9,400 years of cosmic radiation and solar activity from ice cores and tree rings

Abstract

Understanding the temporal variation of cosmic radiation and solar activity during the Holocene is essential for studies of the solar-terrestrial relationship. Cosmic-ray produced radionuclides, such as ¹⁰Be and ¹⁴C which are stored in polar ice cores and tree rings, offer the unique opportunity to reconstruct the history of cosmic radiation and solar activity over many millennia. Although records from different archives basically agree, they also show some deviations during certain periods. So far most reconstructions were based on only one single radionuclide record, which makes detection and correction of these deviations impossible. Here we combine different ¹⁰Be ice core records from Greenland and Antarctica with the global ¹⁴C tree ring record using principal component analysis. This approach is only possible due to a new high-resolution ¹⁰Be record from Dronning Maud Land obtained within the European Project for Ice Coring in Antarctica in Antarctica. The new cosmic radiation record enables us to derive total solar irradiance, which is then used as a proxy of solar activity to identify the solar imprint in an Asian climate record. Though generally the agreement between solar forcing and Asian climate is good, there are also periods without any coherence, pointing to other forcings like volcanoes and greenhouse gases and their corresponding feedbacks. The newly derived records have the potential to improve our understanding of the solar dynamics and to quantify the solar influence on climate.

Most of the cosmic ray particles reaching the Earth are produced outside the solar system during supernova explosions. Mainly two magnetic shields, the solar magnetic field and the geomagnetic field, modulate the cosmic ray flux approaching the Earth. Both shields vary in time, and so does the cosmic radiation at Earth. The weaker these fields, the higher is the cosmic ray intensity at Earth. Ionization chambers and neutron monitors have directly monitored the intensity of cosmic radiation since the 1930s (1). Prior to this time, no direct measurements exist, and cosmogenic radionuclides, which can be considered as a kind of a “natural neutron monitor,” can be used as a proxy for cosmic radiation (2). In this context the two most commonly used radionuclides are ¹⁰Be and ¹⁴C. ¹⁰Be and ¹⁴C are produced in the Earth’s atmosphere by nuclear reactions of cosmic ray particles with atmospheric nitrogen and oxygen (3) (Fig. 1). Thus, the production rates of these nuclides are directly related to the flux of the incoming cosmic ray particles. Although their production processes are very similar, their geochemical behavior in the Earth system is completely different (Fig. 1). ¹⁴C enters the global carbon cycle, and therefore fluctuations of the atmospheric ¹⁴C concentration (Fig. 2B) measured as Δ¹⁴C in tree rings are damped, smoothed, and delayed relative to the ¹⁴C production. The effect of the carbon cycle can be removed by inverse carbon cycle modeling. The resulting ¹⁴C production rate (p¹⁴C, Fig. 2 C and D) calculated with a box-diffusion carbon cycle model (4) is a better measure of the cosmic radiation, but it still contains a climate signal component due to unknown temporal changes of the carbon cycle and a memory effect of the production history. In contrast to ¹⁴C, aerosol-borne ¹⁰Be is removed from the atmosphere relatively fast within a few years and stored in natural archives such as polar ice sheets. Because of its short atmospheric residence time, ¹⁰Be directly reflects cosmic ray intensity variations with almost no attenuation and a delay of 1–2 y. Uncertainties are introduced mainly on annual time scales by atmospheric mixing processes and wet and dry deposition from the atmosphere to the ice (5).

Fig. 1.

Cartoon illustrating some basics of the radionuclides ¹⁴C and ¹⁰Be in the Earth’s system. Both radionuclides are produced in a very similar way by nuclear reactions of cosmic ray particles with the atmospheric gases (3). After production, their fate is very different (system effects). ¹⁰Be attaches to aerosols and is transported within a few years to ground (34). ¹⁴C oxidizes to CO₂ and enters the global carbon cycle, exchanging between atmosphere, biosphere, and the oceans (4).

Fig. 2.

Radionuclide records used for this study. ¹⁰Be concentration records are green (Greenland) and red (Antarctica) and ¹⁴C is black. Time is given as year before present (BP where present refers to 1950 AD). All records are mean normalized (divided by the mean) 22-year averages. (A) New ¹⁰Be record from the EDML ice core (red) and of the existing ¹⁰Be record from the GRIP (GR) ice core (green). The data are plotted on the time scales EDML1 (35) for EDML and GICC05 (36, 37) for GRIP. (B) Δ¹⁴C, the deviation of the atmospheric ¹⁴C/¹²C ratio from a standard value, measured in tree rings (14) and (C) ¹⁴C production rate p¹⁴C (PC), calculated with a box-diffusion carbon cycle model (4) from Δ¹⁴C (14) over the last 9,400 y. (D) ¹⁴C production rate p¹⁴C (PC) and ¹⁰Be concentrations for several available ice cores (GR: GRIP; D3: Dye-3; NG: NorthGRIP; MI: Milcent; SP: South Pole; DF: Dome Fuji) over the last 1,200 y. Grand solar minima are marked by yellow bands and capital letters: O: Oort, W: Wolf, S: Spörer, M: Maunder, D: Dalton, G: Gleissberg.

For the Holocene there are several radionuclide records with high temporal resolution: p¹⁴C (Fig. 2 C and D) derived from Δ¹⁴C measured in tree rings (Fig. 2B), and ¹⁰Be from different ice cores from Greenland and Antarctica (Fig. 2 A and D). The high similarity between the two different radionuclides ¹⁴C and ¹⁰Be and the fact that the well known grand solar minima such as the Maunder and the Dalton minima are well represented in the radionuclide records (Fig. 2D) confirms that these records reflect primarily production changes caused by variations of the cosmic ray flux. However, closer inspection reveals that often, the shapes during these periods differ slightly in their details. Because the cosmic ray induced production changes are almost identical for ¹⁰Be and ¹⁴C, the differences observed in radionuclide records stored in geological archives are the result of so-called system effects. The term “system effects” is used here for all processes influencing the radionuclide signal (e.g., climate-induced transport and deposition changes, changes in the carbon cycle) other than cosmic ray-induced production (heliomagnetic and geomagnetic modulation). A prominent system effect, for example, is the variation of the snow accumulation rate, which directly affects the ¹⁰Be concentration stored in the ice. To some extent system effects can be corrected. For example, snow accumulation effects can be partly removed by calculating ¹⁰Be fluxes (concentration times snow accumulation rate). However, due to the combination of both wet and dry deposition on the total aerosol deposition, neither concentration nor fluxes are truly independent of accumulation rate (6). The determination of the snow accumulation rate of an ice core is not trivial and is associated with additional uncertainties. In the polar regions of Greenland and Antarctica, snow accumulation rates have been rather stable during the Holocene so that ¹⁰Be concentrations and fluxes yield similar results (SI Appendix, Section S7). For the sake of simplicity we use the concentrations below.

From this discussion it is obvious that individual radionuclide records may contain a significant system component which, if not eliminated, would be incorrectly attributed to variations of the cosmic radiation. In fact, system effects are probably the main reason for the observed differences between the existing reconstructions of cosmic radiation based on single radionuclide records only (7–10).

Results

Here we present a cosmic ray record for the past 9,400 y for which the system effects were minimized. This high-resolution and low-noise paleocosmic ray record is used to derive solar activity that in turn provides a powerful tool to search for the solar fingerprint in climate records.

Our approach first combines several hemispheric ¹⁰Be records with the global ¹⁴C record with high temporal resolution (Fig. 2 A, C, and D and Table 1) and then follows the study of Abreu, et al. (11) by using principal component analysis (PCA) to extract the common production signal. With PCA it is possible to disentangle the cosmic ray-induced production signal from the system effects. The successful separation is based on the assumption that the production signal is common to all records (correlated), whereas the system effects are different (uncorrelated) in each record. If system effects affected the radionuclide records spatially and temporarily in a similar way, PCA would “misinterpret” these effects as common variation which in turn would be wrongly attributed to cosmic radiation changes. However, it is known that climate effects differ spatially; hence it seems justified to assume that they affect the different radionuclide records from different hemispheres and archives differently.

Table 1.

Radionuclide records used in the study and their characteristics (3)

Record	Nuclide	Source	Time span (year before present)	Mean value used to normalize data in Fig. 2 (104 atoms/g)	Reference
INTCAL09	14C	tree rings	9,400 to 150	1*	(14)
EDML†	10Be	ice Antarctica	9,400 to 1,200	4.1	New
GRIP ‡	10Be	ice Greenland	9,400 to 300	1.5	(12), (13)
Dye-3	10Be	ice Greenland	526 to −35	1.1	(16)
Milcent	10Be	ice Greenland	769 to 148	1.1	(17)
NorthGRIP	10Be	ice Greenland	561 to −44	1.8	(15)
South Pole	10Be	ice Antarctica	1,107 to −32	3.8	(19)
Dome Fuji	10Be	ice Antarctica	1,255 to 75	9.4	(18)

^*The ¹⁴C production rate in the carbon cycle model is 1 (idealized units) for present day (Δ¹⁴C = 0‰). The global average ¹⁴C production is approximately 2 atoms cm^-2 s^-1 (3).

^†European Project for Ice Coring in Antarctica Dronning Maud Land area

^‡European Greenland Ice Core Project

The three main records of our reconstruction [¹⁰Be from European Project for Ice Coring in Antarctica-Dronning Maid Land (EPICA-DML), Antarctica; ¹⁰Be from European Greenland Ice Core Project (GRIP) (12, 13), Greenland; and global p¹⁴C from tree rings (14)] represent both hemispheres and are measured in different archives. These three records cover a time range from 1,194 to 9,400 y before present (BP) (SI Appendix, Section S1) and have a temporal resolution of a few years. The new ¹⁰Be dataset from the EPICA-DML ice core consists of 1,846 data points with an average temporal resolution of about 4.5 y and therefore provides an essential part for the reconstruction. To homogenize the different records, 22-year averages are calculated for each individual record. Then PCA is applied to the 22-year averages. The first principal component explains 69% of the total variance (SI Appendix, Section S6). This is a striking result because it implies that the ¹⁴C production changes throughout the Holocene are very similar to the ¹⁰Be changes found in polar ice cores from the two different hemispheres. Again, because ¹⁰Be and ¹⁴C are produced in a very similar way and because system effects are not expected to correlate with the production signal, we conclude that the first principal component is the best representation of the global production rate and herewith of the cosmic ray intensity at Earth. The production signal obtained from ¹⁰Be from GRIP and EPICA Dronning Maud Land area (EDML) ice cores and ¹⁴C from tree rings for the period 1,194–9,400 BP is extended to the present day by applying PCA to several other high-resolution ¹⁰Be records from Greenland [NorthGRIP (15), Dye3 (16), Milcent (17)] and Antarctica [Dome Fuji (18), South Pole (19)], and to ¹⁴C from tree rings (14). In this time interval the production signal is also dominant, explaining more than 67% of the total variance. The entire record is normalized to the two most recent 22-year averages centered in 1955 and 1977 AD (SI Appendix, Section S9). By combining several radionuclide records with PCA as done in this study, an assessment of the systematic uncertainty can be done. The common signal in radionuclide records describes about 70% of the variance, implying that the system effects cause the remaining 30%. These system effects are removed by using only the first principal component. The robustness of PCA was tested by applying a jackknife method, applying PCA to subsets of radionuclide records by leaving out single records. The jackknife uncertainty is on average 5% (SI Appendix, Section S8), which is significantly smaller than the large (greater than 50%) variations due to changes in solar activity between periods of low solar activity like the Maunder minimum (20) and of high solar activity like the past decades.

We note that there are several sources of systematic error that are not removed by PCA. These are (i) uncertainties in the time scales of the individual records, (ii) uncertainties in the production rate calculations of ¹⁰Be and ¹⁴C, (iii) the assumption that relative variations of ¹⁰Be concentrations and ¹⁰Be production rates are equal, (iv) uncertainty in the assumption of linear independence between production and system effects.

Variations on different time scales are evident in the final low-noise cosmic radiation (radionuclide production) record (Fig. 3 B, C, and D). A comparison with changes in the geomagnetic dipole field strength (21) (Fig. 3A) shows that the geomagnetic dipole shielding is the main cause of the observed multimillennial variability; the stronger the geomagnetic field, the lower is the cosmic radiation. On multidecadal to centennial time scales the cosmic radiation variations are mostly due to solar modulation (Fig. 3 C and D) as indicated by the coincidence of cosmic radiation maxima and grand solar (sunspot) minima like the Maunder minimum (Fig. 3D).

Fig. 3.

(A) Geomagnetic dipole field strength relative to today (21). (B) Cosmic radiation based on the first principal component of several radionuclide records, 22-year averages, over the last 8,000 y. Time is given as year BP. The gray band represents the standard deviation of the individual radionuclide records without applying PCA (SI Appendix, Section S8). The black dashed line represents the average cosmic ray intensity for 1944–1988 AD. (C) Same as (B), but zoom-in of the past millennium. Capital letters mark grand solar minima: O: Oort, W: Wolf, S: Spörer, M: Maunder, D: Dalton, G: Gleissberg. (D) Same as (C), but zoom-in of the past 350 y. Time is given as year AD. Red circles and green curve are 22-year averages and yearly averages of cosmic ray intensity calculated with (3) using the solar modulation potential (38) obtained from neutron monitor and ionization chamber data (SI Appendix, Section S9). At the bottom the annual sunspot number is plotted (39).

In a second step, we use the new low-noise cosmic radiation record to reconstruct total solar irradiance (TSI) by first removing the shielding effect of the geomagnetic field (3) (Fig. 3A) and then applying the procedure described in refs. 22–24 (SI Appendix, Section S10). The wavelet coherence between TSI of this study and two other recent TSI reconstructions (22, 25) based on single radionuclide records show agreement in the low-frequency part (periods longer than 100 y) and differences in the high-frequency part (periods shorter than 100 y) (SI Appendix, Section S13). This finding indicates that the PCA has indeed reduced the high-frequency noise present in the individual single radionuclide records.

In a last step, TSI is compared with the δ¹⁸O record from Dongge cave, China, supposedly reflecting variations in precipitation amount related to changes in the Asian monsoon (AM) (26) (Fig. 4).

Fig. 4.

Comparison of solar activity (total solar irradiance [TSI]) in blue and δ¹⁸O from Dongge cave, China, in green representing changes of the Asian climate. possibly the Asian monsoon (AM) (low δ¹⁸O corresponds to strong AM monsoon and vice versa). TSI has been reconstructed from the cosmic ray intensity reconstruction (SI Appendix, Section S10). Both records have been normalized (subtraction of mean value and division by the standard deviation), linearly detrended and high-pass filtered with 2,000 y. (A) Time series of solar activity (TSI) and δ¹⁸O. Solar activity (TSI) is plotted on a reversed scale. (B) Wavelet of solar activity (TSI). De Vries cycle at approximately 210 y and Eddy cycle at approximately 1,000 y are marked with horizontal, gray dashed lines. Black boundaries mark 95% significance level. (C) Wavelet coherence of solar activity (TSI) and δ¹⁸O. De Vries cycle at approximately 210 y and Eddy cycle at approximately 1,000 y are marked with horizontal, gray dashed lines. Arrows pointing to the right indicate that the records are in phase. Black boundaries mark the 95% significance level.

Note, however, that a recent climate modeling study (27) did not find a correlation between δ¹⁸O in Chinese precipitation and the amount of precipitation over the Holocene. However, the study (27) showed that changes in monsoon are related to changes in δ¹⁸O, although not via variations in the amount of precipitation.

Climate records have been often compared with Δ¹⁴C as a proxy of solar activity. The Dongge Cave record has been previously compared with Δ¹⁴C (21). As discussed above, Δ¹⁴C is not directly showing the solar forcing signal but a solar and geomagnetic production signal filtered by the carbon cycle system, which leads to a frequency dependent attenuation of the amplitude and a phase shift. Thus, a repetition using the solar activity record derived in our study appears to be justified. This climate record was chosen for two reasons: First, it is one of the few accurately dated climate records with high resolution that covers a similar period as the derived TSI. Second, it is a good example to demonstrate how to detect a solar signal in a climate record. The climate record has a large long-term trend due to orbital forcing, which we removed by subtracting the linear trend and applying a 2,000-year high-pass filter. We used the same data processing (subtracting the linear trend and applying a 2,000-year high-pass filter) for TSI.

We emphasize that changes in TSI are phase-locked with changes in solar spectral irradiance and with other changes related with solar activity, and therefore here we cannot distinguish between these different solar forcings. Hence, in this context TSI is taken as a proxy of solar activity, which is the reason we use the term solar activity instead of TSI in the following discussion.

The Pearson correlation between solar activity and AM is R = -0.29, meaning, during periods of low solar activity the AM strength has been weaker (higher δ¹⁸O) (Fig. 4A). Despite being significantly different from zero (p < 10^-6), this correlation, however, can only explain 10% in the total decadal to centennial variance in the AM record. We did a correlation analysis and found that the strongest correlation is found for a slightly negative lag (22 y), which is consistent with no lag within the uncertainty (SI Appendix, Section S12). The power spectrum of solar activity (SI Appendix, Section S11) shows significant periodicities (p < 0.05) already known from individual ¹⁴C and ¹⁰Be records such as the de Vries cycle (around 210 y), the Eddy cycle (around 1,000 y) (28), and an unnamed cycle at approximately 350 y, as well as other less significant unnamed cycles at approximately 500 and 710 y. These periodicities, although not significant, are also found in the Asian climate (SI Appendix, Section S11). The wavelet spectrum of solar activity (Fig. 4B) shows that the amplitudes of these periodicities have varied in time, that is, the de Vries cycle amplitude has varied with a period of about 2,200 y, called the Hallstatt cycle (29). The largest amplitudes of the de Vries cycle are found during Hallstatt cycle minima centered at approximately 8,200; 5,500; 2,500; and 500 BP. Comparison of the time series of solar activity and its wavelet spectrum (Fig. 4 A and B) show that grand solar minima occur preferentially at minima of the de Vries cycles (note that solar activity is plotted on a reversed scale in Fig. 4A). Comparison of the time series of solar activity and climate (Fig. 4A) and their wavelet coherence (Fig. 4C) shows that in general during the Hallstatt cycle minima of solar activity (again characterized by large de Vries cycle amplitudes and a frequent occurrence of grand solar minima) the AM is weaker. However, there is some discrepancy during the Hallstatt cycle minima between 5,000 and 6,000 BP and in the past 1,500 y. During these periods the wavelet coherence for the periodicity of the de Vries cycle (around 210 y) is low, although several grand solar minima are visible. A similar pattern is visible at the Eddy periodicity (around 1,000 y), which has high power in the wavelet coherence, except in the period 3,000 to 5,000 BP. Such temporal differences are expected because the Sun is not the only driver of the climate system. Other forcing factors such as volcanic aerosols and greenhouse gases have changed in time, obscuring temporally the solar fingerprint.

Conclusions

We combined a new ¹⁰Be record from Dronning Maud Land, Antarctica, comprising more than 1,800 data points with several other already existing radionuclide records (¹⁴C from tree rings and ¹⁰Be analyzed in polar ice cores of Greenland and Antarctica) covering the Holocene. Using principal component analysis, we separated the common radionuclide production signal due to solar and geomagnetic activity from the system effects signal due to the different transport and deposition processes. The common signal represents a low-noise record of cosmic radiation, particularly for high frequencies, compared to earlier reconstructions, which are only based on single radionuclide records. On the basis of this record, we then derived a reconstruction of total solar irradiance for the Holocene, which overall agrees well with two existing records but shows less high-frequency noise. A comparison of the derived solar activity with a record of Asian climate derived from δ¹⁸O in a Chinese stalagmite reveals a significant correlation. The correlation is remarkable because the Earth’s climate has not been driven by the Sun alone. Other forcings like volcanoes, greenhouse gas concentrations, and internal variability also have played an important role. To quantify the solar influence on the Earth’s climate and to distinguish between the different forcings, climate model simulations are required for the Holocene, employing the new dataset of total solar irradiance. The dataset will be available online at the National Oceanic and Atmospheric Administration paleo server (http://www.ncdc.noaa.gov/paleo/forcing.html).

Materials and Methods

Records Used in This Study.

Table 1 summarizes the characteristics of the records used in this study. We applied a decay correction to the ¹⁰Be records, using a half-life of (1.387 ± 0.012) million years (30, 31). The production rate p¹⁴C from the atmospheric ¹⁴C concentrations as measured in tree rings (14) has been calculated with a box-diffusion carbon cycle model (4).

¹⁰Be EDML Record.

The EDML ice core, located on the high plateau of the Atlantic sector of the East Antarctic ice sheet in Antarctic Dronning Maud Land, Antarctica, was drilled within the EPICA framework (32). The ¹⁰Be record used covers the period from 1,193 to 9,400 BP. The chemical preparation of the ¹⁰Be samples was done at Eawag in Dübendorf, Switzerland. First the ice was cut into samples of 25 cm length, corresponding to an average time resolution of about 4.5 y. Then the ice was melted in a microwave oven, mixed with 0.125 mg of ⁹Be carrier, and passed through a cation ion exchange resin. The chemical preparation was done in the same way as for the ¹⁰Be samples from the GRIP ice core (12, 13). The samples were measured at the accelerator mass spectrometry (AMS) facility of Ion Beam Physics group at Swiss Federal Institute of Technology Zurich, Switzerland. The typical AMS uncertainty is 5%. The results were normalized to the ETH standard S555N (33).

Computational Analysis.

The radionuclide records were combined into one using principal component analysis (PCA) to obtain a record covering the entire period of interest. First we divided each record into a series of consecutive 22-year intervals, with the interval centers being the same for all records. The data points in each interval were averaged, and this average was assigned to the central age. The principal components (PCs) were built from eight blocks of records with each block containing two to six individual radionuclide records (SI Appendix, Section S6). The first PC describes most of the total variance and is interpreted as the production signal due to solar and geomagnetic modulation. The first PC of one block is then normalized in such a way that its mean is equal to the mean of the first PC of the previous block in the overlapping interval covered by the two blocks. Note that we changed only the mean but not the variance. Finally, the curves of the eight blocks are combined to represent the common production signal variation. The robustness was tested with a jackknife method by repeating PCA and systematically leaving out one radionuclide record.