Bicentennial Volcanic Activity Cycles and Their Long-Term Impact on Northern Hemisphere Climate

Abstract

Six up-to-date reconstructions of hemispheric and global temperatures and two indices of volcanic activity were analyzed using both Fourier and wavelet approaches over time intervals of up to 1500 years. A cyclicity with a period of 188–250 years was found to be present in both the Northern Hemisphere temperature and volcanic activity. These cycles were found to be negatively correlated with the correlation coefficients, reaching values of −0.61–−0.68 over the entire time interval. The maximum correlation coefficient is reached when volcanic variations lead to temperature variations by 20 years. The combined probability of the null hypothesis—the proposition that in the Northern Hemisphere, there is no real association between the bicentennial change in volcanic activity and temperature and that the revealed correlations arose purely by chance—was found to be no more than 1.2 × 10⁻² over the entire time interval and less than 10⁻² over the time interval of AD 1270–1980. The effect was weaker for the global temperature and was not detected in the Southern Hemisphere. It is shown that the identified bicentennial temperature variation could have made a certain contribution to the warming in the first half of the 20th century. However, this cycle has probably been in decline for the last 40–50 years and the decline should continue for the next few decades. Possible origins of the revealed bicentennial correlations are discussed, and guidelines for further research are proposed.

1. Introduction

Global climate change, due to its great impact on all aspects of human activity, has become one of the most important problems of modern natural science. One of the main challenges in global change research is isolating the variations caused by external forcings, as they need to be separated from the background of natural climate variability. For example, identifying the anthropogenic contribution to global warming over the past 120 years represents an important scientific task. The main difficulty of such an analysis is related to the brevity of the available instrumental data sets. Direct temperature records typically span no more than the last 100–150 years, so they are insufficient for setting limits on the amplitude of long-term (multidecadal or longer) climate variability. Paleoclimatology, which has been actively developing in recent years, makes it possible to overcome these difficulties. The use of long-term paleocrecords provides a unique opportunity to expand our knowledge of long-term climate change. Modern long-term climate reconstructions from natural climate archives span up to the last few millennia. Thus, they seem to be a potential source of information on climatic variations, with periods up to a century or even more. The most commonly used source of information on temperatures in the past is tree rings since they provide annually resolved and precisely dated climate signal that correlates well with the observed temperature and precipitation records. One of the major challenges in dendroclimatology is that the standardization procedure, namely biological trend removal, can suppress long-term variations and, thus, limit the preservation of medium- and low-frequency variability. However, recent advances in standardization techniques make it possible to retain much more low-frequency climate information [1]. Therefore, modern temperature reconstructions have become suitable for analyzing long-term variations, with periods of up to several centuries.

In the last 15 years, a large body of evidence has been obtained for the presence of a bicentennial (about 200 years) variability in the Earth’s climate. This variation has already been identified in (a) the temperature in Central Asia [2], (b) the temperature in Northeastern Alaska [3], (c) the monsoon precipitation on the Qinghai–Tibetan Plateau [4], (d) the monsoon precipitation in South America [5], and (e) the temperature in the Southern Hemisphere [6,7]. Breitenmoser et al. [8] analyzed a near-global collection of 17 annually resolved tree-ring-based climate proxies and found bicentennial periodicity in 10 of them. This work continues the study of such variability of the Earth’s climate. It is known that solar and volcanic activity can influence Earth’s climate on different time scales. Explosive volcanic eruptions are powerful events that have many effects on the Earth’s environmental and climatic processes. Volcanic explosions cause short-term perturbations of the global energy balance, which can activate complex climate feedbacks that operate over fairly long periods of time. The aim of this work is to search for quasi-200-year climate variations on a hemispheric scale and to study their possible links with long-term changes in solar and volcanic activity.

2. Materials and Methods

In this work, I used the four most recent Northern Hemisphere (NH) temperature reconstructions by Schneider et al. [9], Wilson et al. [10], Guillet et al. [11], and Büntgen et al. [12] obtained during the last 10 years. The Southern Hemisphere (SH) temperature reconstruction by Neukom et al. [13] and the median of the full ensemble of the PAGES2k global mean temperature reconstructions [14] were also used. This series of aggregate information, from tens to thousands of individual records, is most relevant to understanding the evolution of temperature over two millennia. These indicators include data on tree-ring width (TRW), maximum latewood density (MXD), stable isotopes in terrestrial archives, marine and lake sediments, documentary information, borehole temperature, and speleothems, and they cover the period up to the last 2000 years. They were obtained using advanced standardization methods that preserve low-frequency information. Thus, they are suitable for studying temperature variations over periods of up to several centuries. Most of the time series are used to reconstruct summer temperatures. However, given that instrumentally measured summer temperatures in the Northern Hemisphere correlate well with yearly averages (Rl≈0.9), the used reconstructions should also reflect long-term variations in annual average temperatures well.

Volcanic sulfate aerosols are deposited with snow and preserved in polar ice. Thus, signals of past volcanic eruptions can be distinguished and quantified by measuring the sulfate in or the acidity of polar ice cores. In this work, I used the two longest records of volcanic activity obtained by Gao et al. [15] and Crowley and Unterman [16] from the Greenland and Antarctic cores. The frequency of volcanic activity is not constant but increases during certain periods, forming repeated clusters of volcanic eruptions, and is, thus, subject to long-term fluctuations. (see Figure 1A,B). Volcanic series averaged over 40 years were used as indicators of the long-term changes in volcanic activity. All temperature and volcanic series used in the work cover time intervals of at least 1200 years. Thus, they are suitable for studying the correlation between long-term variations in volcanic activity and temperature over at least 5–6 two-hundred-year cycles. All data used are described in

Table 1. A list of paleorecords used in analysis.

Source	Abbreviation	Time Span	Reconstructed Value	Geographic Area	Data Type
Crowley and Unterman [16]	SLF	501–2000	aerosol optical depth	Greenland, Antarctica	SO4 concentration, conductivity, 21 records
Gao et al. [15]	AOD	800–2000	stratospheric sulfate aerosol injection	Greenland, Antarctica	SO4 concentration, conductivity, 36 records
Schneider et al. [9]	NHS	600–2002	June–August temperature	Extratropical part of the Hemisphere (Φ > 30° N)	Tree-ring (MXD), 15 regional records
Wilson et al. [10]	NHW	800–2010	May–August temperature	Northern Hemisphere	Tree-ring (MXD), 54 records
Guillet et al. [11]	NHG	500–2000	June–August temperature	Northern Hemisphere	Multi-proxy (TRW, MXD, δ18O), 27 records
Büntgen et al. [12]	NHB	1–2016	June–August temperature	Northern Hemisphere	TRW, 9 regional records
The median of the full ensemble of the PAGES2k [14]	GLB	1–2017	Annual	Globe	Multi-proxy (TRW, MXD, ice core, corals, historic documents, sediments, boreholes, speleothems) 692 individual records
Neukom et al. [13]	SHN		May–April temperature	Southern Hemisphere	Multi-proxy (TRW, MXD, ice core, corals, historic documents, sediments), 111 individual records

and shown in Figure 1.

Fourier and wavelet analyses were used for the study of the spectral properties of climatic and volcanic series. Levels of significance of the peaks of Fourier and global wavelet spectra were estimated relative to the red noise background according to Torrence and Compo [17].

To select quasi-bicentennial cycles, all analyzed records were wavelet filtered in the 171–259-year band using the real-valued MHAT (Mexican hat) basis (see Torrence and Compo [17]). The wavelet transform differs from the well-known Fourier transform in that the analyzed signal is decomposed not into infinite sinusoidal harmonics but into a series of orthogonal waves of the soliton type. They are called wavelets and are well localized in both the frequency and time domains, while Fourier harmonics are localized only in the frequency domain. Due to these properties, the wavelet transform is suitable for analyzing non-stationary time series, including non-periodic inhomogeneities, deterministic chaos, and local periodic structures. Therefore, for the analysis of climate series, most of which are highly non-stationary, wavelet filtering is preferable. This is especially true for signals with irregular shapes.

I assessed the significance of the correlation between the band-pass-filtered records using a statistical experiment involving an ensemble of Monte Carlo simulations. Each simulation contained (a) generating random copies of the analyzed signals by randomizing the phases of their Fourier transforms, (b) wavelet filtering of the random copies in the range of 171–272 years, and (c) calculating the correlation coefficient between two filtered random copies and comparing them with the coefficient of correlation between filtered actual data sets. If the coefficient of correlation between filtered random copies was greater than the coefficient of correlation between the filtered actual records, the result of the simulation was considered positive. The probability of the null hypothesis, namely the assumption that the observed band-pass correlation between the time series under study occurs by chance, was estimated as a ratio of the number of positive results to the total number of simulations. The statistical test applied uses the nonparametric random phase method of Ebisuzaki [18], based on the generation of a large number of random series with the same power spectra as the analyzed signal but with random phases in the Fourier modes (see also [19,20]). Poluianov and Usoskin [20] showed that the nonparametric method provides a fairly conservative estimate of significance and avoids its overestimation, which can occur if the signals under study contain strong variations with pronounced peaks in the spectra.

Fisher’s method for assessing the combined probability was used to estimate the probability of the overall null hypothesis using a number of temperature series.

3. Results

The Fourier spectra of the analyzed time series obtained using the Morlet basis are shown in Figure 2.

Figure 2 shows that almost all analyzed data sets, both volcanic and temperature, have a variation, with a period of 188–250 significant at 0.05 level. This means that the probability that the studied series are random processes of the red-noise type (first-order autoregressive processes) and the observed peaks arose by chance is no more than 0.05. Only the global temperature has a variation, with a period of 170 years, and the probability of the null hypothesis’ p exceeds 0.05. The global wavelet spectra of these records also show the presence of a bicentennial (periods of 195–250 years) variation. The quasi-bicentennial variation in volcanic activity is ensured not only by the increase in the number of eruptions approximately every 200 years but also by the fact that particularly powerful explosions (Samalas (1257), Kuwae (1453), Raung (1593), Huainaputina (1600), and Tambora (1815)) coincide in time with the periods of these increases (see Figure 1C). Thus, a chain of powerful volcanic events can serve as an additional rhythm driver for the bicentennial cycle. The period of the revealed variation is close to the period of the Suess solar cycle, which is 195–250 years [21,22]. To examine a possible relationship between the quasi-200-year temperature variations and the corresponding change in solar and volcanic activity, I calculated the correlation coefficients between the time series wavelet filtered in the bicentennial frequency band (171–282 years). No significant correlation between the quasi-200-year periodicities in hemispheric–global climate proxies and the millennial solar activity reconstructions obtained in [23,24,25,26,27,28] was found. Change in volcanic activity is another potential contributor to the bicentennial hemispheric-scale cyclicity [8]. The coefficients of correlation between the wavelet-filtered volcanic and temperature data sets are shown in

Table 2. Bandpass (171–282 years) correlations between volcanic and temperature indices shifted by 20 years calculated over the entire time interval.

Source	Temperature in the NH [9]	Temperature in the NH [10]	Temperature in the NH [11]	Temperature in the NH [12]	Global Temperature [14]	Temperature in the SH [13]
Sulfate injection [15]	−0.38 ** (0.21)	−0.67 * (0.011)	−0.05 *** (0.75)	−0.55 *** (0.039)	−0.35 *** (0.174)	0.30 + (0.292)
Aerosol optical depth [16]	−0.43 * (0.137)	−0.68 * (0.009)	−0.06 * (0.87)	−0.61 * (0.018)	−0.43 * (0.123)	0.09 + (0.703)

*—coefficient of correlation, calculated over AD 800–1980, **—AD 600–1980, ***—AD 501–1980, ⁺—AD 1000–1980. Big bold figures display coefficients of correlation with significance p < 0.05.

, together with their significances (p-values) in brackets.

The analysis carried out showed that the two-hundred-year variations in the Northern Hemisphere and global temperatures are negatively correlated with the corresponding change in volcanic activity. In two cases, NHW and NHB, the correlation is significant, with p values of less than 0.05. This means that the probability that there is no connection between the bicentennial variations in these two series and the corresponding cycles of volcanic activity, and that the observed correlation arose purely by chance, is no more than 0.05. On the other hand, the NHG series correlates very weakly with the volcanic series over a two-century scale. In most cases, the correlation coefficient is maximum when the lag between volcanic activity and temperature is 20 years. That is, the two-hundred-year cycle in volcanic activity is 20 years ahead of the corresponding cycle in temperature. In some cases, the maximum is reached with a shift of 10 and 30 years, but the difference is small. In the Southern Hemisphere, this effect is not detected. Figure 3 illustrates these correlations.

If we consider the significance of the volcano–climatic correlation, obtained by using reconstructions [9,10,11,12], as the results of the four independent tests of the null hypothesis for the actual temperature variability in the Northern Hemisphere, then we can use Fisher’s method [29] to estimate the joint probability. Using Fisher’s test, I calculated the test statistic:

$${\mathsf{\chi }}_{\mathrm{F}}=-2{\sum }_{\mathrm{i}=1}^{\mathrm{k}}\ln \left({\mathrm{p}}_{\mathrm{i}}\right),$$

(1)

where $p_i$ is the probability of the null hypothesis obtained by the i-test and k is the number of tests. For bicentennial correlations with the SLF and AOD series, we have:

$${\mathsf{\chi }}_{\mathrm{F}}^{\mathrm{S}\mathrm{L}\mathrm{F}}=-2\mathrm{l}\mathrm{n}(0.21\times 0.011\times 0.751\times 0.029)=19.9,$$

$${\mathsf{\chi }}_{\mathrm{F}}^{\mathrm{A}\mathrm{O}\mathrm{D}}=-2\mathrm{l}\mathrm{n}(0.137\times 0.002\times 0.87\times 0.018)=23.7$$

(2)

According to Fisher [29], when all the tests are independent and all of the null hypotheses are true, ${\chi }_F$ has a ${\chi }_{2k}^2$ distribution. The probabilities of the null hypothesis for the obtained ${\chi }_8^2$ values are:

P^SLF (19.9) = 1.1 × 10⁻²,

P^AOD (23.7) = 2.5 × 10⁻³.

(3)

Thus, the study, performed using (a) wavelet filtering, (b) correlation analysis, and (c) Fisher’s statistical approach, showed that the probability that the bicentennial periodicity of temperature in the Northern Hemisphere is not associated with corresponding changes in volcanic activity is very small. Figure 3 shows that, in some cases, the coherency between the two-hundred-year variations in volcanic activity and the temperature is more pronounced in the second part of the analyzed time interval. Taking this into account, I calculated the correlation coefficients for the time interval AD 1270–1980. They are shown in

Table 3. Bandpass (171–282 years) correlations between volcanic and temperature indices shifted by 20 years calculated over the time interval AD 1270–1980.

Source	Temperature in the NH [9]	Temperature in the NH [10]	Temperature in the NH [11]	Temperature in the NH [12]	Global Temperature [14]	Temperature in the SH [13]
Sulfate injection [15]	−0.52 (0.078)	−0.70 (0.027)	−0.38 (0.191)	−0.58 (0.067)	−0.50 (0.131)	−0.05 (0.901)
Aerosol optical depth [16]	−0.59 (0.068)	−0.72 (0.021)	−0.50 (0.113)	−0.72 (0.019)	−0.58 (0.068)	−0.20 (0.534)

Bold figures display coefficients of correlation with significance p < 0.10.

.

Using Fisher’s statistical test, we now obtain the following results:

$${\chi }_F^{SLF}=-2ln(0.078\times 0.027\times 0.191\times 0.067)=21.0,$$

$${\chi }_F^{AOD}=-2ln(0.068\times 0.021\times 0.113\times 0.019)=25.4$$

(4)

P^SLF (21.0) = 7.1 × 10⁻³,

P^AOD (25.4) = 1.3 × 10⁻³.

Thus, in the last seven centuries, the correlation between the bicentennial cycles in temperature and volcanic activity has been increasing. The negative correlation in this interval is manifested both in the reconstruction NHG and in the global temperature. The overall probability of the null hypothesis in the Northern Hemisphere is now less than 0.01. However, in the Southern Hemisphere, the temperature–volcanic correlation does not appear after the middle of the 13th century.

Calculations using volcanic data averaged over 30 and 50 years as indicators of long-term volcanic activity yielded similar results.

4. Discussion and Conclusions

An analysis using four recent paleoreconstructions of Northern Hemisphere temperature found a negative correlation between bicentennial cycles in temperature and volcanic activity, although not all were highly significant. It should be noted, however, that we should not expect a perfect correlation because (a) the temperature series do not reconstruct real temperature with absolute accuracy, with reconstruction errors increasing in the past, and (b) the ice core volcanic records may not accurately reflect volcanic activity because small eruptions may not be detected in high-latitude ice cores. It is all the more remarkable that the two-century variation in the NHB series correlates significantly with the corresponding cycle in volcanic activity over a period of almost 1500 years. Fisher’s combined probability test performed using the obtained separate p values shows that the probability of the null hypothesis (the assumption that there is no relationship between the two-century variations in volcanic activity and temperature) for the entire Northern Hemisphere is no more than 1.1 × 10⁻².

The short-term influence of volcanic eruptions on climate is well-known. Volcanic eruptions release ash and sulfur gases that become sulfate aerosols. Large ash particles fall quickly, but the aerosols can remain in the stratosphere for several years, reducing the amount of solar radiation reaching the Earth’s surface, lowering tropospheric temperatures, and changing atmospheric circulation patterns. The direct cooling effect can last for several years depending on the eruption. However, the cooling trend probably can be further extended by some feedback mechanisms. There is evidence that clusters of volcanic eruptions can maintain surface cooling for periods of up to centuries [30,31,32,33] and even millennia [34] due to ocean–sea ice feedbacks. The quasi-bicentennial variation in volcanic activity is most likely a manifestation of internal geotectonic oscillations, but further research is needed to provide a more reliable answer to this question

In most cases, the correlation is greatest when the bicentennial periodicity in volcanic activity is 20 years ahead of the corresponding temperature variation. Such a delay time probably reflects the characteristic times of oceanic response to volcanic radiative forcing. For example, the model experiments of Zhong et al. [31] showed that maximum cooling in the upper 100 meters of the subpolar North Atlantic reaches a maximum at about 30 years after a cluster of four eruptions. The results of the model simulation performed by Miller et al. [35] show that explosive volcanism can produce abrupt summer cooling, which can be maintained by sea ice–ocean feedbacks long after volcanic aerosols are removed. A similar result was obtained by Waple et al. [36], who studied the climatic reaction to solar radiative forcing and found that the lag between long-term changes in solar forcing and temperature response is 1–15 years for global temperature and up to 30 years for the temperature of separate large regions. The mechanisms that provide delays in the temperature response to the corresponding forcing in cases of solar and volcanic activity may be similar.

The volcano–temperature correlation has been more pronounced over the last 700 years. Fisher’s exact test shows that, in the Northern Hemisphere, the overall probability of the null hypothesis is less than 10−2 over the period 1270–1980 AD. This result is not surprising since it is from the middle of the 13th century that the amplitude of long-term variations in volcanic activity increased significantly; see Figure 1A–C. Until the 13th century, volcanic eruptions occurred less frequently and were distributed more evenly over time.

In the Southern Hemisphere, no correlation between volcanic activity and temperature is observed, either over the entire interval or over the last seven centuries. This is consistent with the modeling results of [31], which showed that the temperature response to volcanic forcing is significantly weaker in the Southern Hemisphere than in the Northern Hemisphere. Monerie et al. [37] also found that volcanic activity has a generally weaker impact on the Southern Hemisphere than the Northern Hemisphere climate. This is most likely due to the fact that the Southern Hemisphere climate is potentially influenced more by internal than external variability [13,38]. And this, in turn, may be due to the fact that the Southern Hemisphere is significantly more covered by oceans, whose large heat capacity can buffer external forcing [39].

On a global scale, the two-century volcano–temperature correlation is weaker than in the Northern Hemisphere. Obviously, this is due to the fact that the global temperature summarizes the data of both hemispheres, namely the Northern, where the effect is noticeable, and the Southern, where it is absent.

The existence of a hemispheric-scale bicentennial temperature periodicity associated with changes in volcanic activity is consistent with the results of Chim et al. [40], who concluded that volcanic forcing on large-scale climate indices may currently be underestimated. The average peak-to-peak amplitude of the two-hundred-year temperature periodicity in the Northern Hemisphere is 0.07–0.10 °C. Although some individual variations have a greater range. Importantly, the quasi-200-year cycle can provide a temperature increase from the mid-19th century to the 1970s of the 20th century of up to 0.20 °C (see Figure 3). Thus, this variation can have a certain contribution to global warming in the first half of the 20th century. However, the last 4–5 decades are likely a period of decline in this cycle, which should have resulted in a decrease in warming. This makes the significant temperature increases in the Northern Hemisphere over the past 30–40 years even more anomalous and suggests that the anthropogenic contribution to global warming in recent decades may have been greater than current estimates suggest. The natural bicentennial cycle should continue to decline over the next few decades.

The correlation found between bicentennial cycles of volcanic activity and Northern Hemisphere temperatures provides new evidence for the long-term effects of volcanic forcing on the climate. This result highlights the need for a further study of feedback mechanisms that may prolong cooling trends and for their careful consideration in climate modeling.

It should be noted that there is a problem with Fisher’s test, which is that the four Northern Hemisphere data sets may not be completely independent. Some of the reconstructed series may have a portion of identical individual temperature indices. For example, in the reconstruction of the NHB, four of the fifteen regional series used, namely the Tornetrask, Jamtland, Athabasca, and Zhaschiviersk, were taken from regions geographically coinciding with the areas that were also used in the NHS, namely Northern Scandinavia, the Great Basin, and North Yakutia. In such cases, Fisher’s approach cannot be considered fully justified, and this point requires further study. Therefore, for more reliable conclusions about the influence of long-term variations in volcanic activity on the Earth’s temperature, it is desirable to obtain new paleoreconstructions based on completely independent individual indicators. This will allow for the use of advanced statistical methods, including Fisher’s approach, which will facilitate more complex and in-depth studies of the effect. It would also be desirable to expand the spatial coverage of the individual proxy records used for hemispheric and global temperature reconstructions, thereby making these reconstructions more representative. Even some of the most recent hemispheric reconstructions are based on data from a fairly limited area. For example, tree-ring series usually are taken from areas close to the northern and altitude timberline. Expanding the geographic coverage, especially the inclusion of more marine proxies, can improve our knowledge of past climate variations and their causes.