A 2700-year annual timescale and accumulation history for an ice core from Roosevelt Island, West Antarctica

4.1 Overview of the annual-layer-counting strategy

The uppermost section (0–42.34 m) of the core was dated by manual identification of annual cycles in records of water isotopes and chemical impurities from the RICE main core as well as the RICE-12/13-B shallow core. For this most recent period, several distinct marker horizons from well-known historical events were used to constrain the chronology (Sect. 4.2).

Below 42.34 m (1885 CE), the timescale was augmented using the StratiCounter layer-counting algorithm (Winstrup et al., 2012; Winstrup, 2016) applied to multiple CFA impurity records from the RICE main core (Sect. 4.3). A previously dated tephra layer at 165 m (Pleiades; 1251.6±2 CE according to WD2014) was used to optimize the algorithm settings, but other than that, RICE17 is a fully independent layer-counted ice-core chronology.

The layer-counted part of RICE17 stops at 343.72 m (700 BCE). At this depth, the annual layers became too thin (< 6 cm, i.e., fewer than eight independent data points per year in the best-resolved records) for reliable layer identification in data produced by the RICE CFA setup. The timescale was extended back to 83 000 years BP using the gas-derived timescale, which covers the entire core with lower resolution (Lee et al., 2018). Excellent agreement (±3 years) between the layer-counted timescale and the independent gas-derived age at 343.7 m allows us to produce the combined Roosevelt Island Ice Core Chronology 2017 (RICE17) by joining the the two without any further adjustments.

4.2 Manual layer interpretation with historical constraints (0–42.34 m; 2012–1885 CE)

The top 42.34 m of the RICE17 chronology was obtained by manually counting annual layers in the combined set of discretely measured IC and ICP-MS data, when available, as well as the continuous water isotope and chemistry records produced by the RICE CFA system. The RICE main core starts at 8.65 m of depth, so the top part of the timescale is based exclusively on the RICE-12/13-B shallow core. At 12.5 m, both cores display a distinct peak in their isotope profiles, showing that they can be spliced directly without the need for any depth adjustments. Layer marks for the top 12.5 m were placed according to the RICE-12/13-B shallow core; lower layer marks refer to the main core. In the overlap section (8.65–19.55 m), we used the combined data set from both cores to reduce the risk of timescale errors caused by core breaks or bad data sections.

Layer identification in this section of the core relied predominantly on annual signals in non-sea-salt sulfate (nss ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$), acidity (H⁺), and iodine (I), as these records displayed the most consistent annual signals (Fig. 4). Extreme sea-salt influx events occasionally caused large sulfate peaks, necessitating the removal of the sea-salt sulfate fraction before layer identification. For the top 20 m, the water isotope records also significantly strengthened the annual-layer interpretations. Smoothing due to the diffusion of water molecules in the firn, however, caused the annual signal in the water isotope records to diminish with depth, resulting in a loss of annual signals below 20 m.

Figure 4Manual assignment of annual layers in an upper section of the RICE core. All units are in ppb, except for δ¹⁸O (in ‰), H⁺ (in µeq L⁻¹), and conductivity (in µS cm⁻¹). The CFA chemistry records are smoothed with a 3 cm moving-average filter. The timescale is constrained by two tie points within this interval: the isotope match to RID-75 (blue bar; 14.6 m) and the Raoul tephra horizon (red bar; 18.1 m). At two locations, the annual layering is unclear (shorter bars) but can be mostly resolved based on the tie-point ages. As a result, the uncertain layer at 16.6 m (short grey bar) is included as a year in the timescale, whereas the uncertain layer at 19.7 m (short white bar) is not. For the second layer, the sulfate record suggests that it is an annual layer, but this is not supported by iodine and δ¹⁸O. As we assume an uncertainty of ±1 year for the tie-point ages, the layer at 16.6 m could possibly be removed from the timescale while still adhering to the age constraints. The existence of this uncertain layer therefore gives rise to a 1-year increase in the timescale uncertainty.

Download

Summers could be identified as periods with high stable water isotope ratios, high concentrations of nss ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ and associated acidity (originating from phytoplankton activity in the surrounding ocean during summer; Legrand et al., 1991; Udisti et al., 1998), and low iodine concentrations (due to summertime photolysis of iodine in the snowpack; Frieß et al., 2010; Spolaor et al., 2014). Layer marks were placed according to the depths of concurrent summer peaks in water isotope ratios, nss ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ concentrations, and acidity levels and assigned a nominal date of 1 January.

The uppermost 42.34 m of the RICE17 chronology was tied to several identifiable marker horizons found in the ice-core records relating to well-known historical events (Sect. 4.2.1–4.2.3; Table 2). The timescale was obtained by identifying the most likely set of annual layers while accounting for age constraints from marker horizons (Fig. 4). We conservatively estimated the age uncertainty of the marker horizons to be ±1 year, thereby allowing for some uncertainty in the timing of deposition of, e.g., volcanic material. We further kept track of uncertain layers (i.e., layers that could possibly be added to or removed from the most likely set of layers while still adhering to the age constraints), thereby producing an uncertainty estimate for the timescale. We interpret it as the 95 % confidence interval for the age at a given depth, similar to that obtained from automated layer identification deeper in the core (Sect. 4.3).

4.2.1 The 1974/75 snow surface

The uppermost age constraint was established by successfully matching the RICE water isotope profile to that from the RID-75 firn core (Fig. 5). Drilled in austral summer 1974/75, the snow surface in RID-75 provided the first tie point for the RICE17 chronology at a depth of 14.62 m (Table 2).

Figure 5(a) RICE water isotope profile (δD) compared to isotope data (δ¹⁸O) from the RID-75 core for the period 1950–1975. Diffusion causes the isotope record to smooth over time, and a smoothed version of the RID-75 isotope profile (thick blue) highlights its similarities to the RICE isotope record (black). (b) Total specific β activity (in disintegrations per hour, dph) for the RID-75 core compared to ²³⁹Pu measurements (normalized intensities) from the RICE main core. Both cores show a sharp increase in nuclear waste deposition starting in 1954 CE and several broader peaks hereafter.

Download

4.3 Automated annual-layer identification (42.34–343.7 m; 1885 CE–700 BCE)

For the section 42.34–343.7 m (1885 CE–700 BCE), the RICE17 annual-layer-counted timescale was produced using the StratiCounter algorithm (Winstrup et al., 2012), extended to interpret the annual signal based on multiple chemistry series in parallel (Winstrup, 2016).

StratiCounter is a Bayesian algorithm built on machine-learning methods for pattern recognition using a hidden Markov model (HMM) framework (Rabiner, 1989; Yu, 2010). StratiCounter computes the most likely timescale and the associated uncertainty by identifying annual layers in overlapping data batches stepwise down the ice core. For each batch, the layering is inferred by combining prior information on layer appearance with the observed data, thereby obtaining a posteriori probability distributions for the age at a given depth. The output of StratiCounter is the most likely annual timescale, along with a 95 % confidence interval for the age as a function of depth. The confidence interval assumes the timescale errors to be unbiased, implying that uncertainties in layer identification partly cancel out over longer distances. Previous research has documented the skill of StratiCounter to produce accurate and unbiased layer-counted ice-core timescales (e.g., Sigl et al., 2015; Winstrup, 2016).

StratiCounter was applied to the full suite of CFA records: black carbon, acidity, insoluble dust particles (42.3–129 m), calcium, and conductivity. See Sect. S2 in the Supplement for the specifics of the algorithm setup. Figure 6 shows three depth sections of CFA data and resulting layer counts. Annual cycles in the high-resolution black carbon record became more distinct prior to 1900 CE, and it was one of the most reliable annual proxies in the core. As observed in the topmost part of the core, acidity also displayed an annual signal, although the lower effective depth resolution of the acidity record (Table S1) made it less useful with depth. The calcium and conductivity records frequently displayed multiple peaks per year but contained complementary information to the other proxies and were hence also useful for layer identification (Fig. 6). From 0 to 129 m, an irregular annual signal was also observed in the insoluble particle record, but data below 129 m were corrupted by the presence of drill fluid in the CFA system, which forced us to exclude the deeper part of this record. The discretely sampled ICP-MS data records did not have sufficient resolution to resolve annual layers.

Figure 6CFA data and StratiCounter annual-layer counts (grey bars) in three 3 m long sections (a–c) of the RICE core. In general, the most distinct annual signal is found in the BC record, but the other records also contain valuable information on the annual layering. Due to contamination from drill fluid, the dust record was not used for layer counting below 129 m. For the top section (a), the preliminary set of manual layer counts used for initializing StratiCounter are shown: crosses (x) indicate certain layers and circles (o) indicate uncertain layers. For this section, the manual layer counts have a bias towards counting too few layers in total. This is a general tendency of the manual timescale, resulting in a manually counted age for the Pleiades tephra that is a few decades younger than observed in WAIS Divide.

Download

Decreasing layer thicknesses caused the annual signal in the impurity records to become increasingly difficult to identify with depth, and the layer-counted timescale stops at 343.7 m (700 BCE).

5.1 Changes in density with depth

Bag-mean densities were measured on the main RICE core for the interval 8–130 m, at which depth ice densities were reached (see Sect. S3 in the Supplement). A steady-state Herron–Langway density model (Herron and Langway, 1980) was used to extend the density profile to the surface. Using observed values for initial snow density (410 kg m⁻³), surface temperature (−23.5 ^∘C), and accumulation rate (0.22 m w.e yr⁻¹), the modeled density profile fits the observed values well, especially for the top 50 m (Fig. S1 in the Supplement). At 8 m of depth, the model agrees with the measured values, providing a smooth transition between observed and modeled densities.

5.2 Thinning of annual layers due to ice flow

To obtain past accumulation rates, the annual-layer thickness profile must be corrected for the cumulative effects of ice flow on the thickness of an ice layer since it was deposited at the surface (this being the ice equivalent accumulation rate at that time). The observed layer thickness relative to its initial thickness as a function of depth, i.e., the “thinning function”, depends on the history of vertical strain at the core site. The vertical strain rates can vary significantly over short distances near ice divides, with near-surface ice flow being more compressive at the divide than at the flanks (Raymond, 1983). As a result, the divide flow causes an upward arch in the internal layers beneath stable ice divides. This is termed a Raymond arch, and one is present beneath the Roosevelt Island ice divide (Fig. 7c; Conway et al., 1999). The vertical strain pattern was measured across Roosevelt Island (Kingslake et al., 2014) using repeat measurements of phase-sensitive radar (pRES). These measurements show significant spatial variation in the vertical velocity pattern and provide important constraints for developing the RICE thinning function.

Figure 7(a) Vertical velocity profiles for divide-like flow and flank flow (Kingslake et al., 2014). For divide flow the preferred Lliboutry fit is plotted along with the measured vertical velocities (blue dots). (b) Thinning function with associated 95 % confidence interval. (c) Radar echogram (Kingslake et al., 2014) with traced layers (red) and location of maximum amplitudes of the stack of Raymond arches (blue circles). The location of the modern topographic ice divide (and the RICE drill site) is marked by the returns from a pole. The core site is located west of the maximum bump amplitudes at depth.

Download

The RICE ice core was drilled at the topographic summit of Roosevelt Island. The shallowest layers in the Raymond arch peak below the current summit, but by mid-depth, the maximum bump amplitude of the arches is offset by approximately 500 m towards the east (Fig. 7c). Following Nereson and Waddington (2002), we interpret this slant in the Raymond arch position as a slight migration of the Roosevelt Island ice divide during recent centuries, with the divide previously being located 500 m east of its present position. The ice core intersects the Raymond arch at mid-depths (∼120 m; ∼1500 CE) at about 70 % of the maximum arch amplitude. Thus, it has experienced a transitional flow regime, i.e., a mixture between pure divide flow and flank flow, for much of the past 2700 years. We consider a recent migration of the ice divide towards its present position to be the most likely scenario, with ice flow at the core site becoming increasingly divide-like over recent centuries. For this scenario, we prescribe the following history of divide migration history: prior to 500 years before 2013 CE, the divide was located 500 m east of its present position. Since then, the divide migrated westward, arriving at its present position 250 years ago. During the migration period, the ice-flow regime linearly transitioned to full divide flow and subsequently remained as full divide flow until present. An alternative, although less likely, scenario is that the location of maximum divide flow has always been offset from the topographic summit, in which case the ice in the core has experienced transitional flow throughout the entire period. The uncertainty of the thinning function associated with the divide history is explained in detail in Sect. S4 in the Supplement.

We used a one-dimensional transient ice-flow model with annual time steps to integrate the vertical strain experienced by an ice layer and thereby track the cumulative layer thinning as a function of time and depth. At each time step, the full-depth vertical velocity profile was found by scaling the shape of the vertical profile (as discussed below) with the surface velocity: the vertical surface velocity was determined as the sum of the time-varying accumulation rate and the rate of change in ice-sheet thickness. The ice-sheet thickness change was prescribed and assumed constant in time. Model iterations were run to get an accumulation history consistent with the cumulative thinning computed by the model.

The shape of the vertical velocity profile was found by fitting an ice-flow parameterization (Lliboutry, 1979) to the measured englacial velocities (Kingslake et al., 2014) corresponding to divide flow and flank flow, respectively (Fig. 7a). Details are given in Sect. S4 in the Supplement. The present vertical velocity profile was constrained to match within uncertainty the current vertical velocity at the surface. During the period of transitional flow, the shape of the vertical velocity profile was taken as a linear combination of the velocity profiles for divide and flank flow, following Nereson and Waddington (2002). We used the relative amplitude of the Raymond arch as indicator for the importance of the two flow regimes, and during the early period of transitional flow, the vertical velocity profile was weighted as 70 % divide-type and 30 % flank-type flow.

Figure 7b shows the resulting thinning function derived for the RICE core site. It decreases from 1.0 at the surface (no strain thinning) to 0.24 at 344 m of depth. Past annual accumulation rates in water equivalents can be calculated as the annual-layer thicknesses divided by the thinning function and multiplied with the firn or ice density.

5.3 Uncertainties in the accumulation history

Uncertainties in the inferred accumulation history originate from three sources: (i) identification of the annual layers; (ii) the density profile; and (iii) the derived thinning function. Since the RICE17 timescale was found to have negligible bias (Sect. 7), average layer thicknesses are also not biased, and uncertainties in accumulation history due to layer identification are minor. Uncertainty associated with the density correction is also small. Hence, except for the uppermost part of the record (with essentially no strain thinning), uncertainty in the thinning function dominates the total uncertainty, and only this factor will be considered here.

At the surface, uncertainty in the thinning function is zero. Increasing uncertainty with depth (Fig. 7b; grey area) arises from (a) a lack of measurements in the upper 90 m of the ice sheet to better constrain the present near-surface vertical velocity, (b) variation of the vertical velocity profile over time as the divide may have migrated, and (c) the amount of ice-sheet thickness change that has occurred. Some of the uncertainty is mitigated by the information contained in the internal layering of the Roosevelt Island ice dome. The amplitudes of the Raymond stack indicate that the onset of divide flow was approximately 3000 years ago (Conway et al., 1999; Martín et al., 2006), i.e., prior to the period considered here, that there has been only modest ice-sheet thinning, and that flow conditions have been relatively stable.

We develop an estimate for the uncertainty by calculating the thinning function using an ensemble of model assumptions: two different parameterizations of the vertical velocity profiles, two assumptions about the divide migration history, and three values of ice-sheet thickness change. See Sect. S4 in the Supplement for an extended discussion. The uncertainty was defined as the full range of these 12 scenarios, which we suggest to be a 95 % confidence interval. The uncertainty of the thinning function is substantial for the older portion of the record as the layers have thinned to a quarter of their initial thickness.

6.1 Acidity

The CFA acidity record is driven primarily by the influx of non-sea-salt sulfur-containing compounds, as evident by its high resemblance to the IC non-sea-salt sulfate and ICP-MS non-sea-salt sulfur records in the top part of the core (Fig. 4). Sulfur-containing compounds have a variety of sources, one of which is dimethylsulfide (DMS) emissions from phytoplankton activity during summer (Legrand et al., 1991; Udisti et al., 1998). Correspondingly, acidity displays a regular summer signal, with maximum values in late austral summer (January–February) and minimum values from June through October. This is similar in timing to the seasonal pattern of non-sea-salt sulfate at Law Dome (Curran et al., 1998) and WAIS Divide (Sigl et al., 2016). The acidity contains an annual signal even in the deepest part of the layer-counted timescale (Fig. 8a–c).

6.2 Sea-salt-derived species: calcium and conductivity

As previously noted by Kjær et al. (2016), the RICE conductivity record is almost identical to the mostly sea-salt-derived calcium record (see, e.g., Fig. 6), suggesting sea spray to also be responsible for peaks in liquid conductivity. We hence consider these two records to be representative of sea-salt deposition at Roosevelt Island.

Both records typically peak during early to middle winter (June–July) but with a large spread in magnitude and timing from year to year (Fig. 8d–i), and oftentimes there are multiple peaks per year. The timing of peak values is approximately similar to sea-salt tracers in WAIS Divide (Sigl et al., 2016) and a few months earlier than peak values in Law Dome (Curran et al., 1998). During the most recent period (1900–1990 CE), we observe a summer peak in the average seasonal conductivity signal, which is not present in the calcium record. This is likely the steady summer contribution from biogenic acidity, the seasonality of which is sufficiently prominent to show up in seasonal averages of conductivity in the top part of the core.

6.3 Black carbon

Seasonal deposition of black carbon in Antarctic snow is primarily driven by biomass burning and fossil fuel combustion in the Southern Hemisphere, modulated by changes in the efficiency of long-range atmospheric transport (Bisiaux et al., 2012). Southern hemispheric fossil fuel emissions have increased since the l950s (Lamarque et al., 2010) but are believed to be a minor contributor to total black carbon deposition in Antarctica (Bauer et al., 2013).

Biomass burning in the Southern Hemisphere peaks towards the end of the dry season, e.g., late summer (Schultz et al., 2008). Given the distinct annual signal in the black carbon record (Fig. 8j–l), StratiCounter was set up to assign peaks in BC a nominal date of 1 January (mid-summer). We note that peaks in BC approximately coincide with peaks in acidity (similar to observed in WAIS Divide), thus ensuring the consistency of nominal dates throughout the core. Minimum concentrations of BC are reached in austral winter (June–July).

The annual signal in black carbon changes with depth in the RICE core. During the 20th century, annual cycles exist but are not very prominent (Fig. 8j). Prior to 1900 CE, the signal is much more distinct, with larger seasonal amplitude and higher annual mean concentrations (Fig. 8k–l). A recent decrease in BC concentrations has also been observed in the WAIS Divide and Law Dome ice cores, where it has been attributed to a reduction in biofuel emissions from grass fires (Bisiaux et al., 2012). At WAIS Divide and Law Dome, however, the shift takes place several decades later than observed in RICE.

With increasing depth in the ice core, thinner annual layers cause the amplitude of the seasonal signal to slowly be reduced. Yet, aided by the high effective depth resolution of the black carbon record, the seasonal cycle persists to great depths in the RICE core, with the deepest part of the layer-counted RICE17 timescale primarily relying on the annual signal in this parameter.

6.4 Dust

The seasonal pattern in insoluble dust particle concentrations showed a weak annual signal with a tendency to peak in summer. The simultaneous deposition of black carbon and dust is consistent with both tracers arriving via long-range transport from southern hemispheric continental sources.

7.1 Timescale validation using multi-decadal variability in methane concentrations

Centennial-scale variations in methane concentrations observed in the RICE gas records can also be found in similar records from WAIS Divide (Mitchell et al., 2011; WAIS Divide Project Members, 2015). Stratigraphic matching of these records allowed for a comparison of the respective ice-core timescales.

The gas records from RICE and WAIS Divide were matched using a Monte Carlo technique reported in Lee et al. (2018). The feature-matching routine employed discretely measured records of methane and isotopic composition of molecular oxygen (δ¹⁸O_atm). Over recent millennia, however, the δ¹⁸O_atm concentrations have been stable and hence provided minimal matching constraints. An average spacing of 26 years between successive RICE methane samples contributed to the matching uncertainty. The matching routine identified 18 match points over the past 2700 years, i.e., an average spacing of 150 years. Subsequent visual comparison of the methane profiles suggested minor manual refinements of the match points (8 years on average, maximum 23 years; all within the uncertainty of the automated matching). These adjustments resulted in a slightly improved fit.

Through the methane feature matching, WAIS Divide ages could be transferred to the RICE gas records, i.e., provide an estimate for the RICE gas ages. During the snow densification process, there is a continuous transfer of contemporary air down to the gas lock-in depth, resulting in an offset (Δage) between the ages of ice and gas at a given depth (Schwander and Stauffer, 1984). To obtain the corresponding ice-core ice ages relevant for this study, Δage was calculated using a dynamic Herron–Langway firn densification model (Herron and Langway, 1980) following Buizert et al. (2015). The approach is described in detail in Lee et al. (2018). The model is forced using a site temperature history derived from the RICE stable water isotopes, and the firn column thickness is constrained by the isotopic composition of molecular nitrogen (δ¹⁵N of N₂). In addition to Δage, this formulation of the Herron–Langway densification model produces as output a low-resolution accumulation rate history (Sect. 8.3).

Compared to most other Antarctic sites, the relatively high surface temperature and accumulation rate at Roosevelt Island give rise to low Δage values (averaging 160 years over recent millennia) associated with small uncertainties (∼57 years; 1σ). Combined with the feature matching uncertainty (average: 48 years), total age uncertainty (1σ) in the transfer of WD2014 to the RICE core is on average 80 years (maximum: 111 years) over the last 2700 years.

The RICE17 timescale is consistent with the WD2014 age of the methane match points (Fig. 9b). Based on the automatic matching routine, the agreement of RICE17 with the gas-matched WD2014 ages is better than 33 years for all age markers, with a root mean square (RMS) difference of 17 years. Agreement between the two timescales is even better when using the manually adjusted match points, for which the RMS difference is reduced to 13 years. We observe, however, that all methane match points below 275 m are associated with older ages in WD2014 than in RICE, suggesting a small bias in the deeper part of RICE17.

7.2 Timescale evaluation from volcanic matching

Using the layer counts in RICE17 as a guide, volcanic horizons identified in RICE could be linked to the WAIS Divide volcanic record (Sigl et al., 2013, 2015), allowing for a detailed comparison of their respective timescales. However, a high background level of marine biogenic sulfuric acids precluded a straightforward identification of volcanic eruptions in the RICE records. Based on multiple volcanic proxy records, including two new volcanic tracers described below, match points were found by identifying common sequential patterns of acidity spikes in the two cores.

7.2.1 New and conventional ice-core tracers for volcanic activity

With its coastal location and low altitude, the RICE drilling site receives a significant seasonal influx of sulfuric acids from biological oceanic sources. Biogenic peak values of up to 200 ppb of non-sea-salt sulfate are of the same order as the expected sulfate deposition from large volcanic eruptions, causing the seasonal signal to partly obscure the episodic deposition of sulfuric acids from volcanic eruptions. Volcanic signals appeared more distinct in the acidity record (Fig. S4), primarily due to the higher resolution of this record.

Traditional volcanic ice-core tracers, ECM and sulfur, were not sufficient for identifying volcanic horizons. The ECM record was very noisy, with few peaks extending above the noise level. The resolution of the discretely sampled sulfur record was low (below 67 m: 5 cm, i.e., < 4 samples yr⁻¹), and it was therefore most valuable in the top. Even here, large volcanoes only left a vague imprint in the form of slightly increased sulfur levels over a multiyear period (e.g., Fig. 10a), with their most distinct feature being elevated sulfur concentrations also during winter. Detection of volcanic horizons in the RICE core therefore primarily relied on two new high-resolution tracers for volcanic activity: direct measurements of total acidity (Kjær et al., 2016) and estimated non-sea-salt liquid conductivity.

Figure 10(a) The RICE volcanic proxy records: non-sea-salt sulfur (nss S; orange), ECM (purple), acidity (H⁺; red), and non-sea-salt conductivity (nss cond; blue) based on the conductivity-to-calcium (grey, green) excess. Green and blue areas are sections of positive excess. (b) Matching of the RICE records to the WAIS Divide non-sea-salt sulfur record (Sigl et al., 2015). Vertical bars indicate volcanic match points (Table 2), with the number of annual layers between match points in the two records given according to their respective timescales. The red bar represents the Pleiades tephra horizon (1251 CE). Two match points (grey bars) had prominent peaks in most of the RICE volcanic records and had an average significance exceeding 2 standard deviations of the general variability of the signal, while the peaks corresponding to two other match points (white bars) were less significant. However, direct comparison of the calcium and conductivity records revealed significant conductivity excess also at these depths.

Download

Peaks in the RICE liquid conductivity record were caused primarily by sea salts, and thus this record could not be used as a volcanic tracer on its own. However, from the close similarity of the conductivity and the mostly sea-salt-derived calcium record (e.g., Fig. 6), we could extract a signal of non-sea-salt conductivity, obtained as the conductivity-to-calcium excess (nss conductivity = conductivity − (a⋅ [Ca²⁺] +b), with a and b calculated from linear regression). As a secondary product, this tracer is prone to measurement errors, calibration and co-registration uncertainties, and is further complicated by differences in measurement resolution. We therefore always double-checked by direct comparison between the two records (Fig. 10a; green and grey lines), allowing peaks caused by misalignment and obvious measurement issues to be identified. We observed high consistency between peaks in the non-sea-salt conductivity and the total acidity records.

A sequence of volcanic signals is shown in Fig. 10a. Additional sections are found in Sect. S5 in the Supplement. Compared to acidic peaks resulting from unusually high biogenic summer activity, volcanic imprints could be distinguished as more prominent and/or broader features. Small and short-lived volcanic eruptions, however, were not easily identified.

7.2.4 Quality assessment of RICE17

The volcanic matching to WAIS Divide allowed for a detailed evaluation of RICE17. The WD2014 counting uncertainty is merely 7 years over the last 2700 years, much less than the uncertainty associated with RICE17, and we hence consider it to be the more accurate of the two. Within their respective uncertainties, the RICE17 and WD2014 chronologies are in full agreement at all volcanic marker horizons (Table 2; Fig. 9b). Indeed, the age differences are much less than the accumulated RICE17 age uncertainties. We hence conclude that the inferred confidence bounds on the RICE17 chronology are reliable, if somewhat conservative.

Agreement between the two ice-core timescales is particularly remarkable down to 285 m (∼150 CE). For this most recent part of the RICE17 timescale, the age discrepancy is less than 7 years at all marker horizons, with an RMS age difference of 3 years. Below 285 m, the volcanic matches indicate that RICE17 has a slight (∼3 %) bias towards younger ages. This is corroborated by the methane match points (Sect. 7.1). At 285 m, the effective depth resolution of the CFA impurity records (1–2 cm) becomes marginal compared to the annual-layer thicknesses (7 cm at 285 m), and we suspect that this has caused the thinnest fraction of annual layers to be indiscernible in the ice-core records.

Table 3Mean value and trends in RICE accumulation rates during various time periods. Change points and trends are found using break-fit regression (Mudelsee, 2009). The most likely change points and trend values are provided, as are the associated confidence intervals (in parenthesis: median and median absolute deviation) determined from block bootstrap analysis. Uncertainties (95 % confidence intervals) of mean accumulation rates are calculated based on the uncertainty in the accumulation reconstruction. Accumulation trend estimates from the bootstrap analysis (in parenthesis) include uncertainties in the determination of the change point but not uncertainties associated with the derived accumulation rate history. The analysis does not account for a potential bias due to ice divide migration, which may slightly affect the mean accumulation rate values prior to 1750 CE and the trend during the period of divide migration (∼1500–1750 CE).

* Change point not well determined from bootstrap analysis.

Download Print Version | Download XLSX

Consequently, the deepest section of the layer-counted RICE17 chronology slowly diverges from WD2014, reaching a maximum age difference of 30 years at the oldest identified volcanic marker horizon (343.3 m, 691 BCE ± 45 years; Table 2). This age offset is of similar magnitude as the uncertainty of the methane-derived RICE17 ages at this depth.

8.1 Long-term accumulation trends

Mean accumulation rates at Roosevelt Island over the entire 2700-year period were 0.25±0.02 m w.e. yr⁻¹. From 700 BCE to ∼1300 CE, the accumulation rate at Roosevelt Island shows a slightly increasing trend (Fig. 11f; Table 3), and in 1250 CE, the 20-year running mean accumulation rate reached its maximum over the last 2700 years (0.32 m w.e. yr⁻¹). Since then, accumulation rates have decreased: very slowly until ∼1650 CE, then more rapidly (0.10 mm yr⁻² from 1650–1965 CE; Table 3). A continued acceleration in the decline of accumulation rates is observed towards the present. Change points and trend estimates with their associated uncertainties were identified using a break-function regression analysis (Mudelsee, 2009). However, the high interannual variability in accumulation rates prohibited a very accurate determination of the break points.

8.2 Significant decrease in recent accumulation rates

The Roosevelt Island accumulation history reveals a distinct and rapid accumulation decrease in recent decades: since the mid-1960s, the annual accumulation has decreased with a rate corresponding to 0.8 mm yr⁻², i.e., 8 times faster than over previous centuries. We note, however, that the relatively short time period for conducting the trend analysis, combined with the large interannual variability in accumulation rates, causes significant uncertainty in the precise timing of the change point and the trend estimate.

Considering the period since 1700 CE (Fig. 12), the lowest decadal mean value of the accumulation rate is observed in the 1990s (0.194±0.001 m w.e. yr⁻¹). Except for very low accumulation rates during the 1850s and 1800s, the remaining top six decades of lowest mean accumulation take place after 1950 CE (Table 4). Indeed, over the last 50 years, only one decade (1960s) stands out as not having experienced below-average accumulation at Roosevelt Island.

Figure 12Decadal accumulation rates at Roosevelt Island since 1700 CE. Grey shadows indicate the 95 % uncertainty bounds on the accumulation rate reconstruction due to uncertainties in the thinning function.

Download

Table 4Ranking of decades since 1700 CE according to lowest mean accumulation. Uncertainties (95 % confidence interval) of the mean values are due to uncertainties in the accumulation reconstruction.

Download Print Version | Download XLSX

The current accumulation rate at Roosevelt Island of 0.210±0.002 m w.e. yr⁻¹ (average since 1965 CE, ±2σ) is 34 % less than the peak accumulation rates around 1250 CE and 16 % below the average of the last 2700 years.

8.3 Comparison to the gas-based accumulation rates

The RICE17 accumulation history is in reasonable agreement with the low-resolution accumulation rate output from the dynamic Herron–Langway firn densification model (Fig. 11f, dashed line). The gas-based accumulation rate history does not resolve high-frequency variations, but it shows a slow increase in accumulation rates of 0.02 mm yr⁻², similar to the trend obtained from the annual-layer thicknesses prior to 1300 CE. However, in contrast to the accumulation rate history derived here, the firn-based accumulation rates continue to increase until present. Further, the absolute values of the inferred gas-based accumulation rates tend to generally underestimate the accumulation rates by ∼0.04 m w.e. yr⁻¹ (16 %).

We speculate that the discrepancies may have to do with the shift in RICE water isotope levels occurring around 1500 CE (Fig. 11g), which in the firn model is used to represent temperature change. It has been suggested that this shift is due to factors other than temperature, e.g., changes in atmospheric circulation patterns and/or regional sea-ice extent (Bertler et al., 2018), and the shift also coincides with commencement of the divide migration period. By using δD to estimate temperature change, the firn densification model will produce an increase in accumulation rates towards present in order to preserve a constant thickness of the firn column, as indicated by relatively steady values of δ¹⁵N–N₂ (Fig. 11f, black dots). Further, the model showed a tendency to underestimate the firn column thickness during the earlier part of the period, which may explain the generally lower level of the modeled accumulation rates here.

8.4 Spatial consistency in recent accumulation rates

The spatial representativeness of the RICE accumulation rates was evaluated by comparing year-to-year profiles of layer thicknesses obtained for the overlap sections of the three available cores: RICE main core, RICE-12/13-B, and RID-75 (Fig. 13), with RID-75 located less than 1 km away from the two RICE cores. All cores were corrected for density changes and ice-flow thinning using the density and thinning profile from the main RICE core.

Figure 13Accumulation reconstructions since 1950 CE for the three Roosevelt Island ice cores described in Table 1.

Download

Annual accumulation records from the three cores are very strongly correlated (correlation coefficients ranging between 0.85 and 0.87), indicating the spatial accumulation pattern across Roosevelt Island to be stable through recent time. The spatial consistency of snow deposition at Roosevelt Island is corroborated by the agreement between their measured water isotope records (Fig. 5a). We may therefore disregard depositional noise and consider the temporal variability in RICE annual-layer thicknesses as representative of local snow accumulation.

This consistency in accumulation history is in contrast to a high spatial variability in mean accumulation rates across Roosevelt Island ice divide. Repeat surveys over 3 years (2010–2013) of 144 survey stakes set across Roosevelt Island showed a strong spatial gradient in snow accumulation across the ice divide: accumulation rates of up to 0.32 m w.e. yr⁻¹ were measured on the northeastern flank, decreasing to 0.09 m w.e. yr⁻¹ on the southwestern flank (Bertler et al., 2018). In accordance with these stake measurements, the absolute accumulation rate is found to be significantly less (∼7 %) for RID-75 than for the RICE cores. Differences in accumulation rate between the two RICE cores were insignificant. Spatial variability in mean accumulation rates, combined with different averaging periods, may explain why previous estimates of Roosevelt Island accumulation rates have varied quite significantly (Herron and Langway, 1980; Conway et al., 1999; Kingslake et al., 2014; Bertler et al., 2018).

The representativeness of the Roosevelt Island accumulation rates is corroborated by the high spatial correlation of the RICE accumulation rates to gridded ERA-Interim precipitation data during recent decades (Bertler et al., 2018). These results suggest that precipitation variability at Roosevelt Island, as observed in RICE, is representative for an extended area, which includes the Ross Ice Shelf, the southern Ross Sea, and the western part of West Antarctica.

8.5 Influence of ice divide migration on the accumulation history

The recent period (∼1500–1750 CE) of divide migration at Roosevelt Island may impact the interpretation of climate records from the RICE core. Ice recovered in the deeper part of the RICE core, deposited before divide migration, originated west of the ice divide. Present accumulation rates show a distinct decrease on the downwind (western) side of the ice divide with a gradient of $\sim \mathrm{5}\times {\mathrm{10}}^{-\mathrm{3}}$ m w.e. yr⁻¹ km⁻¹ (Bertler et al., 2018), although it is muted around the summit area. Assuming a stable snowfall pattern through time relative to the divide, the divide migration would have caused reduced accumulation rates to be observed during the early part (until 1500 CE) of the RICE accumulation history. With the ice originating up to 500 m west of the divide at the time of deposition, our estimates of Roosevelt Island accumulation rates during this early period may therefore have a small negative bias of up to $\mathrm{2.5}\times {\mathrm{10}}^{-\mathrm{3}}$ m w.e. yr⁻¹.

Correcting for the influence of ice divide migration, the main impact on the Roosevelt Island accumulation history is an earlier onset of the period with a more rapid decrease in accumulation rates. The differences are small, however, and the overall pattern of trends in the accumulation rate through time remains the same. In particular, ice divide migration has no impact on accumulation rate trends observed before and after the migration period.

9.1 The RICE accumulation history in a regional perspective

9.2 The RICE volcanic record