1. Introduction
Studying impurities in ice is essential for current research in glaciology. Issues that have to be addressed for analysis of ice cores are the fragility of the ice material, its evolving structure and the low concentrations of impurities.
Analysis of melted samples by ion chromatography gives the average ionic concentrations of soluble (mineral and organic) species, even at very low concentrations. Such information is used for studying past atmospheric composition, for example. However, due to the possible interaction between chemical impurities and physical properties it is of primary interest to know the microstructural localization of impurities. Depending on their localization, either dispersed in the ice matrix or concentrated at grain boundaries, the structure of the sample may evolve from its original in situ state. For example, reactions between soluble sulphuric acid and carbonate particulates may alter the record of atmospheric CO2 present in air bubbles trapped in the ice (Reference Legrand and MayewskiLegrand and Mayewski, 1997).
Several analytical methods have already been successfully developed to locate impurities in ice (Reference Baker and CullenBaker and others, 2003) or to study their effect on properties of ice (Reference Wolff, Mulvaney and OatesWolff and Mulvaney, 1988; Reference Iliescu, Baker and LiIliescu and others, 2003). Acidic ions at a triple junction of Antarctic ice have been studied by the Raman scattering technique (Reference Fukazawa, Sugiyama, Mae, Narita and HondohFukazawa and others, 1998). Reference Trickett, Baker and PradhanTrickett and others (2000) studied the effect of H2SO4 on the mechanical properties of ice by synchrotron X-ray topography. The major technique implemented to observe impurities is scanning electron microscopy with energy-dispersive X-ray microanalysis. Reference Mulvaney, Wolff and OatesMulvaney and others (1988) found sulphuric acid at the grain boundaries in Antarctic ice. In the last few years, two teams have successfully developed this technique (Reference Barnes, Mulvaney, Wolff and RobinsonBarnes and others, 2002, Reference Barnes, Wolff, Mallard and Mader2003; Reference Cullen and BakerCullen and Baker, 2002).
Whilst sublimation concentrates impurities and facilitates detection, it may also introduce artefacts: for example filaments observed in the grain boundaries of polar ice are due to the sublimation of the surrounding ice (Reference Baker, Cullen and IliescuBaker and Cullen, 2003). Developing a new experimental method that enables mapping and analysis of impurities in a nondestructive way is seen as valuable for research in glaciology. To this end we looked at the X-ray microscopy analysis technique developed at the ID21 beamline at the European Synchrotron Radiation Facility (ESRF) in Grenoble, France (Reference Susini, Salome, Fayard, Ortega and KaulichSusini and others, 2002). This technique enables hydrated frozen samples to be investigated without pretreatment. It allows the study of thick samples, and, in the high-energy range produced by this technique, fluorescence yield is high enough to be mapped with significant contrast and to identify low Z elements. With these capabilities the micro-X-ray fluorescence (μXRF) technique appears a suitable basis for developing a new experimental method for mapping and analyzing impurities in a non-invasive way. In this paper, we report the technical and methodological adaptations implemented, the validation of the technique and the first results obtained.
2. Experimental Procedure
2.1. Equipment
An analysis chamber (internal part shown in Fig. 1a) is closed by a tight door, as a high vacuum (~10–5mbar) is necessary to lower the detection limit and to have access to low Z elements.
(a) Internal part of the analysis chamber; a: Dewar flask; b: copper cylinder; c: copper tress; d: thermocouple; e: support of piezometer electric scanner. (b) The sample cell and its support; a: the front window; b: the sample cell support; c: Peek®; d: screw; e: magnets. (c) Details of the sample cell.
To preserve the ice sample from any transformation or sublimation, it must be maintained at low temperature (~120K during analysis). The cooling system consists of a special 3 l Dewar flask (‘a’ in Fig. 1a), automatically filled with liquid nitrogen (LN2) and hermetically fixed to the roof of the chamber by a DN100 flange. The base of the Dewar flask is made of a thick copper cylinder (‘b’ in Fig. 1a) coupled to a magnetized copper plate by a flexible copper tress (Fig. 1c and ‘c’ in Fig. 1a), the sample cell being fixed to this plate by magnets. This maintains the temperature of the ice sample near to the same low temperature as the bottom of the Dewar flask. A thermocouple is fixed on the copper plate to measure the temperature during experiments (‘d’ in Fig. 1a). The upper part of Figure 1b shows the front window of the cell in which the sample is set up. The window is a transparent film of Ultralene®. Ultralene® was selected for its good X-ray transmission, its strength and purity. The film is pressed between the lid and the sample cell (Fig. 1c) with a Viton O-ring to ensure it is airtight. The sample cell is thermally isolated from its support by Peek® pieces (‘c’ in Fig. 1b). The sample cell and its support are made of aluminium. Magnets (‘e’ in Fig. 1b) fix the sample cell and the support into the piezometer-based scanner (Fig. 1a). This piezometer-based scanner moves the support of the sample cell by steps along the x and y directions. Steps are adjustable according to the lateral resolution desired. To obtain a scan rapidly, the total mass of the sample cell device associated with the piezometer has to be <100 g. Hence the choice of aluminium for the sample cell device.
2.2. Sample selection
Artificial samples
In order to estimate the contamination introduced by the sample preparation and to evaluate the background signal of the measurement procedure, one artificial ice sample (S1) was prepared from ultra-pure water.
Natural samples
Samples should be selected that provide the opportunity to evaluate the technique regarding, in particular, localization of impurities, analysis of large aggregates and speciation of elements. Two technical constraints must be dealt with: (i) the μXRF detection limit of the instrument, expected to be about 1 ppm, required samples with a significant amount of impurities; (ii) the small size of the sample cell (diameter only 11 mm) required ice with small crystals, in order to observe several grain boundaries. Accordingly, European Project for Ice Coring in Antarctica (EPICA) Dome C samples, from the Last Glacial Maximum, at 566 m depth, were selected. In addition, samples of accreted ice from Vostok lake, Antarctica at about 3570 m depth were selected to analyze aggregates.
2.3. Sample preparation
Samples were prepared at Laboratoire de Glaciologie et Géophysique de l’Environnement, Grenoble, France, in a cold room at 258 K. To prevent contamination, every tool was new and washed with methanol. Ice cores were cut into samples 7 cm long. Each sample was machined to a cylinder with the same diameter as the sample cell. The surface of each cylinder was adjusted with a microtome to exactly fit in the cell and to be flush with the cell surface on which the Ultralene® film was spread. Once the sample cell was sealed with screws, it was put into a polycarbonate box to avoid contamination of the Ultralene® film. Samples were then transferred to the ESRF at a temperature of 235 K and cooled to the temperature of LN2 just before introduction into the analysis chamber.
2.4. Sample set up in the analysis chamber
It is critical that water vapour be prevented from entering the chamber, as this could turn into hoar on the cell window. For this reason, the chamber was put under a vacuum (~10–5mbar). The Dewar flask was then filled with LN2. Once the temperature of the copper plate was lowered to 113 K, the analysis chamber was brought back to atmospheric pressure with dry N2. Outside the chamber, at the level of the top of the door, a laminar flow of dry N2 was blown downwards to keep humidity low in the chamber at the door aperture. The door was opened, and the sample box removed from the LN2, introduced into the chamber and opened. The sample cell and its support (Fig. 1b) were placed in the X-ray beam and fixed by magnets. The copper plate and its tress were put in contact with the sample cell, and the door was closed. The vacuum was re-pumped and the temperature became stable at 123 K after 1.5 hours. The door was left open <1 min. The maximum temperature observed for the sample cell during set-up was 203 K. There was no hoar in the chamber except for some traces on the copper tress.
2.5. Analysis
Initially, the surface of the polar ice sample was observed using an optical microscope, in order to check the absence of hoar on the Ultralene® film, verify that the ice was intact and identify grain boundaries. Once a grain boundary or an interesting spot was selected, the ice sample was examined using the X-ray microscopy beamline ID21 (Reference Susini, Salome, Fayard, Ortega and KaulichSusini and others, 2002). The X-ray beam was focused down to a microprobe of 0.5 × 0.5 μm2 by means of a Fresnel zone plate (Reference Di FabrizioDi Fabrizio and others, 1999). Two Si (111) crystals provided a monochromatic beam (DE/E = 10–4). The fluorescence signal was analyzed using a high energy resolution germanium solid-state detector. The incident beam energy was fixed at 4500eV to ensure a good fluorescence yield for elements ranging from calcium to sulphur. For sulphur speciation, the excitation beam was tuned at 2482 eV. For these configurations, 4500 and 2482 eV, the probed depth (z axis) is 50 and 25 μm respectively. The lateral resolution, defined by the geometric projection of the focused beam onto the sample (x and y axes), was adapted to the size of the image and ranged from 3 × 3μm2 to 0.5 × 0.5μm2. The integration time, tuned from 100mspixeΓ1 and 5spixel–1, was determined by both the concentration of the emitting elements and the signal-to-noise ratio of the fluorescence signal.
Ice adjacent to the samples was analyzed by ion chromatography, in order to establish the average concentrations of the main ionic species of the melted ice (Table 1) (Reference Saigne, Kirchner and LegrandSaigne and others, 1987). The concentrations ranged from 0.13 to 938ngg–1.
Concentrations (ngg 1) measured by ion chromatography for artificial sample (S1), Dome C (DC1) and Vostok samples (V1, V2)
3. Results and Discussion
3.1. Technique validation
Control of the experimental procedure and background estimation
To estimate the contamination introduced by the sample preparation and to test the influence of the Ultralene® film, two XRF spectra (Fig. 2) were taken under the same experimental conditions as for the polar ice samples: one using a sample cell filled with air and the other using an artificial sample (S1), which was also analyzed by ion chromatography (see Table 1). To take account of the variations in the incoming intensity (Io) during experiments, all the figures presented in this paper show the relative intensity, i.e. the ratio (I/Io) of the emitted signal (I) to the incoming intensity. The results for both the air-filled cell and S1 (Table 2) are similar for the chloride peak. Since Cl was not detected by ion chromatography of S1, the Ultralene® film, which is very close to the sample surface, is the most likely source of the Cl peak.
μXRF spectra from the sample cell filled with air and from the artificial sample (S1). The spectrum from the sample cell filled with air is characterized by a high peak from argon. This result shows the presence of air in the sample cell.
μXRF results from test samples (artificial ice (S1) and sample cell filled with air)
Regarding elements other than Cl (Table 2), the Al peak was lower for S1 than for the cell filled with air. This may be due to the partial absorption of the cell signal by the S1 sample. It was decided that the μXRF signal for an element in S1 would be taken as the background signal. To consider an element as present in polar ice, the μXRF signal should be more than two times higher than the element’s signal in S1.
Sample conservation during the experiment
The major risk of transformation of samples is sublimation. Sublimation can be due to a lack of closure of the sample cell in the vacuum chamber. Figure 2 shows the μXRF spectrum for the sample cell filled with air;there is a high peak for argon. Although present as traces in air, Ar has been detected due to its atomic number (18) which explains its good fluorescence response. This Ar peak confirms that the sample cell is airtight.
To check both the stability of the sample and the reproducibility of the technique, an inclusion containing Si in the Vostok sample at 3572 m depth (V2) has been examined three times at 2.5 hour intervals. The μXRF spectra (only Si Ka line relative intensity) and shape of the Si peaks along the cross-section of the inclusion are shown in Figure 3a and b. Whilst the shape of the inclusion is unchanged (Fig. 3b), a slight decrease of I/Io is observed (Fig. 3a). This suggests a slow evolution of the sample. The decrease is limited to about 20% after >7 hours. This confirms the experiment is non-destructive and can be reproduced.
(a) μXRF spectra (only Si Ka line shown) from an inclusion observed in sample V2 (Vostok at 3572 m) for different radiation exposure times (2.5, 5 and 7.5 hours). (b) Relative-intensity profiles of the Si Ka line along the cross-section of the inclusion. The absence of significant variations confirms the preservation of the sample, even after long exposure times.
Several experiments were carried out on the same part of an inclusion containing Cl in the Vostok sample at 3572 m depth (V1). From optical microscopy, this inclusion appears like a spherical grain (2 μm). Table 3 shows the stability of I/Io for Cl in this inclusion after use of different integration times. This stability means that integration times can be increased (at least by up to 5 s) to improve the sensitivity of the technique without causing elemental repartition. Between each scan, optical microscopy showed that the inclusion was unchanged and that the surface of the sample was not altered by the radiation from the microscope.
Evolution of Cl relative intensity in an aggregate for different integration times
3.2. First results
Results from DC1 at 566 m depth
The observation of sample DC1 by optical microscopy shows a homogeneous surface, apart from the grain boundary. The total area mapped by μXRF was 200 × 188 μm2, with a pixel size of 2.6 (x) x 2.6 (y) μm2, an integration time of 3.3 s and an excitation energy of 4500eV. Note that the signal includes information from a thickness of about 50 μm below each pixel. The potassium map (Fig. 4) shows a strong concentration of K along the grain boundary. This result is observed under the non-destructive conditions of the technique. This can be explained by the concept that K, which in most cases is in an easily soluble form, should be preferentially found along the grain boundaries or associated with insoluble particles. Other elements, such as Si, Al, S and Ca, were also detected. However, instead of being, like K, highly concentrated in a specific area, they are either concentrated only on a few points or distributed in a homogeneous way throughout the mapped area. As a result, their maps show either a few dark points or a grey area without contrast. This is why, instead of their maps, we show their relative intensity profiles in three selected regions of the map identified as Z1, Z2 and Z3 (Fig. 5).
Map of fluorescence yield of potassium (DC1 sample). Pixel size 2.6 × 2.6μm2 integration time 3.3 s. Z1 scan refers to the A particle, Z2 to B, Z3 to C. K is clearly found along the grain boundaries (the Y shape on the right of the map).
Relative intensity profiles (Ca, K, S, Si, Al Ka line) along cross-sections Z1, Z2 and Z3 (see Fig. 4; gb indicates the grain boundary). Si and Al are highly concentrated on two specific points (A and B). Si and Al could be within clay particles and associated with Ca (A, B). S is associated with Ca in one highly concentrated spot (A).
The I/Io signals for Si, Al, Ca and K are < 0.04 except from specific spots. For example, Ca is present specifically at points A and B close to the triple junction (Figs 4 and 5) and point C. Except for a very intense point (A in Fig. 4; I/Io ~ 16), the signal for S (Fig. 5; Z2, Z3) is about four times higher than the I/I o signal for the other elements. The origin of S is likely to be H2SO4 throughout the matrix. The intense point may originate from an insoluble particle of CaSO4. Owing to the long integration time of 3.3 s, and in spite of the sensitivity for Si and Al being lower than for Ca, it is noteworthy that Si and Al were detected at points A–C.
Results from Vostok samples V1 and V2
The V1 sample (3551 m depth) shows very large crystals (Reference Duval, Lipenkov, Barkov and De La Chapelle.Duval and others, 1998; Reference JouzelJouzel and others, 1999), with some inclusions visible by optical microscopy and a large aggregate visible to the naked eye (Fig. 6a). Such an aggregate, not observed in normal glacier ice, is formed by the concentration and compression of very small particles locally accumulated during the accretion of ice. The gas around the aggregate was initially at ~300 bar, making the aggregate very unstable. To make use of the non-destructive character of the technique, three adjacent elemental maps were drawn. The area of each map was 100 × 100 μm2, with a pixel size of 1 (x) × 1 (y) μm2 and an integration time of 0.5 s, resulting in 1.5 hours per map. Note that the signal includes information from a thickness of about 50 μm below each pixel.
(a) Aggregate observed on the V1 sample (Vostok at 3551 m). The lighter regions (visible at the periphery) may be gas bubbles, while the dark regions look like the solid part of the aggregate. The rectangular region indicated was analyzed by μXRF. (b) Results of the analysis: maps of Ca and K distributed on the aggregate (image size 300 × 100 μm2; pixel size: 1 × 1 μm2; integration time 0.5 s). Ca is distributed as large particles. Their shapes and compositions are different inside (circles) and outside (dashed circle) the aggregate. Ca seems anticorrelated with K (white circles). (c) μXRF spectra from the upper particles inside and outside the aggregate (circled in (b)).
Figure 6b shows the Ca map distributed as large particles of ~15 μm, with their shapes and spectra being different inside and outside the aggregate (Fig. 6c). Inside the aggregate, Ca appears to be anticorrelated with K (Fig. 6b). Ca predominates outside the aggregate, probably within one CaCO3 inclusion (dashed circle). In case the aggregate might disappear due to its unstable state, several rapid scans (<1.5 hours each) were made to capture the maximum information. S was observed, concentrated at the border and probably associated with Al, although the sensitivity for Al was low. Work is being carried out to confirm and refine these results (Reference de Angelis, Morel Fourcade, Barnola, Susini and Duvalde Angelis and others, 2005).
Apart from the one large aggregate, some further inclusions were observed by optical microscopy, and then scanned. The signal intensity (I) for each element as well as its ratio (I/Io) to the intensity measured on the artificial sample S1 are reported in Table 4. Results for V1 are not significantly different than for S1. As ion chromatography of V1 had shown high concentrations of the species of interest (Table 1), this confirmed that the matrix was pure (de Reference de Angelis, Petit, Savarino, Souchez and ThiemensAngelis and others, 2004) and suggests that all impurities were strongly concentrated in inclusions. S was mainly found in the large aggregate, while Cl was in the inclusions in the form of NaCl.
Relative intensities for various elements measured in natural samples
It should be noted that, by assessing the number of inclusions, their volumes and the volume explored in the map, a rough evaluation of the NaCl concentration in the inclusions may be derived from the average concentration measured by ion chromatography.
The V2 sample (3572 m depth) was used to test the potential of the new technique to obtain speciation by selecting the energy of the incident beam according to the oxidation number of the target element. Absorption spectroscopy techniques are based on the measurement of the absorption coefficient of variation by tuning the energy of the probing photons through an absorption edge. This variation is physically related to the excitation cross-section of the core electrons into unoccupied electronic states or into vacuum continua. The spectral features observed close to the absorption edge – referred to as X-ray absorption nearedge structure (XANES) – reflect the molecular environment of a given absorbing atom and provide the basic mechanism for imaging with chemical sensitivity (Reference Stöhr and SpectroscopyStöhr, 1992). Obtaining information on different chemical states within systems having the same elemental composition is possible (Reference Cuif, Dauphin, Doucet, Salome and SusiniCuif and others, 2003).
An observation was made of a spot showing a particle including Si, Al, P and S (Fig. 7). The size of the image is 50 × 40 μm2, with a pixel size of 0.5 (x) x 0.5 (y), and an integration time of 0.5 s. Note that the signal includes information from a thickness of about 25 μm below each pixel. The size of the particle is 20 × 13.5 μm2. Figure 7a–c, respectively, show oxidized sulphur using an excitation energy of 2.482 keV, reduced sulphur using an excitation energy of 2.473 keV, and their different localizations. By supplying cartographies of elements for their different oxidation numbers, the technique yields information that can help to determine the sources of the impurities.
(a) Map of S (oxidized form) on the inclusion observed in sample V2 (Vostok at 3572 m) using 2.482 keV, image size 50 × 40 μm2, pixel size 0.5 × 0.5 μm2, integration time 0.5 s. (The dashed line indicates the cross-section analyzed in (c).) (b) Map of S (reduced form) on the inclusion observed in sample V2 using 2.473 keV, image size 50 × 40 μm2, pixel size: 0.5 × 0.5 μm2, integration time 0.5 s. (The dashed line indicates the cross-section analyzed in (c).) (c) Relative intensity profiles from the oxidized and reduced forms of sulphur along the cross-section.
Conclusions
μXRF analyses of polar ice samples were carried out after adaptations of both equipment and methodology. Results obtained from Dome C and Vostok samples are the first results obtained from the technique. The thermal control system keeps ice samples unchanged during analysis. Integration times can be increased without alteration of samples. Sensitivity can be optimized by increasing integration times, and for a given beam energy the set of identified elements is expanded. For a given time window, optimization results from a trade-off between integration time and resolution. For the selected samples, impurities were not altered by the analyses. As the elements within impurities were kept highly concentrated, they were detectable, although their average concentrations, measured on melted samples by ion chromatography, are far below the detection limit of μXRF.
By selecting integration time, resolution and size of the area to be mapped, analysis times can be shortened to about 1 hour which enables us to scrutinize unstable impurities. Being non-destructive, this μXRF technique makes it possible to reproduce experiments to confirm or refine results.
Beyond the detection of elements within impurities, the technique enables their speciation, which may help with finding the origins of impurities. In conjunction with the technical adaptations, the use of artificial ice samples was included in the analysis process. By measuring background μXRF signals from artificial samples, it was possible to detect artefacts or contamination due to experimental conditions. Combined with the reproducibility of experiments, this brings reliability to the results and to the related forthcoming glaciological interpretations.
Acknowledgements
We thank the ESRF for providing the X-ray facilities ME481 and CH1568, and the EPICA and Vostok programmes for ice samples. We thank M. Salome´ for assistance on beamline ID21, R. Pieritz for help during the experiment and C. David, T. Weitkamp and E. Di Fabrizio for providing zone plates. We thank S. Nagorski for improving the English. We also thank Centre National de la Recherche Scientifique, France/Institut National des Sciences de l’Univers (CNRS/INSU) (Groupement de Recherche Européen (GDRE) Vostok) for supporting this work.
