Predicting future climate change depends on comprehensive climate models incorporating all forcings in the climatic system. To estimate the anthropogenic impact, a crucial aspect of climate modeling is the accurate representation of natural forcings including volcanism [e.g., Crowley, 2000]. Volcanic eruptions impact the climate by producing sulfuric acid (H2SO4) aerosols that alter the radiative properties of the atmosphere [Robock, 2000]. The sulfuric acid results from the rapid oxidation of sulfur dioxide (SO2) emitted into the atmosphere by a volcano. From the view of climatic impact, volcanic eruptions can be classified in two groups, tropospheric or stratospheric, depending on if the volcanic plume rises above the tropopause. A tropospheric eruption in general emits small amounts of SO2 and does not have a significant climatic impact as the sulfuric acid aerosols are contained in a relatively small geographic location and tend to be rapidly washed out from the atmosphere. In a stratospheric eruption, the sulfuric acid aerosol layer formed at stratospheric altitudes (e.g., approximately 14 km above the equator) may persist for up to several years while reflecting solar radiation and changing the energy balance of the atmospheric system. The result is usually a brief (a few months to a few years) cooling of the troposphere and Earth surface with amplitude depending upon the optical depth and residence time of the sulfuric acid aerosols, themselves depending upon the amount of SO2 (mass loading) injected into the stratosphere, the location of the volcano and the time of the year of the eruption.
 All volcanic sulfuric acid aerosols eventually fall out from the atmosphere and settle onto the Earth surface including the polar ice sheets. The volcanic signals resulting from the fallout can be detected and measured in glaciological archives, i.e., polar ice cores. Up to the present, the common method of reconstructing volcanic records consisted in measuring the amount of volcanic sulfate in the ice cores and calculating a stratospheric mass loading and the increase of the atmospheric optical depths as a result of the eruptions [Gao et al., 2007; Sato et al., 1993; Zielinski, 1995, 2000]. In general, a very large sulfate signal corresponds to a large stratospheric eruption with significant climatic impact. However, a major limit of estimating the climatic impact using this methodology is the lack of an objective means to determine if a detected signal represents a stratospheric or tropospheric eruption. For example, if a tropospheric eruption occurs in South America, the sulfate signal in Antarctica snow may be very strong owing to the proximity of the sampling site to the volcano and this could lead to the erroneous conclusion that the eruption was stratospheric and therefore had a significant climatic impact. Therefore, the magnitude of volcanic sulfate signals in ice cores alone does not allow the differentiation between tropospheric and stratospheric eruptions and, as a consequence, the accurate estimate of the climatic impact of past eruptions.
 In this work, we use anomalous sulfur isotopic compositions of volcanic sulfate from ice cores to provide new and independent information on the type of past volcanic eruptions that may have significantly impacted climate. Mass-dependent isotopic fractionation processes are governed by relative mass differences between the four sulfur isotopes and are described by δ33S = 1000*[(1 + δ34S/1000)0.515 − 1] and δ36S = 1000*[(1 + δ34S/1000)1.91 − 1]. The deviation from the two mass-dependent relationships is termed anomalous. The anomalous isotope composition is quantified by Δ33S = δ33S − 1000*[(1 + δ34S/1000)0.515 − 1] and Δ36S = δ36S − 1000*[(1 + δ34S/1000)1.91 − 1]. The sulfur isotope anomaly is created by UV photolysis on gas molecules such as sulfur dioxide (SO2) at wavelengths lower than 310 nm [Farquhar et al., 2000, 2001]. This source of sulfur isotope anomaly was further demonstrated by Savarino et al.  who showed that volcanic sulfate formed in the stratosphere and later deposited on the Antarctic ice sheet, acquired the anomalous signature as UV radiation lower than 310 nm is available only above the tropopause in the modern atmosphere. A direct implication of the previous work was that stratospheric eruptions recorded in ice cores could be characterized by their anomalous sulfur isotope composition and differentiated from tropospheric eruptions. Savarino et al.  studied two well-known stratospheric volcanic eruptions, Pinatubo (Philippines, June 1991) and the 1259 A.D. Unknown Event (the location of the volcano is unknown) in Antarctica snow and ice samples, and observed anomalous sulfur isotope compositions in both cases (Δ33S ≠ 0‰). In comparison, the essentially tropospheric eruption [Doiron et al., 1991] of the Cerro Hudson volcano (Chile, August 1991), does not exhibit any sulfur isotope anomaly. Evidence from several other recent studies also supports this hypothesis. Sulfate aerosols collected from the vents of the Masaya volcano (Nicaragua) are found to contain no sulfur isotope anomaly [Mather et al., 2006]. Bindeman et al.  analyzed sulfate from the 1991 Pinatubo eruption, in gypsum samples taken near the volcano (Luzon, Philippines), and did not detect any sulfur isotope anomaly. These results can be explained by the fact that in each case the sulfur did not reach the stratosphere and therefore the sulfate formed near the volcano was probably tropospheric or biological-made. Although small sulfur isotope anomalies can be generated by mass-dependent processes [Ono et al., 2006a; Rouxel et al., 2008], it is unlikely that such processes are responsible for our observed sulfur anomaly owing to the very low levels of biological activities on the Antarctic ice sheet. Likewise, nonvolcanic sulfate formed in the troposphere and found in the snow or soil of Antarctica does not show any sulfur isotope anomaly [Alexander et al., 2003; Baroni et al., 2007; Romero and Thiemens, 2003; Savarino et al., 2003].
 Following the study by Savarino et al.  on two stratospheric and one tropospheric volcanic eruptions, Pavlov et al.  proposed a model to explain the origin of sulfur isotope anomaly detected in volcanic sulfate in Antarctic snow. Their model involves a dynamic process in which the sulfate formed at the beginning of the conversion of the volcanic SO2 carries an anomalous signature different from that in subsequently formed sulfate during the same volcanic event. This suggests that the Δ33S values would change gradually with time during the sulfate deposition following a volcanic eruption. To test this hypothesis, a study of the Pinatubo (Philippines, June 1991) and the Agung (Indonesia, March 1963) stratospheric eruptions was undertaken with a high time resolution sampling of the volcanic sulfate in Antarctic snow [Baroni et al., 2007]. In both cases, the sulfur isotope anomaly of the volcanic sulfate is found to change from positive Δ33S values at the beginning of the sulfate deposition (∼ 1‰) to negative values at the end (∼ −1‰). These results and the Pavlov model indicate that both photochemistry and atmospheric dynamics are involved in producing sulfur isotope anomaly in the sulfate of stratospheric volcanic eruptions, although different photochemical reaction mechanisms are proposed by Pavlov et al.  and Baroni et al.  to explain the formation of the sulfur isotope anomaly. Both of our previous studies [Baroni et al., 2007; Savarino et al., 2003] were based on only a few (2 or 3) volcanic eruptions in Antarctic snow. In this work, we use the methodology developed and knowledge gained in the previous studies to investigate ten volcanic events found in Antarctic ice cores. The main objectives of this study are (1) to identify stratospheric eruptions over the last 1000 years and (2) to determine the advantages and the limits of the sulfur isotope anomaly technique when applied to ice core volcanic records. Owing to the limited volcanic sulfate mass available in our ice core samples, a high time resolution sampling similar to that used by Baroni et al.  was impossible. Therefore, the approach here is similar to that of [Savarino et al., 2003]; that is, each volcanic event had to be sampled and analyzed singularly.