Volume 113, Issue D12
Climate and Dynamics
Free Access

Impact of variations in solar activity on hydrological decadal patterns in northern Italy

D. Zanchettin

D. Zanchettin

Euro Mediterranean Centre for Climate Change, Venice, Italy

Department of Environmental Sciences, University of Venice, Venice, Italy

Search for more papers by this author
A. Rubino

A. Rubino

Department of Environmental Sciences, University of Venice, Venice, Italy

Search for more papers by this author
P. Traverso

P. Traverso

Department of Environmental Sciences, University of Venice, Venice, Italy

Search for more papers by this author
M. Tomasino

M. Tomasino

Euro Mediterranean Centre for Climate Change, Venice, Italy

Department of Environmental Sciences, University of Venice, Venice, Italy

Search for more papers by this author
First published: 19 June 2008
Citations: 61


[1] Using spectral and statistical analyses of discharges and basin average precipitation rates acquired over the Po River since the early 1800s, we investigate the impact of variations in solar activity on hydrological decadal patterns over northern Italy. Wet and dry periods appear to alternate in accordance with polarized sunspot cycles. Intriguingly, a solar signature on Po River discharges is detected to be highly significant since the late 1800s, before the onset of sunspots hyperactivity established by the middle 1900s. In particular, observed hydrological patterns over northern Italy are significantly correlated, under periods of quiet sunspot activity, with parameters characterizing the Sun's orbital motion, specifically with the time derivative of the solar angular momentum (τ) which is thought to modulate the strength of the solar wind and sunspot dynamics under weak sunspot activity. The North Atlantic Oscillation (NAO) is detected as potential link between the Sun and Po River discharges, since it is significantly correlated with both solar activity and the decadal variability in the north Italian climate. In particular, positive (negative) NAO anomalies, which are associated with comparatively lower (higher) Po River discharges, are assessed to alternatively correlate at decadal timescales either with τ or with the Earth's geomagnetic activity (GA), which closely follows sunspot activity. This changing correlation seems to be regulated by the strength of sunspot activity: under periods of quiet sunspot activity, a weakening of the GA-NAO connection and a reinforcement of the τ-NAO connection is observed. In this sense, the strength of solar activity apparently modulates the connection between the NAO and Po River discharges.

1. Introduction

[2] Although most of the Sun-climate mechanisms are not well understood yet, a host of empirical evidence suggests that solar energy changes alter the Earth's climate significantly [Eddy, 1976; Haigh, 1996; Shindell et al., 1999; Marsh and Svensmark, 2003; Lambert et al., 2004; Georgieva et al., 2005] and that climate sensitivity to solar variations obeys a frequency-dependent transfer function of solar energy, so that the damping effect of the oceanic and atmospheric thermal inertia make the climate more sensitive to slower solar variations [Wigley, 1988; Foukal et al., 2004; Scafetta and West, 2006]. Under a global perspective, the Earth's climate is driven by the net sunlight deposited in the terrestrial atmosphere, which primarily depends on solar irradiance and on the Earth's albedo. As warming/cooling of the atmosphere causes, e.g., changes in the moisture-holding capacity of the atmosphere, which alter the hydrological cycle and the characteristics of precipitation, it is conceivable that some Sun-Earth connections entail dynamical alterations of the general circulation rather than simple energy balance [North, 2004]. This also contributes to explain the fact that the response of the Earth's climate to solar cycles is rather large only in some areas. Under this perspective, regional climatic factors such as changes in precipitation or local temperature and altered weather patterns are determined by jet stream relocation, whose patterns are modified by large-scale climate structures (teleconnections) induced by variations in the absorption of solar energy in the atmosphere and in the ocean's surface layers [Weng, 2005; Perry, 1994, 2007]. Recently, Tinsley and Fangqun [2004] reviewed the forcings by space particle fluxes controlled by solar wind variations and highlighted a series of key processes that may be involved in the dynamical causal chain linking solar activity and Earth's atmospheric dynamics. Roughly, they depicted a connection between the Earth's global electric field, which is modulated by space particle fluxes such as galactic cosmic rays, the microphysics of clouds (in particular the processes controlling the cloud condensation nuclei) and the undergoing precipitation, whose dynamics in turn entails diabatic cooling/heating of the atmosphere and, eventually, changes in the meridional transport of heat and momentum with the associated feedbacks on large-scale tropospheric dynamics.

[3] So, the Sun affects considerably the hydrological cycle through multiscale mechanisms of feedback possibly originating an intricate geographical distribution of multispectral solar-related signals in hydrological parameters. In large river basins, the hydrological inertia generated by land surface processes may cause a remarkable frequency-dependent amplification/damping of precipitation signals into streamflow variability [Milly and Wetherald, 2002]. Therefore, under a scenario of an at least partly externally forced hydrological system, the integrated nature of river discharge can unveil substantial solar-induced pulses even if the connections identified between solar activity and precipitation are tenuous at best.

[4] A natural foreground for assessing the physical connection between solar activity and regional climates is the Po River basin (northern Italy, 75,000 km2). In particular, solar-type signals have been detected over the last two centuries in temperature series of the alpine region [Brunetti, 2003] as well as in Po River discharge series [see, e.g., Tomasino and Dalla Valle, 2000; Landscheidt, 2000; Tomasino et al., 2004]. As stressed above, the hypothesis of a Sun-regional climate connection is supported by an apparent causal chain involving dynamical alterations of the general circulation. In fact, the meteorology of this midlatitude area is dominated by the passage of synoptic weather systems modulated by the poleward/equatorward relocation of the North Atlantic storm tracks, upon which the teleconnection named the North Atlantic Oscillation (NAO) has a marked signature [see, e.g., Cacciamani et al., 1994; Quadrelli et al., 2001; Tomozeiu et al., 2002; Trigo et al., 2002]. The NAO-related oscillation in atmospheric masses over the Atlantic, which is the prominent mode of wintertime variability in the Northern Hemisphere [Marshall et al., 2001], is recognized as being part of a more global pattern extending from the Earth's surface to the stratospheric ozone layer [Appenzeller et al., 2000]. A host of studies [e.g., Boberg and Lundstedt, 2002; Thejll et al., 2003; Palamara and Bryant, 2004; Lukianova and Alekseev, 2004] increasingly is suggesting that enhanced/reduced solar activity can be the precursor of at least some NAO variability by means of interactions between heliospheric parameters like the solar wind/interplanetary magnetic field with the Earth's magnetosphere-ionosphere system. Accepting this causality scenario as a working hypothesis, the primary influence of the NAO on the Po River region could therefore explain the presence of solar-type signals in hydrometeorological parameters. Furthermore, the apparent dependence of the NAO on the Earth's geomagnetic activity (GA), which is characterized by the occurrence of polarity changes addressed to a solar origin [Mursula and Zieger, 2001], could explain the presence of phase reversals in the Sun-Earth connection over areas dominated by the NAO, such as the phase inversions detected by Brunetti [2003] between sunspot cycles and temperature over the alpine region.

[5] The spectral characteristics of NAO and GA data show a common occurrence of a peak at approximately 96 months, i.e., ∼7.8 years [Paluš and Novotná, 2007]. Specifically, this periodic component dominates the decadal variability of the NAO [e.g., Loewe and Kolowski, 1998; Brunetti, 2003; Jevrejeva et al., 2005]. Conversely, it is less discernible in the GA spectrum, whose prominent feature is the ∼11-year periodicity of sunspots [e.g., Vennerstrom and Friis-Christensen, 1996]. The ∼7.8-year periodicity has also been recognized among the secondary features of sunspot variability at interannual-to-decadal scales [see, e.g., Kane, 1999]. It has been addressed as a subharmonic of the 1.3-year periodicity [Krivova and Solanki, 2002], which characterizes the solar rotation rate near the base of the solar convective zone [Howe et al., 2000] as well as the solar wind speed [Mursula and Zieger, 2000]. Interestingly, periodicities around 1.3 years, that have been named “midterm quasi periodicities,” are detected in GA with strong spectral power during high solar activity [Mursula et al., 2003]. Nonetheless, the ∼7.8-year periodicity characterizes also the variation of the absolute value of the time derivative of the orbital angular momentum (L) associated to the inertial motion of the Sun about the center of mass of the solar system [Landscheidt, 1988], which is apparently driven by the configuration of the major planets. A connection between L and sunspot activity was first hypothesized in the 1960s [Jose, 1965]. Nowadays, an increasing host of empirical evidence [e.g., Landscheidt, 1999; Charvàtovà, 2000; Juckett, 2003; Javaraiah, 2005] as well as model studies [Zaqarashvili, 1997; Juckett, 2000] suggests that the solar spin-orbit coupling is a viable proposition. Although the connection between L (or other descriptors of the solar motion) with the internal dynamo processes, and consequentially with sunspot activity, is still controversial in some aspects [de Jager and Versteegh, 2005; Shirley, 2006], solar and climatic investigations based on the solar inertial motion approach are particularly appealing because the deterministic nature of its parameterizations would provide a solid basis for climate forecasting if they are proven to deeply affect, at least, some features of the Sun-Earth connection.

[6] On this basis, the primary goal of this study is to evidence statistically significant correlations between solar signals and climatically relevant phenomena influencing the north Italian climate as well as to reassess the correlations between the hydrological variability in the Po River basin and solar activity over the last two centuries. In particular, it is explored how reversals in the polarity of sunspot cycles (i.e., Hale-type cycles) and variations in the solar torque are possibly manifested in hydrological data of the Po River basin through solar-induced variations in the geomagnetic activity and in the NAO. While it is beyond this scope to discuss in detail the physical mechanisms originating such correlations, this study is intended as a first step toward a deeper understanding of aspects of climatic processes like those related to the solar forcing, driving climatic variability over northern Italy, including the Po River catchment area.

2. Data Set Description

[7] This study makes use of the following monthly data series, spanning various periods since the early 19th century:

[8] 1. The 1807–2006 Po River discharge data (hereinafter discharge or D) at Pontelagoscuro, i.e., at the basin's closing section, are described by Zanchettin et al. [2008].

[9] 2. The 1831–2003 basin average precipitation (hereinafter P) is described by Zanchettin et al. [2008]. Precipitation data were evaluated from local monthly data at three representative stations at low elevations (Milan, Turin and Parma). We are confident that these data can adequately describe the multiyear and decadal precipitation variability dealt with in this study. In fact, Zanchettin et al. [2008] showed the consistence between the 5-year smoothed patterns evaluated from these data and from the secular precipitation grid data by Brunetti et al. [2006].

[10] 3. The 1825–2006 Gibraltar-Iceland NAO index was originally produced by Jones et al. [1997].

[11] 4. The 1868–2003 aa (Kpa) index of geomagnetic activity is available at the International Service of Geomagnetic Indices homepage (http://isgi.cetp.ipsl.fr/).

[12] 5. The angular momentum of the Sun's orbit about the center of mass of the solar system (L) and its time derivative (τ = dL/dt, hereinafter also referred as “solar torque”) were originally produced by Ferenc Varadi (F. Varadi, personal communication, 2007) and recently adopted by Javaraiah [2005]. Varadi determined the L and τ values on a 10-day basis by using the Jet Propulsion Laboratory DE405 ephemeris for the period 1600–2099 [Standish, 1998] and considering the constellation of the four major planets of the solar system (J. Javaraiah, personal communication, 2008). Here, monthly mean values of L and τ are evaluated for the period 1749–2006 from the original time series.

[13] 6. The 1749–2006 sunspot number data were provided by the Solar Influences Data Analysis Center, RWC Belgium, World Data Center for the Sunspot Index, Royal Observatory of Belgium (R. A. M. Van der Linden and the SIDC Team, Online catalogue of the sunspot index, 2006, available at http://www.sidc.be/sunspot-data/). Sunspot cycles (SC) are numbered from the cycle that began in 1755 (cycle 1). The times for sunspot minima and maxima of cycles 8–23 are those reported by Mursula and Ulich [1998]. Additionally, Hale Cycles (HC) are obtained by associating successive sunspots cycles to alternating positive and negative polarity, according to the Hale's mechanism of polarity reversal [Hale and Nicholson, 1938].

3. Methodologies

[14] The investigation is focused on two aspects: (1) the detection of high-energy solar-type periodicities in climatological and hydrological variables by studying the frequency representation of each variable and (2) the assessment of linear correlations between low-frequency (smoothed) patterns of hydrological variables and solar-related parameters.

[15] A frequency representation of a given process x(t) is provided by its power spectrum Φx(ω), which is defined as the Fourier transform of the autocorrelation function of x(t). Significance levels for the power spectrum Φx(ω) are determined by comparing Φx(ω) with the background spectrum of an appropriate univariate lag-1 autoregressive process of the type:
equation image
where ρ is the lag-1 autocorrelation of the process (ρ = 0 is a white noise process) and ɛt is a random variable. The discrete Fourier transform of yt, after normalization, is:
equation image
where ω = 1,…, N/2 is the frequency index and N the number of points in the time series. Peaks in Φx(ω) that are significantly above Φy(ω) are assumed to be a true feature of the process. Significance (set at the 5% level in this study) is calculated by multiplying the background spectrum by the 95th percentile value for a chi-square distribution χ2. Before the evaluation of the Fourier confidence spectrum, it is verified that x(t) is normally distributed.

[16] A time-frequency representation of a given process is provided by its wavelet spectrum: wavelet analysis (some background of which is provided in the paper by Torrence and Compo [1998]) allows to get information on both the amplitude and the phase of any “periodic” signal in a time series, and how this amplitude and phase vary with time. The mother wavelet function adopted is the Morlet, with nondimensional frequency ω0 = 6 to satisfy the admissibility condition [see Torrence and Compo, 1998, Figure 2a]. Similarly to the Fourier spectrum, significance levels are established by comparing the wavelet spectrum of the variable under study with background theoretical spectra of a red-noise (univariate lag-1 autoregressive) process.

[17] Concerning the correlations between smoothed (hence autocorrelated) hydrological variables and solar-related parameters, the statistical significance of the correlation statistics (r) must account for the effect of serial correlation, which unavoidably affects the interpretation in a too optimistic sense. The significance of r is therefore tested by evaluating the likelihood of a nonrandom occurrence of the result. This nonparametric estimate is gained by generating an adequate number of surrogate Fourier realizations of one series (i.e., random time series although suitably similar to the series in question) and then by investigating the distribution of the set of r values obtained by correlating each surrogate series with the Fourier realization of the other series. Suitable surrogate data are generated by the Phase Scrambling Fourier Transform (PSFT) bootstrap method formerly adopted by Thejll et al. [2003] and Lukianova and Alekseev [2004]. PSFT preserves the amplitudes of the Fourier spectrum of the original series while replacing the phase values with values drawn randomly from a uniform distribution between 0 and 2π. The phase-randomized sequences have the same linear temporal correlation as their original sequences but lack any phase-dependent (or deterministic) structures that may be present in the original sequences.

4. Results and Discussion

4.1. Solar Activity: Sunspots and Solar Inertial Motion

[18] The spectral power of τ and HC depicted in Figure 1 outlines the existence of a multiscale relationship between sunspot activity and the solar inertial motion. Two periodicities, at least, match with powerful signals in both spectra: at ∼7.2 years (τ: 7.37; HC: 7.17) and at ∼20 years (τ: 19.8; HC: 21.5). The τ spectrum shows a further peak at 12.9 years (∼13), which tests in the HC spectrum as not significant against a random occurrence. The 7.17-year periodicity of HC seems reasonably interpretable as a higher harmonics of the fundamental 21.5-year periodicity. Concerning τ, the 7.37-year periodicity seems hardly addressable as resonant to the 19.8-year periodicity, the 3rd harmonics of which is about 6.6 years. Concerning high-frequency bands, the spectral features of τ unveil a strong quasi-annual periodicity (specifically significant peaks are observed at 0.92–1.09 years) and periodicities in the band of the lowest “midterm periodicities,” specifically at 1.61, 1.68 and 1.86 years.

Details are in the caption following the image
Spectral density of monthly time series of Hale cycles (HC) and solar torque (τ) over the period 1749–2006 with significance levels at p = 0.05 (thick dotted line) and at p = 0.10 (thin dotted line). Arrows mark strong periodicities at ∼7, ∼13, and ∼20 years.

[19] Figure 2 further supports the hypothesis of an intermodulation between solar inertial motion and sunspot activity which varies in time. Specifically, Figure 2 unveils a remarkable synchrony in the occurrence of extrema in τ and HC patterns over SC 13–18 (or either HC 7–9), i.e., over the period 1890–1955 (marked by the vertical dotted lines in Figure 2 (top)). This qualitative relationship is corroborated by the remarkable τ-HC linear correlation assessed in this period: r = 0.758 (Ncases = 792). A PSFT analysis over 10,000 surrogate Fourier series of HC data indicates that the chance associated to r ≥ 0.758 over a 65-year window is about 140/10,000, hence the τ–HC correlation above tests as significant at p = 0.014. Notably, SC 12 matches the onset of a period of prolonged quietness in sunspot activity (i.e., a series of cycles with low sunspot number). Conversely, SC 18 matches the beginning of the period of unusually high sunspot activity which has been reported to be unprecedented for the last thousand years, at least, of Earth's history [Usoskin et al., 2003; Solanki et al., 2004]. Moreover, Lundstedt et al. [2005] individuated a dramatic increase in the average variance in the 8–16 years band of the wavelet spectrum of sunspot number data during the 1940s, with variance peaking in the early 1950s. No τ–HC correlation is detected in the following period (r1956–2006 = 0.121, Ncases = 612, pPSFT = 0.830) and in the immediate preceding period encompassing SC 8–12 (r1834–1889 = 0.302, Ncases = 792, pPSFT = 0.493). So, this relationship leads to the hypothesis that variations in the solar angular momentum influence significantly sunspot dynamics under weak sunspot activity, this influence being manifested with a gap of one SC since the sunspot activity has weakened.

Details are in the caption following the image
(top) Monthly sunspot number polarized accordingly to Hale Cycles (HC) and solar torque (τ) pattern for the period 1749–2006. Numbers and numbers in brackets near HC extrema indicate the number of the sunspot cycle and of the Hale cycle, respectively. Arrows indicate perturbations in τ cycles. (bottom) Cosine of the difference of phase (ΔΦ) between the wavelet power of τ and HC at periods of 86 months (∼7.2 years) and 238 months (∼20 years). The Morlet (ω0 = 6) is used as mother wavelet.

[20] This seems only an approximate rule, since the τ-HC correlation is missing around the Dalton Minimum (DM), which is the period encompassing the low-intensity sunspot cycles observed between 1790s and 1820s. The τ-HC correlation for the period 1795–1833 (i.e., SC 5–7) is low (r = −0.214, Ncases = 468) and lacks statistical significance (pPSFT = 0.661). Nonetheless, some facts suggest that this missing connection does not contradict the above hypothesis. A retrograde phase of L, which is a steep drop of L from its usual positive state to a weakly negative state [Jose, 1965; Landscheidt, 1999], falls between SC 5 and 6, i.e., within the DM. Retrograde phases of L are also associated to epochs of steep drops both in the distance of the Sun's center from the barycenter of the solar system and in the solar orbital velocity [Javaraiah, 2005]. So, when L is retrograde it is reasonable to expect that the transfer of momentum from the orbital motion to the rotation is limited, hence that the dL/dt signature on sunspot dynamics is missing. Moreover, the time-varying phase relationship between τ and HC, which is evaluated at periods of ∼7.2 years (88 months) and ∼20 years (250 months) as the difference of phase (ΔΦ) between the respective wavelet power at the given timescale (Figure 2), unveils remarkable differences in the multiscale τ-HC correlation during SC 5–7 and during SC 13–18. Specifically, the ΔΦ curve for the ∼20-year periodicity shows a sinusoidal pattern which implies a cophase relationship (here assumed as: cos(ΔΦ) > 0.8) between 1895 and 1945 and an approximate antiphase relationship (assumed as: cos(ΔΦ) < −0.8) between 1815 and 1845. This pattern is plausibly the result of the increasing lag originated by the difference (∼1.7 years) between the fundamental ∼20-year periodicities of τ and HC. The ΔΦ curve for the ∼7.2-year periodicity is characterized by shifts/discontinuities that seem related with the phase reversals (perturbations) that quasiperiodically disturb the sinusoidal course of τ at intervals with a mean length of 36 years (arrows in Figure 2). Changes in the solar spin-orbit triggered by such perturbations could be responsible of variations in the dynamics of the Sun's convection zone, hence in the Sun's time varying radiant flux and correlated climatic events recorded on Earth [e.g., Landscheidt, 2000]. In particular, it seems an intriguing observation that the τ perturbation around 1899 preceded the anomalously high ratios of the sunspots' umbral and penumbral areas in 1900–1902 [Hoyt, 1979]. Similarly, it is intriguing that the τ perturbation in the mid 1930s falls within the nearest temporal interval of the secular maximum in the umbral/penumbral ratio observed in 1932 [Hoyt, 1979]. According to the ΔΦ curve, the τ-HC correlation at the ∼7.2-year band moved around 1790 (when a τ perturbation occurred) from a nearly continuous anticorrelation to a nonstationary evolutionary pattern. A prolonged period of rough overall correlation started in 1870s, i.e., after the strong τ perturbation around 1866, and lasted until 1930s, when another perturbation affected the τ pattern (the weak τ perturbation around 1899 could be the cause of the phase lag observed during the 1910s). So, the strong manifestation of τ on HC between 1890 and 1955 may be interpreted as resulting from a multiscale cophase between fundamental modes of τ and HC. Conversely, the missing connection during the DM could be addressed as consequence of the lack of a coherent multiscale phase relationship.

4.2. Po River Hydrology: Evidence of a Strong Sun-Earth Correlation

[21] Figure 3 illustrates the Fourier power spectrum and the Wavelet spectrum of D and P for the period 1831–2003. In order to produce a spectrum cleaned from the strong annual peak, prior to the evaluation of the spectra, the 12-month periodicity was removed from data (deseasoning) by applying the additive seasonal decomposition algorithm [Makridakis and Wheelwright, 1989]. The Fourier spectrum of D exhibits a distinct red noise character, which is seen as a gradual decrease in power with increasing frequency. The most prominent peaks with significant power are detected around 1.71 years (or ∼1.7 years), at 3.2, at 8.2 (or ∼8), at 12.9 (∼13) and at 19.2 (∼20) years. Consistently, the Fourier spectrum of P, which approximates a uniform “white noise” spectrum, shows peaks at ∼1.7, 3.2 and 19.2 years exceeding the significance level at p = 0.05. Notably, the power associated to such peaks shows a dampening in the D spectrum with respect to the P spectrum. This reshaping of the spectral energy reflects the global effect of hydrological processes linking rainfalls to discharge (such as snowpack/glaciers dynamics and basin storage) as well as of anthropic disturbances superimposed on natural fluctuations. The overall consistence between the spectral features of P and D indicates that Po River discharges essentially preserve the fundamental modes of precipitation variability upon unexplained variations. Furthermore, the well-identifiable presence of solar-type periodicities (especially at ∼8, ∼13 and ∼20 years) in the D spectrum suggests that the Sun may be one of the precursors to hydrological processes in the Po River basin, although the ∼11-year periodicity of SC is apparently not detected.

Details are in the caption following the image
(left) Power density (Fourier) spectrum of deseasoned monthly Po River discharge (D) and precipitation (P) for the period 1831–2003 with significance levels at p = 0.05. (right) Wavelet power spectrum of D and P. The Morlet (ω0 = 6) is used as mother wavelet. Contours localize the regions in the time-frequency domain where the wavelet signal is significant at p = 0.05 for a red-noise AR(1) process with lag-1 of 0.54 (D) and 0.03 (P). The shaded region is the cone of influence, where edge effects occur. Arrows indicate solar-type periodicities.

[22] The wavelet spectrum of P and D (Figure 3, right) unveils how the solar-type periodicities are manifested in the time-frequency domain. The most prominent feature of the wavelet spectra is the large area associated to a significant manifestation of the ∼20-year periodicity, which appears as the dominant low-frequency mode in both P and D spectra since 1920. Its synchrony with the intensification of sunspot activity (see Figure 2) seems more than a clue that these oscillations have a solar origin. Further significant signals in D data are manifested at ∼8 years and at ∼13 years: the first periodicity is detected intermittently since the middle 1880s; the latter is observed between the middle 1880s and 1920, when the region of significance moves toward a band of higher periodicities (∼11-year). Under the hypothesis that these decadal τ-type oscillations have a solar origin, their temporal domain would coherently reflect the above observation that variations in the solar orbital momentum are among the major sources of variations in solar activity under periods of weak sunspot activity.

[23] Figure 4 further unveils the time varying relationship between HC and the low-frequency patterns of P and D. The smoothed hydrological data are obtained by averaging the original monthly data of P and D over an ∼8-year moving window by using Gaussian weights (using a ∼13-year moving window would have not altered Figure 4 substantially). Recurrent peaks and troughs are assessed to alternate regularly in P and D patterns according to a typical interval of ∼20 years. These fluctuations, which are discernible since the late 1800s although most clearly visible since 1920, are synchronic with the alternating polarity of sunspots since HC 7. Before HC 7, both the HC-P and the HC-D connection are missing, although the synchrony of sunspot maxima in odd-numbered sunspot cycles with comparatively drier conditions could be suggestive of a phase reversal before HC 6. These qualitative observations are corroborated by the correlation statistics detailed in Table 1 (since P and D data are time-averaged, years affected by border effect are not considered for the evaluation of r). A 24-month delay on smoothed data of P and D (not incorporated in the patterns of Figure 4) was empirically proven to produce the highest correlations with HC data. In particular, since HC 7 (i.e., period 1890–2002) the HC-P and HC-D correlations test as significant after consideration for serial correlation, which ascertains that the observed shaping of D and P patterns by means of HC since the late 1800s is hardly addressable as spurious. Nonetheless, a question arises from the fact that a 24-month delay on P and D data produces the highest HC-P and HC-D correlations: this empirical observation has still to be physically justified. Furthermore, the precipitation to runoff relation seems stronger after 1890, which could also indicate possible inaccuracies inherent in the time series reconstruction, particularly in early data. Finally, the significance analyses beg the question of the role of anthropogenic forcing as a mean of intensifying/reorganizing the relationship between solar activity and the Earth's climate in the recent decades. Still, the following analyses contribute to reinforce the correlation picture outlined by the HC-P and HC-D statistics above.

Details are in the caption following the image
Moving average (∼8-year centered window with Gaussian weights: μ = 0, σ = 47 months) of monthly precipitation (P) and discharge (D) compared with the Hale cycles (HC). Stars mark the occurrence of maxima in the solar torque (τ), and circles mark the occurrence of perturbations in τ cycles.
Table 1. Pearson's Coefficient for Linear Correlation Between Polarized Sunspot Cycles (Hale Cycles or HC) and Precipitation (P) or Po River Discharge (D) Since/Until 1890a
Before 1890 (Cases): 1811–1889 (960) Since 1890 (Cases): 1890–2002 (1356)
HC-D (8) 0.114 (0.727) 0.510 (0.042)
HC-D (13) 0.116 (0.734) 0.535 (0.042)
Before 1890 (Cases): 1835–1889 (936) Since 1890 (Cases): 1890–2002 (1320)
HC-P (8) −0.414 (0.909) 0.564 (0.030)
HC-P (13) −0.009 (0.984) 0.531 (0.027)
  • a P and D data are monthly data smoothed according to moving centered windows with Gaussian weights and amplitudes of ∼8 years and ∼13 years. A 24-month delay was empirically assessed to produce the strongest correlations and is therefore imposed to P and D data. Significance (p values in brackets) is calculated accounting for serial correlation in P or D series (10,000 surrogate data generated by the PSFT algorithm).

[24] The correlation statistics in Table 2 summarize the connection between τ and the smoothed data of P and D. Some degree of interrelation is indeed observed since 1890, especially as long as the ∼13-year patterns of P and D are considered. A closer analysis unveils that the strongest τ-P and τ-D correlations are detected between 1910 and 1956, i.e., mostly under weak sunspot activity. As stressed above, this observation further supports the hypothesis that τ is among the strongest contributors to solar-related variability under weak sunspot activity. Interestingly, Figure 4 unveils a qualitative relationship between τ and P or D: maxima and perturbations in τ cycles apparently occur within a small temporal interval from extrema in P and D. Specifically, τ maxima are generally coeval with peaks in P and D patterns, with the exception of the maxima around 1896 and 1955, while τ perturbations mark troughs in P and D patterns, with the exception of the perturbations around 1900 and 1935. Concerning τ perturbations, the above exceptions have a plausible explanation since these perturbations are very close to a maximum, which counteracted the expected inversion of the τ pattern (see Figure 2). Nonetheless, these reversals may be also connected to reversals in the Sun-Earth connection which anyway depend on the strength of solar activity, for instance the phase reversal observed in the semiannual alternating structure of GA between 1870s and 1930s [Mursula and Zieger, 2001]. Notably, this phase reversal, which was addressed as a consequence of the Earth's annual passage through an asymmetric solar wind speed distribution rather than a consequence of solar internal variations [Mursula and Zieger, 2001], approximates the inversions in the correlation between the Earth's temperature and sunspot activity observed by Georgieva et al. [2005]. These interesting coincidences, coupled with the GA-NAO-Po River correlations outlined above, encourage to investigate the τ-GA relationship for assessing the role of GA and of the NAO in this apparently changing Sun-Earth relationship.

Table 2. Pearson's Coefficient for Linear Correlation Between Solar Torque (τ) and Precipitation (P) or Po River Discharge (D) Since/Until 1890 and Between 1910 and 1955a
Before 1890 (Cases): 1811–1889 (960) Since 1890 (Cases): 1890–2002 (1356) 1910–1955 (552)
τ-D (8) 0.219 (>0.5) 0.357 (0.136) 0.507 (0.149); 0.587 (0.089)
τ-D (13) 0.173 (>0.5) 0.414 (0.032) 0.539 (0.113); 0.726 (0.024)
Before 1890 (Cases): 1835–1889 (936) Since 1890 (Cases): 1890–2002 (1320) 1910–1955 (552)
τ-P (8) 0.173 (>0.5) 0.379 (0.156) 0.631 (0.079)
τ-P (13) 0.113 (>0.5) 0.373 (0.086) 0.703 (0.019)
  • a P and D data are monthly data smoothed according to moving centered windows with Gaussian weights and amplitudes of ∼8 years and ∼13 years. If a trend component is detected in a subperiod, then the correlation between detrended data is also evaluated and reported in italics. Significance (p values in brackets) is calculated accounting for serial correlation in P or D series (10,000 surrogate data generated by the PSFT algorithm).

4.3. Solar-Modulated NAO: A Key Link Between the Sun and the Po River?

[25] Figure 5 illustrates the density spectra of the monthly deseasoned series (the 12-month periodicity was preliminarily removed) of the geomagnetic aa index and of the NAO index evaluated over their whole time domain (periods 1868–2003 and 1825–2006, respectively). The shape of the aa spectrum is dominated by the peak at ∼11 years, but still two peaks at 7.8 years and at 21.5 years are discernible and test as significant against a random occurrence. The working hypothesis is that the 7.8-year peak in the aa spectrum represents a resonant oscillatory mode of the typical 1.3-year periodicity of the solar dynamo strength, of which it is the 6th subharmonics (the power of the 1.3-year peak is low since midterm periodicities in GA follow the overall solar activity). The 7.8-year peak dominates the NAO spectrum, which shows a further peak at 13.3 (∼13) years with significance at p < 0.05. The spectral power at 7.8 years shows a considerable enhancement in the NAO with respect to the aa index, which suggests that even if the associated modes are indeed connected, this connection is only partial. Since the NAO and τ share the presence of dominant periodicities around ∼7–8 years and ∼13 years, it seems a reasonable hypothesis that τ is a key for assessing a robust connection between solar activity and the NAO.

Details are in the caption following the image
(top) Power density spectrum of deseasoned monthly aa index (GA) for the period 1868–2003 and (bottom) NAO index for the period 1825–2006 with significance levels at p = 0.05. Triangles indicate solar-type periodicities.

[26] Figure 6 presents the ∼8-year patterns of the aa index (previously linearly detrended) and of the NAO to further enlighten their time-varying connection. The detrend of the aa index was operated through subtracting the linear component y = 0.093·t–179.8 from the original monthly series (t is the time in years plus monthly fractions of the year). Apparently, the two curves are in good synchrony before 1910 and since 1950s. Conversely, between 1910 and 1950s the two patterns seem to evolve gradually toward an inverse peak-trough connection, which is manifested during 1930s and 1940s. The hypothesis of an inversion in the aa-NAO phase relationship during 1950s seems corroborated by apparent coeval variations in ΔΦ evaluated for periodicities at ∼8 years and at ∼11 years (see Figure 6, bottom). Specifically, ΔΦ curves indicate that GA and NAO fluctuations have been in a multiscale cophase since the early 1950s. Although in the preceding period the phase relationship seems less clear, GA and NAO fluctuations seem to be roughly in cophase until 1910 and in antiphase during 1930s and 1940s. It is worth noting that the apparent inversion in the early 1950s approximate the observed weakening of the τ-HC connection since SC 19. The correlation statistics detailed in Table 3 summarize the connections between aa, NAO and τ until/since 1956 (since aa and NAO data are time averaged, years affected by border effect are discarded). The aa index is not correlated with τ throughout the whole period, which indicates that τ and GA are linearly independent climate predictors. So, the NAO possibly explains how both τ and GA are related with north Italian hydrological variability: the aa-NAO connection is excellent since 1956, while it is lacking before 1956; conversely, the NAO-τ connection is strong before 1956 and is missing after 1956. Overall, these results point to the hypotheses that the GA-NAO connection and the τ–NAO connection (if any) are mutually exclusive and that this mutual exclusion is regulated by the strength of sunspot activity, so that under periods of quiet sunspot activity a weakening of the GA-NAO correlation as well as a reinforcement of the τ-NAO correlation should be observed. This hypothesis makes sense as long as it is considered that GA is strongly dependent on sunspot activity and it is assumed that τ influences the NAO not only as a precursor of sunspot dynamics, hence of GA variability. This hypothesis seems further supported by the correlation between τ and the ∼13-year pattern of the NAO (evaluated as averages over a moving window with Gaussian weights with parameters μ = 0 and σ = 77 months): since 1956 the correlation is poor (r1956–2002 = 0.053, pPSFT = 0.818, Ncases = 564), before 1956 the correlation tests as statistically significant (r1872–1955 = 0.321, pPSFT = 0.044, Ncases = 1524).

Details are in the caption following the image
(top) Moving average (∼8-year centered window with Gaussian weights: μ = 0, σ = 47 months) of monthly NAO index (1825–2006) and detrended aa index (1868–2003). Points affected by border effect were removed. (bottom) Cosine of the difference of phase (ΔΦ) between the wavelet power of NAO and GA at periods of 95 months (∼8 years) and 132 months (∼11 year). The arrow marks an apparent phase reversal.
Table 3. Pearson's Coefficient for Linear Correlation Between Solar Torque (τ), NAO, and aa Index During Subperiods Since the 19th Centurya
Before 1956 (Cases): 1872–1955 (1008) Since 1956 (Cases): 1956–1999 (516)
τ-aa −0.082 (0.742) 0.083 (0.836)
(detrended) 0.125 (0.773)
Before 1956 (Cases): 1832–1955 (1488) Since 1956 (Cases): 1956–2002 (564)
τ-NAO −0.445 (0.008) 0.052 (0.871)
Before 1956 (Cases) Since 1956 (Cases): 1956–1999 (516)
1872–1909 (456) 1910–1955 (554)
aa-NAO −0.059 (0.893) −0.309 (0.335) 0.755 (0.006)
(detrended) 0.502 (0.200) −0.199 (0.570) 0.830 (∼0)
  • a NAO and aa data are smoothed monthly data according to a moving centered window with Gaussian weights and amplitude of ∼8 years; the 1868–2003 linear trend component is removed from aa data. If a trend component is detected in a subperiod, then the correlation between detrended data is also reported. Significance (p values in brackets) is calculated accounting for serial correlation in NAO or aa series (10,000 surrogate data generated by the PSFT algorithm).

[27] At this stage, the key issue is proving that this apparent time-varying Sun-NAO connection concurs to explain the Sun-Po River connection detailed above. Figure 7 compares the ∼8-year patterns of D and of the NAO that were originally presented in Figures 4 and 6, respectively. Figure7 enlightens a rough long-term NAO-D anticorrelation: strong positive (weak positive or negative) anomalies of the NAO generally result in lower (higher) discharges. The connection between the two patterns is strengthened when a delay of 24 months is imposed to D, for which correlation tests as highly significant after consideration for serial correlation (r1829–2002 = −0.338, Ncases = 2040 pPSFT = 0.018). However, this NAO-D relationship does not seem a completely robust feature of discharge variability throughout the nineteenth and twentieth centuries, since incongruence is detected during 1840s–1860s, around 1910 and between the late 1950s and 1980s. Changes in the NAO-D phase relationship evaluated at ∼8 and ∼13 years (Figure 7, bottom) confirm that the NAO-D connection is in fact time varying. Interestingly, the ΔΦ curve at ∼8 years indicates a rough NAO-D cophase until the middle 1940s, when a sudden phase reversal occurred, while the ΔΦ curve at ∼13 years indicates a strong NAO-D antiphase between the middle 1960s and 1930. So, the two modes in concern appear to be characterized by NAO-D phase relationships of opposite sign, at least since 1860s.

Details are in the caption following the image
(top) Moving average (∼8-year centered window with Gaussian weights: μ = 0, σ = 47 months) of monthly Po River discharge (D, 1807–2006) and NAO index (1825–2006). Points affected by border effect were removed and the NAO axis is reversed to facilitate comparison. (bottom) Cosine of the difference of phase (ΔΦ) between the wavelet power spectra of NAO and D at periods of 95 months (∼8 years) and 155 months (∼13 year).

[28] The close gap between the apparent change in the NAO-D connection around 1945 and the intensification of solar activity corroborates the hypothesis that the strength of solar activity indeed modulates the NAO-D connection. Considering the characteristic temporal intervals assessed above, the NAO-D correlation at ∼8-year is well detectable under the period dominated by low solar activity (r1890–1955 = −0.439, pPSFT = 0.134; detrended data: r = −0.550, pPSFT = 0.035). So, in this period solar variations appear to be eventually manifested in regional hydrological processes through their effect on large-scale atmospheric dynamics. Conversely, the NAO-D correlation is weak during periods encompassing strong solar activity (r1956–2002 = −0.285, pPSFT = 0.370; r1834–1889 = −0.353, pPSFT = 0.228; detrended data between 1834 and 1889: r = −0.416, pPSFT = 0.177). Interestingly, in periods of solar hyperactivity the ∼8-year pattern of the NAO closely follows GA (hence sunspot cycles), while the ∼8-year patterns of P and D apparently follow Hale cycles (hence polarized sunspot cycles). Some clarification to this incongruence may be provided by the semiannual oscillations of inverse polarity observed in GA in concert with Hale cycles [Mursula and Zieger, 2001]. In fact, perturbations in the Northern Hemisphere grow to their largest magnitude in autumn, so that the autumnal precipitation guarantee the biggest fraction to the annual total inflow for riverine systems. Therefore, the 22-year alternation of enhanced/attenuated GA in autumn could indeed result in changes on the observed decadal behavior of Po River discharges.

4.4. Further Discussion

[29] The results stressed above suggest that solar activity puts identifiable constrains to the regional hydrological cycle in the Po River basin, which are essentially manifested as an alternation of dry and wet periods every ∼20 years. This Sun-regional climate correlations further indicate that variations in the solar magnetic activity may represent the key to unlock the relationships between solar variability and historical records of Earth's climate [e.g., Eddy, 1976]. The apparent time-varying interplay between solar inertial motion/solar magnetic activity an the response of Earth's geomagnetic activity, large-scale atmospheric structures and the north Italian climate may have profound implications for climatology if similar connections are assessed in other Earth's regions. For instance, Alexander [2005] detected a linkage with solar activity for the 21-year periodicity in the alternating wet and dry sequences of rainfalls and river flows in southern Africa. Furthermore, cycles of solar magnetic activity could provide a further explanation to the apparent multidecadal alternation of wet and dry phases observed in the Paranà River basin (South America) identified by Mauas and Flamenco [2005]. In the case of Northern Hemisphere's climates, the hypothesis of a solar connection at the ∼20-year timescale via the variability of large pressure systems (as the NAO for the Po River basin) is further supported by the identification, in the band of Hale-type periodicities, of a strong solar radiative forcing on the sea level pressure fields over the Arctic and Europe [Lohmann et al., 2004], in particular those reflecting the annular mode/Arctic Oscillation described by Thompson and Wallace [1998] which resembles the NAO pattern.

[30] This interplay also opens a perspective on evolutionary terms over decadal or even shorter timescales, since decadal patterns constitute the background upon which events on progressively shorter timescales are embedded in and constrained by. A number of predictions of the intensity of solar activity during the next sunspot cycles (namely the 24th and the 25th) suggest that an abnormally low activity could take place in the next decades [see Badalyan et al., 2001; Schatten and Tobiska, 2003; Landscheidt, 2003]. The observed sequence of alternating wet and dry periods associated to Hale Cycles, coupled with these indications of the next sunspot cycles being less intense (but longer) than the most recent ones, suggests that the hypothetical dry (wet) period associated to SC 24 (SC 25) is comparatively weaker but longer. Nonetheless, a weakening of sunspot activity could also lead to a loss of HC-type signals in north Italian precipitation and Po River discharges. In fact, the severe droughts of the Po River in summers 2003, 2005 and 2006 seem to indicate the onset of a very strong dry period. The sharpening of hydrological extremes may provide evidence for the climatic effect in the region of the anthropic increase of greenhouse gas concentrations. This hypothesis makes sense, since such increase is expected to progressively strengthen the NAO [e.g., Krichak and Alpert, 2005], hence progressively reduce precipitation over northern Italy.

5. Conclusions

[31] The modulation of regional precipitation and Po River discharge documented in this study indicates that variations in solar activity do appear to be correlated with hydrological decadal patterns for northern Italy, which generally consists of wet and dry periods alternating in accordance with polarized sunspot cycles (Hale cycles). Interestingly, the solar signature on Po River discharges appears to have been highly significant since the late 1800s (specifically since 1890), i.e., well before the onset of sunspots hyperactivity in the middle 1900s. In fact, north Italian hydrological patterns are significantly correlated under weak sunspot activity (specifically between 1910 and 1955) with parameters of the Sun's orbital motion, specifically with the time derivative of the solar angular momentum (τ) which is thought to modulate the strength of the solar wind as well as sunspot dynamics under weak sunspot activity. The hypothesis that the mechanism of solar forcing involves both the magnetic activity of sunspots and the Sun's orbital dynamics seems corroborated by the detection of consistent solar signals in the NAO, which is assessed to significantly correlate with both the Sun and the decadal variability of the north Italian climate. In particular, positive (negative) NAO anomalies are associated to comparatively lower (higher) Po River discharges. At decadal timescales, periods when the NAO is correlated with τ apparently alternate with periods when the NAO is correlated with the Earth's geomagnetic activity (GA), which closely follows sunspot activity. This behavior seems to be regulated by the strength of sunspot activity, so that under periods of quiet sunspot activity a weakening of the GA-NAO connection as well as a reinforcement of the τ-NAO correlation is observed. Consistently, the strength of solar activity apparently modulates the connection between the NAO and Po River discharge. This latter hypothesis seems supported by the synchrony between the intensification/attenuation of solar activity and apparent changes in the connection between the NAO and Po River discharge detected around 1890 and 1955.


[32] This study was funded by the Euro-Mediterranean Centre for Climate Change (CMCC), Venice (Italy). The authors thank the two engaged anonymous referees for pointing out criticisms and providing constructive suggestions.