regional precipitation occurs in summer, originates from the Indian summer monsoon. Therefore, there is a significant anticorrelation between summer precipitation and temperature on the TP. Thus, deficient summer precipitation on the TP can induce drought and trigger high temperatures, while abundant rainfall can dampen the summer high temperature. The reduced temperature effect in response to the strong tropical eruption of Tambora in April 1815 is perhaps a typical case for the TP.
This research was supported by the National Key R&D Program of China (2016YFA0600404), the National Natural Science Foundation of China (Grants 41471035, 41101043, and 41405054). We thank Dr. Hongbin Liu for assistance in MXD measurement, and also greatly appreciate the Laboratory for Climate Studies, China Meteorological Administration for allowing us to use the densitometric devices. We thank the National Meteorological Information Center of China Meteorological Administration for making meteorological station data available. Jianping Duan acknowledges support from the Alexander von Humboldt Foundation. Lun Li was also supported from the Basic Scientific Research and Operation Foundation of CAMS (2016Y001). Jürg Luterbacher was supported from the German Science Foundation (DFG) project AFICHE and the Belmont Forum and JPI-Climate Collaborative Research Action “INTEGRATE: An integrated data-model study of interactions between tropical monsoons and extra-tropical climate variability and extremes.” The meteorological stations data used in this study are available for application at China Meteorological Data Sharing Service System (http://data.cma.cn/data/cdcdetail/dataCode/SURF_CLI_CHN_MUL_MON.html). The regional tree-ring series and temperature reconstruction will be archived at the NOAA paleodataset. Additional data related to this paper may be requested from the corresponding author.