The first scientific work in the study area was conducted during the British–Norwegian–Swedish Antarctic Expedition in 1949–1952, in which detailed descriptions of both the geology and morphology of the Jutulstraumen basin were made, including the ice sheet and ice shelf (Swithinbank, 1957). Work in the area was continued during the International Geophysical Year (IGY) 1956/57, at the Norway Station (later renamed SANAE), on the western edge of the ice shelf between 1956 and 1960 (Lunde, 1961; Neethling, 1970). In the last three decades Norwegian groups have worked on FIS and Jutulstraumen under the auspices of the Norwegian Antarctic Research Expedition (NARE), focusing on spatial and temporal variability of SMB using shallow firn cores and Ground Penetrating Radar (GPR) (e.g. Melvold et al., 1998; Melvold, 1999; Isaksson and Melvold, 2002; Sinisalo et al., 2013; Schlosser et al., 2012, 2014).

As part of the EPICA project, a 100 m-deep ice core (labelled S100) was drilled on the eastern part of FIS during NARE 2000/01 (Fig. 1). This core covers the period AD 1737–2000 ± 3 years and shows higher SMB values in the 19th century than in the 18th and 20th centuries, but otherwise no significant trends (Kaczmarska et al., 2004). This core is the longest available high-resolution climate record from this part of coastal DML.

Rotschky et al. (2007) compiled a SMB map for western DML, including FIS, but data were not available to resolve fine-scale variability in the area of the ice rises. More recently, Sinisalo et al. (2013) and Lenaerts et al. (2014) used field and model data to show that the ice rises have a substantial role in shaping both local SMB and meteorological conditions. Finally, Altnau et al. (2015) compile available oxygen stable isotopes and SMB data for the last three decades; they find a negative SMB trend for the coastal regions, but a positive trend on the polar plateau over the same time period. They conclude that atmospheric dynamic effects are more important at the coast than thermodynamics, the latter being the dominant factor on the polar plateau, where changes in SMB and stable isotope ratios occur mostly in parallel.

Three shallow (ca. 20 m) firn cores were retrieved at FIS (Fig. 1, Table 1) in January 2012 (KC) and January 2014 (KM and BI) during field expeditions organised by the Norwegian Polar Institute (NPI). Table 1 presents the location of the drill sites, maximum elevation of the ice rises, and recovered core lengths. Each core was drilled from the bottom of a 2 m snow pit; the pit wall was sampled at 5 cm intervals for water stable isotope analysis. Bulk core density was determined for each subcore piece (average length ∼ 45 cm) and for each snow pit sample (20 cm). Snow and firn samples were collected following clean protocols (Twickler and Whitlow, 1997), shipped frozen to NPI, and later to the Paul Scherrer Institute (PSI), Switzerland, for cutting and chemical analysis. Sample length ranged from 4 to 8 cm, depending on the sample depth and density. The presence and thickness of ice lenses were recorded during cold room analysis of the KC core. Major ions (methane sulfonic acid, MSA), SO42-, and Na+) were analysed at PSI. Subsamples for water stable isotopes analysis were shipped to the Institute of Geology at Tallinn University of Technology (TUT), Estonia.

Water stable isotope ratios (18O / 16O and 2H / 1H) were measured at TUT using a Picarro L2120-i water isotope analyser (cavity ring-down spectroscopy technology) with a high-precision A0211 vaporiser. Measurements were calibrated against both the Vienna Standard Mean Ocean Water (VSMOW) and the Vienna Standard Light Antarctic Precipitation (VSLAP) standards. Reproducibility of δ18O and δ2H measurements was ±0.1 and ±1 ‰ (for 4–6 replicate measurements), respectively. Measuring both oxygen and hydrogen water stable isotopes in the ice rises cores yields deuterium excess (d=δ2H-8δ18O).

Major ions (MSA, SO42-, and Na+, Table 2) were analysed at PSI using a Metrohm ProfIC 850 ion chromatography system combined with an 872 Extension Module and autosampler. The precision of the method was around 5 % for all ions (Wendl et al., 2015). In this study we use records of major ions Na+ and MSA to corroborate the dating of the three cores (see Sect. 4.1), performed by identifying seasonal cycles in the oxygen isotope record; a detailed palaeoenvironmental analysis at the ice rise sites using the ion data is the subject of a separate paper in progress.

Due to higher accumulation rates, the δ18O seasonal variability in the KM and BI cores is better defined than at the more inland KC site (Fig. 2), and dating uncertainty for the KM and BI cores is, therefore, lower. Dating of the firn cores is performed by annual layer counting, using the seasonality of the water stable isotope signal. Since the KM and BI cores were drilled from the bottom of a snow pit (0.9 mw.e.), the snow pit data are used to reconstruct the period between winter 2012 and summer 2014. Winter minima and summer maxima in the δ18O record are identified to obtain a timescale with subannual resolution. Assuming a uniform distribution of precipitation throughout the year, an equidistant timescale is adapted between the summer maxima (January) and the winter minima (July). Well-pronounced seasonal cycles of major ion concentrations (e.g. MSA and Na+, Fig. 3) are used to corroborate the dating. Based on annual layer counting, the KM and BI cores cover the periods from winter 1995 to summer 2014 and winter 1996 to summer 2014, respectively. The error in the dating is estimated as ±1 year for both of these cores.

Counting δ18O winter minima in the KC core is not as straightforward as for the KM and BI cores, due to the lower accumulation and the lower amplitude of the seasonal signal (Fig. 2a). Using δ18O snow pit data available for the surface layers at KC (Fig. 2a), a SMB of 0.19 mw.e.yr-1 is estimated for the period 2007–2011. Accordingly, when interpreting the seasonal variability of the δ18O stratigraphy, this mean SMB 2007–2011 value was considered a guideline. Counting the δ18O winter minima in the deeper section of the core suggested 1958 as the date for the bottom of the core (12.93 mw.e.). Using the identified winter minima, we can identify tentative depths for summer maxima (Fig. 2a). Most of the horizontal dashed lines in Fig. 2a coincide with maxima in the δ18O, indicating a good estimate of the annual cycle using winter minima and SMB 2007–2011 as a reference.

Furthermore, volcanic horizons are used to corroborate the dating and estimate the dating uncertainty. We use maxima (values above the mean + 2σ level) in the non-sea-salt sulfate (nssSO42-) concentrations of the KC core to identify volcanic horizons (Fig. 4). NssSO42- was calculated from the mean seawater composition using Na+ as standard ion. nssSO42-=SO42-total-k×Na+total, where k=SO42-seawaterNa+seawater=0.06, when ion concentrations are in µmolL-1. Peaks in nssSO42- are assigned to known volcanic eruptions that could alter snow composition at the drilling site, using the Volcanic Explosivity Index (VEI) (Global Volcanism Program, Smithsonian National Museum of Natural History, http://www.volcano.si.edu/). The VEI is a relative measure of the explosiveness of a volcanic eruption based on the volume of released material, plume height, and qualitative remarks (e.g. by rating an eruption from gentle to mega-colossal); these parameters are used to construct an open-ended logarithmic scale starting with VEI = 0. Only eruptions with VEI ≥ 3 were considered in this analysis (Table 3). We attribute the nssSO42- peaks in the KC core at depths of 3.8 and 4 mw.e. to the Pinatubo volcanic eruption (1991) (peak 1a, Fig. 4, Table 3) and Cerro Hudson (peak 1, Fig. 4, Table 3) in agreement with Cole-Dai et al. (1997). These depths correspond to late 1993 and late 1992 from the annual layer counting timescale, but delays of 1–2 years between eruption and deposition are commonly observed; the Pinatubo signal has been reported in the 1993 Antarctic snow layer (Cole-Dai et al., 1997). Another nssSO42- peak found at 6.41 mw.e. could originate from the El Chichón volcanic eruption (1982), at a depth corresponding to the year 1983 based on the annual layer counting. Both, Pinatubo and El Chichón volcanic horizons have been previously identified in dielectric profiles of other cores drilled in the region (Schlosser et al., 2012, 2014). Additional volcanic eruptions potentially observed in the nssSO42- record and SMB between potential volcanic horizons are listed in Table 3, e.g. Agung (peak 5). The error in the dating by annual layer counting is conservatively estimated to be ±3 years, based on the maximum difference between the Pinatubo volcanic signals found in the nssSO42- record (i.e. peak 1a, Table 3) and the eruption date. The timescale error could be reduced by complementary dating methods, such as counting annual cycles of chemical species (e.g. MSA and Na+), and 3H measurements for future cores drilled at these sites.

Annual SMB in the cores was calculated from distances between summer maxima in the δ18O record (Fig. 5, Table 1). The average annual SMB for the full period covered by the KC, KM, and BI cores is estimated to be 0.24, 0.68, and 0.70 mw.e., and the average SMB for the common period covered by all three cores (1996–2012), is 0.21, 0.70, and 0.71 mw.e., respectively. The lowest inferred annual SMB values at KC, KM, and BI were 0.11 mw.e.yr-1 (1986), 0.39 mw.e.yr-1 (2005), and 0.40 mw.e.yr-1 (2004), respectively, while the highest values were 0.45 mw.e.yr-1 (1982), 0.95 mw.e.yr-1 (2011), and 1.21 mw.e.yr-1 (2011).

These SMB values generally agree with other estimates at FIS (Sinisalo et al., 2013; Schlosser et al., 2014) obtained from shallow cores (1983–2009) and stakes (2010–2012). Furthermore, the anomalously high snowfall in DML during 2009 and 2011, recorded by GRACE satellite data (Boening et al., 2012) and stake data at FIS (Sinisalo et al., 2013), appear to be reflected in the SMB records of KM and BI (KM: 0.78 and 0.95 mw.e.; BI: 1.00 and 1.21 mw.e., in 2009 and 2011, respectively) (Fig. 5).

SMB derived from the stake closest to each core site (40 m to 1 km) at the three ice rises in 2013 are similar to average SMB values from cores at KC (0.22 and 0.24 mw.e.yr-1 from the stake and core data, respectively) and BI (0.73 and 0.70 mw.e.yr-1), but differ at KM (0.38 vs. 0.68 mw.e.yr-1). Differences in point estimates for single years are to be expected, given the spatial variability of snow accumulation. The spatial variability of SMB on the ice rises from stake and GPR data will be presented elsewhere.

In all three cores, there are ice layers of varying thickness, indicating that melt occurs several times per year. We have no evidence, however, for mass transport between annual layers. Figure 6 shows the number of ice lenses and thickness related to density, δ18O, MSA, and nssSO42- concentrations in the KC core. There is no direct correspondence between SMB, δ18O, and the ice layers in the core from KC (Fig. 6), in agreement with previous results from the core S100 (Kaczmarska et al., 2006). We compare melt features found in the KC core with the MSA and nssSO42- profiles, but do not find a systematic association between ice lenses and anomalies in the MSA or nssSO42- concentrations, as we could expect from redistribution of ions by meltwater percolation and refreezing. Some correspondence exists between the thickest ice layers and peaks in the nssSO42- record (e.g. at 21, 20, 18, and 13 m, Fig. 6) but there are no such peaks in the MSA record, as would be expected for an ion that it is just as readily eluted as nssSO42-. Therefore, while redistribution of ions by meltwater cannot be ruled out, it is not likely a dominant post-depositional effect that would significantly influence the seasonal isotopic or chemical signals at the core sites. It is more likely that the development of ice lenses is a local process depending on several factors, including air and snow pack temperatures, and that the combination of post-depositional processes, such as wind scouring, contribute to the perturbation of the subannual signal in the KC core site.

In general, SMB at the sites closest to the coast, KM and BI, is higher than at KC. The topography of the individual ice rises is a key determining factor. While KM and BI are relatively symmetrical domes, KC is more elongated, with a ridge axis stretching from SW to NE (Fig. 1). Therefore, air transported from the NNE during a precipitation event is lifted over a longer and gentler slope at KC than at KM and BI, which can lead to a weaker influence of topography than on the steeper slopes of KM and BI. Wind redistribution is critical for accumulation patterns. Networks of stake measurements across KC and KM show an uneven snow distribution, with 3-fold higher accumulation on the lowest-elevation upwind side, compared to the summit (Lenaerts et al., 2014). This spatial pattern is well replicated with the regional atmospheric climate model RACMO2, although an accurate DEM is critical in such comparisons. Our results suggest that the differences in accumulation at KM and BI compared to KC and the other core sites at FIS, are most likely related to topographical effects. This can be further explored by referring to the study by Altnau et al. (2015) which presents a vast coverage of SMB and δ18O for coastal and inland DML. By inspecting Fig. 2 in Altnau et al. (2015), it can be observed that high annual SMB values, similar to those measured at the KM and BI sites, occur in locations associated with pronounced topographic features, e.g. mountain ranges and troughs, i.e. anything where orographic lift may induce precipitation in comparison to the flat areas in the proximities.

Previous studies from coastal sites in the same area of DML have reported large temporal and spatial SMB variability (Melvold, 1999; Kaczmarska et al., 2004; Schlosser et al., 2014). The SMB records from the KM and BI cores reveal high interannual variability and no significant long-term trend during the period 1995(1996)–2014. At the more inland KC site, SMB variability is lower, but there is also a weak, yet significant (at 95 % confidence level) negative SMB trend of -0.002 mw.e.yr-1 for the period 1958–2012. This is in agreement with previous SMB studies done in the whole region of western DML (Isaksson and Melvold, 2002; Kaczmarska et al., 2004; Divine et al., 2009; Schlosser et al., 2014), which also document a negative trend in SMB during the 20th century. A comparison between annual SMB calculated in the cores taken at KC, KM, BI, S100 (Kaczmarska et al., 2004), and a composite core constructed averaging the annual SMB from four firn cores (M2, G3, G4, and G5) retrieved at Trolltunga and Jutulstraumen (Schlosser et al., 2014) (Table 1), is shown in Fig. 5 (individual SMB profiles of the cores conforming the composite are shown in Fig. S1a in the Supplement). Overall, SMB values from the KM and BI cores are higher than for the KC, S100 cores, and the composite record. Furthermore, the negative trend at KC agrees well with that found in the S100 and in FIS composite core. One proposed mechanism for decreasing SMB since the 1980s is the occurrence of stronger zonal wind flow with lower-amplitude long waves during the same period. The flow was characterised by the Southern Annular Mode (SAM) index (Schlosser et al., 2014). During high SAM index phases meridional moisture flux decreases; consequently, there is less precipitation and SMB in Antarctica (Schlosser et al., 2014).

Frezzotti et al. (2013) investigated Antarctic SMB over the last 800 years, and found that there were statistically non-significant changes in SMB over most of Antarctica, with no overall clear temporal trend over the longest timescale. However, they also report a clear increase in SMB (> 10 %) since the 1960s in regions where the SMB is high, i.e. coastal regions, and over the highest part of the East Antarctic ice divide. The authors attribute these dissimilar trends between high SMB locations and the rest of Antarctica to a higher frequency of blocking anticyclones. These anticyclones increase precipitation at coastal sites and lead to the advection of moist air at the highest areas. Strong winds producing snow redistribution and erosion would account for the reduction on SMB at windy sites. As discussed above, our results show that the SMB trends at KC is similar to SMB trends reported elsewhere at FIS and western DML (Isaksson and Melvold, 2002; Kaczmarska et al., 2004; Divine et al., 2009; Schlosser et al., 2014). No significant temporal trends in SMB are found at KM and BI.

Figure 7 shows the raw and annual δ18O, δ2H, and d (deuterium excess) data for the KC, KM, and BI sites. The higher accumulation rates at KM and BI allow high-resolution water stable isotope records (average 20 data points per year), whereas resolution at KC is much lower (average 7 data points per year). Table 1 shows the median values of δ18O, δ2H, and d of the KC, KM, and BI cores. The δ18O and δ2H signals at KM and BI (Fig. 7b, c) show pronounced seasonality, with seasonal amplitudes up to 10 and 78 ‰, for δ18O and δ2H, and up to 10 ‰ for d. Stable isotope ratios for KM and BI are similar, while KC has generally lower values of δ18O and δ2H than the two other cores. The core site at BI is 130 m higher than that of KC and KM (see Table 1), such that differences in isotope ratios can be attributed to local effects. Similar results were found by Fernandoy et al. (2010) for firn cores drilled in the hinterland of Neumayer Station, where higher-elevation cores did not always have lower δ18O values.

The present study is among the few that also includes deuterium, and the first involving such data from FIS. It is difficult to reliably determine the seasonal cycle of d for the analysed cores because we cannot date the cores accurately at the subannual level, post-depositional processes such as water vapour diffusion in the firn column may alter the d profile, and, finally, the d time series are relatively short. Nevertheless, we use the derived age models to estimate the intraannual variability in d for the two higher-resolution cores at KM and BI. The results (Fig. 8) suggest that the absolute values and magnitudes of the seasonal cycle of d are in reasonable agreement with observed and modelled d at other coastal Antarctic ice core locations (Schlosser et al., 2008; Inoue et al., 2016; Schoenemann and Steig, 2016). At both KM and BI, maximum values for d occur in austral autumn, preceding the corresponding seasonal minima in δ18O by 3–4 months and, most likely, air temperatures (Fig. 8). On the other hand, fresh snow sampling at nearby Neumayer Station suggested a spring maximum for d (Schlosser et al., 2008). We note, however, that, because of the method we use to construct the core chronologies, the seasonal curve of δ18O is fit to have a maximum (minimum) in the first (sixth) month of the year. Nonetheless, our timescale does not alter the fact that there is a lag between d and δ18O peaks. Such an offset is also reported at coastal locations by Schlosser et al. (2008) and Inoue et al. (2016), and contradicts the conventional interpretation of winter maxima in d being the result of shifting moisture-source regions to higher latitudes (e.g. Delmotte et al., 2000; Pfahl and Sodemann, 2014).

Recent studies show that the relationship between d and moisture source parameters is more complex than previously thought since d is also strongly sensitive to both equilibrium and kinetic fractionation during precipitation formation (Steen-Larsen et al., 2014; Dittmann et al., 2016; Schoenemann and Steig, 2016). The d measured in precipitation is controlled by different processes, exerting opposite effects. We hypothesise that the observed d to δ18O offsets for coastal locations, which have a relatively mild climate compared to the plateau, occur due to the effects of high-latitude moisture entrainment during winter. The influence of local moisture sources with low d may outweigh the thermal effects on equilibrium and kinetic fractionation during precipitation. However, the seasonal balance of these effects is likely site specific and the lack of studies on d controls at coastal locations precludes us from proposing a definitive explanation to this phenomenon.

High-correlation coefficients between the annual d and austral spring to summer SAM indices of -0.55 (significant at the 95 % confidence level) for KM and -0.33 (not significant) for BI cores point to the possible role of seasonal changes in moisture transport and precipitation in the area in shaping the annual isotopic signal. A positive SAM index is generally associated with stronger zonal westerlies and comparatively little exchange of moisture and energy between middle and high latitudes (Marshall et al., 2013; Schlosser et al., 2016), hence increasing the contribution of local, less depleted of δ18O, moisture sources in precipitation (Noone and Simmonds, 2002). Decreased meridional southward moisture transport during the positive SAM phase may vary the annual moisture balance towards a higher fraction of local spring and summer moisture. Compared to moisture originating from more remote lower-latitude sources, local sources in spring and summer typically have lower d values, leading to generally negative annual d anomalies preserved in the snow. The multidecadal positive trend in SAM (e.g. Marshall, 2003), which is especially pronounced for austral summers, may in turn drive the weak negative trend of 0.1 ‰decade-1 (not significant at the 95 % confidence level) in d detected in the longer KC core, also contributing to an observed positive trend in δ18O in the regional core network.

Figure 9 shows the mean annual δ18O for the KC, KM, and BI cores compared to the S100 and composite core from Schlosser et al. (2014) (individual δ18O profiles of the cores conforming the composite are shown in Fig. S1b). Overall, annual δ18O values at the three ice rise cores are higher than at S100 or for the composite core. However, both the inferred multiannual means and the standard deviations of δ18O for the three ice rise cores fall within the typical range of variability for other cores from the coastal DML (Altnau et al., 2015).

The positive linear trends in δ18O observed in the KC and BI cores also agree well with linear trends reported for the S100 and FIS composite cores (Kaczmarska et al., 2004; Divine et al., 2009; Schlosser et al., 2014). However, none of the linear trends observed in the KC, KM or BI cores are significant at the 95 % confidence level. Similar to earlier studies (e.g. Schlosser et al., 2014) no correlation is found between δ18O of the ice rise cores and measured air temperature at Neumayer Station, the closest station suitable for comparison. Neumayer is situated on a small ice shelf, with synoptic conditions similar to FIS; no temporal trend is found in air temperature since the founding of the station in 1981. Likewise, no relationship between stable isotopes and SMB is seen in the ice rise cores (compare with Fig. 5). This confirms previous studies, which find poor correspondence between SMB and proxy temperatures, suggesting that it is large-scale atmospheric circulation rather than the thermodynamic relationship between SMB and temperature that is the determining factor here. This was also found in a recent study by Fudge et al. (2016), who investigated the temperature–SMB relationship using data from the WAIS Divide ice core.