Introduction
In a context of climate change, the study of
paleoclimate is an important tool for understanding the interactions between
climate and atmospheric conditions . Ice cores have been
used to retrieve climatic and atmospheric conditions back to 800 000
years before present (BP) . Notably, ancient atmospheric gases get
enclosed within bubbles in the ice and allow us to reconstruct the past
history of atmospheric composition . The trapping
of air in ice is due to the transformation of firn (porous compacted snow)
into airtight ice at depths ranging from ∼50 to ∼120m depending on temperature and accumulation conditions. It is
characterized by an increase in bulk density and a decrease in porosity with
depth along the firn column. It is only at the bottom of the firn column that
the porosity of the medium gets closed and traps the interstitial air. From a
gas point of view the firn is traditionally divided in three main parts from
the surface to the bottom: the convective zone, the diffusive zone, and the trapping
zone e.g.,. The
convective zone is characterized by the mixing of air in the snow with
atmospheric air through wind action . In the diffusive
zone the dominant gas transport processes are molecular diffusion and
gravitational settling. Finally, the trapping zone corresponds to the
enclosure of air into bubbles through the closure of the porosity. The
process of densification and pore closure can last for thousands of years at
the most arid sites in Antarctica.
Air trapping affects the recording of atmospheric variability in ice cores.
One known effect of the gas enclosure mechanism is the damping of fast variations
in the atmosphere, also called smoothing . This smoothing arises for
two reasons: (i) the gas diffusion in the firn mixes air from different
dates, and thus a bubble does not enclose gases with a single age but rather
an age range
;
(ii) in a given horizontal layer, bubble enclosure takes place over a range
of time rather than instantaneously. These two phenomena combined mean that
at a given depth, the air enclosed is represented by a gas age distribution
and not by a single age
. Gas enclosure
mechanisms thus act as a low-pass filter, attenuating signals whose periods
are too short compared to the span of the distribution.
reported the only existing observations of the
smoothing effect under low-accumulation conditions. They concluded that the
abrupt methane variation during the cold event of 8.2kyrBP
recorded in the EPICA Dome C ice core, compared with its counterpart from the
Greenland GRIP ice core, had experienced an attenuation of 34 to 59%.
Sites with low accumulation tend to have broader age distributions leading to
a stronger damping effect
.
A heuristic explanation is that the span of the age distribution is directly
related to the densification speed of a firn layer, which is slow at the low-temperature and arid sites of the Antarctic plateau. For the most arid sites
the impact of diffusive mixing is negligible compared to progressive
trapping, and the smoothing is hence mainly driven by the speed of porosity
closure.
Even if the bulk behavior in firn is the increase in density and decrease in
open porosity with depth, local physical heterogeneities affect firn
densification and gas trapping .
Working on ice cores and firn from high-accumulation sites,
, , and
have discussed the influence of centimeter-scale
physical variability in firn on recorded gas concentrations. They argue that
physical heterogeneities can lead to variations in closure depth for
juxtaposed ice layers. For instance, a given layer could reach bubble
enclosure at a shallower depth and earlier (respectively deeper and later) than
the surrounding layers in the firn, thus trapping relatively older gases
(respectively younger gases). In periods of atmospheric variations in trace
gas composition occurring on a similar timescale as the trapping process,
this mechanism can lead to gas concentration anomalies along depth in an ice
core and has been called layered bubble trapping. Based on observations in
high-accumulation Greenland ice cores and modeling for the WAIS Divide ice
core, report that such artifacts can reach 40ppbv in the methane (CH4) record during the industrial time. In
addition, the amplitude of the artifacts increases with lower accumulation
rates.
Here we investigate for the first time the existence and impacts of
heterogeneous trapping and smoothing in very low-accumulation conditions
using continuous measurements of trace gases. High-resolution methane
concentration (combined with carbon monoxide) measurements were performed
along a section of the Vostok 4G-2 ice core drilled in the Antarctic
plateau. The section studied corresponds to the Dansgaard–Oeschger event
number 17 (DO-17, ∼60000yrBP), a climatic event
associated with particularly fast and large atmospheric methane variations
. This makes
this event especially adapted for the quantification of both gas record
smoothing and layered trapping. To interpret our data we compare them with
the much less smoothed methane record measured in the WAIS Divide ice core
WDC,, where the accumulation rate is an order
of magnitude larger than at Vostok.
Ice core samples and analytical methods
Vostok ice samples
The ice core analyzed in this study is the 4G-2 core drilled at Vostok, East
Antarctica in the 1980s . Measured depths range from
895 to 931m, with a cumulative length of 27.5m
due to several missing portions in the archived ice at Vostok station. The
ice core sections analyzed have been stored at Vostok station since the
drilling and were transported to the Institut des Geosiences de
l'Environnement (IGE, Grenoble, France; formerly LGGE) 3 months before
analyses. Although stored at Vostok at temperatures of ∼-50∘C, the samples showed clathrate relaxation cavities. The gas
age over this depth interval spans a 3000-year interval
centered on 59400±1700yrBP
. The estimated snow
accumulation rate at the Vostok core site for this period is 1.3±0.1cmiceyr-1 .
Even though DO events are associated with large warmings in the Northern
Hemisphere, isotopic records indicate that DO-17 temperatures on the
Antarctic plateau remain at least 5 ∘C below modern temperatures
Fig. 2 in.
Continuous methane measurements
The Vostok 4G-2 ice core sections were analyzed at high resolution for
methane concentration (and carbon monoxide as a by-product) at IGE
over a 5-day period and using a continuous ice core melting system with
online gas measurements (CFA, continuous-flow analysis). Detailed
descriptions of this method have been reported before
.
Ice core sticks of 34 by 34mm were melted at IGE at a
mean rate of 3.8cmmin-1 using a melt head as described by
, and the water and gas bubble mixture was
pumped toward a low-volume T-shaped glass debubbler. All the gas bubbles and
approximately 15% of the water flow were transferred from the debubbler
to a gas extraction unit maintained at 30 ∘C. The gas was extracted
by applying a pressure gradient across a gas-permeable membrane (optimized
IDEX in-line degasser; internal volume 1mL). The gas pressure
recorded downstream of the IDEX degasser was typically
500–600mbar and was sufficiently low to extract all visible air
bubbles from the sample mixture. A homemade Nafion dryer with a 30mLmin-1 purge flow of ultrapure nitrogen (Air Liquide;
99.9995% purity) dried the humid gas sample before entry into the laser
spectrometer. Online gas measurements of methane were conducted with a SARA
laser spectrometer developed at Laboratoire Interdisciplinaire de Physique
(Grenoble, France) based on optical-feedback cavity-enhanced absorption
spectroscopy OF-CEAS;. Such a
laser spectrometer has been used before for continuous-flow gas analyses
e.g.,;
however, the IGE CFA system was specifically optimized to reduce experimental
smoothing by limiting all possible dead and mixing volumes along the sample
line. For this study the rate of OF-CEAS spectrum acquisition was 6Hz. The 12cm3 optical cavity of the spectrometer was
maintained at 30mbar of internal pressure, which corresponds to an
equivalent cavity volume of only 0.36cm3 at STP and allows
for a fast transit time of the gaseous sample in the cavity. Consequently,
the SARA instrument introduces a significantly lower smoothing than the CFA
setup. The SARA spectrometer was carefully calibrated onto the NOAA2004 scale
before the CFA analyses using three
synthetic air standards with known methane concentrations (Scott-Marrin, Inc.;
Table S1 in the Supplement). CH4 concentrations measured
during the calibration agreed with NOAA measurements within 0.1% over a
360–1790ppbv range. A linear calibration law was derived and
applied to all CH4 data (Fig. S1 in the Supplement).
Allan variance tests were
conducted using mixtures of degassed deionized water and synthetic air
standard to evaluate both the stability and the precision of the
measurements. The best Allan variance was obtained with an integration time
larger than 1000s, illustrating the very good stability of the CFA
system. However, in order to optimize the depth resolution of our
measurements, we used an integration time of 1s for which a
precision of 2.4ppbv (1σ) was observed. This corresponds to
a peak-to-peak CH4 variability of ∼10ppbv. Hereafter, this
variability will be referred to as analytical noise.
The mixing of gases and meltwater during the sample transfer from the melt
head to the laser spectrometer induces a CFA experimental smoothing of the
signal. The extent of the CFA-based damping was determined by performing a
step test (left panel of Fig. S2), i.e., a switch between two synthetic
mixtures of degassed DI water and synthetic air standards of different
methane concentrations, following the method of
. It shows that the CFA system can resolve
signals down to the centimeter scale. We were also able to extract the
impulse response of the system that will be used in Sect. to
emulate CFA smoothing. A more detailed discussion of the frequency response
of the system can be found in Sect. S2. Breaks along the
core regularly let ambient air enter the system, resulting in strong positive
spikes in methane concentration. In order to remove these contamination
artifacts, exact times corresponding to a break running through the melt head
were recorded during the measurements and later used to identify and clean
the data from contamination.
Nitrogen isotopes
The ratio of stable nitrogen isotopes, 15N / 14N, was measured
at the Laboratoire des Sciences du Climat et de l'Environnement (LSCE),
France. Briefly, a melting technique followed by gas condensation in
successive cold traps was used to extract the air from the ice, and the air
samples were then transferred to a dual-inlet mass spectrometer (Delta V
Plus; Thermo Scientific). The analytical method and corrections applied to
the results are described in and the references therein.
The results are expressed as deviations from the nitrogen isotopic ratio in
dry atmospheric air (δ15N). Discrete samples every 50 cm and
duplicates were analyzed when possible. A total of 96 data points, including
39 duplicates were obtained. The pooled standard deviation over duplicate
samples is 0.011‰.
Experimental results
Methane record
The methane record spanning the DO-17 event extracted from the Vostok
4G-2 ice core is presented in blue in Fig. . Two
corrections were applied to these data: (i) data screening and removal of
kerosene contamination and (ii) full dataset calibration to account for the
preferential dissolution of methane during the melting process.
Kerosene, used as drilling fluid for the Vostok 4G-2 ice core extraction, was
detected in some of the meltwater from our continuous-flow analysis. This
contamination induces surface iridescent colors and a strong characteristic
smell; it was detected not only in the meltwater from the outer part of our
ice sticks but also in some of the meltwater from the center of the ice
samples. However, the continuous flow of the meltwater does not allow us to
clearly identify the contaminated ice core sections. Carbon monoxide (CO) was
measured simultaneously with methane by using our laser spectrometer
. We attributed simultaneous anomalies in CH4
(increase of about 20ppbv or more) and CO (increase of about
100ppbv or more) mixing ratios to kerosene contamination and
suppressed the corresponding data by visual inspection of the dataset. An
example of such a kerosene contamination is visible in Fig. S3.
indicate that methane contamination lower than 40ppbv was observed by discrete measurements in the brittle zone of
the Vostok 3G core, which is consistent with our observation in 4G-2. The impact of
kerosene contamination on CO in ice cores has not been quantified so far.
Adding the length of all kerosene-affected ice core sections, a total of 2.1m of data was removed. The calibration of the methane mixing ratio for
preferential solubility was achieved by matching
our continuous methane measurements with the already calibrated WDC methane
dataset , as described in Sect. S1.2. The resulting methane record has a high resolution, but presents
numerous discontinuities due to missing ice, ambient air infiltration, and
kerosene contamination. The signal displays two distinct scales of
variability.
Atmospheric-history-relevant variability. The general shape of the
signal can be divided in two parts, a stable zone extending from 931 to
915m of depth and two consecutive methane variations of
approximately 100ppbv each, extending respectively from 915 to
907m and from 907 to 895m. They respectively
correspond to the plateau preceding the DO-17 event and the DO-17 event
itself.
Centimeter-scale variability.
The signal also displays centimeter-scale methane variations. A portion of
these variations is explained by the 10ppbv analytical noise of
the CFA system. However, in the upper part of the core (above 915m) the signal also exhibits abrupt variations with amplitudes
reaching up to 50ppbv and widths of about 2cm. Most
of those spikes are negatively orientated and therefore laboratory air or
kerosene contamination can be ruled out. It should be noted that the width of
the spikes is in the attenuation range of the CFA system, meaning that the
true signal in the core has a somewhat larger amplitude than the measured
signal. Moreover, the spikes exhibit a specific distribution with depth. For
instance, no spike is observed in the lower part of the ice core where the
methane concentration is essentially flat, and only negative spikes appear
between 900 and 905m of depth as seen in the zoomed part of
Fig. .
(a) Methane concentration along the Vostok 4G-2 ice core.
In blue: data cleaned from ambient air and kerosene contamination and
calibrated. In black: data cleaned from layered trapping. (b) Zoom
of the section from 902.0 to 902.7m; (c)
δ15N of N2 as a function of depth in the Vostok 4G-2 ice core.
Orange dots: isotopic measurements. The vertical error bars correspond to the
pooled standard deviation. In blue: linear regression.
Revised age scale using Nitrogen isotopes
The current reference chronology for the Vostok ice core is the Antarctic Ice
Core Chronology 2012 AICC2012;. However, only two gas stratigraphic links between Vostok
and other cores are available for the DO-17 period in AICC2012, leading to
relatively large uncertainties in the Vostok gas age scale over this period.
The δ15N of N2 profile over the DO-17 event in the Vostok core is
shown Fig. . We fitted the experimental values with a
linear regression (slope of 9.63×10-4‰m-1
and intercept of -0.417‰). Considering the diffusive zone
of the firn to be stratified according to a barometric equilibrium
, its height can be
expressed as H=(RT/gΔM)ln(1+δ15N), where R is the ideal gas
constant, T the temperature, g the gravitational acceleration, and
ΔM the difference in molar mass between 14N and 15N. With a
firn temperature of 217K , the mean
δ15N value of 0.46‰ translates into a diffusive
column height of 85m and an LIDIE (lock-in depth in ice
equivalent) of 59m (using a mean firn relative density of
D=0.7). This value lies in the lower range of the AICC2012 LIDIE
estimations for this depth range in the Vostok ice core: 58 to 70m .
The age difference between the ice and the enclosed gases (ΔAge) can
be estimated using the height of the firn with
ΔAge =(H+Hconv)D/accu, where H and Hconv are the heights
of the diffusive and convective zones, respectively, D is the average
density of the firn column, and “accu” the accumulation rate. Present-day
observations report a convective zone spanning down to 13m at
Vostok . We used this value as an estimate for the
convective zone depth during DO-17. In Fig. , the ΔAge
values inferred from our δ15N record using D=0.7 and an
accumulation rate of 1.3cmiceyr-1 are compared
with the values from AICC2012 .
The AICC2012 ΔAge values display a variability of several centuries as
shown by the dashed black line in Fig. . These variations are
sufficient to induce significant distortions in the duration of methane
events. These distortions affect the comparison between our measurements and
the WDC record from , as seen in Fig. S11. Furthermore, the amplitude of the ΔAge variations is
similar to the uncertainty in gas age (1479 to 1841 years). The studied
period is fairly stable in terms of temperature and accumulation at Vostok
; thus the
ΔAge changes in the AICC2012 chronology are likely to result from
artifacts of the optimization method rather than to correspond with actual
variations. We hence revised the AICC2012 gas age scale by deriving a new
smooth gas age using the AICC2012 ice age scale and our ΔAge values
inferred from the linear interpolation of δ15N data
(Fig. ). This new smooth chronology enables us to visually
identify the different subparts of the DO-17 event between the Vostok and
WDC methane records. It is important to note that this gas age chronology
will again be improved by matching the Vostok and WDC methane records (see
Sect. ). The corresponding ΔAge of this final
chronology is displayed as the green line in Fig. .
ΔAge along the Vostok record. Orange dots: ΔAge
directly estimated from δ15N measurements. In blue: ΔAge
derived from the linear regression on isotopic measurements. Black dashed
line: ΔAge as given by AICC2012. In green: ΔAge after matching
with the WDC CH4 record. Black dots: tie points (minima, maxima, and
mid-slope points) used to match the WDC record (see
Sect. ).
Layered bubble trapping
Conceptual considerations of the layered trapping mechanism
Due to heterogeneities in firn density and porosity, an ice layer may undergo
early gas trapping . Thus during gas trapping, the
corresponding layer is at a more advanced state of closure than the
surrounding bulk layers. Similarly, some layers may undergo a late closure.
If gases can circulate through the open porosity surrounding the anomalous
layers, the early closed layers will contain abnormally ancient gas with
respect to the surrounding layers. On the other hand, layers closed late will
contain abnormally recent gas. This leads to very local inversions of the gas
age scale along depth. As explained in , such a
mechanism affects trace gas records only during periods of variations in
the concentration of atmospheric gases. Then, abnormal layers contain air
significantly different in composition from surrounding layers and appear as
spikes in the record. On the other hand, during periods without atmospheric
variations, the abnormal layers do not contain air significantly different in
composition from their surroundings, and the gas record is not affected.
The orientation of layered trapping spikes depends on the type of atmospheric
variations, as illustrated in Fig. . For instance, in a
period of local maximum in methane concentrations, both early and late
closures tend to enclose air with lower mixing ratios, as displayed in case A
in Fig. . Similarly, in periods of methane minima,
abnormal layers tend to enclose air with larger mixing ratios, as displayed
in case C in Fig. . In the case of monotonous
increase or decrease, early and late closures lead to artifacts with opposite
signs, as represented by case B in Fig. . It should be
noted that early and late closures are not expected to affect the record with
the same importance. Indeed, a late pore closure means that the surrounding
firn is sealed and prevents long-distance gas transport. The latest closure
layers will not be able to trap young air if gas transport is impossible in
the surrounding firn layers, resulting in less important artifacts.
Expected orientation of layered trapping artifacts depending on the
characteristics of atmospheric variations. Black curves correspond to a
normal chronological trapping, blue to early pore closure, and orange to late
pore closure. Cases A, B, and C respectively represent local maximum,
monotonous trend, and local minimum situations.
Observed layered trapping in the Vostok 4G-2 ice core
The positive and negative spikes observed in the Vostok 4G-2 methane record
introduced in Sect. are consistent with the expected
impacts of layered trapping. First, the absence of spikes in the lower part of
the record below 915m is consistent with the absence of an
overall methane trend over the corresponding period. Moreover, in periods of
methane local maxima at around 903 and 910m of depth, most of the
spikes are negatively oriented as expected with the conceptual mechanism of
layered bubble trapping (cf. case A in Fig. ).
Thin sections of ice covering the depth range between 902.0 and 902.42m (zoomed range in Fig. ) have been analyzed to
investigate whether structural anomalies were associated with anomalous trapping.
The method is described in detail in Sect. S4. We were not able to observe
any link between the grain sizes and abnormal layers in the methane record.
Nonetheless, structural anomalies may have existed at the time of pore closure
before disappearing with ∼60000 years of grain evolution.
Explanations for the methane anomalies other than layered trapping were considered
as well. Looking for a correlation between ice quality and methane anomalies
was also a motivation for the above thin section analysis. Although the
samples showed small clathrate relaxation cavities, the CFA sticks did not
reveal visual signs of stratification possibly associated with abnormal
layers. Examples of a CFA stick picture and thin section results are provided
in the Supplement. The ice samples were not large enough to allow for CFA
duplicate analysis, but the sticks were not melted in a regular depth order so
that instabilities in the measurement system could be more easily detected.
As contamination cannot explain negative methane concentration anomalies, we
could not find a convincing alternate explanation for layered bubble trapping
in our results.
Simple model of layered trapping
A major difficulty for understanding the gas trapping in ice is to relate
the structural properties measured in small samples to the three-dimensional
behavior of the whole firn. For example, pore closure anomalies have been
associated with tortuosity anomalies, with more tortuous layers closing earlier
, or to density anomalies, with denser layers
closing earlier . In this section we used the latter hypothesis supported by
observed relationships between local density and closed porosity
e.g., to test whether
density-driven anomalies could result in artifacts as observed in the Vostok
methane record.
In our simple model, the ice core is discretized in layers of 2cm width. Abnormal layers are stochastically distributed along the
ice core. Based on the characteristics of our Vostok methane signal, we use a
density of 10 abnormal layers per meter. They are given a random density
anomaly (Δρ, normally distributed) representing the density
variability at the bottom of the firn. Based on various sites,
propose linear regressions in which the
close-off density variability declines for declining accumulation and
temperature. Their lowest-accumulation site is Dome C, with an accumulation
of 2.5cmiceyr-1 and a density variability (Δρ)
of 4.6kgm-3. Applied to Vostok DO-17 conditions, the
accumulation-based extrapolation leads to a variability of 7kgm-3 and the temperature-based extrapolation leads to a
variability of 2.7kgm-3. This defines our extreme values
(7 and 3kgm-3), and we chose the middle number of 5kgm-3 as the best-guess value. Hence, in the model, the abnormal
layers are given a firn density anomaly distributed according to a
zero-centered Gaussian distribution of standard deviation of 5kgm-3. In order to convert density anomalies into a closure
depth anomaly (the difference in pore closure depth between an abnormal layer
and an adjacent layer following the bulk behavior), we assume that all layers
have similar densification rates (dρ/dz). Using the data-based
density profiles at Dome C, Vostok, and Dome A in
, dρ/dz in deep firn is estimated to
be in the range 1.7 to 2.5kgm-4. Thus, the gradient is
set to be 2kgm-4. Specifically, a layer closing in advance
(or late) closes higher (or lower) in the firn. Dividing the above typical
density anomaly (Δρ) by the depth gradient (dρ/dz),
the characteristic depth anomaly in deep firn of anomalous layers is about
2.5 m. Using the estimated accumulation rate of 1.3cmiceyr-1 for this period, it translates into an age anomaly
(the gas age difference between an abnormal layer and an adjacent layer
following the bulk behavior) of about 207 years. As explained in Sect. 4.1,
late pore closure tends to produce weaker age anomalies than early closure
due to the sealing of the surrounding firn. To take into account the lack of
explicit gas transport in the model, we reduce the standard deviation of late
closure age anomalies to 52years, i.e., 25 % of the value used
for early closure artifacts. The value of 25% has been chosen to limit
late trapping artifacts in a visually consistent manner with the
observations. The methane mixing ratio at a given depth is computed using an
atmospheric trend history and a gas age distribution (GAD) of trapped gases
. The atmospheric methane scenario used
is the high-resolution methane record from the WAIS Divide ice core
. The WDC gas age chronology (WD2014) was scaled to
the GICC05 chronology (with present defined as 1950) dividing by a factor of
1.0063 as in . For the rest of the paper we use this
scaled WD2014 chronology to express WDC gas ages. All layers are assumed to
have the same GAD, simply centered on different ages. The GAD used here is
the one derived in Sect. specifically for the Vostok ice
core during the DO-17 event. A sensitivity test using a very different GAD is
described in the next paragraph. Finally, in order to reproduce the gas
mixing in the CFA system discussed in Sect. , the modeled
concentrations have been smoothed by convolving the signal with an estimated
impulse response of the CFA system (Fig. S2). The smoothing characteristics
of our measurement system were determined experimentally as in
. The CFA smoothing induces a damping of about
18% of the modeled artifacts.
Modeled layered trapping artifacts. The black curve represents the
results of smooth trapping. Spikes correspond to a single stochastic
realization of the layered trapping with CFA smoothing. Blue stands for
early closure and yellow for late closure. Blue shaded areas correspond to
the range of concentration anomalies for early closure anomalies up to 2
standard deviations (depth anomaly of 5m corresponding to an age
anomaly of 415 years). Yellow shaded areas correspond to late closure
anomalies with 25% of the early closure extent (depth anomaly of
1.25m corresponding to an age anomaly of 104years).
The modeled artifacts (Fig. ) globally reproduce
the depth distribution and amplitude of the methane anomalies observed in the
Vostok ice core (Fig. and Sect. ). To test
the robustness and sensitivity of our model to uncertainties and underlying
assumptions, we modified several model parameters. First, the limitation of
late closure trapping was removed, hence simulating symmetrical behavior
between early and late trapping. The results displayed in Fig. S7 show a
clear increase in the amplitude of late closure artifacts. In particular, the
enhanced late trapping produces artifacts of about 50ppbv before
the onset of the DO-17 (in the 914 to 917m depth range). Their
absence in the CFA measurements confirms our assumption of the predominance of
early closure artifacts. On the other hand, as shown in case B of
Fig. , some limited late trapping is required to
reproduce what appear as positive anomalies at the onset of the DO-17 event
(912 to 913m depth range). We also estimated the sensitivity of
the model to the density variability (Δρ) and densification rate
(dρ/dz). The extremal values for these two parameters provided at
the beginning of this section result in typical depth anomalies of 1.2 and
4.1m, corresponding to age anomalies of 99years and
341years. The model results are displayed in Figs. S8 and S9. Using
a reduced depth anomaly of the anomalous layers leads to largely reduced
amplitudes of the anomalies. Using an increased depth anomaly of the
anomalous layers leads to overestimated amplitudes of the anomalies,
especially between 903 and 910m of depth.
As using a Gaussian distribution of density anomalies is equivalent to using
a random depth anomaly; the smallest anomalies produced by the model do not
exceed the analytical noise. We imposed a density of 10 anomalies per meter,
which results in about 5 significant anomalies per meter (exceeding 10ppb) in the 895 to 915m depth range. About 70% of
these significant artifacts correspond to early closure layers. The width of
the anomalous layers also influences the amplitude of the modeled anomalies
because it is in the attenuation range of the CFA system. While 2cm layers experience a damping of 18%, an attenuation of about
30% is observed with 1cm layers. The anomalies observed in
the Vostok signal have widths ranging between 1 and a few centimeters.
Their smoothing by the CFA system is thus limited. We also tested an
alternative to the homogeneous GAD hypothesis, assuming that anomalous
layers have a strongly reduced GAD similar to the gas age distribution in
the WDC core. The results are displayed in Fig. S10. As the WDC record of
the DO-17 event is less smooth than the Vostok record, the reduced GAD assumption
leads to large positive artifacts, especially around 912m of depth,
which are not observed in the Vostok signal.
Finally, under the hypothesis of density-based layering, age anomalies
strongly depend on accumulation as explained by . A
lower accumulation leads to a weaker density variability in the firn
, but at the same time leads to a larger age
difference between successive firn layers due to a steeper age–depth slope.
The second effect tends to dominate and the net effect of a lower
accumulation is an increase in age anomalies due to layered trapping.
Moreover, it is important to note that the good agreement between our density-driven model and observations does not imply that tortuosity is not an
important factor in anomalous trapping. High-resolution air content
measurements could potentially help us better understand the physical
properties of anomalous layers at closure time.
Layering model parameters and resulting depth anomaly, age anomaly,
and associated figure. The first row corresponds to the reference simulation
and the sensitivity tests are below. The depth and age anomaly values refer to
the standard deviation (1σ) of early trapping artifacts. These
1σ values are half the 2σ values mentioned in the corresponding
figure captions.
dρ/dz
Δρ
Limit late
Narrow
Depth anomaly
Age anomaly
Figure
(kgm-4)
(kgm-3)
anomalies
GADs
(m)
(yr)
2
5
Yes
No
2.5
207
2
5
No
No
2.5
207
S7
2.5
3
Yes
No
1.2
99
S8
1.7
7
Yes
No
4.1
341
S9
2
5
Yes
Yes
2.5
207
S10
Removing layering artifacts in the methane record
To extract an undisturbed (chronologically monotonous and representative of
atmospheric variability only) methane signal from the Vostok 4G-2 core,
layered trapping artifacts need to be removed from the high-resolution CFA
record. Some sections of the core exhibit mainly positive or negative
artifacts. Hence removing them using a running average would bias the signal.
To account for this specificity, a cleaning algorithm has been developed. The
underlying assumptions are that the chronological signal is a slowly varying
signal with superimposed noise composed of the analytical noise and the
layered trapping artifacts. Using a looping procedure, the artifacts are
progressively trimmed until the resulting noise is free of spikes. The
detailed algorithm is described in the following.
Using the CFA signal (with or without already partially removed layering
artifacts during the cleaning process) a running median is computed with a
window of 15cm. Then a binned mean is computed with bins of 50cm. The goal of this step is to remove noise without introducing a
bias due to layering artifacts.
A spline of degree 3 is used to interpolate between the binned points on
the original CFA depth scale. This interpolating spline does not further
smooth the signal and is used as an approximation of the chronologically
monotonous signal, free of layering artifacts.
By removing the spline from the CFA signal we obtain the detrended noise of
the signal composed of the analytical noise and the remaining artifacts.
We then compute the normalized median absolute deviation (NMAD) of the
detrended noise. The expression of the NMAD is
1.4826×med(|ximed(xi)|), where xi represents the noise values and
“med” the median. This is a robust estimator of variability, weakly
sensitive to outliers . It
enables the estimation of the variability of the noise without the artifacts, i.e.,
the analytical noise.
The detrended noise is cut off with a threshold of 2.5 times the NMAD.
We then check whether the noise is free of spikes. For this we compare the NMAD
(estimation of the variability without spikes) and the standard deviation
(estimation of variability with spikes) of the detrended noise. If these two
quantities are similar, the noise is free of anomalous layers. Once the
standard deviation is lower than 1.5 times the NMAD, the procedure is
finished. Otherwise, the algorithm is looped.
This algorithm does not require an estimation of the analytical noise
beforehand, since this value is dynamically computed. However, it is
sensitive to the value of 1.5 used to compare the NMAD and standard
deviation to test for the presence of artifacts. The remaining signal after
cutting off the layered trapping anomalies has a noise amplitude of ±16 ppbv and is represented in black in Fig. .
With our method 15% of the methane data points have been removed. As
expected, the signal is almost unmodified below 915m, with a
portion of removed points of only 1.3%. On the other hand, the
variability above 915m is greatly reduced and about 26% of
the methane data points have been removed.
Smoothing and age distribution in the Vostok 4G-2 ice core
The smoothing of the methane record
Once the methane signal is cleaned from layered trapping artifacts, we
consider the data to be a chronologically ordered and unbiased signal
recorded in the core. It is smoothed (high frequencies are damped) with
respect to the true atmospheric signal and can be used to infer the degree
of smoothing in the Vostok ice core. The damping can be visualized in
Fig. by comparing the Vostok record with the WDC record.
High-frequency atmospheric variability is much better preserved in the WAIS
Divide ice core because the accumulation rate is more than an order of
magnitude higher in the range from 18 to
22cmiceyr-1 for the studied period;,
and
thus the firn densification and gas trapping are faster. For instance, the
methane variation spanning between 59000 and 58800yrBP is
damped by ∼50% in the Vostok record compared to WDC. Moreover, a
20ppbv sub-centennial variation is present in the WDC record
between 58700 and 58600yrBP. In the Vostok record, however,
this short-scale variability event has been smoothed out. On the other hand
the multi-centennial variability visible between 58700 and 58400yrBP is well preserved with only a slight damping. From the
comparison between WDC and Vostok, we can infer that the smoothing in Vostok
4G-2 prevents us from retrieving information below the centennial scale
during the DO-17 period.
WDC CH4 signal convolved with different GADs: the Dome C GAD
estimated for the Bølling–Allerød by in red, the Dome
C GAD estimated for the Last Glacial Maximum by in
green, a lognormal fit to the modern Vostok GAD from
in yellow, and the Vostok DO-17 GAD estimated in Sect. in
black (uncertainty envelope shown in light blue). The WDC record
is displayed in dashed green, and the CFA Vostok
measurements in blue. Yellow dots show the tie points used to match the WDC
and Vostok records.
Estimate of the gas age distribution
The smoothing of gas concentrations in ice core records is the direct
consequence of the broad gas age distributions in the ice
. We call absolute GAD the age distribution expressed on an
absolute timescale in years before present. The relative GAD is the
distribution expressed relatively to its mean age. For a given layer,
absolute and relative GAD thus only differ by a translation in age. Here we
assume that all layers densified under the same physical conditions hence
share the same relative GAD. Since computing concentrations along an ice core
using GADs is equivalent to a convolution product
, the resulting concentrations will be
called convolved signals.
The climatic conditions of the glacial period on the Antarctic plateau have
no modern analogue, and thus relevant GADs cannot be inferred from modern firn
observations. High-resolution CFA-based gas records offer a new opportunity
to estimate GADs without modern analogue. We thus developed such a method,
which requires a reference atmospheric scenario with much higher frequencies
resolved. The method can be extended to gases other than methane or to
low-accumulation records other than the Vostok 4G-2 core. The principle of the
method is to determine a GAD able to convolve the high-accumulation record
(in our case, WAIS Divide) into a smoothed signal that minimizes the
differences with the observed low-accumulation record (in our case, Vostok).
It can be seen as an inverse problem. Two assumptions are made to reduce the
number of adjusted parameters and thus ensure that the problem is
well defined in a mathematical sense. First, all ice layers have the same
relative GAD over the considered period. Second, following
, this relative GAD is assumed to be a lognormal
distribution that is fully characterized by two free parameters (for
instance, its mean and standard deviation). Due to the asymmetry of the GAD,
the resulting convolved signal displays age shifts when compared with the
original atmospheric scenario. Hence for a valid comparison between the
record and convolved signals, it is necessary to modify the age scale and to
optimize the GAD in an iterative process. Using an initial age scale, the
steps are as follows.
First, a new gas age scale is derived. Tie points are manually selected
between the low-accumulation record and the convolved high-accumulation
record. The tie points we selected correspond to minima, maxima, and
mid-slopes points of the methane record. For the initialization, since no GAD
has been optimized yet, we use the atmospheric scenario instead of the
convolved signal. The new gas chronology is then generated by interpolation
and extrapolation between tie points.
A new lognormal GAD is optimized by modifying its two parameters in
order to minimize differences between the simulated and observed smoothed
signals. We performed this optimization with a differential evolution
algorithm .
If the definition of a new chronology and a new GAD
does not improve the RMSD (root mean square deviation) between the convolved
signal and the measurements five times in a row, then the algorithm is stopped.
The above methodology can be applied to different ice drilling sites. Here we
describe the specific aspects to match the Vostok record with WDC.
state the following: “Only at gas ages >60kaBP is there a possibility that the continuous measurement system caused
dampening of the CH4 signal greater than that already imparted by
firn-based smoothing processes”. Moreover, Fig. S1 of the supplement to
Rhodes et al. (2015) predicts a GAD width of about 40 years for the DO-17 event, which is much
smaller than the width of the Vostok GAD. This ensures that the WDC signal
resolves enough high frequencies to be used as the weakly smoothed
atmospheric scenario compared to the Vostok record and that the convolving
function given by the algorithm is close to the actual Vostok GAD. As
explained in Sect. , the WD2014 gas chronology is converted to
the GICC05 scale and not further modified. The
algorithm only adjusts the Vostok gas ages, which remain well within AICC2012
uncertainties. The initial gas ages used are the ones derived from nitrogen
isotope measurements in Sect. , and the optimization
has been performed on data ranging from 900 to 915m of depth.
This depth interval has been chosen since it corresponds to a significantly
dampened event in the Vostok record, which is sensitive to the choice of the
GAD. The optimized gas age distribution is displayed in
Fig. in black, with uncertainty intervals shown as light
blue shaded area. The uncertainty envelope encloses all the distributions
resulting in simulated Vostok signals with an RMSD from the measurements lower
than 150% of the optimal RMSD. The optimal lognormal parameters are
given in Table . The chosen tie points are displayed in
Fig. , and the optimized ΔAge values along the Vostok
core are depicted in green in Fig. . The optimal convolution of
the WDC methane record from into a Vostok signal
can be seen in black in Fig. , with the impact of the
uncertainty on the GAD displayed as the light blue envelope. The convolution
fits the methane measurements within the analytical noise. The overall
consistency between the measured and simulated Vostok signals confirms that
the Vostok record is a smoothed version of the WDC record and that the
choice of a single GAD for the whole DO-17 record is a credible hypothesis.
This last point is consistent with the fairly stable climatic conditions on
the Antarctic plateau over this time period
.
Gas age distributions. In black: the Vostok GAD during the DO-17
estimated with our optimization scheme; the uncertainty envelope is shown in
light blue. In yellow: a lognormal fit to the modern-condition Vostok GAD
from . In red: the estimated Dome C GAD during B/A from
. In green: the estimated Dome C GAD during LGM from
.
Discussion
Understanding the smoothing of ice core signals under low-accumulation conditions
In Fig. , our GAD adjusted to produce the expected smoothing
rate for the DO-17 event in the Vostok ice core (in black) is compared to
other available gas age distributions for low-accumulation-rate conditions.
The different parameters of the lognormal GADs used in this section are
displayed in Table . For modern ice cores, GADs can be
estimated with gas transport models constrained by firn air composition data
. However, the results directly depend
on the closed versus total porosity parameterization used, which is
insufficiently constrained e.g.,. We
performed a comparison of our optimized GAD for Vostok during DO-17 with the
lognormal fit to a GAD constrained with modern-condition firn air
measurements at Vostok in yellow in
Fig. . Note that using the modern GAD from
or a lognormal fit to this GAD leads to the same
smoothing, but the lognormal GAD enables us to provide simple GAD parameters
in Table 2. The comparison with our optimized DO-17 GAD suggests a slightly
narrower distribution for the glacial period, despite lower temperatures. On
the other hand, the GAD estimate from for Dome C
during the Bølling–Allerød (B/A; accumulation of about 1.5cmiceyr-1) is narrower (in red Fig. ) and
results in a slightly too-weakly smoothed methane record in
Fig. . Finally, the GAD proposed by
for Dome C during the Last Glacial Maximum (LGM) is broader than the other
presented GADs (in green Fig. ) and thus leads to a
stronger smoothing in the record in Fig. . The GADs calculated
for modern conditions from at Dome C and
at Vostok are very similar, which is consistent with
the comparable accumulation rates of the two sites: 2.7cmiceyr-1 at Dome C and
2.4cmiceyr-1 at Vostok . We
therefore do not observe a systematic broadening of GADs for lower
accumulation rates, even at a given site. This questions either the
relationship between GAD widths and accumulation rate or the consistency
between GADs derived from gas transport models in firn and the GAD obtained
with our method of record comparison.
Parameters defining the lognormal distributions used as GADs for
Vostok DO-17 (this study), lognormal fit to modern Vostok
, Dome C during the Bølling–Allerød Dome C
B/A;, and Dome C during the Last Glacial Maximum
Dome C LGM;. Location and scale respectively
refer to the parameters μ and σ used in Eq. (1) in
. SD stands for standard deviation.
Site and
Location
Scale
Mean
SD
period
(yr)
(yr)
Vostok DO-17
4.337
1.561
259
835
Vostok modern
4.886
1.029
226
308
Dome C B/A
4.886
0.5
150
79
Dome C LGM
5.880
1
590
773
The most likely reason for an inconsistency between GADs inferred from firn
models and from CFA data is the large uncertainty in the representation of
gas trapping in firn models. As mentioned above, the closed versus total
porosity ratio is very uncertain, as it was measured only at a few sites and
in small size samples. Better constraints on the physics of gas trapping
would thus be helpful. However, there is no modern analogue of the central
Antarctic plateau sites (such as Vostok or Dome C) under glacial conditions.
Thus, using CFA high-resolution gas measurements at different sites to
constrain Holocene GADs at low-accumulation sites would be the only way to
check the consistency of the two methods. Previous comparisons between sites
indicate that the smoothing is larger for low-accumulation conditions
. Indeed, a simple argument is that the lower the
accumulation and the temperature, the slower a firn layer will densify, and
thus the broader the GAD. The comparison of the DO-17 records between WDC and
Vostok 4G-2 corroborates this relationship: the higher-accumulation WDC
signal is less smooth than the Vostok signal (Fig. ).
The weaker-than-expected smoothing during DO-17 at Vostok could be due to the
presence of a strong layering preventing air renewal and mixing, as suggested
in .
From a paleoclimatic point of view, an important conclusion of this work is
that the smoothing of atmospheric trace gases recorded in ice cores from the
central Antarctic plateau could be less than expected under glacial
conditions, resulting in more retrievable information about past atmospheric
conditions. Ice cores with the oldest enclosed gases, such as in the Oldest
Ice project , will be retrieved from very low-accumulation sites. They could thus potentially provide meaningful
information down to the multi-centennial scale.
Layered trapping and atmospheric trend reconstructions
The anomalous layers in the Vostok methane record discussed in
Sect. are 1 to a few centimeters thick, and discrete
samples used for methane measurements in ice cores are typically also a few
centimeters thick. In our study, the use of high-resolution continuous
analysis made it possible to identify abnormal methane values that appeared
as spikes in the record. However, in the case of discrete measurements, the
absence of continuous information makes it hard to discriminate between
normal and abnormal layers. For instance, the comparison of the WDC
continuous record and the EPICA Dome C (EDC) discrete methane record
indicates a potential artifact during the onset
of the Dansgaard–Oeschger event 8 (∼38000yrBP), as
displayed in Fig. . One of the EDC samples shows a reduced
methane concentration, which should be visible in the less smooth WDC record
as well if this corresponded to a true atmospheric feature. Moreover, the
measured mixing ratio in this EDC sample is consistent with an artifact
resulting from early gas trapping. As mentioned in
and confirmed by our study, it is important for paleoclimatic studies to
avoid interpreting such abnormal values as fast atmospheric events.
Discrete EDC methane record (blue) and continuous WDC methane record
(orange). The WDC record was put on the GICC05 timescale and then shifted
by 250 years to improve matching. We suggest that the circled point
corresponds to a layered trapping artifact.
However, continuous-flow analysis may not always allow us to distinguish
between layering artifacts and the chronologically ordered signal. The deep
parts of ice cores with low accumulation and high thinning are of particular
interest in paleoclimatology since they enclose very old gases
. However, with a strong thinning,
the width of abnormal layers may shrink below the spatial resolution limit of
analytical systems. In such a case, an average mixing ratio over several
layers is measured. Since layered trapping artifacts are unevenly distributed
in terms of sign, they bias the measured average signal. In the very simple
case of a record with artifacts that are all negatively orientated, cover 15%
of the ice core, and all reach 50ppbv, this bias is about -7ppbv. In the case of records with lower accumulation or stronger
methane variations the bias will be even more important. The development of
very high-resolution gas measurement techniques thus offers important
perspectives for analyzing the deepest part of ice cores. In intermediate
situations in which anomalous layers could be distinguished but a high-accumulation record is not available (before the last glacial–interglacial
cycle), the effect of smoothing is more difficult to constrain, but the
presence of layered trapping artifacts is in itself an indication that some
smoothing may occur because layered trapping occurs only under fast
atmospheric change conditions.
Conclusions
We presented the first very high-resolution record of methane in an ice core
sequence formed under very low-accumulation-rate conditions. It covers the
gas record of Dansgaard–Oeschger event 17, chosen for its abrupt atmospheric
methane changes on a similar timescale as gas trapping.
The continuous-flow analysis system, optimized to reduce gas mixing, allowed
us to reveal numerous centimeter-scale methane concentration anomalies.
Positive anomalies affecting both the methane and carbon monoxide records
were attributed to kerosene contamination and discarded. The remaining
anomalies are unevenly distributed, a few centimeters wide, and mostly
negatively oriented with dips as low as -50ppbv. The anomalies
occur only during time periods of fast atmospheric methane variability. The
main characteristics of the size and distribution of the anomalies could be
reproduced with a simple model based on relating realistic firn density
anomalies to early or (to a lesser extent) late trapping. Such layered
trapping anomalies may be confused with the climatic signal in discrete
climate records or bias the signal if too narrow to be detected by a CFA
system (e.g., under the high-thinning conditions of the deep part of ice
cores). It is important for future paleoclimatic studies not to interpret
those abrupt variations as fast chronologically ordered atmospheric
variations. Further use of high-resolution continuous analysis will allow us
to discriminate layered trapping artifacts and to better identify their
statistical characteristics. Moreover, the sign of the trapping artifacts is
not random: some sections of the record display only positive or negative
artifacts. Thus, simple averaging would result in a systematic bias of the
signal. Hence, we developed a cleaning algorithm aiming at minimizing this
bias.
After removing the centimeter-scale anomalies, the remaining Vostok methane
signal is distinctly smoother than the WDC record .
The snow accumulation rate being more than 1 order of magnitude higher at
WDC than at Vostok, the WDC signal contains higher-frequency features. The
comparison of the two signals opens the possibility to estimate gas age
distributions for conditions of the East Antarctic plateau during the last
glacial period, which have no modern analogue. For the DO-17 event at Vostok,
the resulting gas age distribution is narrower than expected from a
comparison with modern firns . It may
be due to an incorrect prediction of gas trapping by firn models and/or an
incorrect extrapolation of the firn behavior to very low-temperature and
low-accumulation conditions. The apparently similar smoothing at Vostok under
DO-17 and present conditions contradicts the expected primary effect of
temperature and accumulation rate: lower temperature and accumulation rates
induce a longer gas trapping duration and thus a stronger smoothing.
On the other hand, point out the lack of firn
layering representation in most firn models and conclude that firn layering
narrows gas age distribution in ice. From a paleoclimatic point of view, ice
cores with the lowest accumulations contain very old gases. The smoothing under glacial conditions being less
important than expected implies that
atmospheric information on a shorter timescale than previously expected might
be retrieved. However, similar measurements need to be performed on other low-accumulation records to confirm our results for different sites and/or
periods. For the DO-17 event at Vostok, multi-centennial atmospheric
variations are still accessible in the record. Further comparisons of high-
and low-accumulation records of the last glacial cycle will allow us to
better constrain the relationship between ice cores and atmospheric gas
signals, even with no modern analogue conditions.