Continuous flow analysis (CFA) of the KCC core was performed with the setup at the Division for Climate and Environmental Physics, Physics Institute, at the University of Bern. Analyses performed on the meltwater flow included meltwater conductivity, insoluble particle concentration and size distribution as well as selected ion species (Ca2+, NH4+, NO3-, Na+; see Table ). In addition, stable water isotopes were analysed using a Picarro instrument coupled directly to the meltwater flow. The size distribution of insoluble particles recorded by the optical particle sensor was used to derive a profile of the “coarse particle percentage” (CPP). The CPP was calculated based on particle volume, and represents the percentage of particles exceeding a threshold of 4.0 µm. The threshold was chosen such that it corresponds to the expected median particle diameter of Saharan dust particles at CG, which was shown to be distinguishable from background sources . Deviations from a CPP of 50 % indicate higher or lower contribution of large and small particles respectively. The melt rate was adjusted to provide the necessary amount of water for all analyses, resulting in an effective depth resolution ranging from 1.2 cm at the very top of the core to about 0.5 cm for all depth below approximately 25 m w.e. Electrical conductivity measurements (ECM) performed at the Institute of Environmental Physics, Heidelberg University, were used primarily to obtain a qualitative record of the acidity of the ice in connection to the detection of Saharan dust events.

The KCI core was analysed using the reduced CFA setup at the Institute of Environmental Physics, Heidelberg University. Meltwater conductivity and insoluble particle concentration were measured by CFA at about 0.7 cm effective resolution. Continuous sub-sampling of the core for stable water isotope analyses was conducted at a depth resolution typically ranging between 5 and 10 cm. Due to the relatively high firn temperature at CG, isotope smoothing is much faster compared to polar sites with similar annual layer thickness. Hence re-sampling most of KCI even at 1.5 cm depth resolution did not significantly restore any high-frequency isotope variability .

Laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) was conducted in the WM Keck Laser Ice Facility at the Climate Change Institute (University of Maine) and used to analyse 44Ca at ultra-high depth resolution (better than 120 µm). The more abundant 40Ca is blocked by mass interference from 40Ar used as carrier gas. Details regarding the method, sample preparation and calibration routine can be found in and . Briefly, the components of this system include a Thermo Element 2 ICP-MS, a New Wave UP-213 laser, and a cryo-cell chamber, designed to seal a 1 m ice core from the surrounding air while maintaining a uniform temperature of -15 ∘C. In order to ensure a complete seal of the ablation chamber, porous firn parts could not be measured. From 29.5 m w.e. to bedrock, the KCC ice core was analysed for 44Ca along a single ablation track. The 44Ca signal comprises contributions of soluble and insoluble Ca . Crucial for further deployment for annual layer counting, the trend components in the LA-ICP-MS measured Ca signal have been shown to be in good correspondence with the lower-resolution CFA Ca signal, as shown in Fig.  and previously by and .

For KCI, insoluble particle concentration and meltwater conductivity were used for annual layer counting, extending down to about 26 m w.e. Below 26 m w.e. the identification of annual layers became ambiguous and was abandoned. This depth corresponds (taking the uncertainty in layer counting into account) to (AD 1492±30). A two-parameter model (based on a simple analytical expression for the decrease of the annual layer thickness with depth) was used to extrapolate a continuous age–depth relation to greater depth . Note that high-resolution annual layer counting could only be performed in KCC (see below), since only small sections of KCI have been analysed by LA-ICP-MS so far .

All impurity species measured by CFA (Table ) were used in combination for annual layer counting. Annual layers were defined as local maxima in at least two of the six impurity signals, with special emphasis on NH4+ featuring the largest seasonal amplitude. An example of counting annual layers in the CFA profiles in shown in Fig. a. In order to identify highly thinned, sub-centimetre annual layers expected to dominate the deeper core sections, an independent counting was established using the LA-ICP-MS Ca profile starting at 29.5 m w.e. (corresponding to AD 1760). At this depth, the average annual layer thickness was estimated from CFA-based counting as around 3 cm. The LA-ICP-MS Ca record was investigated at full resolution and as a smoothed version (using the leading components in singular spectrum analysis or Gaussian smoothing). In its upper section, the LA-ICP-MS Ca profile is characterized by regular occurrence of several distinct peaks grouped together, along with an elevated baseline of Ca concentration (Fig. b). The groups of peaks are separated by a comparatively stable signal of low Ca concentrations. The latter is interpreted as resulting from the varying degree of winter snow being included in the record otherwise dominated by summer snow. Accordingly, the grouped peaks correspond to sub-annual snow deposition events of elevated Ca concentration during the summer period, which is also observed in the most shallow parts of KCC (Supplement). For the depth interval 29.5–32.5 m w.e., counting separated groups of peaks (typically 3–5 peaks per annual layer) in the LA-ICP-MS Ca record results in good agreement with counting performed on the CFA profile (typically within ±1 year per 10 counted years; see Fig. b). Below 32.5 m w.e., average annual layer thickness becomes close to 1 cm and counting in the CFA profile becomes increasingly difficult (i.e. frequent “shoulder type” annual layers merged into a single impurity peak). The LA-ICP-MS Ca profile continues to show distinct groups of peaks that become increasingly closely spaced and eventually merge into single broad peak events (Fig. c). Accordingly, LA-ICPMS Ca was the dominant source of annual layer counting after around 32.5 m w.e. (AD 1600). The annual layer signal remains clearly identifiable for the remaining part of the depth range investigated here (e.g. apparently not affected by diffusion of soluble Ca).

For KCC, results from 14C analysis are found to back the annual layer counted age scale. Five of six 14C dates agree with the counting within their 1σ range (Fig. ), corresponding to a root mean square deviation of 118 years (227 years including the outlier). The outlier 14C point contradicts a monotonic increase of age with depth and is thus disregarded. This is justified, because the 14C age of this sample matches with a very sensitive section of the 14C-calibration curve. Therefore already a small, unknown blank contribution would be able to shift the calibrated age of this sample significantly. Within the 2σ error range it also hits the error range of the annual layer counting chronology in the present configuration. Its deviation is therefore not of consequence.

For the KCI ice core, the radiocarbon ages are found to agree with the extension of the existing age scale based on the two-parameter model. It seems worth noting, however, that four out of six 14C points lie systematically above the extrapolated age scale (albeit in agreement within their 1σ range). Only the sample at 28.4 m w.e. shows an age that is significantly older than expected. This might be due to the extremely small (also compared to the other KCI samples) sample size of only 2.2 µgC, making this sample prone to even very small potential blank contributions. In this context a potential influence of aged organic material (e.g. from Saharan dust) has to also be considered. At present, the age of this sample is therefore regarded as an outlier. Additional radiocarbon measurements of this core section above and below the critical sample are planned to further refine the match, and to test if the systematic deviation of the 14C ages persists. The 14C ages of the KCC and KCI samples are summarized in the Supplement.

Potential sources of uncertainty in annual layer counting stem from (i) erroneously identifying or missing of existing annual layers, (ii) interpolating data gaps and (iii) an incomplete stratigraphy missing years due to annual snowfall fully eroded from the surface. Regarding (i), we estimated the likelihood of miscounting layers by marking “uncertain years” (Fig. ). In view of the high snow erosion at CG, uncertain layers were defined as additional peaks in close proximity to an annual layer (e.g. “shoulder type” peaks). To quantify counting uncertainty from uncertain layers, we followed the approach successfully employed for Greenland ice cores. This is to count uncertain layers as 0.5 ± 0.5 years and to estimate the maximum counting error (MCE) from N uncertain layers as N × 0.5 years . With 144 uncertain layers detected within the upper 40 m w.e. of KCC, this corresponds to an uncertainty of ±72 years at AD 1000. With respect to (ii), the depth interval considered in this work was completely recovered without core loss. The ends of the CFA core sections were trimmed in case of irregular core breaks. This resulted typically in less than a centimetre of missing CFA data, thus not interfering with annual layer counting. The ECM profile was used as an alternative backup across these short CFA data gaps. Likewise, the CFA data were used as an alternative indicator where the LA-ICP-MS profile was incomplete, which only concerned one major instance of missing LA-ICP-MS data between about 33.8 and 34.24 m w.e.

Contribution (iii) constitutes a fundamental difference relative to Greenland conditions, since CG is not a closed system with respect to precipitation and loss of the annual snowfall in selected years can occur. The frequency of occurrence in these total snow loss events is, however, extremely hard to quantify. Counting annual layers in between the above-mentioned (dust) horizons within the last century reveals an offset of typically only 1–2 years as compared to the known age of the horizons. Thus, the counting appears not to be systematically flawed by missing years. Hence we regard uncertainty (i) as dominant and use the MCE as an uncertainty estimation of the KCC age scale. Notably, the uncertainty refers to the unconstrained counting approach used here and could be further refined in the future with new absolute dating horizons, especially for the pre-AD 1900 period.

The uncertainty of the KCI age scale was obtained in a consistent manner, using the MCE for the annual layer counted interval and extrapolating the upper and lower uncertainty limits with the two-parameter model. Figure  shows the resulting age–depth relation and uncertainty bands for KCC and KCI, together with 14C dates (shown with their 1σ uncertainty range) available for the respective depth interval. Table  gives a complementary summary of the age–depth relation, uncertainties and annual layer thickness for 10 m depth intervals. It is important to note that we are less confident about the age–depth relation of KCI compared to KCC, due to KCI featuring (i) annual layering counting using only two bulk parameters and only down to 26 m w.e. and (ii) the extrapolation by the two-parameter model.

To investigate potential offsets between the KCI chronology and the presumably more reliably dated KCC, we compared the stable water isotope time series of the two cores. This is motivated by the fact that the decadal isotope trends among the CG ice cores have been previously shown to agree over the last 250 years . Without substantial dating offset between the cores, this inter-core agreement should hold also on longer time intervals. It is important to note that due to the strong effect of isotope diffusion at CG, inter-annual or even seasonal isotope variability is effectively eliminated. As a consequence, the records (except for the last 100 years in KCC) resolve only decadal-scale variability at best. Hence we did not apply any further smoothing to the time series. In order to avoid potential biases from increasing sampling resolution, both time series were sub-sampled to nominal biennial resolution. Figure  shows the comparison of the respective time series on their original timescales for the last 1000 years.

The two original time series of KCC and KCI already feature striking similarities, although frequently separated by a lag between the two time series (e.g. note the distinct isotope minima around AD 1360). The direction and magnitude of this lag varies with time, hampering an absolutely straightforward adjustment to match the two records. Aiming to adjust the KCI record to KCC time series, we employed the powerful algorithm developed by for correlating paleoclimate time series. In doing so, we left the last 150 years of the KCI age scale unchanged (since considered reliably dated) but did not prescribe any further user-defined tie points to the algorithm. The result shows the original lag between the two time series eliminated (Fig. ). A maximum shift of around 15 years is needed to align the two records (e.g. around the AD 1360 isotope minimum), which is within the estimated dating uncertainty of KCI (Fig. ). As a result, the aligned time series are significantly correlated (r=0.47). This degree of correlation is within the typical range of correlating CG isotope time series on the decadal scale within the last 250 years . However, this is the first time that the correlation holds to this extent also for the comparatively old core sections of CG ice cores, e.g. we find a correlation coefficient of r=0.50 when considering the interval AD 1500–1000 only. In the following, all KCI time series are considered on their aligned timescale.

The covariation of the δ18O time series between KCI and KCC strongly suggests a common atmospheric driver, i.e. temperature. At first glance Fig.  shows an increasing trend over the last 100 years but also generally higher mean isotope levels prior to about AD 1900. This is in line with earlier findings suggesting an “early instrumental period” warmer than instrumental data by about +0.4 ∘C derived from the CG isotope signal . Here we find the generally higher average isotope levels to persist over much of the pre-industrial period – for instance, the mean δ18O level in KCC between AD 1860 and 1000 is higher by about 0.75 ‰than the AD 2000–1860 average. A quantitative use of the common isotope signal would therefore require addressing systematic so-called “upstream effects” and a reliable calibration of the isotope signal against instrumental temperature.

Upstream effects concern the systematic variation in seasonality of the net accumulation upstream of the drilling site and have the potential to bias long-term core averages. Quantifying this effect requires accurate identification of the upstream catchment area (typically by sophisticated flow modelling) and evaluating the spatial variability in mean isotope levels. Dedicated efforts to evaluate the upstream effect for the KCI–KCC flow line are currently underway (C. Licciulli and J. Lier, IUP Heidelberg, personal communication, 2017). From a preliminary inspection of snow pit data recently obtained for the KCI-KCC flow line, there is no clear indication of a systematic trend in mean δ18O levels upstream of KCI, however – comprising roughly the years 2016–2014, three snow pits evaluated thus far show mean δ18O levels of -13.22, -11.94 and -15.04‰ at about 60, 195 and 300 m distance upstream of KCI, respectively (KCC is located roughly 110 m upstream of KCI; see Fig. ).

In order to calibrate the stable water isotope signal, we used the instrumental temperature dataset compiled in an earlier study . This temperature dataset (referred to here as “CG modified temperature”) was specifically adjusted to the CG ice core conditions, taking into account the summer bias in precipitation and snow deposition. To calculate an isotope/temperature sensitivity, we considered both KCC and KCI individually as well as a stack of the two stable isotope records (calculated as their simple average at nominal annual resolution). Using AD 2000–1860 as calibration period (thus deliberately avoiding the “early instrumental period” prior to 1860) our results reproduce earlier findings of . This specifically includes showing (i) an overall agreement between the isotope and temperature record interrupted by characteristic decadal mismatch periods (Fig. ), (ii) an increase in isotope/temperature correlation for multi-annual and decadal averages (e.g. r=0.33,0.47,0.64 for discretely binned annual, 5- and 10-year averages, respectively, in the case of KCC) and (iii) higher correlations obtained from the stack vs. the individual time series (e.g. r=0.48,0.67,0.79 for annual, 5- and 10-year averages, respectively).

Regarding sensitivity values, we also find an increase with length in averaging period as well as substantially higher sensitivity values for KCI than KCC, revealing 2.3 vs. 1.4 ‰ ∘C-1, respectively, when using discretely binned 10-year averages (and 1.8 ‰ ∘C-1 for the stacked record). Hence we obtain sensitivity values about 3-fold higher than what is expected, e.g. based on the isotope/temperature relationship of 0.65 ‰ ∘C-1 reported by for European temporal trends in precipitation. Since a stricter confinement towards sampling the high summer season can be expected for the lower-accumulation KCI, the sensitivity difference between KCI and KCC points towards the seasonal bias in snow sampling being connected with enhanced sensitivity. It is important to note that the high isotope-sensitivity deserves a separate thorough investigation, ideally comprising regional climate–isotope modelling and taking into account post-depositional effects such as snow preservation, upstream effects and isotope diffusion, which is outside the scope of this study. Until an adequate long-term calibration of the CG isotope signal is achieved, however, we do not attempt a quantitative temperature reconstruction based on the isotope composite record so far.

Figure  shows the time series of δ18O in comparison with Ca2+, deuterium excess and CPP. Due to the logarithmic distribution of the Ca2+ data, we generally use a log scale to show the Ca2+ time series. For KCC, we find δ18O and Ca2+ to be significantly correlated over the last 1000 years (at r=0.49; see Fig. ). This suggests that the common decadal-scale signal driver behind the shared variability between the δ18O time series of KCC and KCI also holds for Ca2+. Prior to about AD 1860, the δ18O time series constitutes nearly an upper envelope signal as compared to Ca2+. Short excursions to low Ca2+ concentrations missing a respective counterpart in δ18O may have been smoothed out by isotope diffusion. It is worth noting in this respect that the firn–ice transition in KCC coincides roughly with the last 100 years in the record (Tables  and ).

Following our general distinction (Sect. ), two signal components of the Ca2+ time series are investigated separately: (i) the frequency of occurrence in Saharan dust deposition, and (ii) the long-term average Ca2+ concentration. In order to investigate to what extent the time series agreement observed for δ18O also holds in the case of mineral-dust-related species, we use the insoluble particle signal as a surrogate for Ca2+ (since Ca2+ has not been measured for KCI). Figure  shows an overview of the insoluble particle datasets of KCC and KCI. The KCC data show that the insoluble particle signal and Ca2+ concentrations are generally highly correlated (e.g. r=0.9 within the last 1000 years). The KCC–KCI inter-core comparison of insoluble particle records reveals agreement of decadal-scale features, as well as similarities regarding periods of low concentrations and higher peak abundance (e.g. 1800–1820 vs. 1780–1800, respectively). Differences in the magnitude of individual peak events as well as mean levels of particle concentrations can be explained in light of (i) the KCI particle signal measured on diluted sample meltwater, (ii) potential calibration differences in the optical particle sensor and (iii) inter-site snow deposition variability. Accordingly, it was not attempted to construct a composite record of insoluble particles from the two cores.

Based on the above considerations and the agreement within the instrumental period (Fig. ), the potential of the Ca2+ signal to quantitatively record temperature variability is explored further over the full 1000-year period. For this purpose the biennial logarithmic Ca2+ is calibrated tentatively against instrumental temperature using linear regression within the time period AD 2006–1860. The respective 90 % confidence intervals are used to calculate a temperature reconstruction with uncertainty bands (0.7–1–1.8 ∘ / log Ca2+ (ppb)). Decadal trends are again highlighted by Gaussian smoothing in Fig. . The resulting 1000-year record is shown with a tentative 200-year extension. Regarding its overall features and in view of remaining dating uncertainties, the record provides evidence of “Little Ice Age conditions” systematically cooler than the reference and calibration period, with an average of -0.3 ∘C between AD 1800 and 1200. A shorter warm interval of about +0.3 ∘C is found in the late 1100s These features are especially noteworthy considering the above-average mean of the δ18O values with respect to AD 2006–1860 (Fig. ). The comparison with δ18O suggests that the Ca2+–temperature coupling may be less affected by non-stationary sensitivity (and upstream) effects.

In this context we further explored our Ca2+-based temperature reconstruction attempt in comparison with other proxy reconstructions of European summer temperature. We show here results from using the mean European summer temperature anomalies reconstructed by , considering their composite-plus-scaling method (CPS) adjusted to biennial resolution and our reference time period of AD 2006–1860. The decadal trends (represented by Gaussian smoothing) of both reconstructions shown in Fig.  are generally consistent in their overall features (and formally correlate at r=0.4). These features comprise the recent warming trend and below-average conditions during the “Little Ice Age” (LIA), and a warm episode in the late 1100s (“Medieval Climate Anomaly”, MCA). The only major feature of disagreement occurs around the already noted minimum around AD 1250–1230. The overall low levels of impurities (especially NH4+) and δ18O may point to increased deposition of winter snow during this time, or exceptionally cold summer conditions. It is worth noting that if, tentatively, extending the comparison for another 200 years beyond AD 1000, the agreement between the two reconstructions continues to last until AD 800, consistently showing a relatively warm interval lasting between about AD 1000 and 850. The overall agreement is especially noteworthy in the light of (i) only small offsets exist between the general features of the two records, which may stem from the remaining dating uncertainty of KCC (the comparison is not intended as a dating validation, however), and (ii) the absence of evidence of a non-stationary sensitivity. The magnitude of the general features (LIA, MCA) and decadal-scale temperature variability derived from our ice core record is consistent overall with the other proxy reconstruction.

A potential systematic bias to the observed Ca2+–temperature coupling could arise from strong and long-term changes at the dust source, e.g. increased dust mobilization, which could also affect the long-term Ca2+ variability. At the same time, increased dust mobilization would likely also influence the Saharan dust spikes and their frequency of occurrence. However, we only find one instance of long-term changes in dust occurrence rate – the period AD 1100–1200. Being the most outstanding period in the dust event occurrence, it warrants being looked at more closely. This outstanding period is characterized by (i) an increased frequency of Saharan dust events, (ii) above average levels, both in CPP and deuterium excess, starting to rise around AD 1100, and (iii) a delayed relative increase in δ18O and Ca2+, e.g. first showing minimum concentrations between AD 1130 and 1170 followed by a shorter maximum around AD 1170–1200. The CPP maximum constitutes the dominant feature of the entire record (notably, this result does not depend on the exact threshold chosen to calculate the CPP). The median of the CPP record within this AD 1100–1200 time period (0.61 ± 0.11, reported with one median absolute deviation) indicates an increase in coarse particles by about 12 % relative to the median of the rest of the 1000-year time period (0.48 ± 0.05). While an increase in dust mobilization cannot be ruled out, the connection of the distinct increase in coarse particles with enhanced dust event frequency rather suggests an increase in the direct transport of Saharan dust (as opposed to indirect advection with longer pathway and thus stronger decrease in coarse particles). Finding increased values of deuterium excess supports the view of a relative increase in direct Saharan dust advection, as increased deuterium excess could be expected from warm and dry air masses collecting moisture over the Mediterranean e.g.. Overall, these findings motivate getting an even more detailed picture of this outstanding period, e.g. using elemental and isotopic fingerprinting of the dust layers for a precise provenance investigation e.g., which is left for future investigations.