The SPICEcore (SP19) timescale was developed by combining annual layer counting with volcanic event matching between SPICEcore and the WAIS Divide chronology. We identified 251 volcanic tie points that are clearly visible in both SPICEcore and WAIS Divide (Sigl et al., 2016). These tie points link SP19 with the WAIS Divide chronology, resulting in one of the most precisely dated interior East Antarctic records. Above 798 m, ages are interpolated between volcanic tie points using layer counts. Below 798 m, ages are interpolated between tie points by finding the smoothest annual layer thickness profile (minimizing the second derivative) that satisfies at least 95 % of the tie points (following Fudge et al., 2014).

Although it is possible to create an independent, annually layer-counted SPICEcore timescale during the Holocene, we linked the entire SP19 chronology to the WAIS Divide chronology for several reasons: (1) annual layers are insufficiently thick below 798 m (approximately 11 341 BP) to consistently resolve individual years, requiring synchronization to another ice core to achieve the best possible dating accuracy – tying the entire SP19 chronology to the WAIS Divide core ensures consistent temporal relationships between these two records; (2) although annual layers are remarkably well-preserved in SPICEcore chemistry, WAIS Divide has a higher accumulation rate (Banta et al., 2008; Fudge et al., 2016b; Koutnik et al., 2016) and stronger seasonality in chemical constituents (Sigl et al., 2016), producing more robust annual layering (Fig. 4); (3) it is expected that some years at the South Pole experience very low accumulation, resulting in a lack of an annually resolvable record during those years (Hamilton et al., 2004; Van der Veen et al., 1999; Mosley-Thompson et al., 1995, 1999); (4) an attempt to independently date the Holocene annual layers created drift of several percent at stratigraphic tie points. We therefore elected to anchor the SP19 timescale to WD2014 and use the annual layer counts as a means of interpolating between WD2014 tie points during the Holocene. The SP19 timescale spans -64 BP (2014 CE) to 54 302±519 BP, with the annually dated Holocene section of the core extending to 11 341 BP (798 m depth).

The matching of volcanic events in sulfate and ECM records is commonly used to synchronize ice core timescales (e.g., Severi et al., 2007, 2012; Sigl et al., 2014; Fujita et al., 2015), including the recent extension of the annually resolved WAIS Divide timescale to East Antarctic cores (Buizert et al., 2018). Volcanic matching is based on the depth pattern of events more than the magnitude of the events because the magnitude in individual ice cores can vary significantly across Antarctica depending on the location of the volcano and atmospheric transport to the ice core site. The volcanic matching between SPICEcore and WAIS Divide is based primarily on the sulfate record for SPICEcore and the combined sulfur and sulfate records for WAIS Divide (Buizert et al., 2018). AC-ECM from SPICEcore and WAIS Divide was used as a secondary data set and to fill small data gaps in the sulfate record. An example of the four data sets is shown in Fig. 5.

The volcanic matches were performed independently by two interpreters (T.J. Fudge and David Ferris) and then reconciled by one (T.J. Fudge) with concurrence from the other (David Ferris). The position of each match was defined as the inception of the sulfate rise in order to most consistently reflect the timing of the volcanic event itself. Of the final 251 tie points, 229 were identified in the sulfate data by both interpreters. Of the remaining matches, 14 were made by one interpreter in the sulfate data and at least one interpreter in the ECM data. One of the other matches was made only with ECM because of a gap in the sulfate data for SPICEcore. The last seven matches were part of sequences not initially picked by one interpreter but deemed to be sufficiently distinct from the other events in the sequence to be included.

We note that the purpose of the volcanic matching was to develop a robust SPICEcore timescale, not to assess volcanic forcing. Thus, there are many potential volcanic matches that were not included either because they did not have the same level of certainty as the final 251 matches or because they were in close proximity to the final matches and thus did not provide additional timescale constraints.

For the pre-Holocene section of the core, ages between the volcanic matches are interpolated by finding the smoothest annual layer thickness by minimizing the second derivative (Fudge et al., 2014). The goal of finding the smoothest annual layer thickness time series is to prevent sharp changes affecting the apparent duration of climate events on either side of a volcanic match point. The method allows the ages of the volcanic matches to vary within a threshold to produce a smoother annual layer thickness interpolation. The degree of smoothness was set such that 95 % of the tie points are shifted by 1 year or less, which is a reasonable uncertainty on the precision of the volcanic matches.

The SP19 chronology extends from 2014 CE (-64 BP) at the surface to 54 302 BP at 1751 m depth. The timescale and volcanic tie points are depicted in Fig. 7 with volcanic tie points pinning the timescale also shown. Annual layer thicknesses near the surface are roughly 20 cm thick (owing to the low density of firn), decreasing rapidly to ∼8 cm yr-1 by the firn–ice transition. The timescale is annually resolved between -64 and 11 341 BP, below which resolution varies based on the distance between tie points. Using the methods in Sect. 3.2 (Fudge et al., 2014), we report timescale values interpolated at 10-year resolution. The longest distance between tie points is 2476 years between 16 348 and 19 872 BP.

In discussing uncertainty values for SP19, the reported values are uncertainty estimates rather than rigorously quantified 1σ or 2σ values. There are several reasons for this: (1) the chemicals used to count annual layers have similar cyclicity and are not independent; (2) while each of the five interpreters counted layers independently, they were likely employing similar strategies; (3) certain years may not be well-represented in the data, providing insufficient information for accurate dating or quantifying uncertainty; (4) volcanic events were identified in clusters such that each event is not necessarily independent; (5) it is difficult to assign a numerical index of confidence to specific volcanic tie points. Instead, we discuss timescale uncertainties as uncertainty estimates, which are intended to approximate 2σ uncertainties but cannot be precisely defined as such. This approach follows that of Sigl et al. (2016).

We assess the SP19 timescale uncertainty with respect to the previously published WD2014 timescale (Sigl et al., 2016; Buizert et al., 2015). The absolute age uncertainty will always be equal to or greater than the uncertainty already associated with WD2014 (Buizert et al., 2015; Sigl et al., 2016; Fig. 8). In addition to the uncertainty in WD2014, there is also uncertainty in our ability to interpolate between stratigraphic tie points. During the Holocene, our layer counting of sodium and magnesium concentration improves the timescale accuracy between tie points. Interpolation uncertainty can be estimated using the drift among the five different interpreters. We calculate the number of years picked by each interpreter in running intervals of 500 years in the final WD2014 synchronized timescale. Under ideal conditions, each interpreter would also pick 500 years within each interval, but on average the number of years picked by interpreters differs from the final timescale by 6.7 %, usually by undercounting. This is similar to the metric described in Sect. 3.3, wherein the average change in years needed to reconcile the layer counts and volcanic tie points was 5.6 % of the interval length. Here, we report the larger and more conservative value of 6.7 %. If our layer counting skill drifts by ±6.7 % while unconstrained by volcanic tie points, then the interpolation uncertainties remain within ±18 years of WAIS Divide throughout the Holocene with the exception of a poorly constrained interval between approximately 1800 and 3100 BP. The maximum uncertainty within the Holocene is ±25 years, occurring at roughly 2750 BP, where the nearest tie points are 373 years away at 2376 and 3123 BP. This relationship can be applied across the Holocene, with layers accumulating an uncertainty value equal to 6.7 % of the distance to the nearest tie point (Fig. 8; blue).

Below 798 m depth (start of the Holocene), there were no annual layers to aid in our interpolation of the timescale, leading to larger uncertainties. Our assumption of the smoothest annual layer thickness (Fudge et al., 2014) satisfying tie points is the most accurate interpolation method in the absence of additional information, at least in Antarctic ice (Fudge et al., 2014). Using the WAIS Divide ice core as a test case, Fudge et al. (2014) estimated that the interpolation method accumulates uncertainties at a rate of 10 % of the distance to the nearest tie-point, roughly 50 % faster than the uncertainty of periods with identifiable annual layers. The longest interval with no volcanic constraints is between 16 348 and 19 872 BP. At 18 110 BP, the center of the interval, the interpolation uncertainty reaches a maximum of 124 years, although uncertainties are proportionally lower in other intervals with closer volcanic tie points.

Figure 8 shows the total uncertainty estimates associated with the SP19 chronology, with interpolation uncertainties added to the published WAIS Divide uncertainties. The WD2014 and interpolation uncertainties are added in quadrature since the two sources of uncertainty are independent. The maximum estimated uncertainty in SP19 is 533 years at 34 050 BP, the majority of which is attributed to uncertainties in WD2014. While it is not possible to rigorously quantify uncertainties throughout SP19, we believe these estimates provide reasonable and conservative values suitable for most paleoclimate applications. We acknowledge there is additional uncertainty related to the accuracy of our assigned stratigraphic tie points. Because of the conservative procedures discussed in Sect. 3.1, wherein only unambiguous matches were used in linking the WAIS Divide and SPICEcore timescales, it is unlikely that any of these matches are in error. In previous work (Ruth et al., 2007), potential errors associated with tie points have been estimated by removing each tie point one at a time and interpolating between the new series of tie points (with one point missing). If this procedure is repeated for each tie point and for each depth, the maximum error in age resulting from the erroneous inclusion of a tie point is approximately 83 years. However, because clusters of volcanic events were used to match the WAIS Divide and SPICEcore records, each tie point is not necessarily independent. Therefore, this method is more useful at sections of widely spaced tie points with greater potential uncertainties but underestimates the uncertainties surrounding closely spaced events in SPICEcore and WAIS Divide. Examining calcium records from WAIS Divide (Markle et al., 2018) and SPICEcore shows concurrent timing in calcium variations between the two cores (Fig. S5), further supporting the choices of tie points.

Visual stratigraphy in SPICEcore provides an independent check on the glaciochemical layer counting we used to interpolate the Holocene depth–age scale between tie points. Visual layer counting was conducted to a depth of 735 m (∼10 250 BP; Fegyveresi et al., 2019). We calculate the offset between the visual stratigraphic timescale and a linear interpolation between tie points and do the same for the chemistry layer counts (Fig. 9). If both the chemical and visual layer counting methods were capturing the true variability in layer thickness within intervals, then both would show the same structure within each interval.

There is broad correspondence between visual and chemical stratigraphy at all depths, which, given their almost completely independent origin and measurements techniques, is highly reassuring. In detail, though, there is little high-frequency correspondence between visual and chemical layer counts below 1400 BP (150 m depth), although a direct comparison is not possible since visible layer counts were not linked to stratigraphic tie points between 1400–2400 BP and 8400–9500 BP. Furthermore, visible layer counts were matched to the tie points within error of the WAIS Divide timescale, whereas the chemistry layer counts were forced to match within ±1 year of each tie point. In counting visible layers, occasional under- and overcounting of depth hoar layers within annual strata is likely, especially in deeper ice where thinning will make adjacent layers appear even closer. There were some intervals (e.g., 2000–2500 BP) in the core that appeared more homogeneous during viewing, and therefore annual layer choices have a higher level of uncertainty. Because of the differences between methodologies in matching to tie points and because of the uncertainties in visual counting below 2000 BP (200 m), we did not attempt to reconcile the visible and chemical layer counts, but instead rely only on the annual layers in the chemistry data.

Between 100 and 1400 BP, both visible and glaciochemical timescales remain remarkably coherent and do not indicate drift of more than ±2 years. Over this interval, the correlation between the visible and chemical layer offsets from constant annual layer thickness (red and blue curves in Fig. 9) is 0.74. The correlation between the two layer counting methods is as high as r=0.85 between the tie points at 841 and 1268 BP. The discrepancy within the top 100 years is due to the tie point at 10.58 m, which was not included at the time of visible layer counting, as well as low layer chemical counting confidence within the firn column. There is no obvious relation between the accumulation rate and statistical agreement among methods.

The SP19 timescale allows us to produce annually resolved estimates of past snow accumulation to 11 341 BP (Fig. 10). We apply a Dansgaard–Johnsen model (Dansgaard et al., 1969) to estimate the amount of thinning undergone by each layer of ice. Since the entirety of the Holocene in SPICEcore is located within the top third of the core (over 1900 m above the bed), the challenges associated with reconstructing surface accumulation are smaller than at sites with records closer to the bed (e.g., Kaspari et al., 2008; Thompson et al., 1998; Winski et al., 2017). Radar measurements indicate a bed depth at the South Pole of 2812 m, giving an ice-equivalent thickness of 2774 m, using the South Pole density function developed by Kuivinen et al. (1982). We used a kink height of 20 % of the ice thickness and an input surface accumulation rate of 8 cm yr-1 (water equivalent), consistent with the parameters used by Lilien et al. (2018). The average Holocene accumulation rate is 7.4 cm yr-1 (water equivalent), in excellent agreement with results of previous studies (Hogan and Gow, 1997 – 7.5 cm yr-1 to 2000 BP; Mosley-Thompson et al., 1999 – 6.5–8.5 cm yr-1 for the late 20th century). The upstream flow dynamics are too complicated for a static 1-D model to accurately determine the thinning function before the Holocene.

As discussed in Lilien et al. (2018), Koutnik et al. (2016) and Waddington et al. (2007), South Pole layer thicknesses are affected by (1) spatial variability in surface accumulation being advected to the South Pole; (2) past climate-related changes in snow accumulation; and (3) post-depositional thinning due to ice flow. Thinning models can account for only the third factor. An understanding of Holocene climate history as recorded at other sites and in other indicators in SPICEcore, combined with knowledge of the modern upglacier variation in accumulation (Lilien et al., 2018), makes it clear that the Holocene SPICEcore time variations in accumulation are primarily from advection of spatial variations. Figure 10 shows the Holocene accumulation rate in SPICEcore (black) compared with geophysically derived accumulation estimates over space using ice-penetrating radar (blue, details in Lilien et al., 2018). Using the present-day surface velocity field and the inferred 15 % increase in flow rate, present-day upstream surface accumulation rates were matched with corresponding ages at the SPICEcore borehole (Lilien et al., 2018). The close match between present-day near-surface accumulation rates upstream and the annual accumulation rate in SPICEcore shows that the millennial-scale signal of accumulation rate in SPICEcore is related to spatial patterns of snow accumulation upstream of the South Pole.

A striking feature in the Holocene accumulation record in SPICEcore is the sharp dip centered on 2400 BP. Annual layers were notably less clear in that portion of SPICEcore because low accumulation rates led to low sampling resolution (five to six samples per year). For instance, in the interval between 228 and 275 m, the interpreters picked between 511 and 670 years, when 747 years are present based on the volcanic tie points. Because the undercounting of layers in the development of SP19 is coincident with low accumulation rates, we are confident that this undercounting is due to poorly resolved layers in SPICEcore rather than to erroneous tie points or errors in the WD2014 chronology.

The cause of the sharp drop in accumulation is not clear. Modern accumulation rates upstream of SPICEcore were measured using a 20 m deep isochron imaged with ice-penetrating radar (Lilien et al., 2018). These results show lower accumulation in the location where the 2400 BP ice originated (Fig. 10). However, the modern upstream spatial pattern of accumulation shows a decline that is both more gradual and less than half the magnitude of the 2400 BP change in SPICEcore. It is possible that this represents a climatic signal, but we note sharp accumulation variations at this time that are not observed in the WAIS Divide core (Fudge et al., 2016b; Koutnik et al., 2016). Instead, we hypothesize that this event was most likely a transient local accumulation anomaly. Farther upstream at ∼75 km from the South Pole, there is an accumulation low where the rate of change is approximately 3 cm yr-1 in 2 km. With the current South Pole ice flow velocity of 10 m yr-1, this could explain a 3 cm yr-1 decrease in 200 years, similar to what is observed at 2400 BP. If a climate-driven accumulation anomaly did contribute to this sharp change, these anomalies do not appear to be common, as we see no other large and sustained change in the annual timescale.

On sub-centennial timescales, the effects of upstream advection of spatial accumulation patterns are likely smaller, such that annual-to-decadal patterns in snow accumulation in SPICEcore may be indicative of climate conditions. Previous studies have used a snow stake field 400 m to the east (upwind) of South Pole station to assess recent trends in accumulation rate with differing results. Mosley-Thompson et al. (1995, 1999) found a trend of increasing snow accumulation during the late 20th century, while Monaghan et al. (2006) and Lazzara et al. (2012) found decreasing snow accumulation trends between 1985–2005 and 1983–2010, respectively. No significant trends exist in the SPICEcore accumulation record within the last 50 years, although there is a significant (p=0.046) increasing trend in snow accumulation in SPICEcore since 1900. Note that errors in measured firn density would influence this accumulation trend.

SPICEcore nitrate concentrations provide independent support for the Holocene accumulation rate history implied by the SP19 timescale. Previous studies have recognized an association between accumulation rate and nitrate concentration among ice core sites (Rothlisberger et al., 2002). Nitrate in surface snow, exposed to sunlight, results in photolytic reactions that volatilize nitrate and release it to the atmosphere (Erbland et al., 2013; Grannas et al., 2007; Rothlisberger et al., 2000). Evaporation of HNO3 may also significantly contribute to nitrate loss in the surface snow (Munger et al., 1999; Grannas et al., 2007). Under low-accumulation conditions such as in East Antarctica, the amount of time snow exposed at the surface is the dominant control on nitrate concentration, such that with more accumulation, snow is more rapidly buried and retains higher nitrate concentrations (Rothlisberger et al., 2000).

There is close correspondence between accumulation rate and nitrate concentration in SPICEcore (Fig. 11a). This association is strongest on multidecadal to multicentennial timescales with correlation coefficients between accumulation rate and nitrate reaching peak values after a 512-year smoothing (r=0.60; Fig. 11 inset). Although the smoothing makes standard metrics of statistical significance inapplicable, the similarity between time series is expected given the previous work described above. Among sites, an inverse relationship exists between seasonal amplitude of nitrate concentration and accumulation rate. High-accumulation sites such as Summit, Greenland, exhibit strong annual nitrate layering, whereas low-accumulation sites such as Vostok (∼2 cm w.e. yr-1; Ekaykin et al., 2004) and Dome C (∼3.6 cm w.e. yr-1; Petit et al., 1982) do not show annual nitrate layers at all (Rothlisberger et al., 2000). SPICEcore has much higher accumulation rates than Vostok or Dome C and retains weak intra-annual variability in nitrate. While minor compared with multi-annual and longer variability, nitrate seasonal cyclicity, wherein nitrate often peaks in the summer months (described in Grannas et al., 2007; Davis et al., 2004), is discernable in the SPICEcore nitrate record. As expected, the seasonal amplitude of nitrate over the Holocene closely follows nitrate concentration and accumulation rate (Fig. 11b) and is even more highly correlated with accumulation than nitrate concentration itself, especially on multicentennial to millennial timescales (r=0.80 at 512-year smoothing). Nitrate and accumulation rate are entirely independent variables in terms of their measurement, adding confidence to the annual layer counting and tie points underlying the SP19 chronology.

The relationship between inferred variations in accumulation rate and nitrate concentration breaks down prior to the Holocene, but a relationship between nitrate and calcium concentrations emerges. During the Pleistocene, the correlation between the centennial median of calcium and nitrate is r=0.80 (p<0.01; Fig. 12), compared with r=0.26 (p<0.01) during the Holocene. Rothlisberger et al. (2000, 2002) observed the same pattern at Dome C and attributed it to the stabilization of nitrate through interaction with calcium and dust. They proposed that CaCO3 and HNO3 react to form Ca(NO3)2, which is more resistant to photolysis and consequently leads to higher concentrations of nitrate in the glacial age snowpack despite lower accumulation rates. The stabilization effect of calcium apparently overtakes photolysis and evaporation of nitrate in terms of importance only at the very high calcium concentrations as seen in the pre-Holocene ice.

Stable isotope ratios of atmospheric diatomic nitrogen (δ15N–N2) in trapped air in SPICEcore show a pattern similar to accumulation rate within the Holocene (Fig. 11c). δ15N–N2 values were measured using the procedures described by Petrenko et al. (2006). The δ15N–N2 in ice cores is driven by gravitational enrichment and is a proxy for past thickness of the firn column (Sowers et al., 1992). Firn densification rates depend primarily on temperature and overburden pressure, with the second parameter closely linked to the accumulation rate at the site. Low temperatures and high accumulation rates both act to thicken the firn, thereby increasing δ15N–N2 (Herron and Langway, 1980; Goujon, 2003).

We perform a simple attribution study to see whether δ15N–N2 variations can be explained by reconstructed accumulation history or variable temperature. We compare three climatic scenarios in a dynamical version of the Herron–Langway densification model (Buizert et al., 2014). The first uses variable temperature (from δ18O using a scaling ratio of 0.8 ‰ ∘C-1) and variable accumulation (from annual layer thickness) forcing; a second uses constant temperature (-51.5 ∘C) and the variable accumulation forcing; a third uses variable temperature and constant accumulation (7.8 cm yr-1) forcing. The correlations between the δ15N–N2 data and each model run are displayed in Fig. 13 for both raw and detrended time series. The model scenario forced by both temperature and accumulation has the best correspondence with the δ15N–N2 data (r=0.65; p<0.01). While secular changes in temperature appear to be driving the decreasing trend in δ15N–N2, millennial-scale fluctuations in δ15N–N2 appear to be driven by accumulation, supported by the high correlation (r=0.64; p<0.01) with the accumulation-only model run using detrended time series. In particular, a sharp drop in δ15N–N2 is present at approximately 2400 BP, coincident with (and driven by) the local minimum in accumulation. These experiments provide additional confidence in the reconstructed accumulation history. To our knowledge, these data represent the best observation of accumulation-driven δ15N–N2 variation, making it a valuable target for benchmarking firn densification model performance (Lundin et al., 2017).