Current volcanic reconstructions based on ice core analysis have
significantly improved over the past few decades by incorporating
multiple-core analyses with a high temporal resolution from different parts
of the polar regions into a composite common volcanic eruption record.
Regional patterns of volcanic deposition are based on composite records,
built from cores taken at both poles. However, in many cases only a single
record at a given site is used for these reconstructions. This assumes that
transport and regional meteorological patterns are the only source of the
dispersion of the volcanic products. Here we evaluate the local-scale
variability of a sulfate profile in a low-accumulation site (Dome C,
Antarctica), in order to assess the representativeness of one core for such a
reconstruction. We evaluate the variability with depth, statistical
occurrence, and sulfate flux deposition variability of volcanic eruptions
detected in five ice cores, drilled 1 m apart from each other. Local-scale
variability, essentially attributed to snow drift and surface roughness at
Dome C, can lead to a non-exhaustive record of volcanic events when a single
core is used as the site reference, with a bulk probability of 30 % of
missing volcanic events and close to 65 % uncertainty on one volcanic
flux measurement (based on the standard deviation obtained from a five-core
comparison). Averaging
When a large and powerful volcanic eruption occurs, the energy of the blast
is sufficient to inject megatons of material directly into the upper
atmosphere (Robock, 2000). While ashes and pyroclastic materials fall
rapidly to the ground because of gravity, gases remain in the atmosphere over longer timescales. Among gases, SO
In polar regions, the deposition of the sulfuric acid particles on pristine snow will generate an acidic snow layer, enriched in sulfate. The continuous falling of snow, the absence of melting and the ice thickness make the polar snowpack the best records of the Earth's volcanic eruptions. Hammer (1977) was the first to recognize the polar ice's propensity to record such volcanic history. Built on the seminal work of Hammer et al. (1977), paleovolcanism developed around this discovery and has two aims.
The first relies on the idea that the ice record can reveal past volcanic activity and, to a large extent, its impact on Earth's climate history (Robock, 2000; Timmreck, 2012). Indeed, on a millennium timescale, volcanoes and solar activity are the two main recognized natural climate forcings (Stocker et al., 2013). Based on ice records, many attempts are made to extract the climate forcing induced by a volcanic eruption (Crowley and Unterman, 2013; Gao et al., 2007, 2008; Sigl et al., 2013, 2014; Zielinski, 1995). However, such an approach is inevitably prone to large uncertainty pertaining to the quality of the ice record and nonlinear effects between deposition fluxes and source emissions (Pfeffer et al., 2006).
The second aim of paleovolcanism is to provide an absolute dating scale when clear volcanic events in differently located ice cores can be unambiguously attributed to the same dated event (Severi et al., 2007). The time synchronization of different proxy records is possible, allowing the study of the phasing response of different environmental parameters to climate perturbation (Ortega et al., 2015; Sigl et al., 2015) or estimating the snow deposition over time (Parrenin et al., 2007). Whatever the aim, paleovolcanism should rely on robust and statistically relevant ice core records.
Work undertaken to date to establish a volcanic index has assumed that volcanic events are clearly identified, without any false signal from background variations induced by other sulfur sources (e.g., marine, anthropogenic). Seasonal layer counting is used whenever possible, bipolar comparison of ice sulfate records has become the method of choice to establish an absolute dated volcanic index (Langway et al., 1988). Both known and unknown events can be used to synchronize different cores. However, only a limited number of peaks, with characteristic shape or intensity, known to be associated with a dated eruption, can be used to set a reliable timescale (Parrenin et al., 2007). This restriction is partly fueled by the poor and/or unknown representativeness of most volcanic events found in ice cores. Most of the time, a single core is drilled at a given site and used for cross comparison with other sites. This approach is clearly insufficient for ambiguous events.
On a large scale, sulfate deposition is highly variable in space and mainly
associated with atmospheric transport and precipitation patterns. On a local
scale (ca. 1 m), variability can emerge from post-deposition processes.
While sulfate is a nonvolatile species supposed to be well preserved in snow,
spatial variability is induced by drifted snow and wind erosion leading to
surface roughness heterogeneities (Libois et al., 2014). These effects are
amplified at low-accumulation sites where most of the deep drilling is
performed (EPICA community members, 2004; Jouzel, 2013; Lorius et al.,
1985). To the best of our knowledge, only one study has used multiple
drillings at a given site to analyze the representativeness of the ice core
record (Wolff et al., 2005). This study took advantage of the two EPICA
(European Project for Ice Coring in Antarctica) cores drilled at Dome C,
10 m apart (Antarctica; 75
In the present study we took advantage of the drilling of five ice cores at
Dome C, initially intended for the analysis of sulfur isotopes of the
volcanic sulfate. Putting aside the number of records, our approach is
similar in many points to the work of Wolff et al. (2005). However, it has
the advantage of relying on highly resolved sulfate profiles. In addition,
the spatial scale is slightly smaller as the five cores were drilled 1 m
apart. The comparison of five identically processed cores is a chance to
approach the representativeness of a single-core reconstruction at a
low-accumulation site, the most prone to spatial variability. The
representativeness of a volcanic record can be assessed by isolating the
volcanic peaks in different records, as done in Wolff's work and in this
study, or by a global comparison of the sulfate concentration records as
proposed in Gfeller et al. (2014). In the latter case, the full individual
profiles (background
New constraints on the variability of sulfate deposition recorded by spatial heterogeneity at such sites are expected from the present work. Even if recent publications (Sigl et al., 2014), underline the need to use multiple records at low-accumulation sites to overcome the spatial variability issue, such records are not always available. This lack of records adds uncertainty to the volcanic flux reconstruction based on polar depositional patterns. Our study should help to better constrain the error associated with local-scale variability and, ultimately, the statistical significance of volcanic reconstructions. The present study discusses the depth shift, occurrence of events and deposition flux variability observed in the five cores drilled.
The project VOLSOL (VOLcanic and SOLar natural climatic forcings), initiated
in 2009, aimed at constraining the estimation of the natural part of
radiative forcing, composed of both volcanic and solar contributions using
ice core records of sulfate and
Analyses were performed directly in the field during two consecutive summer campaigns. Thirty meters were analyzed in 2011, and the rest was processed the following year. The protocol was identical for each core and the steps followed were as follows:
decontamination of the external layer by scalpel scrapping; longitudinal cutting with a band saw of a 2 cm stick out of the most external
layer; sampling of the ice stick at a 2 cm resolution (ca. 23 600 samples); thawing the samples in 50 mL centrifuge tubes and transferring them into 15 mL
centrifuge tubes positioned in an autosampler; automatic analysis with a Metrohm IC 850 in suppressed mode (NaOH at 7 mM,
suppressor H
Due to the fragility of snow cores, the first 4 m were only analyzed on a
single core (Fig. 1). We will thus not discuss the variability of the
Pinatubo and Agung eruptions present in these first 4 m. Concentration data
are deposited in the public domain and are made freely available at the NOAA
National Centers for Environmental Information, Paleoclimatology Data
(
Sulfate profiles on the five replicate cores obtained during a
drilling operation at Dome C – Antarctica in 2011. Data are available at
As with most algorithms used for peak detection, the principle is to detect
anomalous sulfate concentration peaks from a background noise (stationary or
not), which could potentially indicate a volcanic event. The estimation of
the background value should therefore be as accurate as possible. Using core
2 as our reference core, we observed a background average value stationary
and close to 85 ppb
Age vs. depth in core 1 drilled in 2011 CE, Dome C – Antarctica. Dates are given as CE dates, with negative figures indicating dates BCE.
Core 1 was entirely dated with respect to the recently published volcanic ice
core database (Sigl et al., 2015) using the Analyseries 2.0.8 software
(
Tie points used to set the timescale and synchronize the cores.
Volcanic events are named “ev
Through the routine described above, the five cores are depth-synchronized using the 23 tie points, and other potential volcanic events in each core cores are detected independently. Therefore, the number of peaks detected in each core is different (between 47 and 54) and their depth (with the exception of the tie points used) is slightly different from each of the other cores due to the sampling scheme and position of the maximum concentration. After correcting the depth shift between cores, a composite profile was built by summing all the peaks identified in the five cores. In this composite, sulfate peaks from different cores are associated with a same event as soon as their respective depth (corresponding to the maximum concentration) is included in a 20 cm depth window. This level of tolerance is consistent with the dispersion in width and shape of peaks observed (Fig. 3). A number of occurrences is then attributed to each sulfate peak, reflecting the number of times it has been detected in the five-core data set (Fig. 4).
Kuwae (
Panel
Depth offsets between cores are the result of the surface roughness at the time of drilling, variability in snow accumulation, heterogeneous compaction during the burying of snow layers and logging uncertainty. This aspect has been discussed previously, over a similar timescale (Wolff et al. 2005) and over a longer timescale (Barnes et al. 2006) in Dome C. Surface roughness, attributed to wind speed, temperatures and accumulation rate, is highly variable in time and space. These small features hardly contribute to the depth offset on a larger spatial scale, in which case glacial flow can control the offset between synchronized peaks, as seems to be the case at the South Pole site (Bay et al. 2010). However, in Dome C, and on the very local spatial scale we are considering in the present work, roughness is significant regarding the accumulation rate. It is therefore expected that synchronized peaks should be found at different depths. The offset trend fluctuates with depth, due to a variable wind speed (Barnes et al., 2006). To estimate the variability in the depth shift for identical volcanic events, we used the tie points listed in Table 1. For each peak maximum, we evaluate the depth offset of core 1, 3, 4 and 5, with respect to core 2. To avoid logging uncertainty due to poor snow compaction in the first meters of the cores and surface roughness at the time of the drilling, we used the UE 1809 depth in core 2 (13.30 m) as a depth reference horizon from which all other depth cores were anchored using the same 1809 event. For this reason, only eruptions prior to 1809 were used to evaluate the offset variability, that is 18 eruptions instead of the 23 used for the core synchronization. Figure 5 shows the distribution of the depth shift of the cores with respect to core 2. While the first 40 m appear to be stochastic in nature, a feature consistent with the random local accumulation variations associated with snow drift at the Dome C site, it is surprising that at greater depth, offset increases (note that the positive or negative trends are purely arbitrary and depend only on the reference used, here core 2). The maximum offset, obtained between cores 3 and 5 is about 40 cm. Such accrued offsets with depth were also observed by Wolff et al. (2005) and were attributed to the process of logging despite the stringent guidelines used during EPICA drilling. Similarly, discontinuities in the depth offset, observed by Barnes et al. (2006), were interpreted as resulting from logging errors. As no physical processes can explain a trend in the offsets, we should also admit that the accrued offset is certainly the result of the logging process. In the field, different operators were involved, but a common procedure was used for the logging. Two successive cores extracted from the drill were reassembled on a bench to match the nonuniform drill cut and then hand-sawed meter by meter to get the most precise depth core, as neither the drill depth recorder nor the length of the drilled core section can be used for establishing the depth scale. This methodology involving different operators should have randomized systematic errors, but obviously this was not the case. Despite the systematic depth offset observed, synchronization did not pose fundamental issues as the maximum offset in rescaled profiles never exceeds the peak width (ca. 20 cm) thanks to the 10 possible comparisons when a pair of cores is compared. Confusion of events or missing events are thus very limited in our analysis (see next section).
Depth offset of 18 common and securely identified volcanic events in cores 1, 3, 4 and 5 relative to core 2. To overcome offset due to the drilling process and poor core quality in the first meters, UE 1809 (depth: ca. 13 m) is taken as the origin and horizon reference.
The variability in event occurrence in the five ice cores has been evaluated through the construction of a composite record (Fig. 4) and the counting of events in each core as described in the Methods section. By combining the five ice cores, we listed a total of 91 sulfate peaks (Pinatubo and Agung not included), which are not necessarily from volcanic sources. Some peaks can be due to post deposition effects affecting the background deposition or even contamination. When it comes to defining a robust volcanic index, peak detection issues emerge. The risk of misinterpreting a sulfate peak and assigning it, by mistake, to a volcanic eruption, as well as the risk of missing a volcanic peak, can be examined through a statistical analysis conducted on our five cores.
We try to evaluate to what extent multiple-core comparison facilitates the identification of volcanic peaks among all sulfate peaks that can be detected in a core. To do so, we assumed that a peak is of volcanic origin as soon as it is detected in at least two cores. In other words, the probability of having two nonvolcanic peaks synchronized in two different cores is zero. It is expected that combining an increasing number of cores will increasingly reveal the real pattern of the volcanic events. All possible combinations from two- to five-core comparisons were analyzed, totalizing 26 possibilities for the entire population. The results of each comparison were averaged, giving a statistic on the average number of volcanic peaks identified per number of cores compared. The results of the statistical analysis are presented in Fig. 6. As expected, in a composite made up of one to five cores, the number of sulfate peaks identified as volcanic peaks (through being detected at least twice) increases with the number of cores combined in the composite. Thus, while only 30 peaks can be identified as volcanic from a two-core study, a study based on five cores can yields 62 such peaks. The five-core comparison results in the composite profile given in Fig. 4a. The initial composite of 93 peaks is reduced to 64 volcanic peaks (Pinatubo and Agung included) after removing the single peaks (Fig. 4b). Each characteristic of the retained peaks is given in Table 2. The main conclusion observing the final composite record is that only 17 of the 64 peaks were detected in all of the five cores and 68 % of all peaks were at least present in two cores. On the other end of the spectrum, two-core analysis reveals that only 33 % (30 peaks on average) of the peaks are identified as possible eruptions. A two-core comparison still presents a high risk of not extracting the most robust volcanic profile at low-accumulation sites, a conclusion similar to that of Wolff et al. (2005). Surprisingly, it may also be noted that this five-core comparison does not result in an asymptotic ratio of identified volcanic peaks, suggesting that five cores are not sufficient either to produce a full picture. High-accumulation sites should be prone to less uncertainty; however, this conclusion remains an a priori that still requires confirmation.
Black dots on the red line (left axis) represent the number of
sulfate peaks that can be identified as volcanic peaks in a composite
profile made up of
Sulfate peaks (maximum concentration in nanograms per gram and flux
of volcanic sulfate deposited in kilograms per square kilometer) considered
to be volcanic eruptions based on the statistical analysis of the five cores.
Flux is calculated by integrating the peak, using the density profile
obtained during the logging process. Volcanic flux values are corrected from
background sulfate (calculated separately for each sulfate peak). Zero stands
for non-detected events in the cores. Agung (3.77 m) and Pinatubo (1.52 m)
were not included in the statistical analysis because they were analyzed only
in core 1 and thus are marked as not applicable (n/a). The estimation of the
average volcanic flux takes into account undetected peaks, for which the flux
is considered to be 0. The relative error in the flux (estimated as 10 %)
takes into account the IC measurement relative standard deviation (below
4 % based on standards runs), the error in firn density (relative error
estimated as 2 %) and the error in sample time length (10 %). The
last column displays data obtained from Castellano et al. (2005) for
identical volcanic peaks. For similar peaks Castellano's flux generally falls
into the average flux
Peaks probability to be detected in two, three, four or five cores,
as a function of their flux. The three categories of fluxes are defined by peak
flux value, relatively to the average background flux, and quantified by
Close look at UE 1809 and Tambora (1815) events showing the absence of the Tambora event in two out of the five cores. This figure illustrates the possibility of missing major volcanic eruptions when a single core is used.
Large and small events are not equally affected by these statistics. Figure 7
shows that the probability of presence is highly dependent on peak flux
and the risk of missing a small peak (maximum flux in the window (
To compare peak height variability, detected peaks were corrected by
subtracting the background from peak maxima. We considered
Statistics on sulfate signal for identical peaks in cores 1, 2, 3, 4 and 5. Geometric standard deviations are calculated on peak heights (i.e., maximum concentration reached in nanograms per gram) and on peak sulfate flux (i.e., total mass of volcanic sulfate deposited after the eruption). Background corrections are based on background values calculated separately for each volcanic event.
This study confirms in many ways previous work on multiple drilling variability (Wolff et al., 2005). As already discussed, peaks flux uncertainty can be significantly reduced (65 to 29 %) by averaging five ice core signals. A five-core composite profile was built using the criterion that a peak is considered volcanic if present in at least two cores. We observed that the number of volcanic peaks listed in a composite profile increases with the number of cores considered. With two cores, only 33 % of the peaks present in the composite profile are tagged as volcanoes. This percentage increases to 68 % with five cores. However, we did not observe an asymptotic value, even with five cores. A single record at a low-accumulation site is therefore very unlikely to be a robust volcanic record. Of course, peaks presenting the largest flux are more likely to be detected in any drilling, but the example of the Tambora eruption shows that surface topography is variable enough to erase even the most significant signal, although this occurs rarely. This variability in snow surface is evidenced in the depth offset between two cores drilled less than 5 m from each other, as peaks can easily be situated 40 cm apart.
At low-accumulation sites such as Dome C, where surface roughness can be on the order of the snow accumulation and highly variable, indices based on chemical records should be considered with respect to the timescale of the proxy studied. Large timescale trends are only a little sensitive to this effect. By contrast, a study on episodic events such as volcanic eruptions or biomass burning, with a deposition time on the order of magnitude of the surface variability scale should be based on a multiple-drilling analysis. A network of several cores is needed to obtain a representative record, at least in terms of recorded events. However, although lowered by the number of cores, the flux remains highly variable, and the mean flux obtained from five cores is still uncertain by almost 30 %. This point is particularly critical in volcanic reconstructions that rely on the deposited flux to estimate the mass of aerosols loaded in the stratosphere and, to a larger extent, the climatic forcing induced. Recent reconstructions largely take into account flux variability associated with a regional pattern of deposition, but this study underlines the necessity of not neglecting local-scale variability at low-accumulation sites. Less variability is expected with a higher accumulation rate, but this still has to be demonstrated. Sulfate flux is clearly one of the indicators of eruption strength, but due to transport, deposition and post-deposition effects, such a direct link should not be taken for granted.
With such statistical analysis performed systematically at other sites, we should be able to reveal even the smallest volcanoes imprinted in ice cores, extending the absolute ice core dating, the teleconnection between climate and volcanic events and improving the time resolution of the mass balance calculation of ice sheets.
Some of this work would not have been possible without the technical support
from the C2FN (French National Center for Coring and Drilling, run by
INSU). Financial support was provided by LEFE-IMAGO, a scientific program
of the Institut National des Sciences de l'Univers (INSU/CNRS), by the Agence
Nationale de la Recherche (ANR) via contract NT09-431976- VOLSOL and by a
grant from Labex OSUG@2020 (Investissements d'avenir – ANR10 LABX56). E. Gautier
sincerely thanks the Fulbright commission for providing the PhD Fulbright
fellowship. The Institute Polaire Paul-Emile Victor (IPEV) supported the
research and polar logistics through the program SUNITEDC No. 1011. We would
also like to thank all the field team members who were present during the VOLSOL
campaign and helped us. Data are available at the World Data Center for
paleoclimatology
(