Four ice cores were drilled during the 2017–2018 austral summer field season at the coastal East Antarctic Mount Brown South site (MBS1718-Main, -Alpha, -Bravo, and -Charlie). Here we present CFA data for the MBS1718-Main core only.

CFA is particularly well suited for measuring trace chemical species and impurities at high resolution (millimeter to centimeter scale) for long ice cores . This allows for the production of seasonal-scale records of aerosol species spanning thousands of years at sites with sufficiently high accumulation. CFA is advantageous compared to discrete methods such as ion chromatography, as it provides the ability to analyze ice core samples at speeds of up to four centimeters per minute and with minimal time required for decontamination and sample changeover .

The CFA process involves melting a vertical section of an ice core (typically from top to bottom) by placing it on a purpose-built heated plate (melthead) connected to a melt water extraction system. The melthead used in the Copenhagen CFA system is designed to prevent contamination, such that the external portion of the core section is diverted to waste and thus manual decontamination is only required along the horizontal surfaces at core breaks . The inner decontaminated melt water is debubbled and subsequently diverted through a system of measurement instrumentation using a series of peristaltic pumps. Measurements are taken continuously, with data recorded each second . The detectors used are chosen for their short response time and ability to detect the low impurity concentrations observed in polar ice .

The modular nature of CFA allows it to be simplified for deployment to the field for in-situ measurements . It may also be expanded for additional gas and isotope measurements in addition to chemistry and impurities . The Copenhagen CFA setups used in the 2018 and 2019 MBS campaigns both build on the CFA setup described in and . Both campaigns used the same system components, however, the entire CFA system was relocated to a new building between campaigns. The move resulted in some differences between the setups, which are detailed below.

The chemistry and impurity measurements from both campaigns include insoluble microparticles (dust > 1 µm), electrolytic conductivity, calcium (Ca²⁺), sodium (Na⁺), ammonium (NH4+), hydrogen peroxide (H₂O₂), and meltwater acidity. Additionally, sulfate (SO42-) was analysed in 2018. The acidity, sulfate, and calcium records, however, are not presented here, due to independent problems with the measurement systems for those species.

The 2018 CFA measurements took place at the Center for Ice and Climate at the University of Copenhagen in the laboratory described also by (which at that point was located at Juliane Maries Vej Copenhagen) during October-November 2018.

An optical depth registration system (Waycon LLD-150-R5232-50H) was used for registration of the melt speed during the 2018 campaign, in a setup similar to that used in . While the laser distance meter allows for precise melt speed and sample length measurements (± 2 mm accuracy, and 0.5 mm resolution), it can be temperamental as vibrations from ambient activity (heavy footfalls or accidental knocking of the freezer enclosure) can disrupt measurements. Additionally, the ice is never a perfect fit to the frame and as it melts it changes from leaning on one side of the frame to the next causing some additional noise seen in the very precise laser measurement. Additionally, the space required for the laser system limits the length of the frames used in the freezer, thus limiting the sample stick length to 50 cm. For the campaign in 2018, 10 sample sticks (5 m) were melted continuously per run, book-ended by standards runs (see Sect. 4.3 and Table  in the Appendix).

A sulfate measurement system following the method used for NorthGRIP was trialed in the 2018 system, but due to a high limit of detection of the system, sulfate measurements were unsuccessful, and thus are not presented here. Discrete sulfate measurements of the full MBS record via ion chromatography can be found in .

In October–November 2019 the Physics of Ice, Climate, and Earth at the University of Copenhagen moved to a new address at Tagensvej 16. Due to the relocation the 2018 system was dismantled and rebuilt in the new location. During reassembly changes were made to the system, described herein.

A cable-driven rotary encoder (draw wire position transducer, SX80, WayCon; sensitivity: γ = 25 counts mm⁻¹, linearity: ± 0.15 %) was implemented for depth registration in 2019 similar to that described in . This encoder setup freed up more space in the freezer unit, allowing for melting of meter long sample sticks and thus the data from the 2019 campaign has fewer sample breaks per run as 5–10 samples (5–10 m) were melted continuously per run. Further, all chemistry lines were optimized in length to minimize dispersion and thus smoothing of the final record.

As part of the 2019 rebuild, a gas extraction system (3M, Liqui-Cel MM-0.75 × 1 Series) coupled to the mixed air-water stream from the top of the debubbler unit was added to the CFA. While the gas extraction system was not used for data collection during the MBS melting campaign, it was sporadically implemented for testing and refinement throughout the 2019 measurements. Further detail on the gas extraction system and its impact on measurement quality follows in Sect. below.

As described above, the two system setups used in the MBS CFA measurements are largely similar, using much of the same equipment and instrumentation. The primary differences, and their potential impact on measurements between the two setups are as follows.

Depth registration system: The 2018 setup utilized a laser distance meter reflecting off of a weight placed on top of the ice, while the 2019 setup used a cable-driven rotary encoder attached to a weight resting on the top of the ice. Additionally, the two instruments result in different amounts of “downtime” in the system during ice stick frame changeover, with slightly more time required when changing frames with the rotary encoder system. While these factors might influence the resulting CFA depth scale, these offsets can be cross-checked with the lengths measured in the freezer during sample preparation. These differences are minimal, and likely within the scope of the measurement error of the instrumentation.

Precise depth information is crucial for ice core data. The position-derived melt rate data collected during CFA campaigns is used to reconstruct the total length of ice melted, to which the recorded length of ice removed during preparation must be re-introduced at the appropriate depths. To account for the time when the encoder is removed during the frame changes, for both the laser and draw-wire encoder, the position data is used to identify the times just before and after each frame change period, and the average melt rate over the previous period of uninterrupted melting is used to reconstruct a continuous melt rate. The down-time that occurs during frame changeover is slightly longer for the draw-wire encoder than that of the laser encoder, as the draw-wire assembly that is attached to the encoder must be fully removed from the remaining ice and subsequently replaced once the new frame has been mounted.

The information recorded during sample preparation is used (including ice removed for decontamination at sample ends and breaks and poor-quality ice not analyzed), together with the position of each break as logged during melting, to create appropriately sized gaps in the depth scale corresponding to the missing ice. The depth scale is finalized by assigning the top of each CFA data run to the recorded top depth of the corresponding bag from the accepted field depth measurements. Manual decontamination of the cores can introduce slight (sub-millimeter) discrepancies due to mis-reading lengths on the standard ruler used in preparation as well as slight misjudgments and biases in identification of break positions during melting. These unavoidable inaccuracies are estimated to be on the scale of one or two millimeters at most.

In rare cases, measured lengths of the same stick varied from one measurement to another (for example, when measured in Hobart prior to shipping to Copenhagen). This was found to be due to slightly slanted cuts where one core meets the next, with one measurement taken from the long side and another taken from the short side. These discrepancies were found to amount to less than 0.01 % of the length of the record . Careful consultation of comprehensive line scan images of the core sections was used to correct for these errors and determine the most accurate core length across all measurements. This process required the depth scale for the entire Main core to be developed, taking into account small differences in field to lab core lengths as well as any small discrepancies in the IC (discrete chemistry) stick lengths. This is described in detail in . Any such differences were accounted for in the development of the CFA depth scale in the precise assignment of the top depths at approximately every meter. This top depth assignment was done to prevent any small discrepancies from propagating through the length of the core. Additional description of the depth scale procedures can be found in .

The absorption and fluorescence spectrometric methods used for chemistry concentration measurements require calibration to convert from instrument signal voltage/light intensity to concentration. In order to properly calibrate these instruments, standard solutions with predetermined chemical concentrations are passed through the system before and after each sample run. A multi-element standard solution is used for a three step calibration of the Ca²⁺, Na⁺, and NH4+ measurements (Merck Certipur^® IC Multi-element Standard VII). A single component standard solution is used for a two step calibration of H₂O₂ (Sigma-Aldrich Hydrogen Peroxide Solution 30 wt % in H₂O). The multi-element standard solutions are prepared prior to each run, and the peroxide standards are prepared prior to each run from a first dilution prepared at the start of each measurement day. All standards are prepared using ultra-pure deionized water (Merck RiOs™ 16 Milli-Q^®). Standard recipes and resulting concentrations are presented in Table .

Calibrations are calculated from the standards, which were run at the beginning and end of each measurement run, as described in and . This allows for monitoring of system drift and provides ample standards data for each run, in case any issues arise during a standards run. Each standard solution is passed through the system in series and the resulting signal voltage was recorded both manually and by the LabVIEW software used to operate the system. For the fluorescence method (Ca²⁺, NH4+, and H₂O₂ measurements), the standard calibration is based on the linear relationship between concentration and fluorescence signal voltage. Sodium is a pseudo-absorption method, and calibration is based on a curve fit to the standard signal response . Calibrations are computed using a semi-automated script written in MATLAB. The script isolates the signal voltage at each standard input concentration and calculates the appropriate calibration coefficients. The standards are well fitted, with the average r2 value for the calibration curves for each analyte being greater than 0.99.

Due to differing distances of each of the measurement instruments from the melter, there is a time delay between when the ice passes the melter and when the measurement takes place.

The time delay between the ice on melter and when the measurement is recorded down stream is calculated in two parts. Firstly, there is the elapsed time from when the sample parcel passes the melter to when it is divided into disparate melt streams for each analytical unit. The bulk melt delay time is measured during sample melting as the elapsed time from when the first sample ice passes the melter (recorded as observed by lab operator) and the initial peak response observed in the conductivity measurements. Conductivity is used as it is located closest to the melter (by tubing distance/mixing volume as well as time), and thus the first signal to respond. In both systems, the approximate total time offset between the sample ice reaching the melter and the response seen in the conductivity signal was approximately 40 s.

Secondly, there is a delay time individual to each species measured due to the internal dynamics of each analytical melt stream. These response delay calculations are computed based on each instrument response time during the standards runs. This additional delay time is measured as the time elapsed between when the signal increase is seen in the conductivity measurements and target species. We determined that additional delay time as time at which the derivative of the response curve of each standard reaches a maximum (approximating the midpoint of the signal response rise). The average delay time (from all standards runs used in calibration) is presented in Table .

Delay times vary for each species, but are similar between the two system setups (Table ). Other choices for delay calculations could have been to use the start or end of the signal rise. However, due to smoothing, these points can be hard to reliably identify, whereas the maximum of the derivative of the increase is easily and systematically identified. However, we recognize that this can introduce a minor offset between methods that are highly smoothed in comparison to those that have faster response times.

As the delay times can vary slightly from run to run, due to factors including tubing length, tubing wear, and fluctuations in melt speed, delay times for each species were measured and applied individually to each CFA run of 5–10 m of ice melted. Each delay time estimation was calculated based on the standards data collected either immediately prior to or following the run.

Expected concentrations of calcium at the MBS site are very low (median chemistry concentration from the discrete chemistry measurements is 0.142 ppb). Due to the nature of the system, and the limits of the ultrapure water system used to produce the standards, reagents, and baseline values for measurements produced here, this is significantly below the limit of detection we are reliably able to achieve using the Copenhagen CFA system (see Table ). As the CFA Ca²⁺ data for MBS does not show any discernible seasonal cycles throughout the record, and the calibrated Ca²⁺ concentration values measured close to or below the LOD of our instrumentation, we consider these values too low to report.

Data is registered at one second time steps, and thus the sampling resolution varies with melt rate. The melt rate of the CFA system is selected to optimize resolution, while accounting for density changes from firn to ice, however with the introduction of the gas extraction system in 2019, the melt rate was calibrated to produce optimal gas volumes for simultaneous measurement. The 2018 CFA setup operated with a target melt rate of ∼ 4.5 cm min⁻¹ throughout the upper 35 m depth, decreasing across 10 m to reach ∼ 3 cm min⁻¹ below 45 m depth (median actual melt rate 3.45 cm min⁻¹ across the full 2018 record). The 2019 campaign (∼ 95–295 m depth) had a target melt rate of 4 cm min⁻¹ (median actual melt rate 4.08 cm min⁻¹). These median melt rates produce a direct sampling resolution (independent of smoothing due to response time) of 0.58 and 0.68 mm, respectively.

The internal dynamics of the system (mixing volumes and flow conditions within tubing) lead to signal dispersion, as described by . This gives a smoothing effect, wherein a discrete parcel of sample is measured in the CFA system as a dispersed signal spread across a short time period (on the order of a few seconds) spanning the expected signal response time for that parcel . This dispersion time is calculated as as an e-folding time based on the 10 %–90 % rise time in the signal response during the standards run. Based on the melt rate, it is possible to use this dispersion signal time to calculate the effective smoothing length (and thus a minimum realistic resolution) for each species (Table ). Although the melt rate used in 2019 was faster than in 2018 (4.08 and 3.45 cm min⁻¹ respectively), the shorter response times (due to shorter tubing lengths used in the rebuilt system) resulted in a higher resolution in 2019.

It is worth noting that, as described by , the signal peak under realized (dispersed) flow conditions occurs slightly before the expected signal response under idealized “plug flow” conditions. This effect is linked to the specific dynamics of each CFA system, and similar to , we do not quantify this slight offset for the Copenhagen system, but note that it could lead to a slight systematic offset (on the order of a few millimeters) biased towards shallower sample depths.

During the 2019 CFA melting campaign, the introduction of a new gas extraction system was trialed. This system setup comprised a gas extraction line originating from the upper outlet of the debubbler, coupled to a micro-module which extracts dry air from the sample stream for gas analysis. These system trials influenced the internal pressure balance of the chemistry system tubing. This had a significant impact on the meltwater acidity (pH) measurements , which suffered from back pressure changes influencing the sensitive dye to water ratios, making the pH record unusable. The gas extraction system also impacted the insoluble microparticle data collected by the ABAKUS laser particle counter.

While the specifics of the gas extraction system are beyond the scope of this data description, there is a visible signature in the microparticle record that only occurs when the gas extraction system was on-line with the chemistry CFA. The signature can be seen as a significantly elevated baseline in the microparticle record (measured from Milli-Q^® ultrapure water through the system), as well as an alteration in the amplitude of the dust signal. These changes correspond with the recorded valve switches of the gas system. Due to the nature of the system interference, we are not able to reliably correct for this interference. We therefore only present the insoluble microparticle data measured in 2018 before the gas system testing began. Due to additional concerns with data quality and contamination of the microparticle record in the deepest portion of the dry drilled section, we only present microparticle data down to 85 m depth.

The limit of detection (LOD) varies for each detection channel. Sample concentration ranges, LOD, median, and standard deviation for each of the analytical detection channels are presented in Table . LOD is calculated as 3 times the standard deviation of the blank signal measured on ultrapure (Milli-Q®) water during the standards runs . The standard deviation (SD) of the lowest concentration standard (9.9 ppb for Ca²⁺, Na⁺, and NH4+ 60 ppb for H₂O₂) is also used as a measure of precision of the system. Frequency histograms for each measured species are presented in Fig. .

Measurement uncertainty for the wet chemistry analyses is driven primarily by uncertainty in standards preparation, due to instrument uncertainty of the microliter pipette (20–200 µL Socorex Acura manual^®) and bottle-top dispenser (Dispensette^® III 5–50 mL, Brand GmbH). This uncertainty is discussed in , and as standards used here were prepared in a similar manner, the uncertainty estimate is less than 10 % .

Despite careful decontamination of the ice samples prior to melting, some cleaning of the data is still necessary. Often, very short-duration, high concentration spikes can be seen in the data signal due to occasional air bubbles passing through the measurement cells in the instruments despite the use of debubbler and gas permeable membranes (Accurel^®) immediately prior to each detector. These particular signals are easily identified and removed from the record by a simple 10 s smoothing, or (as implemented here) by using a filter that applies a threshold cutoff to the differential of the signal (due to the characteristics of these features). The passage of air bubbles through the detectors results in a signal which is characterized by a significantly steeper increase (and subsequent decrease) in the signal voltage than anything produced by variability in the ice core sample. Because of this characteristic peak shape, we are able to define for the differential of the dataset, a threshold, values above which we use to define “bubble spikes” in the data, which can be safely removed from the dataset.

Contamination signals at core breaks, due to contamination from drill fluid (Estisol-140) or general laboratory contamination, are also relatively easily removed . This type of contamination is characterized by a steep increase in particle count followed by an exponential decay as the contaminated sample passes through the system. For this record, the data coinciding with signals deemed to be caused by this type of contamination have been manually removed from the dataset.

Due to the very low concentrations in Antarctic ice of all species measured and the sensitivity of instrumentation, some measurements fall very close to or below the limit of detection of the system. Data that fall below baseline values (measured from Milli-Q^® water run through the system before and after each run) have been removed from the dataset.

To verify data quality and depth scale accuracy, and to demonstrate the ability of the CFA record to correlate with signals identified in the discrete dataset, we investigated all conductivity peaks that fall above 3σ from a 90 cm moving mean of the conductivity record. The 90 cm window was chosen as on average it represents an approximation of 3 years, despite the known accumulation variability shown in the MBS cores. Using this method, we are able to identify 72>3σ conductivity peaks. Of these 72 peaks, 16 correspond with volcanic events reported in to within one year age-at-depth uncertainty. Volcanoes identified in this method include Pinatubo (1991), Agung (1963), Tambora (1815), Mount Mélébingóy/Parker Peak (1640), Ruiz (1595), and Samalas (1257), in addition to a number of the unknown peaks identified in MBS (Table ). This method identifies additional peaks above 3σ, however, these have not been matched to volcanoes identified in , and may be attributed to factors other than volcanic signals.

We attribute existence of peaks that are not identified in both datasets to the differences in the data types used (comparing conductivity to non-sea salt sulfate) as well as the method of peak identification. Because the peak identification method used here is based solely on bulk conductivity, it is likely that some peaks are related to sources of other soluble ions, in addition to volcanic sulfate. use the non-sea salt sulfate, calculated from bulk sulfate following , a more specific indicator of volcanic horizons than bulk conductivity, and therefore likely to be better able to distinguish more volcanic horizons. Additionally, identified and matched volcanic events qualitatively based on assessment of the size, shape, and relative concentration of sulfate compared to existing Antarctic records (WAIS divide, Law Dome, and Roosevelt Island), while we use a simple cutoff threshold method (3σ above the moving mean), which is likely to miss lower-magnitude volcanic horizons that would be more easily identifiable by their shape rather than simply the maximum peak conductivity.

Due to the marginal location of the MBS site, volcanic signals stand out less from the background conductivity record (as a measure of bulk ions) than those of more inland sites . This is due to the proximity to the coastline and increased influence of oceanic sources of impurities leading to dilution of the signal . When a total ion budget is considered (based on discrete ions), peaks in the total ions can be seen that are not associated with elevated sulfate (see Fig. ).

use back-trajectory analysis through the satellite-era to demonstrate that meridional transport pathways are disproportionately represented during the extreme precipitation events that influence MBS accumulation. We hypothesize that this meridional transport pathway might also influence impurity transport to MBS. The greater background influx of impurities due to the coastal location is likely to inhibit the ability of the simple threshold cutoff method used here on the conductivity record to `see” all volcanic horizons able to be identified based on volcanic sulfate.

Accounting for the impact of the different methods used, we consider the ability of the CFA conductivity signal to identify half of the volcanic events found in the non-sea salt sulfate signal by across the MBS record to be an external validation of the CFA depth scale and conductivity record described here.

Because discrete measurements have been performed on the full MBS-Main core , we are able to use the discrete record as a comparison dataset to investigate similarities and/or differences in the datasets. The only species measured directly with both methods is sodium, which we use here to cross-check the reliability of our methods and data quality. The published discrete datasets present the raw IC data with minimal data quality assessment beyond the satellite-era section (top ∼ 20 m) of the main core presented in . As such, there appear to be some outlier samples, likely due to lab contamination, however, a quality assessment of the discrete dataset is beyond the scope of this work. The discrete IC data were measured in micro-equivalents per liter (µeq L⁻¹), and have therefore been converted to ppb for comparison to the CFA dataset presented here. In order to produce a direct comparison of the two datasets, the data were resampled (using a simple linear interpolation method) to a common depth scale at 3 cm resolution.

On visual inspection, the two datasets appear to be very well correlated. The seasonal variability of the Na⁺ signal can be clearly identified and peaks can be easily matched between the two datasets. Slight deviations in magnitude can likely be attributed to the differences in measurement method used (ion chromatography vs. fluorescence detection). The CFA method for measuring sodium ions relies on complicated chemical reactions including a reaction column which must be regularly recharged , and thus minimal drift in the signal is expected. While there are also slight discernible depth discrepancies between the two records, these discrepancies are minor (on the order of a few mm) and do not propagate significantly throughout the record. The discrete Na⁺ dataset is included for visual comparison in Fig. as well as Fig. . It can be seen in the histogram comparison of the dataset that the measured concentration ranges are similar, however with a somewhat higher median concentration measured in the CFA dataset. To assess variability between the two datasets, we have compared the records using Pearson's correlation, and the two datasets show a significant, but moderate to weak correlation (r=0.4307, p>0.001).

While an inter-lab comparison across these two methods is helpful, it is important to note that issues such as contamination and measurement error are not inherently more likely in one method than the other. Discrepancies between the two datasets, therefore, do not necessarily indicate problems with the CFA dataset presented here. Regardless, the significance of the correlation between the datasets, both in the measured concentration ranges (Fig. ) and the alignment with depth (Fig. ), do provide validation and reassurance with regards to the quality of the measurements presented here.