LONG-PERIOD CYCLES OF THE SUNS ACTIVITY RECORDED IN

DIRECT SOLAR DATA AND PROXIES

M. G. OGURTSOV1 , YU. A. NAGOVITSYN2 , G. E. KOCHAROV1 and H. JUNGNER3

1 A.F. Ioffe Physical-Technical Institute, 194021, Polytechnicheskaya 26, St.-Petersburg, Russia
(e-mail: maxim.ogurtsov@pop.ioffe.rssi.ru)
2 Pulkovo Astronomical Observatory, 196140, Pulkovskoe shosse 65/1, St.-Petersburg, Russia
(e-mail: nag@gao.spb.ru)
3 University of Helsinki, P.O. Box 64, FIN-00014, Gustaf Hllstrminkatu 2, Helsinki, Finland

(Received 4 June 2002; accepted 6 August 2002)

Abstract. Different records of solar activity (Wolf and group sunspot number, data on cosmogenic
isotopes, historic data) were analyzed by means of modern statistical methods, including one espe-
cially developed for this purpose. It was confirmed that two long-term variations in solar activity
the cycles of Gleissberg and Suess can be distinguished at least during the last millennium. The
results also show that the century-type cycle of Gleissberg has a wide frequency band with a double
structure consisting of 50 80 years and 90 140 year periodicities. The structure of the Suess cycle
is less complex showing a variation with a period of 170 260 years. Strong variability in Gleissberg
and Suess frequency bands was found in northern hemisphere temperature multiproxy that confirms
the existence of a long-term relationship between solar activity and terrestial climate.

1. Introduction

It is well known that the Sun actively influences many phenomena on the Earth.
Due to this link, changes of solar activity affect many aspects of the life of mankind.
The question about the role of solar variability in global climate change today is of
special interest. Therefore, knowledge of the history of solar activity, its time evo-
lution and its documentation in solar-terrestrial relationship is of great importance.
It is evident that such knowledge needs reliable information on the time behavior of
different parameters of solar activity (SA) over maximally long times. However, we
have reliable information on SA (direct observations of a global magnetic field of
the Sun, measurements of galactic cosmic ray intensity and fluxes of solar radiation
in different spectral ranges) only for the last few decades. The knowledge of SA
during the last 100150 years (Zrich series of Wolf numbers, Greenwich series of
sunspot area, Aa and Kp geomagnetic indices) is poorer. Tremendous work, done
by Wolf and Waldmeier (1961), Hoyt and Schatten (1998), who compiled a lot of
sporadic sunspot observations of 17th to the beginning of 19th centuries, allowed
to obtain a reconstruction of sunspot numbers over the last 34 centuries. The
reliability of these reconstructions before the middle of 19th century needs further
investigation. Data on SA over longer time intervals can be obtained only from
solar proxies (historical data, cosmogenic isotopes) and hence are substantially

Solar Physics 211: 371394, 2002.

2002 Kluwer Academic Publishers. Printed in the Netherlands.
372 M. G. OGURTSOV ET AL.

less reliable. But in spite of the limited length of records of SA, some modes of
long secular solar variability are already known. In the second half of the 19th
century Wolf suggested solar cycles with periods of 55 and 160 years (see Schove,
1983). In the following century the ideas of Wolf were developed and significant
evidence of the existence of two longer periodicities of SA were obtained. The
first of them is a century-scale cycle revealed by Gleissberg (1944). Usually this
cycle is considered as a variation with the period 8090 years or, sometimes,
even precisely 88 year. The second one is a variation with a period of about 200
(160270) years (Schove, 1983). This is often called a Suess cycle. Suess (1980)
found a significant 203 year variation in tree-ring radiocarbon records. It should
be noted, however, that there are many evidences that the century-type solar vari-
ability may have spectral power in a wider frequency band than 8090 years.
For example in the work of Cole (1973), the double structure of the century-type
cycle was found both for Zrich sunspot data (59 and 88 year) and for data on
the phase of the 11-year solar cycle (78.5 and 94.5 year). Wittmann (1978) found
92-year and 55-year peaks in the spectrum of yearly averaged sunspot numbers for
AD 17011976. Spectral analyses of the data on aurora borealis after 500 A.D.
show peaks at 58 year and 65 year and a broad peak in the range 80130 years
(Schowe, 1983). Stuiver and Braziunas (1993) analyzed the long decadal 14 C series
and found significant 89 and 148 year periodicities for 60002000 B.C. and a 126-
year variation for 2000 B.C.1840 A.D. Existence of two kinds of century-long
solar variability 115 year and 95 year cycles was claimed by Chistyakov (1986).
A thorough investigation of solar centennial variability over a long time scale was
done by Nagovitsyn (1997, 2001). He used historic data and showed that the length
of the century-long solar cycle is not constant and most likely changes from about
65 to more than 130 years (the 7080-year and 100120-year modes dominate).
In the present work we continue the analyses of long-term (20300 years) SA
variations in the Gleissberg and Suess frequency range, using all the complexity
of direct and indirect solar data and applying modern statistical methods. The link
between solar activity and terrestrial climate is also considered.

2. Data

In order to trace the long-term solar variability in detail we used all available data
on SA direct oberservations and different proxies.

2.1. D IRECT DATA ON SOLAR ACTIVITY

2.1.1. Telescopic Sunspot Observations

Despite the first accurate description of sunspots made by Kepler in 1607, system-
atic and qualified telescopic observations of sunspots begin only at the middle of
the 19th century. But strong efforts of many scientists, and particularly of Rudolf
 LONG-PERIOD CYCLES OF THE SUNS ACTIVITY 373

Wolf (director of the Observatory at Bern and later in Zrich) make it possible to
extend the sunspot record back to the beginning of 18th century. The reliability of
different parts of the Wolf number data set, however, is different: the data are reli-
able since 1848, their reliability is good during 18181848, it is questionable from
17491817 and it is poor during 17001748 (Eddy, 1976). Recently Hoyt and
Schatten (1998) have finished the reconstruction of group sunspot number (GSSN)
in which they used many historical sources missed by previous investigators. Their
data set covers the time interval 16101995 and authors consider GSSN data as
quite reliable during 16531730, 17501788 and after 1795 (Hoyt and Schatten,
1998). Wolf numbers and GSSN show some differences in the 18th century the
values of Wolf numbers during this period are a bit higher. These two series are the
longest direct instrumental records of SA.

2.1.2. Historical Naked-Eye Sunspot Observations

The largest sunspots and sunspot groups (a total area more than 1900 h) can be
seen with unaided eye at sunrise and sunset or through smoke and haze (Wittmann,
1978). These sunspots were observed during the pre-telescopic era by ancient Ori-
ental (in particularly Chinese) astronomers and now the data on ancient sunspot
observations made by naked eye (SONE) is the longest non-proxy record of SA in
the past. The most complete catalogue of SONE, covering more than 18 centuries
and including more than 200 events, was collected by Wittmann and Xu (1987).
The reliability of the information on solar activity, extracted from historical chroni-
cles, was analysed in many works (see Willis, Easterbrook, and Stephenson (1980),
Wittmann and Xu (1987), Nagovitsyn (2001)) and it was shown that the Oriental
historic data really reflect such features of sunspot activity as 11-year and century-
scale cycles, butterfly diagrams, deep Maunder-type minima of SA. Qualification
of ancient Chinese astronomers also was high observations of many lunar and
solar eclipses, novae, supernovae, comets, meteor showers and even a possible ob-
servation of Jupiters satellite Ganymede (364 B.C.) confirm their professionalism.
However, the SONE record have substantial shortcomings:
(1) It is very likely that ancient astronomers often mixed real sunspots with other
celestial or meteorological phenomena. For example, Wittmann and Xu (1987)
compared more than 20 Oriental sunspot sightings with the data of European as-
tronomers since 1848 and they found that only one third of the naked-eye observa-
tions are confirmed by western telescopic records.
(2) Ancient sunspot observations were not systematic and, as a result, are non-
uniform in time. For example, very often sunspots were detected near the day of
new moon. This happened because just in these periods observations were more
frequent and intensive determination of the new moon date had an important
calendar purpose (Wittmann and Xu, 1978). Dating of observations is also not
always quite accurate.
(3) Some climatic effects may be present in different SONE series (Willis,
Easterbrook, and Stephenson, 1980). Nevertheless, SONE data are rather valuable
374 M. G. OGURTSOV ET AL.

for investigation of solar variability over a long time scale and the catalogue of
Wittmann and Xu was intensively used in our work.

2.2. S OLAR ACTIVITY PROXIES

2.2.1. Cosmogenic Isotope 10 Be

The cosmogenic isotope 10 Be is generated in the Earths atmosphere in reactions of
secondary component of galactic cosmic rays (GCR) with the nuclei 14 N and 16 O.
After its generation 10 Be is oxidised to BeO, captured by atmospheric aerosols, re-
moved from atmosphere by precipitation (the atmosphere residence time is
12 years (Beer et al., 1990)) and finally fixed in polar ice and bottom sediments.
The concentration of beryllium in polar ice is directly influenced by the intensity of
GCR. Because high-density magnetic regions of solar wind, which affect scattering
and transport of GCR in the heliosphere, are governed by activity of the Sun, the
intensity of GCR is well controlled by SA. Therefore 10 Be concentration in ice
contains important information about SA in the past. Investigations of past SA
activity based on 10 Be data have been presented in many works (Beer et al., 1990,
1994). It has been shown that records of cosmogenic beryllium clearly report SA.
It should be noted, however, that the concentration of beryllium in ice depends not
only on GCR intensity but on the rate of ice accumulation. This feature determines
the main shortcoming of the 10 Be record its dependence on local climate, in
particular via precipitation rate (see, e.g., Lal, 1987; Monaghan, 1987; Beer et al.,
1988; Damon and Peristykh, 1997). In the present work we used the data on 10 Be
concentration in the South Pole ice. The record was obtained by Bard et al. (1997)
and covers the time interval after A.D. 850.

2.2.2. Cosmogenic Isotope 14 C

The cosmogenic isotope 14 C is produced in the atmosphere by the reaction of
cosmic-ray-induced thermal neutrons with nitrogen 14 N(n, p)14 C. Then 14 C is ox-
idized to CO2 and assimilated into a number of geochemical and geophysical
processes and distributed between the main carbon reservoirs (atmosphere,
biosphere, ocean). Records of radiocarbon concentration in the past are mainly
based on measurements of 14 C concentration in tree rings. The ability of 14 C to re-
flect long-term changes of SA has been established many years ago (see, for exam-
ple, Stuiver and Braziunas, 1989). The basic problems with the use of radiocarbon
data are the following:
(1) As in the case of beryllium, the concentration of 14 C in tree-rings is sensitive
not only to changes of activity of the Sun but also to changes of climate (Damon,
1970; Stuiver and Braziunas, 1993; Peristykh and Damon, 2000).
(2) The carbon exchange system works as a low-pass filter, which attenuates
the short-term variations of the rate of radiocarbon production Q and amplifies
long-term changes. This results in essential deformations of primary variations of
Q (and hence of GCR intensity) obtained from radiocarbon tree-ring records.
 LONG-PERIOD CYCLES OF THE SUNS ACTIVITY 375

A well-known way to overcome the second difficulty is to use not 14 C data but
a value Q, reconstructed by means of a many-reservoir carbon exchange system.
However, the value of the reconstructed radiocarbon production rate substantially
depends on the rate of exchange between different carbon reservoirs (see, for exam-
ple, Stuiver and Quay, 1980). In the present work we therefore used the extended,
decadal 14 C record derived from tree-rings (Stuiver and Pearson, 1993), covering
the time interval back to 7950 BP. This is the one of the longest S.A. proxies
existing today. It should be noted that GCR intensity is controlled by interplan-
etary magnetic field and the sunspot number is not a very precise indicator of
this field. For example, Solanki, Schssler, and Fligge (2000) have shown that
the time-averaged level of interplanetary magnetic flux depends on the Schwabe
(11 year) cycle length, too. So, the concentration of both cosmogenic isotopes
(14 C and 10 Be) in natural archives likely reflects mainly the solar modulation of
heliospheric conditions.

2.2.3. Historical Auroral Observation

Aurora borealis is a result of fluorescence of ionized atmospheric gases, mainly
nitrogen. This ionization is caused by currents in the upper atmosphere, which in
turn are results from the interaction of solar wind with the Earths magnetosphere.
Hence, the frequency of aurorae occurrences is closely connected with the activity
of the Sun. The polar lights are easily observed by the unaided eye and for this
reason we have historical information on auroral activity up to two thousand years
back. Schove (1979) has determined dates of maxima of the 11-year solar cycles
and estimated approximate Wolf numbers at each maximum during more than
2000 years using catalogues of ancient auroral observations and some preliminary
assumptions. Nagovitsyn (1997), in turn, reconstructed Wolf numbers for A.D.
11001700 using the data of Schove and a specially developed nonlinear model of
solar cyclicity. Obviously, the reconstruction made by Nagovitsyn is really a proxy
of auroral activity rather than SA. It should be noted that these data substantially
overestimate SA in periods of its deep minima. For example, Wolf numbers, de-
rived from auroral observations, reach a value of 2030 during the Maunder min-
imum. This is probably connected with the attenuation of a link between sunspot
activity and aurorae in very low activity epochs.
All the proxies, used in our analysis (filtered or averaged), are plotted in Figure 1
together with data on historic minima and maxima of SA established by Eddy
(1976) and Schove (1983). The similarity between long-term variations of all these
SA indicators is quite evident and can be taken as proof for the reliability of the
chosen proxies.

2.3. C LIMATIC PROXIES

Methods of dendrochronology, intensively developed lately, allow reconstructing

4060% of the variance of the temperature signal. In the present work, we used
376 M. G. OGURTSOV ET AL.

Figure 1. (A) Number of sunspot observations made by the naked eye (35 year averaged)).
(B) Band-pass filtered decadal 14 C data. (C) Data on 10 Be at South Pole (relative fluctuation),
wavelet filtered in 64 128 scale band. (D) Historic maxima and minima of solar activity: (1) Late
Roman maximum, (2) Byzantine maximum, (3) Dark Age minimum, (4) Medieval maximum,
(5) Wolf minimum, (6) Late Medieval maximum, (7) Spoerer minimum, (8) Maunder minimum.
(E) Reconstructed Wolf numbers, wavelet filtered in 64 128 scale band.
 LONG-PERIOD CYCLES OF THE SUNS ACTIVITY 377

tree-ring reconstruction of mean annual temperature in the northern hemisphere

(Mann, Bradley, and Hughes, 1999) to show a possible solarclimate link over the
last millennium.

3. Methods

The main aim of the present work was to analyse the spectral content of SA and
its evolution in time. For this reason a complex analysis using both Fourier and
wavelet approaches was performed. The basic properties of Fourier and wavelet
transforms and the advantage of the wavelet one in the analyses of non-stationary
time series have been described in many works (see, for example, Astafieva, 1996;
Torrence and Compo, 1998; Ogurtsov et al., 2001). We note that in the present
work as in the previous one by Ogurtsov et al. (2001) we used the Morlet basis
for the time-spectral analysis and the MHAT basis for filtration. One of the more
important problems of wavelet analysis is the determination of confidence levels
for local and global wavelet spectra. In this work we solved this problem using a
method described in Appendix 1.
In order to analyse the significance of both Fourier and wavelet spectra we must
know the value of factor of red (AR (1)) noise present in the data. These fac-
tors were estimated by means of the following procedure. First, the low-frequency
component (usually with periods more than 50 years) were filtered out of the series
under investigation y(t) and the residual part y res (t) was considered as consisting
mainly of noise. Secondly, the function y res (t) = y res (t 1) was constructed
and was estimated. Numerical experiments, made with artificial data sets, show
that this method allows the determination of the factor AR(1) with uncertainty less
than 20%.
Analysis of SONE data is another serious problem. The quality of information,
contained in Oriental chronicles is not good enough to use them for accurate quan-
titative estimations of SA level. Therefore we converted SONE data into a binary
form: S(t) = 1 if an event (sunspot) was recorded in year t, S(t) = 0 if an event
was not recorded. But statistical analysis of such binary series is not easy, because
the majority of statistical methods cannot be applied in that case. Therefore, we
searched for the characteristic periodicities in SONE data using an approach of
autosimilarity functions, especially carried out for this purpose and described in
Appendix 2.

4. Results and Discussion

In Figure 2 the local Morlet wavelet spectra of GSSN (Figure 2(a)) and Wolf
numbers (Figure 2(b)) are presented. The spectra were calculated for the same
time interval A.D. 17001995 and the 0.99 confidence level (c.l.) was evaluated in
378 M. G. OGURTSOV ET AL.

Figure 2. (A) Local wavelet (Morlet basis) spectrum of GSSN. White domains local wavelet power
< 0.2, black domains local wavelet power > 1.0 (0.99 c.l.), step of local wavelet power 0.8.
(B) Local wavelet (Morlet basis) spectrum of Wolf numbers. White domains local wavelet power
< 0.2, black domains local wavelet power > 1.0 (0.99 c.l.).

assumption about red noise with = 0.3. It is seen from Figure 2 that two signifi-
cant century-scale oscillations really are in the spectrum of sunspot variability, a 90
100-year cycle (since the second part of the 18th century) and 5060-year cycle
(until the first part of the 19th century). The shorter cycle is manifested more clearly
in Wolf numbers. So, both series, Wolf numbers and GSSN, show the bimodal
structure of century-long variation. However, instrumental records are too short to
make more precise conclusions about the character of centennial solar variability
and its time evolution.
A much longer SONE series was analysed using the approach of autosimilarity
function (see Appendix 2). For a more convenient analysis, the series, covering
 LONG-PERIOD CYCLES OF THE SUNS ACTIVITY 379

Figure 3. (A) Binary SONE data. (B) Fourier spectrum of the autosimilarity function of binary SONE
data for 0 600 A.D. (C) Fourier spectrum of the autosimilarity function of binary SONE data for
801 1801 A.D.
380 M. G. OGURTSOV ET AL.

Figure 4. (A) 10 Be concentration in South Pole ice (see Figure 1 in Bard et al. (1997)). (B) Local
wavelet (Morlet basis) spectrum of 10 Be concentration. White domains local wavelet power < 0.2,
black domains local wavelet power >1.0 (0.99 c.l.). (C) Fourier spectrum of 10 Be concentration.
Dotted line: 0.95 c.l. (red noise factor 0.9).
 LONG-PERIOD CYCLES OF THE SUNS ACTIVITY 381

time interval A.D. 01800, was divided into two parts: 0600 A.D. and 8011801
A.D. Fourier spectra of autosimilarity functions of both these data sets are plotted
in Figure 3. Significances of the main peaks in the spectra were estimated using a
statistical experiment, described in Appendix 2. It is seen from Figure 3 that the
significance of the majority of the spectral peaks is low. This is likely a result of
the non-systematic character of the observations and a lot of noise input by ancient
astronomers, who often mixed sunspots with other phenomena, into their record.
Nevertheless, this series contains some information about solar variability in the
Suess and Gleissberg frequency bands. The 250-year cycle is distinctly manifested
in SONE data after AD 800 (c.l. > 2 ) and before AD 800 the 170-year periodicity
is present, though weak. This result is in agreement with that obtained by the Yun-
nan group (1977), who analysed the medieval Chinese sunspot observations and
suggested two cycles in their data the stronger at 240270 years and a weaker one
at 165210 years. Century-scale variability of SONE data is manifested by peaks
at 68, 92, and 126 years, though the more significant of them (68 years) attains
only 1 c.l. Cycles with periods of 61 years and 8191 years were also found
in the Yunnan Observatory (1977). Ancient Oriental sunspot observations thus
confirm the existence of the Suess cycle and give evidence that century-scale solar
variability has a wide frequency band (60130 years). More exhaustive analysis of
SONE data can be found in the work of Nagovitsyn (2001).
A long beryllium record from Antarctica (Bard et al., 1997) (A.D. 10001900,
Figure 4(a)) was analysed using both wavelet and Fourier approaches. The factor of
red noise was estimated as 0.9, and spectra of 10 Be data (local wavelet and Fourier
ones) are shown in Figures 4(b) and 4(c). It is seen from Figure 4 that 200 years,
110135 years and 5070 years variations are present in the beryllium series dur-
ing almost all time intervals. The second oscillation mode is the strongest while the
significances of the other two cycles are not so high.
A reconstruction of Wolf numbers made by Nagovitsyn (1997) using SA es-
timations made by Schove (who, in turn, mainly used auroral data) is shown in
Figure 5(a). Together with the direct Wolf numbers the series covers the time
interval A.D. 11001995. The record was analysed in the same way as the pre-
vious record. Wavelet and Fourier spectra, plotted in Figure 5, show the presence
of highly significant periodicities in the periods of 170220, 90135, and 50
85 years. The Suess cycle is the most significant. The factor of red noise was 0.3.
Finally, we investigated the decadal radiocarbon series of Stuiver and Pearson
(1993). For convenience of analysis, we constructed a series of residuals of the
14 C record from the initial data set filtered in a 801000 year band. This data
set is presented in Figure 6(a). Then, as in the previous cases, Fourier and wavelet
analyses were been performed. The results are shown in the Figures 6(b) and 6(c).
As seen in Figure 6, the radiocarbon spectrum also shows variance in 160270-
year, 80140-year, and 4070-year bands. These cycles are strongest during 2500
6000 B.C. (see Figure 6(b)). It is known that the time interval 25006000 B.C.
was a period of a weak geomagnetic dipole field, which shields the Earth from
382 M. G. OGURTSOV ET AL.

Figure 5. (A) Wolf numbers reconstructed by Nagovitsyn (1997) using data of Schove (1979). After
1700 A.D. direct Zrich data. (B) Local wavelet (Morlet basis) spectrum of Wolf numbers recon-
structed by Nagovitsyn. White domains local wavelet power < 0.2, black domains local wavelet
power >1.0 (0.99 c.l.). (C) Fourier spectrum of Wolf numbers reconstructed by Nagovitsyn. Dotted
line: 0.99 c.l. (red noise factor 0.3).
 LONG-PERIOD CYCLES OF THE SUNS ACTIVITY 383

Figure 6. (A) Residuals of long decadal 14 C record of Stuiver and Pearson (1993). (B) Local
wavelet (Morlet basis) spectrum of the residuals of long decadal 14 C record. White domains local
wavelet power < 0.2, black domains local wavelet power > 1.0 (0.99 c.l.). (D) Fourier spectrum of
the residuals of long decadal 14 C record. Dotted line: 0.95 c.l. (red noise factor 0.9).
384 M. G. OGURTSOV ET AL.

TABLE I
Secular variations in different solar activity indicators.

Gleissberg frequency band Suess frequency band

Data set Time interval 50 80 yrs 90 140 yrs (160 260 yrs)

Wolf number 1700 1995 AD Strongly Strongly ?

manifested manifested

GSSN 1700 1995 AD Distinctly Strongly ?

manifested manifested

SONE 1 0 800 AD Manifested, but Manifested but Manifested but

weakly weakly weakly
significant significant significant

SONE 2 801 1801 AD Absent Manifested, but Distinctly

weakly manifested
significant
10 Be 1000 1900 AD Distinctly Strongly Distinctly
manifested manifested manifested

Wolf number 1100 1995 AD Strongly Strongly Strongly

(reconstructed) manifested manifested manifested

14 C 7748 BC1945 AD Appears Appears Appears

periodically periodically periodically

cosmic rays (see, for example, Dergachev and Veksler, 1991). This means that
this epoch was a period of strong influence by cosmic rays. The more pronounced
14
C variability in the 40270-year range during 25006000 B.C. is a proof for its
solar origin. It can also be noted that periods of amplification of the Suess cycle
( 1600 A.D., 400 B.C., 3100 B.C., 5300 B.C., see Figure 6(b)) coincide
with the maxima of the ca. 2400-year Hallstattzeit cycle that have been found
already by Vasiliev, Dergachev, and Raspopov (1999). The red noise factor for
the long decadal 14 C series was 0.9.
The information extracted from a variety of indicators of SA is summarized in
Table I.
It is seen from Table I that variations with periods 5080 years, 90140 years
and 160260 years are clearly observed in different SA data (both direct and
proxy) including indicators of sunspot and auroral activity and solar modulation
of GCR. It proves the existence of all these modes of the Suns variability over a
time scale of 1000 years and more. Hence, the time variations, mentioned above,
are likely connected with the oscillating system governing the SA fluctuations.
 LONG-PERIOD CYCLES OF THE SUNS ACTIVITY 385

The ratio of periods of the main solar variations (5080 years)/(90140 years)
(90140 years)/(160260 years) (1/2) is very interesting and may reflect fractal
properties of SA. But the quality of our data on SA is not good enough to make any
ultimate conclusion about the nature of the solar oscillator. For example, the data
on Wolf numbers, both directly measured and derived from aurorae observations,
show that 5080 and 90140-year modes rather replace each other and alternate
(see Figures 2 and 5(b)) but in the cosmogenic isotope data these cycles can exist
simultaneously (see Figures 4(b) and 6(b)). So, various indicators of SA give us a
different picture about the time evolution of the main modes of solar variability,
while precise information on this point is very important for understanding of
fractal and other properties of activity of the Sun. Hence, further investigations
of changes of SA over time scales as long as possible, are necessary.
The next question is connected with a possibility of solar-climate link, which
is actively debated. Modern methods of dendrochronology make it possible to
obtain climatic proxies with a length up to 1000 years and more and, hence, to
extract information about climate change over a long time scale. Here we analysed
the annual northern hemisphere temperature reconstruction of Mann, Bradley, and
Hughes (1999), which is presented in Figure 7 together with its Fourier and wavelet
spectra (red noise factor was 0.25). As seen from Figures 7(b) and 7(d) highly
significant time variations with periods 200, 110120, and 5080 years exist in
the temperature proxy of Mann, Bradley, and Hughes (1999). These periodicities
are very similar to those established in SA. Records of global northern hemispheric
temperature and 10 Be in South Pole ice, wavelet filtered in the 56194-year scale
band, are plotted in Figure 7(c). Some negative correlation between these series
(coefficient of correlation 0.51) is seen at least during the last 700 years. It gives us
new evidence of long-term solar activity having an influence on climate. The ques-
tion about the centennial solar modulation of global and regional terrestrial climate
and its possible mechanisms have been discussed in more detail by Ogurtsov et al.
(2001, 2002).

5. Conclusions

Summarising results of our analysis we can conclude that:

(1) Two basic modes of long-term solar variability the cycles of Gleissberg and
Suess really exist at least within the last millennium, and probably during a longer
period (up to 10 000 last years). They are manifested in direct and proxy indicators
of different parameters of solar activity (sunspots, heliospheric solar modulation,
aurorae). It indicates that Gleissberg and Suess cycles are the fundamental features
of SA.
(2) The century-type solar variation the Gleissberg cycle has not a single 80
90-year periodicity (as it was considered till now) but has a wide frequency band
(50140 years) and a complex character. More likely it consists of two oscillation
386 M. G. OGURTSOV ET AL.

Figure 7. (A) Mean annual temperature in the northern hemisphere, reconstructed by Mann et al.
(1999). (B) Local wavelet (Morlet basis) spectrum of the temperature multiproxy of Mann et al.
(1999). White domains local wavelet power < 0.2, black domains local wavelet power > 1.0
(0.99 c.l.). (C) Temperature multiproxy of Mann et al. (1999) and 10 Be concentration at the South
Pole, wavelet filtered (MHAT basis) in 56 194 scale band. (D) Fourier spectrum of the temperature
multiproxy of Mann et al. (1999). Dotted line: 0.999 c.l. (red noise factor 0.25).
 LONG-PERIOD CYCLES OF THE SUNS ACTIVITY 387

modes 5080-years periodicity and 90140-years periodicity. The Suess cycle

is 160260 years and the cycle is more stable and less complex, as Schove (1983)
suggested.
(3) Global northern hemisphere climate has appreciable variability in the Gleiss-
berg and Suess bands at least during the last 1000 years. It confirms an assumption
that climate is modulated by SA during the corresponding time interval.
The work made in this paper using modern statistical methods shows that com-
plex analysis of all variety of direct and indirect information about SA in the past
is a perspective tool for investigation of long-term variability

Acknowledgements

This research was done in the context of an exchange between the Russian and
Finnish Academies (project No. 16) and was supported by INTAS grant No. 2002-
550. Yu. A. Nagovitsyn is grateful to the Federal program Astronomy (15.13)
and INTAS (grant Nos. 2001-543, 2001-752). M. G. Ogurtsov is grateful to S.
Helama for his aid in the preparation of the paper.

Appendix 1. Estimation of Significance of Wavelet Spectra

One of the more important questions of wavelet analyses is the estimation of

significance of details of the calculated spectra. Recently, Torrence and Compo
(1998) suggested a perspective method for the solution of the problem. They rec-
ommended to do this evaluation on the basis of a comparison of normalised local
wavelet spectrum of analysed signal with some kind of background spectrum.
However, the next task arises at once a problem to choose the appropriate back-
ground spectrum. Our analysis showed that the use of a pure noise spectrum as a
background is not the best way and may lead to underestimation of significance
at least in a case when the signal is corrupted by noise. In the present work we
therefore followed the approach of Torrence and Compo (1998) but used a number
of background global wavelet spectra of artificial signals, each of which consisted
of a sinusoid with fixed period and Gaussian red noise and with a fixed significance
X. Here we illustrate our method by means of a test time series of 900 points,
consisting of two sinusoids with the periods 7.8 year and 73.0 year and of Gaussian
red noise with = 0.3 (Figure 8(a)). Both these sinusoids are highly significant, as
is confirmed in the Fourier spectrum (Figure 8(b)). The 0.999 confidence level (c.l.)
in Figure 8(b) was calculated as multiplication of the usual red noise continuum
by a corresponding 2 value (6.4 for 0.999). The X confidence level for the global
wavelet spectrum of signal under investigation was calculated in the following way:
(1) Initially we calculated parameters of the signal Y X (t, k ), which has unit
variance and consists of red (AR(1)) noise and a sinusoid with frequency k , so
388 M. G. OGURTSOV ET AL.

Figure 8. (A) Test signal contains two sinusoids (with periods 73 yr and 8 yr) corrupted by red noise
with = 0.3. (B) Fourier spectrum of the test signal. (C) Local wavelet (Morlet basis) spectrum
of the test signal. White domains local wavelet power < 0.2, black domains local wavelet power
> 1.0 (0.999 c.l.). (D) Global wavelet (Morlet basis) spectrum of the test signal.
 LONG-PERIOD CYCLES OF THE SUNS ACTIVITY 389

that the peak of its Fourier spectrum has a X confidence level. It means that for the
signal

Y X (t, k ) = AX
0 (k ) cos(k t) + R (t),

(1)

where R (t) is a red noise with a factor and variance r2 , AX

0 (k ) is to be cal-
culated so that PN () the normalized Fourier spectrum of Y X (t, k ) at the
frequency k will have the peak PNX (k ) with the confidence level X, i.e.,
2
2,X
PNX (k ) = PR (k ) , (2)
2
where
1 2
PR () =
 is a spectrum of red noise, (3)

1+ 2 2 cos
N0
2
2,X is the value of 22 corresponding to the chosen confidence level X, N0 is
number of points in the data set. It is easy to calculate AX
0 (k ). According to Horne
and Baliunas (1986) for the signal consisting of a sinusoid and normally distributed
random value with the zero mean we have
N0 AX 0 (k )
2
PNX (k ) = . (4)
2 AX
0 ()
2
+ r 2
2
If we work with the signal initially normalized by its variance, i.e.,
2
AX
0 (k )
+ r2 = 1, (5)
2
then

2PNX (k )
AX
0 ( k ) = . (6)
N0
(2) At the next step a number Ns of Monte-Carlo simulations is performed.
In each simulation a signal Y X (t, k ) is generated and a global Morlet wavelet
spectrum of the signal is calculated. These Ns spectra are averaged and a final
mean spectrum P W,X (a, k ) is determined. Apparently the spectrum P W,X (a, k )
has a maximum P max (ak ) at ak = 1/k. In Figures 9(a) and 9(b) the mean global
wavelet spectra, for 3 (a3 = 18 year) and 8 (a3 = 70 year), are shown with thick
lines. The corresponding mean Fourier spectra are shown as thin lines.
(3) After this procedure has been carried out for a number of frequencies 1 ,
2 , . . . , n (signals contain sinusoids with different frequencies) a number of
maxima of global Morlet wavelet spectra P max (a1 ), P max (a2 ), . . . , P max (an ) will
390 M. G. OGURTSOV ET AL.

Figure 9. (A) Mean Fourier (thin line) and Morlet wavelet (thick line) spectra of signal consisting
of 900 points and containing a sinusoid with period 66 years corrupted by red noise with = 0.3.
Significance of 66 yr variation is 0.999. Dotted line: 0.999 c.l. for Fourier spectrum. (B) Mean Fourier
(thin line) and Morlet wavelet (thick line) spectra of signal consisting of 900 points containing a
sinusoid with period 18 years corrupted by red noise with = 0.3. Significance of 18 year variation
is 0.999. Dotted line: 0.999 c.l. for Fourier spectrum. (C) 0.999 confidence level (thick line) for
Morlet wavelet spectrum of a signal consisting of 900 points.
 LONG-PERIOD CYCLES OF THE SUNS ACTIVITY 391

be obtained. Finally, joining the points P max (a1 ), P max (a2 ), . . . , P max (an ) we de-
fined a curve ConX (a) - the X confidence level for the global wavelet spectrum
of the time series consisting of N0 points (see Figure 9). This line was used as a
background spectrum in further analyses of local spectra, i.e., all domains of the
local wavelet spectrum of the analysed signal (consisting of N0 points), which lay
above ConX (a), were considered as true features with the confidence level X. In the
Figures 8(c) and 8(d) local and global wavelet spectra of tested signal are shown.
The confidence level was estimated according to the method described above. The
good agreement between the significance of peaks of Fourier and wavelet spectra
(both global and local) is evident. It should be noted that if we used a pure red noise
spectrum as background the significance of global wavelet spectrum peaks would
be less than 0.99 and the same would be for the local spectrum.

Appendix 2. Function of Autosimilarity and Analysis of Binary Information

For the analysis of binary time series we used an approach, developed by Nagov-
itsyn (1992, 2001) and based on ideas from cluster analyses. Multidimensional
statistics make it possible to classify events proceeding from an intuitively for-
mulated concept of their similarity in a space of features. One of the more popular
measures of similarity of cases i and j , having features k = 1, 2, . . . , is the general
similarity coefficient gij , suggested by Gower (1971):

p
Ui j k
k=1
gij = , (7)
p
Wij k
k=1
where Uij k denotes the contribution provided by the kth variable, Wij k are infor-
mation weight factors. Uij k , Wij k are different for different kinds of data (quantita-
tive, nominal or binary). In case of a binary variable we have:

Case i + +
Case j + +
Uij k 1 0 0 0
Wij k 1 1 1 0

If we are interested in periodicities contained in p-dimension time series, we can

formulate a concept for a function of autosimilarity (AF) a measure of similarity
of a matrix of the data (time features) with itself when shifted by time :
 p 
1 
n 
g( ) = Ui,i+,k /Wi,i+,k . (8)
n + 1 i=0 k=1
392 M. G. OGURTSOV ET AL.

Figure 10. (A) Test series, consisting of a sinusoid with a period of 15 years, corrupted by strong
white noise. Thick grey lines 1.75 and 2.05 levels. (B) Binary series made from the initial one
using 1.75 threshold. (C) Binary series made from the initial one using 2.05 threshold. (D) Fourier
spectrum of the initial series, plotted in (A). (E) Fourier spectrum of the binary series, plotted in (B).
(F) Fourier spectrum of the binary series, plotted in (C).
 LONG-PERIOD CYCLES OF THE SUNS ACTIVITY 393

It is seen that the autosimilarity function g( ) can be considered as a multidimen-

sional analogue of the autocorrelation function. Its first important advantage is its
applicability for analysis of different types of variable, including a binary one. AF
has been applied for analysis of multidimensional data (Nagovitsyn, 1992) and in
present paper we used its one-dimensional version. The second useful property
of the AF is the possibility to use it for the analysis of extreme events. This op-
portunity is illustrated in Figure 10. A time series Y (t), constructed of a sinusoid
with period 15 years and white noise is shown in Figure 10(a). The sinusoid is
so strongly corrupted by noise that its Fourier spectrum (Figure 10(d)) does not
contain any significant peak. Then we constructed a binary series Y B (t) from Y (t)
by a rule
 B
Y (t) = 1, Y (t) > Ythrs,
(9)
Y B(t) = 0, Y (t) < Ythrs.
We made two binary data sets using two thresholds Ythrs a low one (1.75) and a
high one (2.05). These binary series (the first consists of 80 events and the second
of 22 events) are shown in Figures 10(b) and 10(c) and the corresponding Fourier
spectra of their autosimilarity functions are plotted in Figures 10(e) and 10(f). The
significance of the peak of the Fourier spectrum of AF was estimated by means
of a statistical experiment. In this experiment a number of randomly distributed
copies of the analysed data set were constructed (see Bieber et al., 1990) and the
Fourier spectrum of AF of each of these copies was calculated. Then the value
of the strongest peak of the spectrum of each random set was determined and
compared with the peak under investigation. If the main peak of the noisy series
was higher than the analysed peak, the test was considered as having a positive
result. Significance of the peak, of interest to us, was finally estimated as a ratio of
positive results to full number of simulations. It is seen from Figures 10(e) and 10(f)
that analyzing a binary series consisting of extreme events allows one to suppress
noise substantially and to reveal the variation, which cannot be found in the initial
data. The SONE time series, investigated in the present work, is just an example
of an extreme event record. Really, giant sunspots available for observation by the
unaided eye can occur mainly during the periods of high SA. According to Eddy,
Stephenson, and Yau (1989) the Wolf number RZ must exceed 50 in order that so
large a sunspot or sunspot group could appear.

References

Astafieva, N. M.: 1996, UFN 166, 1145 (in Russian).

Bard, E, Raisbeck, G. M., Yiou, F., and Jouzel, J.: 1997, Earth Planetary Sci. Lett. 150, 453.
Beer, J., Siegenthaler, U., Bonani, G., Finkel, R. C., Oeschger, H., Suter, M., and Wlfi, W.: 1988,
Nature 331, 675.
Beer, J., Blinov, A., Bonani, G., Finkel, R. C., Hofman, H. J., Lehmann, B., Oeshger, H., Siggl, A.,
Schwander, J., Staffelbach, T., Stauffer, B., Suter, M., and Wlfli W.: 1990, Nature 347, 164.
394 M. G. OGURTSOV ET AL.

Beer, J., Soos, F., Lukachyk, Ch., Mende, W., Rodrigues, J., Siegenthaler, U., and Stellmacher, R.:
1994, NATO ASI Series 25, 221.
Bieber, J. W., Secker, D., Stanev, T., and Steigmer, G.: 1990, Nature 348, 407.
Chistyakov, V. F.: 1986, Soln. Data No. 6, 78 (in Russian).
Cole, T. W.: 1973, Solar Phys. 30, 103.
Damon, P. E.: 1970, in I. U. Olsson (ed), Radiocarbon Variations and Absolute Chronology, Proc. of
Nobel Symposium 12, 571.
Damon, P. E. and Peristykh, A. N.: 1997, in R. Bostrom et al. (eds.), IAGA 1997 Abstract Book,
Reklam & Katalogtryk, Uppsala, Sweden, 191.
Dergachev, V. A. and Veksler, V. S.: 1991, Application of Radiocarbon Method for the Investigation
of Past Nature, Leningrad, 255 pp. (in Russian).
Eddy, J. A.: 1976, Science 192, 1189.
Eddy, J. A., Stephenson, F. R., and Yau, K. K. C.: 1989, Royal Astron. Soc., Quat. J. 30, 65.
Gleissberg, W.: 1944, Terrest. Mag. 49, 243.
Gower, J. C.: 1971, Biometrics 27, 857.
Horne, J. H. and Baliunas, S. L.: 1986, Astrophys. J. 302, 757.
Hoyt, D. V. and Schatten, K.: 1998, Solar Phys. 179, 189.
Lal, D.: 1987, Geophys. Res. Lett. 14, 785.
Mann, M. E., Bradley, R. S., and Hughes, M. K.: 1999, Geophys. Res. Lett. 26, 759.
Monaghan, M. C.: 1987, Earth Planetary Sci. Lett. 84, 197.
Nagovitsyn, Yu. A.: 1992, Space-Temporal Aspects of Solar Activity, A.F., Ioffe PhTI, St.-Petersburg,
pp. 197 208.
Nagovitsyn, Yu. A.: 1997, Astron. Lett. 23, 742.
Nagovitsyn Yu. A: 2001, Geomagn. Aeron. 41, 711.
Ogurtsov, M. G., Kocharov, G. E., Lindholm, M., Eronen, M., Merilainen, J., and Nagovitsyn, Yu. A.:
2002, Solar Phys. 205, 403.
Ogurtsov, M. G., Kocharov, G. E., Lindholm, M., Eronen, M., Merilainen, J., and Nagovitsyn, Yu. A.:
2002, Holocene, submitted.
Peristykh, A. N. and Damon, P. E.: 2000, Radiocarbon 42, 137.
Schove, D. J.: 1979, Solar Phys. 63, 423.
Schove, D. J.: 1983, Sunspot Cycles, Hutchinson Ross Publ. Co., Stroudsburg, Pennsylvania.
Solanki, S. K., Schssler, M., and Fligge, M.: 2000, Nature 408, 445.
Stuiver, M. and Braziunas, T. F.: 1989, Nature 338, 405.
Stuiver, M. and Braziunas, T. F.: 1993, Holocene 3, 1.
Stuiver, M. and Pearson, P. J.: 1993, Radiocarbon 35, 215.
Stuiver, M. and Quay, P. D.: 1980, Science 207, 11.
Suess, H. E: 1980, Radiocarbon 22, 200.
Torrence, C. and Compo, G. P.: 1998, Bull. Am. Meteorol. Soc. 79, 61.
Vasiliev, S. S., Dergachev, V. A., and Raspopov O. M.: 1999, Geomagn. Aeron. 39, 80 (in Russian).
Waldmeier, M.: 1961, The Sunspot Activity in the Years 1610 1960, Schulthess and Company AG,
Zrich.
Wittmann, A.: 1978, Astron. Astrophys. 66, 93.
Wittmann, A. D. and Xu, Z.: 1987, Astron. Astroph. Suppl. Ser. 70, 83.
Willis, D. M, Easterbrook, M. G., and Stephenson, F. R.: 1980, Nature 287, 617.
Yunnan Observatory: 1977, Chinese Astron. 1, 347.