Dating glacier ice of the last millennium byquantum technologyZhongyi Fenga,1, Pascal Bohleberb,c, Sven Ebsera, Lisa Ringenaa, Maximilian Schmidta,b, Arne Kerstingb, Philip Hopkinsb,Helene Hoffmannb,d, Andrea Fischerc, Werner Aeschbachb,e, and Markus K. Oberthalera

aKirchhoff-Institute for Physics, Heidelberg University, 69120 Heidelberg, Germany; bInstitute of Environmental Physics, Heidelberg University, 69120Heidelberg, Germany; cInstitute for Interdisciplinary Mountain Research, Austrian Academy of Sciences, 6020 Innsbruck, Austria; dAlfred Wegener Institute,Helmholtz Center for Polar and Marine Research, 27570 Bremerhaven, Germany; and eHeidelberg Center for the Environment, Heidelberg University, 69120Heidelberg, Germany

Edited by Philippe Bouyer, Institut d’Optique, Palaiseau, France, and accepted by Editorial Board Member Angel Rubio March 26, 2019 (received for reviewOctober 4, 2018)

Radiometric dating with 39Ar covers a unique time span and offerskey advances in interpreting environmental archives of the lastmillennium. Although this tracer has been acknowledged for de-cades, studies so far have been limited by the low abundance andradioactivity, thus requiring huge sample sizes. Atom trap traceanalysis, an application of techniques from quantum physics suchas laser cooling and trapping, allows us to reduce the sample vol-ume by several orders of magnitude compared with conventionaltechniques. Here we show that the adaptation of this method to39Ar is now available for glaciological applications, by demonstrat-ing the entire process chain for dating of alpine glacier ice by argontrap trace analysis (ArTTA). Ice blocks as small as a few kilograms aresufficient and have been obtained at two artificial glacier caves.Importantly, both sites offer direct access to the stratigraphy atthe glacier base and validation against existing age constraints.The ice blocks obtained at Chli Titlis glacier at 3,030 m asl (SwissAlps) have been dated by state-of-the-art microradiocarbon analysisin a previous study. The unique finding of a bark fragment and alarch needle within the ice of Schaufelferner glacier at 2,870 m asl(Stubai Alps, Austria) allows for conventional radiocarbon dating. Atboth sites the existing age information based on radiocarbon datingand visual stratigraphy corroborates the 39Ar ages. With our results,we establish argon trap trace analysis as the key to decipher so faruntapped glacier archives of the last millennium.

glacier ice dating | argon-39 | atom trap trace analysis

Nonpolar glaciers are dynamic archives of environmentalchange, covering altitudes where other climate records are

sparse. In particular, the European Alps host a unique juxtaposi-tion of glaciers and other climate archives, in close proximity toboth anthropogenic sources of pollutants and the densest networkof long instrumental climate records on Earth. However, only afew glaciers of the highest summit regions, typically above 4,000 mabove sea level (asl), archive snow and thus past climate signals ona quasi-continuous basis such that a stratigraphic chronologybased on layer counting may be obtained. Glaciers at summit lo-cations of lower altitudes are more abundant and have recentlybeen investigated for their potential as climate archives (1). Be-cause periods without net accumulation or even prolonged massloss can occur at these glaciers, their stratigraphy does not includelayers of every single year, making dating by annual layer countingimpossible. Hence, age constraints can only be obtained fromradiometric methods yielding an absolute age information.According to the age range accessible by their half-lives, 3H

and 210Pb are established tools to constrain the age of glacier icewithin the last 100 y (2). Microradiocarbon techniques buildingon analysis of particulate organic carbon extracted from glacierice are now available for dating ice samples older than roughly1,000 y (3, 4), but for ages younger than this, the 14C technique ishampered by ambiguities in the calibration curve as well aslimited sample size (5, 6). Likewise as for glaciers in manynonpolar mountain ranges, a substantial part of the ice volume in

the European Alps may not reach maximum ages that fall withinthe range of the 14C technique. Even if the lowermost layer of aglacier can be dated by radiocarbon methods, the larger portionof the stratigraphy is likely to be substantially younger and hencenot suitable for the application of 14C. As a result, there is animmediate demand in the glaciological community for radio-metric dating of glacier ice within the age range of 100–1,000 y.A particular example is the Little Ice Age period from late 13thto middle 19th century. For climate science, this is a key periodfor understanding climate variability and well suited for modelcalibration as instrumental and historical climate data areavailable for cross validation of climate proxies and model re-sults, in the Alps from about 1500 CE onward (7).A powerful tracer lies within the air bubbles enclosed in gla-

cier ice: the rare isotope 39Ar. It is a unique tracer with a half-lifeof 269 y (8), hence matching the time span of 100–1,000 y. The39Ar is produced by cosmic ray induced spallation of 40Ar in theatmosphere. As such, changes of the cosmic radiation flux overtime do affect the atmospheric abundance of 39Ar (9), althoughthese effects are smaller than the current measurement precision.The keys to practical use of the information stored in climatearchives of glaciers are small ice samples of the order of kilogramsto achieve the required spatial and thus temporal resolution.However, the entrapped gas content and low 39Ar abundance leadsto only 2,000–20,000 39Ar atoms contained in 1 kg of modern ice,making quantitative detection extremely difficult.

Significance

Alpine summit glaciers have a characteristic age range be-tween 100 and 1,000 years. Reliable dating is the key to accessthis valuable environmental archive, including the Little IceAge. Glacier ice contains past air and thus also the rare radio-isotope 39Ar, uniquely suitable as an age tracer for this timespan. Only argon trap trace analysis (ArTTA), the adaptation oftechniques from quantum optics to 39Ar, enables small samplesizes necessary for the application to glacier ice. We presentthe first dating of glacier ice using less than 2 mL STP of argonfrom ∼5 kg of ice, finally opening the door for radioargondating in glaciology.

Author contributions: Z.F., P.B., L.R., M.S., A.K., P.H., A.F., W.A., and M.K.O. performedresearch; Z.F., P.B., S.E., and H.H. analyzed data; and Z.F., P.B., S.E., L.R., M.S., A.K., P.H.,H.H., A.F., W.A., and M.K.O. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission. P.B. is a guest editor invited by theEditorial Board.

Published under the PNAS license.1To whom correspondence may be addressed. Email: iceArTTA@matterwave.de.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1816468116/-/DCSupplemental.

Published online April 17, 2019.

www.pnas.org/cgi/doi/10.1073/pnas.1816468116 PNAS | April 30, 2019 | vol. 116 | no. 18 | 8781–8786

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Nov

embe

r 18

, 202

0

A much larger number of atoms is required for classical analysisby low-level decay counting, implying the need for significantlylarger sample sizes of the order of tons (9, 10). For this reason,environmental routine measurements of 39Ar have until todaymainly been confined to groundwater reservoirs, where nearlyunlimited sampling is possible, e.g., in ref. 11. A strong reductionin the required sample size has become feasible recently by themethod of atom trap trace analysis. This technique utilizes theisotopic shift in optical resonance frequency to capture singleatoms of the desired isotopic species. The required multiphotonscattering for this process yields perfect selectivity. It was originallydeveloped for the isotope 81Kr (12) and has been applied in sev-eral studies (13). Dating old Antarctic ice has been demonstratedusing samples of several hundred kilograms (14).The adaptation of this method to 39Ar poses two main chal-

lenges, namely, the relative abundance that is lower by a factor of1,000 and the lack of a proper reference isotope. We refer to it asargon trap trace analysis (ArTTA), and its application to envi-ronmental studies has been demonstrated with large groundwa-ter (15) and recently with small ocean samples (16). ArTTA isthus the door-opener for broad application of radioargon datingin such research fields as glaciology that have so far been ex-cluded, due to sample size requirements (17).

Site Description and Sample SelectionFor the purpose of this study we selected two glacier sites (Fig. 1)offering artificial glacier caves, which make highly controlledsampling of suitable sample sizes for ArTTA possible. Thecornice-type summit at Chli Titlis (3,030 m asl, central Switzer-land) holds the glacier on its north-facing slope, with a tunneldug for touristic purposes around 100 m into the ice alongbedrock starting at the cable car station (18). Schaufelferner inthe Stubai Alps (Austria) is part of the Stubai glacier ski resortand covers altitudes from 3,270 to 2,810 m asl. To enable accessfor tourists, the glacier cave was drilled close to the cable carstation in 2013 CE.The glacier caves provide direct access to the internal glacier

stratigraphy and thus relative age control. The visual stratigraphyof the ice at both sites shows abundant bubble-rich layers ofwhite appearance, with occasional transparent, nearly bubble-free layers originating from refrozen meltwater. Due to the small

size of these glaciers, their ice needs to be frozen to bedrock, andhence nearly stagnant, to become of substantial age, i.e., a fewhundred years or more. At Chli Titlis cave, ice temperatures arebelow zero throughout the year aided by artificial cooling. AtSchaufelferner cave, initial englacial temperature measurementsalso revealed ice frozen to bedrock. Continuous temperaturemonitoring is currently underway to further investigate the spa-tial distribution of seasonal variability of englacial temperatures.At Chli Titlis and Schaufelferner, ice flow velocities inside thecaves (i.e., near bedrock) are close to zero, indicating the pres-ence of old ice. This also means that the locations have remainednearly unchanged between the two sampling campaigns between2014 and 2018 CE.Both sites have been the subject of previous glaciological in-

vestigations but differ primarily by being located within thenearly stagnant summit region (Chli Titlis) vs. a site having un-dergone substantial ice flow (Schaufelferner). Accordingly, oursampling strategy was guided by the idea to obtain two samples(i) of neighboring layers significantly different in age (Chli Titlis)and (ii) within a single layer of the same age (Schaufelferner).At Chli Titlis, ice blocks of ∼4 kg were cut out by chainsaw.

The outermost layer exposed to the tunnel was removed beforecollecting sample blocks to avoid contamination due to cracks ormelt water. Microradiocarbon dating revealed a strong verticalage gradient (Fig. 2A). For instance, 14C ages of a profile sam-pling 1.9 m of the lowermost ice of the glacier range from 1,246(block 1-2) to 3,138 (block 1-9) y before 2018 CE. For later directcomparison with 39Ar, all calibrated radiocarbon ages have beenadjusted to refer to the year 2018 CE as present. Further detailsregarding the sampling methods and the site characteristics ofthe Chli Titlis glacier cave can be found in ref. 19. Based on the14C age constraints, we selected the two youngest ice blocks forour comparison with 39Ar, blocks 1-1 and 1-2.At Schaufelferner, the ice in the cave has flowed downward

from the top yielding tilted layering without folding. A uniquefeature of this site is the rare finding of two macroscopic particlesof organic origin, a bark and a larch needle, inside the ice (Fig.2B). Both objects have been radiocarbon dated in a previouscampaign. The age range obtained from the bark and needle is sofar considered the best representation of the actual age of this icelayer (20). For 39Ar analysis, two adjacent blocks were obtained by

3000

2800

321

59500

2055

00

59000 60000

2050

0020

6000

2950

3000

3000

675200 675300 675400 675500

1804

0018

0500

1806

00

±

top station

Chli Titlis Glacier (CT)

glacier cave

CT SFIT

ATCH

15°0'0"E5°0'0"E

48°N

46°N

SchaufelfernerGlacier (SF)

glacier cave

10 m

SchaufelfernerGlacier

name date typebark 05.09.2014 14Cneedle 05.09.2014 14Cice (2 samples) 01.05.2017 39Ar

A B C

Fig. 1. Site maps of (A) Chli Titlis glacier cave and (B) Schaufelferner glacier cave with (C) schematic view of the Schaufelferner ice tunnel and sample lo-cations. Due to the summit location, the ice at the Chli Titlis cave is nearly stagnant, showing horizontal layers allowing straightforward sampling and relativeage control via stratigraphy, i.e., older ages at greater depth. The Schaufelferner glacier cave is located downstream of the summit and has undergonesubstantial ice flow. Inclined layers undisturbed by folding are visible within the cave. The GPS coordinates for Chli Titlis and Schaufelferner are reported inthe Swiss grid and Gauss–Krüger system, respectively.

8782 | www.pnas.org/cgi/doi/10.1073/pnas.1816468116 Feng et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 18

, 202

0

chainsaw within one layer at the location indicated in Figs. 1C and2B. The according location is at roughly 4 m horizontal distancefrom the macroscopic organic particles. The 39Ar sampling loca-tion has been chosen as close as possible to the layer containing

the organic objects while permitting us to obtain kilogram-size iceblocks of undisturbed, bubble-rich ice. The layers sampled for 39Arand the organic particles are stratigraphically in close proximity,despite the horizontal distance (as indicated in Fig. 1C).

Discussion of ArTTA ResultsTo ensure reliable performance of the ArTTA setup for thenecessarily small sample sizes, artificial samples of known con-centration have been prepared and analyzed. The results areshown in Fig. 3A (see SI Appendix, Table S1, and ref. 21 for moredetails). Samples with 2 mL STP argon and concentrations of 66,33, and 10 pmAr have been produced by mixing modern argonwith an 39Ar-free sample provided by the dark matter searchcollaboration, Darkside (22). These results confirm the capabil-ity for reliable dating by ArTTA within the age range of 100–1,000 y before 2018 CE.The glacier ice samples were analyzed by ArTTA; Table 1

provides an overview of all important parameters and measure-ment results (see SI Appendix, Table S2, for more details). No-tably, due to very low air content, the samples of the two iceblocks of Chli Titlis yielded as little as 0.5 mL STP argon.However, count rates were still significantly above background(Materials and Methods) and allowed for reliable measurements.The two adjacent ice blocks 17-1 and 17-2 from Schaufelferner

were prepared for ArTTA following two different approaches,yielding samples A and B (Fig. 3B, Inset). For the SchaufelfernerA sample, the core parts of both blocks were obtained by care-fully removing a centimeter-thick layer of every surface exposedto the atmosphere, followed by additional scraping of the cutsurface using a microtome. The Schaufelferner B sample com-bines these removed surface layers and was only cleaned ofsections evidently containing refrozen (i.e., clear, bubble-free)ice. The fact that the results of samples A and B are not sig-nificantly different indicates that contamination by modern air isunlikely and that simple chainsaw sampling is indeed possible asit does not seem to spoil the radioargon dating.Regarding the actual dating results of the ArTTA analyses, we

performed a comparison with age constraints by 14C, carefullytaking into account (i ) the different 14C sample types

A

B

Fig. 2. (A) The sampling site at the summit of Chli Titlis. The wall in the caveshows a distinct horizontal layering, hence no evidence of layer folding. The14C age constraints reveal a strong vertical age gradient within the sampleddepths (19). The two uppermost blocks (1-1 and 1-2) have been analyzed for39Ar. The results of 527+119−156 y before 2018 CE for the uppermost block 1-1 and1, 126+1286−273 y before 2018 CE for block 1-2 are realistic in view of existing ageevidence provided by visual stratigraphy and 14C ages (see Site Descriptionand Sample Selection). (B) Schaufelferner glacier cave. A bark particle (Inset)and a larch needle have been extracted from the wall and allow for mac-roscopic 14C dating (see ref. 20 for more details). The ice blocks of suitablesize used for 39Ar dating have been taken in a later campaign, a few metersapart from the original location. Great care has been applied to select a layerfor 39Ar as close as possible to the layer including the organic objects. The39Ar dating results of 193+53−55 and 198+60−64 y before 2018 CE agree with 14Cfindings (see Discussion of ArTTA Results).

A B

Fig. 3. (A) Proportionality of ArTTA with known samples. Three artificial concentrations have been produced by mixing an 39Ar-free sample with modernargon. The 10 and 33 pmAr samples have been measured once, whereas the mean values of three 66 pmAr and five 100 pmAr measurements (stars) are given(see ref. 21 for more details). The analysis is consistent within the given 1σ confidence interval and confirms the possible dating range of ArTTA. The line ofperfect agreement is shown as guide to the eye. (B) Sequence of ice analysis. In weekly measurement runs, the local mean values of several references (stars)are used to infer the concentration of the samples. Background can be neglected for these measurements due to their 10× modern concentration and shortmeasurement time of 2 h. Contrarily, the counts of sample measurements (circles) have to be corrected for background. This parameter is obtained bybackground measurements with 39Ar-free samples (triangles). To verify the necessary effort for ice sample decontamination, the similar ice blocks 17-1 and17-2 from Schaufelferner have been prepared by two different approaches to yield samples A and B (Inset).

Feng et al. PNAS | April 30, 2019 | vol. 116 | no. 18 | 8783

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Nov

embe

r 18

, 202

0

(macroscopic vs. microscopic) and (ii) relative age informationprovided by the visible stratigraphy in the two caves.The two ice blocks of Chli Titlis are dated at the far old end of

the time span accessible to 39Ar dating. Block 1-1 is 39Ar dated at527+119−156 y before 2018 CE, where the error is mainly given byfinite counting statistics. The adjacent block below, block 1-2, isat least 170 y older, with a best-estimate argon age of 1, 126+1286−273 ybefore 2018 CE. Constraints on the age by 14C dating exist onlyfor block 1-2, where we obtain 1,246–1,378 y before 2018 CE (SIAppendix, Fig. S1). It is important to note that no macroscopicorganic fragments were found at Chli Titlis, and all 14C analyseshad to be performed on microscopic particulate organic carbon,for which known biases toward older ages can exist. Followingthe discussion of potential reservoir effects in ref. 3, the 14C ageis thus regarded as upper age limit (19). Additionally, importantrelative age control is provided by the evident near-horizontallayering; that is, older ice is located at greater depth (Fig. 2A).The 14C results revealed that age differences of several hundredyears occur at close range, even between two adjacent blocks.The large age gradient in the lowermost ice layers is not neces-sarily connected to layer thinning by deformation but caninstead result from past hiatuses in glacier growth or intermittent

melting periods (19). In this context, the 39Ar ages agree withwhat is known to date about the Chli Titlis glacier cave, namely,(i) that age differences of up to several hundred years can occureven between two adjacent blocks of ∼20 cm height and (ii) thatthe 39Ar age of 1, 126+1286−273 y of block 1-2 is a match against thelower carbon age limit of 1,246 y before 2018 CE. The 39Ar agefor block 1-1 of 527+119−156 y adds information and is consistent withthe already known vertical age gradient connected to the inter-mittent ice build-up process.At Schaufelferner glacier cave, the 39Ar results of samples A

and B are consistent and reveal layer ages of 193+53−55 and 198+60−64 ybefore 2018 CE, respectively. The uncertainty in the conven-tional macroscopic 14C dating of the bark particle and larchneedle is caused primarily by ambiguities in calibration of 14Cages within the respective time period. Using OxCal v4.3.2 (23)and the IntCal13 atmospheric calibration curve (6), the mostlikely age ranges assigned to the bark particle and the larchneedle are 375–532 and 505–632 y before 2018 CE at 68 and48% probability (SI Appendix, Figs. S2 and S3), respectively (20).Regardless of the calibration issue, the 14C age range of themacroscopic organic particles has to be considered as an upperestimate of the age of the glacier ice. This is due to the potential

Table 1. Results of ArTTA of glacier ice samples

Sample nameConcentration,

pmAr Argon age, yCountedatoms

Estimatedbackground Time, h

Size,mL STP Weight, kg

Titlis 1-1 25.7+9.2−8.5 527+119−156 50 26 20.00 0.5 4.3Titlis 1-2 5.5+5.6−5.3 1, 1261286−273 31 25 20.00 0.6 4.2Schaufelferner A 60.8+8.9−8.0 193+53−55 92 12 22.30 1.7 6.7Schaufelferner B 60.0+10.1−9.1 198+60−64 66 10 19.50 1.4 5.9

Fig. 4. Setup of ArTTA. At location 1, the sample is compressed into a buffer to compensate for different sample volumes. At location 2, argon is excited into ametastable state for convenient optical transitions at 812 nm. At location 3, transversal laser cooling is applied to increase the 39Ar flux into the detection region. Atlocation 4, 40Ar is deexcited to the ground state and provides a signal for flux monitoring. At location 5, longitudinal laser cooling slows the thermal atoms down totens of meters per second. At location 6, single 39Ar atoms are captured in amagnetooptical trap, and the fluorescence is monitored by a photo diode. At location 7,several turbomolecular pumps (TMP) collect the gas from the apparatus. The sample is cleaned by a getter pump and is recompressed to the buffer volume. Atlocation 8, enriched 39Ar outgasses from the vacuum walls and contaminates the sample. This effect scales with volume and limits the analysis for smaller samples.

8784 | www.pnas.org/cgi/doi/10.1073/pnas.1816468116 Feng et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 18

, 202

0

delay in deposition on the glacier surface after creation of themacroscopic organic fragments, e.g., death of the respective tree.No such age offset exists for 39Ar. In contrast to what is knownfrom polar studies regarding systematic age offsets between theenclosed air and ambient ice matrix (24), no significant effect inthis regard is expected within the 39Ar dating uncertainty. This isdue to the rapid formation of ice at our study sites, typically ofthe order of a decade or less (25). Taking the most probablecalibrated 14C range for comparison with 39Ar results in a dif-ference of at least 85 y between the 14C and the 39Ar ages. Thisage difference is well within what can be explained based onglaciological considerations. The 39Ar and 14C dating resultsrefer to neighboring layers. Likewise as for Chli Titlis, an agedifference of the order of decades is reasonable at Schau-felferner due to hiatuses and melting periods.Based on our results we find no evidence of contamination

related to our sampling method by chainsaw. Thus, sampling iscomparatively simple allowing us to obtain blocks of convenientsize of ∼5 kg. The results of Chli Titlis and Schaufelferner are ademonstration of conclusive 39Ar dating of glacier ice withsamples containing less than 2 mL STP argon. In concert withevidence provided by the visual stratigraphy, the comparisonwith 14C age constraints corroborates the ArTTA age datingmethod, both for its midend (Schaufelferner) and far-end (ChliTitlis) age range. At Chli Titlis, the 39Ar dating results show anage range and a vertical age gradient that reproduce and extendearlier findings obtained from 14C. For Schaufelferner, the mostlikely age ranges assigned by 14C dating of macroscopic organicobjects are determined systematically older than the 39Ar datesbut stay within a range that can be explained based on glacio-logical considerations. Furthermore, both 39Ar and 14C ages in-dicate that the ice at Schaufelferner is likely a remnant of the 1850glacier maximum. The 39Ar dating tool provides a more reliable

age constraint in this case, considering the ambiguity associatedwith the 14C age calibration. Thus, the ice in the Schaufelfernercave originates from the Little Ice Age maximum state.

The Future of Glacier Ice Dating with ArTTASince glaciers at other nonpolar mountain ranges, e.g., CentralAsia, Himalaya, or Andes, are not much different from the Alpsregarding their glaciological characteristics, e.g., size, accumu-lation, and englacial temperature, the impact of this study goesbeyond the Alpine realm. However, the European Alps canprovide a unique combination of (i) glaciers featuring expectedage ranges suitable for 39Ar, (ii) access to kilogram-size icesamples at low cost through excellent infrastructure, and (iii)availability of multiproxy reconstructions of Holocene climate.With the introduction of the ArTTA ice dating technique, we canretrieve hitherto untapped paleoclimate records and validatethem in the nexus of European archives. The main scientificpotential of 39Ar dating of glacier ice is the interpretation of theice layers formed during the last millennium, to reveal the pasthistory of summit glacier growth in the Alps. This history in-cludes a fundamental gap in knowledge in the context of thehighly complex climate patterns during the core period of theLittle Ice Age. It should be noted that in contrast to other gla-ciological dating methods such as surface exposure dating or thedating of wood fragments, 39Ar offers a radiometric datingtechnique of the glacier ice itself. Thus, it can be applied not onlyat glacier tongues but also at summits with access to cold stag-nant ice. Here the ice has undergone no or very little ice flow,which substantially reduces uncertainties in the interpretation ofthe dating results. Even the analysis of ice dynamics during thelast centuries can be done with 39Ar by sampling along thecentral flow line of a glacier. Past ice dynamics can be consideredan important but fairly unknown parameter governing the re-action of glaciers to climate change. Accordingly, developing thefull potential of dating by 39Ar will provide new opportunities forglaciological and glacier-based paleoclimatic research.Because the glacier surface above both the Chli Titlis and

Schaufelferner glacier caves is shrinking rapidly, it is seasonallycovered by sheets to minimize summer ice melt. This providesclear evidence of prolonged negative mass balance and illustrateshow current warming conditions pose an immediate threat tolosing the precious information stored in such glaciers. In thisrespect, the dating tool of 39Ar arrives just in time in modernglaciological research. It also generates a broader impact in thefield of Holocene climate science and other environmental re-search fields. Similar to the introduction of accelerator massspectrometry for radiocarbon dating, the ArTTA technologyopens application fields for 39Ar dating, including glacier ice. Inthis sense, the 39Ar dating technology has the potential to yieldmajor scientific advances in our understanding of several envi-ronmental systems and paleoclimate archives.

Materials and MethodsIce Processing and Argon Extraction. The ice blocks were transported from theglaciers to a −20 °C cold storage at Heidelberg University. Before the ex-traction, they were cleaned by cutting off melted layers. Ice blocks of up to8 kg were then put into a 12.6 L stainless steel container which was evacu-ated with a turbomolecular pump. The ice was melted and the gas wasextracted by freezing it through a water trap onto a liquid nitrogen cooledactivated charcoal trap.

ArTTA is highly selective and thus immune to impurities due to otherelements. Still, a high argon purity and yield was desired to maximize thecounting efficiency and minimize the sample amount. For this purpose, aspecific argon purification system had been designed. The gas compositionwas analyzed with a quadrupole mass spectrometer before the gas wastransferred to a 900 °C titanium sponge getter. With that, all gases wereremoved except for noble gases and hydrogen. At a second titanium spongegetter at room temperature the hydrogen was adsorbed, and the remaininggas fraction, consisting of >99% argon, was captured on a charcoal trap and

Fig. 5. Simulated accuracy for 0.5-mL STP samples. For each given concentra-tion, 10,000 Monte Carlo simulations of the here employed measurement rou-tine have been performed using the experimentally determined parameters, i.e.,outgassing rate and modern count rate. This yields distributions of the mostprobable values of the inferred concentrations indicated by the mode and SD.The range for measurable concentrations is exceeded as soon as the backgroundcontribution cannot be statistically distinguished anymore. For samples of 0.5 mLSTP, this is the case for most measurements below 8 pmAr (Inset).

Feng et al. PNAS | April 30, 2019 | vol. 116 | no. 18 | 8785

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Nov

embe

r 18

, 202

0

transported to the ArTTA setup. With this system, an ice sample was pro-cessed within 4 h with an argon recovery rate of >98%.

ArTTA Setup. A simplified scheme of the ArTTA apparatus is shown in Fig. 4.The purified Ar gas of the ice sample is first compressed into a buffer volumeto compensate for different sample sizes while obtaining a constant flowinto the main apparatus. Laser cooling is building on strong closed dipoletransitions, which are available for metastable argon (Ar*). Thus, metastableargon is produced in an RF-discharge source. Liquid nitrogen cooling is ap-plied to reduce the initial velocity of the atoms.

The flux into the small detection region is increased by two transversallaser cooling stages. The first stage collects the divergent atoms from theeffusive source and collimates them to a beam, whereas the second stagecompresses the beam. Subsequently, the longitudinal velocity is reduced fromthermal velocities to a fewmeters per second by a Zeeman slower. Single 39Aratoms are finally captured and detected in a magnetooptical trap, thusguaranteeing perfect selectivity by millions of resonant photon scatteringprocesses in a spatially confined region. To reduce the off-resonant scat-tering of the huge isotopic background of abundant 40Ar in the detectionregion, this isotope is selectively deexcited from the metastable state to theground state by an additional quench laser. The fluorescence of this processprovides a direct signal for flux monitoring.

Several turbomolecular pumps realize the necessary ultrahigh vacuum inthe apparatus. Their collected gas is cleaned with a nonevaporative getter byremoving any nonnoble gas contributions. The restored gas is compressedinto the buffer volume again, thus enabling full recycling of the sample. Withthis procedure the required sample size is as low as 0.5 mL STP.

The sample size of the current ArTTA setup is limited by outgassing ofembedded argon enriched in 39Ar. The contribution due to this contami-nation is dependent on sample size, concentration, and measurement time.A correction of about 10 pmAr is expected for a volume of 2 mL STP in a 20-hmeasurement.

Measurement Procedure. ArTTA is currently capable of analyzing one sampleper day with relative accuracy dependent on the actual concentration. A fullmeasurement cycle starts with ∼20 h of sample measurement. This is directlyfollowed by ∼2 h of referencing to an artificial sample with 10× enriched39Ar compared with modern concentration. The apparatus is flushed with akrypton discharge while refilling the liquid nitrogen reservoir to remove anyfrozen content on the source and reduce cross sample contamination. Eachsample is framed by at least two reference measurements, but more can beused if appropriate, e.g., weekly averages. It is this temporally local refer-encing which makes the measurement robust against long-term drifts (seeref. 21 for more details).

Data Processing. To infer the sample concentration from the number of atomsdetected, careful estimate of the background is necessary. The contribution

due to embedded argon enriched in 39Ar can be determined by backgroundmeasurements with 39Ar-free samples (22). This long-term memory effect isdescribed by a constant outgassing of 39Ar yielding a time-dependent con-centration

cðtÞ= csample+aout

Vsamplet

with sample concentration csample, sample volume Vsample, and39Ar outgas-

sing rate aout. By integration over time, the detected total atom number Ntot

is given by

Ntot =�csamplet +

12

aoutVsample

t2�λ0

with λ0 describing the count rate of a sample with modern concentration.This parameter is inferred from measurements of reference samples with10× modern 39Ar concentration.

The model is used in a Bayesian analysis to obtain the probability densityfunction for the sample concentration. The reported concentrations are theextracted most probable values and the uncertainties correspond to the 1σ con-fidence interval containing 68.3% of measurements (see ref. 21 for more details).

Notably, the contribution to the detected atoms due to the background isincreasing quadratically in time, which limits the integration time and thusthe accuracy of the inferred concentration.

Accuracy and Limit. To show the potential of our apparatus, we performnumerical simulations of the current measurement routine by using theexperimentally determined parameters λ0 and

39Ar outgassing rate aout. Foreach given concentration, the references and background as well as a totalnumber of counted atoms have been generated numerically in 10,000Monte Carlo simulations and analyzed in the same way as measured data.The most probable values of the inferred concentrations are shown in Fig. 5by indicating the mode of this distribution as well as 1σ, 2σ, and 3σ intervalscontaining 68.3, 95.5, and 99.7% of the values, respectively. The contribu-tion by the embedded contamination is especially dominant for small sam-ple sizes and low concentrations. For the case of 0.5 mL STP samples andbelow 8 pmAr, the background cannot be statistically distinguished in mostof the measurements, but a single measurement can yield concentrationssignificantly, i.e., 1σ, above zero (see ref. 21 for more details).

ACKNOWLEDGMENTS. We thank V. Rädle for carefully reading the manu-script. We further thank the Bergbahnen Titlis-Engelberg and the StubaiGlacier ski resort for their assistance in logistics. This work was supportedby the Deutsche Forschungsgemeinschaft in two joint projects (OB 164/11-1,AE 93/14-1, and OB 164/12-1, AE 93/17-1) and the Austrian Science Fundproject Cold Ice (P29256-N36) as well as by the European Research Commis-sion Advanced Grant EntangleGen (Project ID 694561).

1. Haeberli W, Frauenfelder R, Kääb A, Wagner S (2004) Characteristics and potentialclimatic significance of “miniature ice caps” (crest-and cornice-type low-altitude icearchives). J Glaciol 50:129–136.

2. Gäggeler H, Von Gunten H, Rössler E, Oeschger H, Schotterer U (1983) 210Pb-Dating ofcold alpine firn/ice cores from Colle Gnifetti, Switzerland. J Glaciol 29:165–177.

3. Hoffmann H, et al. (2018) A new sample preparation system for Micro-14C dating ofglacier ice with a first application to a high Alpine ice core from Colle Gnifetti(Switzerland). Radiocarbon 60:517–533.

4. Uglietti C, et al. (2016) Radiocarbon dating of glacier ice: Overview, optimisation,validation and potential. Cryosphere 10:3091–3105.

5. Ramsey CB (1995) Radiocarbon calibration and analysis of stratigraphy: The OxCalprogram. Radiocarbon 37:425–430.

6. Reimer PJ, et al. (2013) IntCal13 and Marine13 radiocarbon age calibration curves 0–50,000 years cal BP. Radiocarbon 55:1869–1887.

7. Dobrovolny P, et al. (2010) Monthly, seasonal and annual temperature reconstruc-tions for Central Europe derived from documentary evidence and instrumental re-cords since AD 1500. Clim Change 101:69–107.

8. Stoenner RW, Schaeffer OA, Katcoff S (1965) Half-lives of Argon-37, Argon-39, andArgon-42. Science 148:1325–1328.

9. Loosli H (1983) A dating method with 39Ar. Earth Planet Sci Lett 63:51–62.10. Oeschger H, Stauffer P, Bucher P, Moell M (1976) Extraction of trace components from

large quantities of ice in bore holes. J Glaciol 17:117–128.11. Corcho Alvarado JA, et al. (2007) Constraining the age distribution of highly mixed

groundwater using 39Ar: A multiple environmental tracer (3H/3He, 85Kr, 39Ar, and 14C)study in the semiconfined Fontainebleau Sands Aquifer (France). Water Resour Res43:W03427.

12. Chen CY, et al. (1999) Ultrasensitive isotope trace analyses with a magneto-opticaltrap. Science 286:1139–1141.

13. Lu ZT, et al. (2014) Tracer applications of noble gas radionuclides in the geosciences.Earth Sci Rev 138:196–214.

14. Buizert C, et al. (2014) Radiometric 81Kr dating identifies 120,000-year-old ice atTaylor Glacier, Antarctica. Proc Natl Acad Sci USA 111:6876–6881.

15. Ritterbusch F, et al. (2014) Groundwater dating with Atom Trap Trace Analysis of39Ar. Geophys Res Lett 41:6758–6764.

16. Ebser S, et al. (2018) 39Ar dating with small samples provides new key constraints onocean ventilation. Nat Commun 9:5046.

17. Collon P, Kutschera W, Lu ZT (2004) Tracing noble gas radionuclides in the environ-ment. Annu Rev Nucl Part Sci 54:39–67.

18. Lorrain R, Haeberli W (1990) Climatic change in a high-altitude alpine area suggestedby the isotopic composition of cold basal glacier ice. Ann Glaciol 14:168–171.

19. Bohleber P, Hoffmann H, Kerch J, Sold L, Fischer A (2018) Investigating cold basedsummit glaciers through direct access to the glacier base: A case study constrainingthe maximum age of Chli Titlis glacier, Switzerland. Cryosphere 12:401–412.

20. Hoffmann HM (2016) Micro radiocarbon dating of the particulate organic carbonfraction in Alpine glacier ice: Method refinement, critical evaluation and dating ap-plications. PhD thesis (Heidelberg University, Heidelberg, Germany).

21. Ebser SC (2018) Dating of ice and ocean samples with Atom Trap Trace Analysis of39Ar. PhD thesis (Heidelberg University, Heidelberg, Germany).

22. Agnes P, et al. (2016) Results from the first use of low radioactivity argon in a darkmatter search. Phys Rev D 93:081101.

23. Bronk Ramsey C (2017) OxCal Version 4.3.2. Available at https://c14.arch.ox.ac.uk/oxcal.html. Accessed April 12, 2019..

24. Sowers T, Bender M, Raynaud D, Korotkevich YS (1992) δ15N of N2 in air trapped inpolar ice: A tracer of gas transport in the firn and a possible constraint on ice age-gasage differences. J Geophys Res 97:15683–15697.

25. Ambach W, Eisner H (1966) Analysis of a 20 m. firn pit on the Kesselwandferner(Ötztal Alps). J Glaciol 6:223–231.

8786 | www.pnas.org/cgi/doi/10.1073/pnas.1816468116 Feng et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 18

, 202

0