Calibrated cryo-cell UV-LA-ICPMS elemental concentrations from NGRIP ice core reveal abrupt , sub-annual variability in dust across the interstadial period GI-21 . 2

Several abrupt shifts from periods of extreme cold (Greenland stadials, GS) to relatively warmer conditions (Greenland interstadials, GI) called Dansgaard-Oeschger events are recorded in the Greenland ice cores. Using cryo-cell UV-laser-ablation inductively-coupled-plasma mass spectrometry (UV-LA-ICPMS), we analysed a 2.85 m NGRIP ice core section (2691.50 – 2688.65 m depth, age interval: 84.86 – 85.09 ka b2k, thus covering ~230 years) across the transitions of GI-21.2, a short-lived 15 interstadial prior to interstadial GI-21.1. GI-21.2 is a ~100-year-long period with δO values 3 – 4‰ higher than the following ~200 years of stadial conditions (GS-21.2), which precede the major GI-21.1 warming. We report concentrations of ‘major’ elements indicative of dust and/or sea salt (Na, Fe, Al, Ca, Mg) at a spatial resolution of ~200 μm, while maintaining detection limits in the low-ppb range, thereby achieving sub-annual time resolution even in deep NGRIP ice. We present an improved external calibration and quantification procedure using a set of five ice standards made from aqueous (international) standard 20 solutions. Our results show that element concentrations decrease drastically (more than tenfold) at the warming onset of GI21.2 at the scale of a single year, followed by relatively low concentrations characterizing the interstadial part before gradually reaching again typical stadial values.

methodology section.
4) In addition, the authors claim to use 24Mg, 27Al, 40Ca, 56Fe. All these masses are highly interfered by spectral and matrix interferences in ICPMS. Despite I think they used a SF-ICP-MS or a collision-cell instrument to reduce the interferences, I think that a better description of the methodology should be given. I know that most of the details are given in Della Lunga (2014), but a minimal description of the methodology is compulsory.
We have added several details regarding the methodology, including a description of the removal of interferences by the use of H 2 in the Agilent 7500cs reaction cell; but we also note than these details are provided in our earlier publication in the Journal of Glaciology which had been included at the outset (Della Lunga et al., 2014). If the ice matrix was contributing to the signals on m/z=24, 27, 40, 56 then we would not obtain no resolvable counts from instrumental background for our ice blanks. 5) Then, I do not fully understand the objective of the MS, since most of the findings are not new at all. I would have rather focussed the MS into a comparison between LA-ICPMS vs CFA, but this would have required a more robust statistical tool.
I would therefore suggest the authors to readdress the MS to a specific target: i) analytical (in this case the paper lacks of many details), describing in detail the new advancement of this powerful technique and duly comparing the data with CFA results; in this case the reproducibility of the analysis on different sections is a key parameter, but as far as I can see there are no evidence of this in the paper; ii) more oriented toward a climatic/environmental interpretation; in this case the real benefit of the LA-ICP-MS approach should have been better explained.
As we also state in the reply to R1, the focus of the manuscript was predominantly methodological yet with one integral case study of deep ice to illustrate the power of the significantly improved, in fact unprecedented, spatial resolution. To make this focus clearer, the revised manuscript now presents a stronger methodological section and a more detailed description of the signal emerging from ablation with 193 nm laser and its possible interpretation. The climatic picture emerging from our analysis of GI21.2 is intended to demonstrate the capability of the technique to recover extremely small scale variability together with Stadial-Interstadial fingerprint in dust and sea salt proxies for one of the most abrupt and short lived transitions in deep ice where ice layer thickness is becoming very small.

Introduction
[……] This mechanism, which is thought to be the primary reason for sea-salt enrichment in ice cores during cooling events, receives further contributions of sea salt from another source. When sea ice is formed, highly saline brine and fragile frost flowers form on top of the frozen surface. This brine represents a further source of aerosol, carried over land by the wind (Wolff et al, 2003). However, a quantitative assessment of the contribution of brine, frost flowers, and blowing snow to the wintertime peak in sea salt aerosol it is still a matter of debate (Huang and Jaegle, 2016).
[……]The aim of the present study is to assess the sensitivity and the phasing of dust/sea-salt proxies as Na + ,

Methods and Calibration
This section corresponds to more than two hundred years, given the observed layer thickness of ~10 mm (Vallelonga et al., 2012). In the flow-model-based GICC05modelext timescale, the section covers an age range of 85.09 -84.86 ka b2k and includes the 100-year long GI-21.2 and the transitions in and out of this period (Wolff et al., 2010;Rasmussen et al., 2014). We utilized samples from a similar position within the ice core cross section as in Della Lunga et al. (2014).
[……] The adopted methodology includes the acquisition of the following mass/charge ratios: 23(Na), 24 ( [.…] Each element has been externally calibrated using a set of four custom-made ice standards chosen from a total of five (SLRS-5, SLRS-5_10, ICP-20, NIST1648a and Water Low), prepared at RHUL from four different standard solutions at different concentrations and different dilutions (Table S1, see supplementary material). All of our Ice standards except SLRS-5 were prepared by dilution between 1:10 and 1:1000 of the certified reference material with ultrapure H 2 O (>18 MΩ·cm); we very mildly acidified these solutions with 1% ultrapure HNO3 to stabilize them before freezing and to align them with the acidity of the multi-elemental standard solution ICP1 (Sigma-Aldrich), which was the only one being originally (before dilution) in 10% HNO 3 , unlike all of our other standard solutions.. CFA-Na, respectively. This seems to indicate that there is not an overall systematic shift between the two techniques (see below). In general, the difference between LA-Na and CFA-Na, could derive from the tendency of Na to show higher concentrations in the proximity of grain boundaries and junctions, as it is described in the following section. Therefore, laser ablation tracks show much higher variability as a result of scanning across several boundaries and junctions at small scale, introducing a factor of differentiation that is also reflected in our calibration since it reduces the homogeneity of our ice standards.
As a further test, we compared the cryo-cell UV-LA-ICPMS data acquired in the frozen state with results from the same three NGRIP samples analysed via solution-ICPMS after melting (10 ml Results show that calibrated solution data are consistent with our LA-ICPMS data and differ by 5 -20 %, which is essentially within our margin of error. Sample 4882B4, representing the last part of GS-21.2, shows the lowest concentrations amongst the three samples and also the consistently largest differences between solution and laser data (see Fig. S4 in the supplementary material).

Origin of Laser ablation elemental signal
The intensity of the LA-signal associated to a certain mass/charge ratio, characteristic to one element, is built up by two different contributions: one from soluble ions present in the ice matrix and the other one from dispersed insoluble mineral particles containing the element in their structure. Micro-particles in the NGRIP ice core have a mean grain size between 1 and 2 µm (Ruth et al., 2003) and therefore are too small to be identified unequivocally with our laser camera. Visual inspection of the sample before, after, and during ablation indicated that no residual spatter of the ablation process was deposited back onto the ice surface after the laser hit the sample, indicating a complete digestion of the material removed by the ablation pulses. This suggests that no fractionation between soluble and insoluble particle is taking place by effect of the laser sampling.
We investigated the spatial distribution of Na, Mg, Al, Ca and Fe over two small horizontal planes (i.e, perpendicular to the core length axis) by analysing 2D maps of concentrations across two of these specific cross sections (Fig 9 and 10). These sections were constructed interpolating several acquisition points obtained via static laser drilling. Fig 9 and 10 both show concentrations spanning over a range of several tens of ppb for each element across the entire sections. The cross-sections intersect few grain boundaries and junctions (as observable in the laser camera image). The grain boundary net has been overlaid in black onto the elemental maps and shows that, in most of the cases, high concentrations areas are located in the proximity of boundaries and junction, broadly mimicking their pattern. In both cases, these patterns are somehow clearer for element like Na and Mg, related to sea salt, and become less defined going from Ca to Al and Fe. This might be associated with the fact that the elemental signal has a relative increasing contribution from micro-particles going from Ca to Al, to Fe, whereas the contribution from micro-particles to the Na and Mg signal would be minimal. This would also suggest that micro-particles are slightly less inclined to be aligned on boundaries and junctions compared to soluble impurities and therefore generate a less defined pattern of concentrations in our maps.

Discussion
[….] Most of the differences between CFA and LA-ICPMS proxies are observed at a small scale and are mainly influenced by few factors, the first of which is the effect of sample volume. In fact, we estimate that every LA-ICPMS data point corresponds to ~120 ng of ablated ice (based on scanning speed and ice crater depth) whereas CFA sampling resolution is about 0.1-1 g for each data point (Vallelonga et al., 2012). This introduces a difference in the sampling volume between the two datasets that can also be influenced by surface effects and especially by the wavy nature of layers at this scale and core depth. This is particularly important for Na, whose lateral variability induced by any non-horizontal layering is also affected by diffusion of Na that has been observed at this depth, resulting in a smoothing of the CFA annual signal (Vallelonga et al., 2012). Furthermore, the CFA insoluble dust data presented here refer to measurements of particles of size >1 µm and therefore do not account for insoluble impurities of sub-micron size (Vallelonga et al., 2012).
The elemental maps shown in Fig. 9 and 10 demonstrate that, at sub-cm scale, the concentrations of impurities it is strongly influenced by the presence of boundaries and junctions even when considering horizontal planes, whose original impurity-input is therefore assumed to be roughly identical. This introduces a main source of differentiation between LA and CFA sampling and can account for some of the small-scale variability we observe in the LA-profiles. This is particularly relevant for element like Na and Mg whose 2D distribution seems to follow closely the grain boundary net, presenting higher concentrations in the proximity of boundary and junction. On the other hand, 'dust'-proxies as Ca, Al and Fe, do not show such a closer overlap of high intensity and presence of boundaries or junctions, possibly as a result of being increasingly associated with insoluble micro-particles dispersed in the ice matrix, which indeed constitutes the CFA-Dust signal. This suggests that micro-particles in the ice matrix are less inclined to reside on boundaries and junction compared to soluble ions and is consistent with previous studies of deep ice cores (Della Lunga et al., 2014;Eichler et al., 2016). As a result, the averaging of LA-signal between two or more parallel tracks spaced by few mm is not only desirable but necessary.

Author contribution
DDL designed the experiment, performed the analysis, interpreted the data and wrote the manuscript. WM helped designing the experiment, performing the analysis and the data interpretation and edited the manuscript.
SOR and AS contributed to the designing of the experiment, the sample preparation, the data interpretation and edited the manuscript. PV provided CFA data for comparison, helped with the data interpretation and edited the manuscript.

Acknowledgements
This work has been supported by a RHUL studentship granted to Damiano Della Lunga, with the analytical costs     Table S1 (supplementary material). LOD indicates limits of detection. See text for details.