1. Introduction
Radiocarbon dating has been performed at Uppsala University since the 1950s (Olsson Reference Olsson1957, Reference Olsson1959). Based on a 6 MV EN tandem accelerator (High Voltage Engineering Europa B.V., Amersfoort, The Netherlands), accelerator mass spectrometry (AMS) was introduced in the early 1980s (Possnert Reference Possnert1984). Using the AMS technique, it was possible to routinely date samples between 0.5 and 1 mg C equivalent (Possnert Reference Possnert1985), a substantial reduction in required material, that opened new avenues for application (see e.g., Linick et al. Reference Linick, Damon, Donahue and Jull1989).
In 2001, a new 5 MV Pelletron tandem accelerator (National Electrostatics Corp., Middleton, WI, USA) was installed at the Ångström Laboratory at Uppsala University. For radiocarbon dating, the system was equipped with two Cs sputter ion sources, each fitted with a multiple-sample holder. One source was dedicated to samples with high 14C content, while the other was optimized for high-throughput measurements of low-14C archaeological samples. Of the six beamlines available on the Pelletron, one was dedicated to AMS measurements and included a bending magnet for suppression of isobaric interference. Further details on this setup can be found in Salehpour et al. (Reference Salehpour, Possnert and Bryhni2008).
The improved measurement statistics and substantially reduced background of the Pelletron system enabled analysis of even smaller sample amounts. This capability led in 2008 to the establishment of a dedicated laboratory for preparing ultra-small samples using the Uppsala Ultra Small Sample technique (UUSS, Salehpour et al. Reference Salehpour, Håkansson and Possnert2013). AMS measurements of ultra-small samples facilitated major advances in life sciences, particularly through bomb-pulse dating of DNA, which enabled assessments of tissue regeneration and pharmaceutical drug tracing (e.g., Salehpour et al. Reference Salehpour, Forsgards and Possnert2010, Reference Salehpour, Håkansson and Possnert2013, Reference Salehpour, Håkansson and Possnert2015; Spalding et al. Reference Spalding, Bergmann, Alkass, Bernard, Salehpour, Huttner, Boström, Westerlund, Vial, Buchholz, Possnert, Mash, Druid and Frisén2013).
To increase sample throughput, improve reliability, and achieve higher precision, while reducing competition for beam time on the large tandem accelerator also employed for ion beam materials analysis and modification, a dedicated accelerator system for radiocarbon analysis, a MICADAS (Ionplus AG, Zurich, Switzerland), was installed in 2014. As MICADAS is dedicated to detection of carbon isotopes exclusively, it requires a much smaller laboratory footprint and allows for replacement of the electromagnets of earlier AMS machines with permanent magnets both on the low and high energy side, drastically reducing power consumption. Its specialization in carbon-isotope measurements also simplifies technical operation of the machine and reduces the risk of contamination.
The first five months of successful operation of the system at the Tandem Laboratory were summarized in Salehpour et al. (Reference Salehpour, Håkansson, Possnert, Wacker and Synal2016). Since this early status report, several technical upgrades have been implemented: The acquisition of a second source in April 2017 facilitates effective maintenance of the machine and reduces down-time significantly. The Automated Control Software (ACS) was installed in September 2019. In December 2021, the valve-driven stripper gas inlet was changed to the new capillary layout, which, once adjusted to the correct capillary length, provides a stable supply of helium stripper gas to the accelerator. To save time during sample revision and reach a better final vacuum, an additional fore-vacuum pump was added to the source chamber in February 2022. In April 2024, a ceiling-mounted crane was permanently installed above the low energy magnet for improved safety and convenience during change of magnet foils in the LE-magnet pulsation system (due to potential sparking from the pulsing HV). The most recent upgrade (April 2025) involved adopting the new nozzle design in the sputter ion source, enabling operation at lower Cs temperatures and reducing clogging of the Cs line, a common cause of operational interruptions. Additional safety improvements were also implemented, including a sputter shield on the accelerator entrance on the low-energy side to prevent carbon buildup and high-voltage shorts (September 2023), as well as a lead shield at the pressure gauge on the source chamber to improve radiation protection (June 2024).
Over the last ten years, the annual number of samples submitted for dating has steadily increased, reaching current levels of about 5000 dated samples per year, of which about 45% originate from charcoals, more than 15% from macrofossils and more than 10% from unburned bones. The demand for dating more unusual materials (such as DNA, ecological fuel and micro-carbonates) or smaller sample quantities is rising, prompting us to seek new solutions and improve our methods and routines. In this contribution, we discuss three specific observations relevant to the goal of providing high-quality measurements for very small samples:
-
1. the target scatter derived from a single magazine run is regularly found to be lower than suggested by long-term statistics;
-
2. the low ion current obtained from small samples is particularly sensitive to contamination and non-linearities in the measurement and may require correction; and
-
3. the δ13C and δ15N measurements from an IRMS were found to depend on the total carbon content of a sample, with δ13C values affecting radiocarbon ages.
2. Descriptive background
At the Tandem Laboratory, a wide variety of sample materials, ranging from solids to gases and liquids, from carbon dead to hypermodern, and from large pieces to nearly invisible quantities are dated. To accommodate this diversity and obtain reliable radiocarbon results, different pretreatment lines are employed together with dedicated sets of standard samples used to correct the measured BP ages.
In general, a MICADAS measurement is continued until a level of 300–350 thousand 14C counts is reached for all “large” oxalic acid standards (SRM-4990C, in the following called OxII, ca. 1 mg C equivalent). With an average 12C current on the high-energy side of approximately 20 μA, this corresponds to almost exactly one day of measurement time for a full magazine, or about 30 minutes of data acquisition per sample. Magazines containing smaller samples are equipped with correspondingly smaller standards and are run longer to obtain sufficient counting statistics, typically 45–60 minutes per sample. A MICADAS sample magazine has 39 available slots, of which 20–50% are filled with standard materials and the remainder with samples of unknown age.
The BP-age or Fraction Modern Carbon (F14C as defined by Reimer et al. Reference Reimer, Brown and Reimer2004) of a sample is determined from the measured ion currents of the relevant isotopes 12C, 13C and 14C followed by analysis using the software package BATS (Wacker et al. Reference Wacker, Christl and Synal2010). To obtain an absolute value, the measurement must be compared to a known standard. Note that in the rest of the present contribution, we focus in our discussion on standard samples and largely refrain from a discussion of processing blanks and other reference samples employed for ensuring appropriate sample preparation procedures.
2.1. Target scatter
The final uncertainty of a dating result (σFinal) within a magazine is typically determined by combining the statistical measurement uncertainty (σMeas) with an empirically derived external uncertainty, the so-called target scatter (σTS) as shown in Equation (1) (Wacker et al. Reference Wacker, Christl and Synal2010):
${\sigma _{Final}} = \sqrt {\sigma _{Meas}^2 + \sigma^2 _{TS}}$
The target scatter reflects uncertainties in the dating process that extend beyond counting statistics and arise from external factors such as sample inhomogeneity, sample handling, or graphitization. In this contribution, we compare two possible approaches for determining the target scatter:
The first method compares the variance of all samples of the same material within each individual magazine. In this approach, every magazine holds six oxalic acid standards (OxII, SRM-4990C). The target scatter is subsequently adjusted such that the uncertainty of each individual measurement corresponds to the spread of the ensemble. This method will be referred to as the “6-Ox-method”.
The second method relies on the observed long-term statistics of a known standard material, in our case the IAEA standard C7. In this approach, the target scatter is adjusted in such a way that the uncertainty of the C7 standard matches its long-term empirically determined uncertainty. This method will be referred to as the “long-term-observation method”.
In the present case, because one C7 standard is included in each magazine, our approach effectively compares the intra-magazine scatter with the scattering between different magazines.
The BATS software does not take δ13C uncertainties into account. The influence of these uncertainties can be explored by examining how isotopic fractionation affects the measured 14C/12C ratio R. A simplified approach (Stenström et al. Reference Stenström, Skog, Georgiadou, Genberg and Johansson2011) gives a correction term that accounts for fractionation effects
${R_{frac}} = R\cdot{\left( {{{1 + {{ - 25} \over {1000}}} \over {1 + {{{\delta ^{13}}C} \over {1000}}}}} \right)^2}$
With a linear approximation valid for δ13C values in the range relevant for 14C dating (–40‰ < δ13C < +10‰) and considering only the δ13C uncertainty of the sample measurement one obtains the following dependency:
${\sigma _{{R_{frac}}}} = \left| {{{\partial \left( {{R_{frac}}} \right)} \over {\partial \left( {{\delta ^{13}}C} \right)}}} \right|\cdot{\sigma _{{\delta ^{13}}C}} \cong R\cdot0.002\cdot1000\cdot{\sigma _{{\delta ^{13}}C}}$
For a C7 sample with a known 14C/12C ratio of RC7 = 0.4953 [F14C], this relation for the influence of δ13C measurement uncertainties on the measured 14C rate further simplifies to
${\sigma _{{R_{frac}}}} \cong 0.001\cdot1000\cdot{\sigma _{{\delta ^{13}}C}}\left[ {{F^{14}}C} \right]$
With
${{\partial \left( {{R_{frac}}} \right)} \over {\partial \left( {{\delta ^{13}}C} \right)}} \lt 0$
, a change in δ13C of ± 1 ‰ leads to a change of ∓ 0.001 [F14C], or translated to age, a change of ± 17 years for a sample of a similar F14C concentration as the IAEA C7 standard’s, while for a modern sample this would translate to ±8 years.
2.2. Small samples
We consider a sample as small if it contains less than 500 μg of carbon and ultra-small if it contains less than 25 μg. The chemical pretreatment for most sample materials is the same as for regular-sized samples. A special case is biological materials such as brain cells, which have been repeatedly measured at the Tandem Laboratory within a collaboration with Karolinska Institute in Stockholm, Sweden, funded by the Swedish Research Council (Linnaeus-Grant 2010–2018). Details of the pretreatment of small biological samples can be found in Salehpour et al. (Reference Salehpour, Håkansson and Possnert2013), and the combustion, gas transfer and graphitization steps are now also applied to other sample materials. Major modifications include pre-baking of all glass ware and CuO a second time directly prior to use to further reduce the risk of contamination. The iron powder used as catalyst in the graphitization process is conditioned by heating to 500 °C for 300 s in presence of air and subsequent reduction with hydrogen at 500 °C for 300 s, for three cycles, with pumping to vacuum in between cycles. This conditioning resulted in a reduction of the 12C current obtained from a pure pressed iron sample by a factor of 10, significantly reducing the contribution from contamination to the measured sample current. The samples are combusted at 900 °C for 300 s. The obtained graphite is pressed into the same type of sample holder (cathode) used for large samples and is not treated differently during the measurement process.
Small samples contain reduced absolute amounts of carbon and therefore only give rise to small ion currents during accelerator measurement. Therefore, they are more vulnerable to the effects of inevitable contamination by trace amounts of foreign carbon with different radiocarbon age. To correct F14C values for small samples, a contamination model as given in Equation (5) can be employed, in which the total sample (t) is treated as a sum of the true sample (s) and the contamination (c), with mt = ms+mc.
${m_t}\cdot{F_t} = {m_s}\cdot\;{F_s} + {m_c}\cdot{F_c}$
By measuring a mass series of a standard material, typically OxII, and determining the background level using a graphite blank, the mass and radiocarbon age of the contaminating contribution introduced during graphitization and measurement process can be quantified as discussed in more detail in Salehpour et al. (Reference Salehpour, Håkansson and Possnert2013, Reference Salehpour, Håkansson and Possnert2015). With this information, it is possible to correct the F14C of unknown samples. Similar to what was previously reported for the Pelletron system (Salehpour et al. Reference Salehpour, Håkansson and Possnert2015), we generally observe old contamination, also for MICADAS.
Depending on the sample materials and preparation lines, different sets of standards are used for age calibration. For all small sample measurements in MICADAS (values given in F14C): size-series of SRM-4990C (OxII, 1.3407) are used for calibration of ages as a function of sample mass (in μg C) and graphite is used to monitor machine background. For small archaeological samples: an additional size series of IAEA-C7 (oxalic acid, 0.4953) is included to control the calibration. For small modern/biological samples, additional size series of IAEA-C6 (sucrose, 1.5061) and salmon DNA (CAS 100403-24-5, Sigma-Aldrich, 1.014) are used. For small carbonates, an additional size series of IAEA-C2 (travertine, 0.4114) is used together with CSTD (coral, 0.9442) and Icelandic spar (carbon dead, to monitor sample processing) instead of graphite.
2.3. Stable isotopes
Subsequently to material-specific preparation protocols, regular archaeological samples (charcoal, macrofossils, collagen, sediments) of sufficient size are processed in an Elemental Analyser (EA, Vario Isotope Select, Elementar Analysesysteme GmbH, Langenselbold, Germany). Approximately 10% of the combusted sample is directed to an Isotope Ratio Mass Spectrometer (IRMS, Isoprime Precision, Elementar Analysensysteme GmbH, Langenselbold, Germany) for determination of the stable isotope ratios δ13C and δ15N. The remaining fraction is sent to an Automated Graphitization Equipment (AGE, Ionplus AG, Zurich, Switzerland) for graphitization using iron powder (Alfa Aesar 99%, CAS:7439-89-6) as a catalyst.
Determination of the stable isotope values δ13C and δ15N is particularly relevant for certain sample types, such as unburned bones and macrofossils, because the isotopic signature can give insights on the diet of past persons or animals (Pauly Reference Pauly2025). Moreover, δ13C is required for accurate radiocarbon age calculation when it is not already obtained through accelerator-based measurement of the 14C/12C and 13C/12C ratios (Stuiver and Polach Reference Stuiver and Polach1977).
3. Results and discussion
3.1. Analysis of target scatter
Throughout all operational years of our MICADAS system, the IAEA standard C7 has been used as a control sample for current correction. Figure 1A presents a histogram of more than a thousand C7-analyses performed between 2015 and 2025. Eight data points that diverted more than 3σ from the mean have been omitted prior fitting using a Gaussian distribution. The resulting long-term C7 value for the Tandem Laboratory was determined to be 5646 BP, with a single measurement deviation of ±33 years and a mean value error of ±1 year.
Panel A: Histogram of 14C age of C7 reference samples measured with MICADAS from 2015 to 2025. The Gaussian fit yields an average of (5646 ± 33) BP. Panel B: Target scatter values for 52 consecutive magazines measured during 2024 and 2025 as determined by using the “6-Ox-method” (short term) or the long-term observation (see section 2.1 for details). Panel C: Comparison of resulting uncertainties of C7-age in calendar years (1σ) determined via “6-Ox-method” (left) and long-term observation (right), respectively. Panel D: Comparison of δ13C values for C7-standards measured in an IRMS and MICADAS. Both distributions illustrate high accuracy (consensus value C7:
${{\rm{\delta }}^{13}}{\rm{C}} = \left( { - 14.48{\rm{\;}} \pm 0.21} \right)$
‰) but the measurement precision of the IRMS is significantly higher than that of the MICADAS.
To assess whether the “6-Ox-method” and the long-term observation approach comparable target scatter values, we selected a sequence of 52 consecutive regular magazines (with AGE-prepared samples). For each magazine, the experimental datasets were analyzed twice, once with each method, and the resulting values of the target scatter are shown in Figure 1B. The “6-Ox-method” yields considerably lower target scatter values (red circles) compared to the long-term method (blue triangles). For the “6-Ox-method”, approximately half of the magazines result in a target scatter value of 0.0%, corresponding to cases where the variation in OxII-dates is lower than the expectation from mere statistics. The remaining values vary broadly between 0.05 and 0.23%.
In contrast, the long-term observation method produces target scatter values tightly clustered between 0.24 and 0.30%. The histograms in the right panel of Figure 1B represent the distinctly different target scatter distributions from the two adjustment methods. These differences directly yield significantly different dating uncertainties as shown in Figure 1C. For the same 52 C7 standards measured, the resulting age uncertainty is 25 years when applying the “6-Ox-method” – only slightly above the purely statistical uncertainty of 24 years (i.e., TS = 0.0) – compared to 33 years when using the long-term method.
The observation that approximately half of the magazines do not require the application of a target scatter correction is consistent with statistical expectations when the true target scatter is very small compared to the statistical uncertainties. In contrast, the long-term observation of the C7 standard indicates an additional contribution to the scatter that exceeds what is observable within single magazines. A variation analysis including temporal autocorrelation and nearest-neighbor-correlation shows no detectable correlation or interdependence between consecutive measurements of C7, i.e., between consecutive magazines. Furthermore, the well-fitting Gaussian distribution of the long-term spread supports the interpretation of a statistical nature of this additionally observed contribution. Setting the target scatter according to the long-term observation thus helps to avoid potentially underestimating the target scatter based on the low number of OxII references within a single magazine and account for potential variances between individual magazines.
As a possible explanation for the higher target scatter required to reproduce the observed long-term variability, we investigate the influence of the uncertainty in δ13C measured by MICADAS. Figure 1D compares the δ13C values of all C7-standards measured using a dedicated IRMS with those of our MICADAS. Although both machines show good accuracy, well incorporating the consensus value of (–14.48 ± 0.21)‰, the precision of the IRMS measurement is significantly higher. According to Equation (4), the effect of δ13C uncertainty on the measured 14C rate can be directly estimated. The measurement uncertainty of MICADAS,
${\sigma _{{\delta ^{13}}C}} = 1.57$
‰ (see Figure 1D) introduces an additional uncertainty in the measured rate due to fractionation correction of
${\sigma _{{R_{frac}}}} \cong 0.001\cdot1000\cdot{\sigma _{{\delta ^{13}}C}}\left[ {{F^{14}}C} \right] = 0.00157\left[ {{F^{14}}C} \right]$
or translated to a relative representation
${{{\sigma _{{R_{frac}}}}} \over {{R_{frac}}}} \cong {{{\sigma _{{R_{frac}}}}} \over R} \cong {{0.00157} \over {0.4951}} = 0.0032$
After estimating the direct influence of the δ13C uncertainty on the final dating result, a relative error contribution of 0.32% is obtained. Comparing this value with the required target scatter range of 0.24–0.30%, it becomes evident that the influence of the δ13C uncertainty measured by MICADAS explains our need for a target scatter value larger than what is expected from the OxII-standards exclusively. Combining this target scatter-related contribution with the uncertainty arising from counting statistics yields the final relative dating uncertainty of
${\sigma _{Final}} = \sqrt {\sigma _{Meas}^2 + \sigma _{{\delta ^{13}}C}^2} = \sqrt {{{0.0030}^2} + {{0.0032}^2}} = 0.0044$
With a value of
${R_{C7}} = 0.4951\;\left[ {{F^{14}}C} \right]\;$
, σ
Final
yields an absolute uncertainty of
${\sigma _{RC7}} = 0.0044\cdot0.4951\;[{F^{14}}C] = 0.0022\;{F^{14}}C$
. So, in total
${R_{C7}} = \left( {0.4951 \pm 0.0022} \right)\left[ {{F^{14}}C} \right]\quad {\rm{and}}\quad Ag{e_{C7}} = \left( {5646 \pm 35} \right)\;years$
This result is in good agreement with the long-term observation based solely on accelerator measurements. Additional correction using IRMS-derived δ13C values could subsequently be applied to further reduce uncertainty.
3.2. Small samples
About 9000 samples with masses below 200 µg C equivalent have so far been prepared and dated at the Tandem Laboratory. The relationship between the measured 13C current on the high-energy side of the accelerator and the corresponding carbon mass of the sample is shown in Figure 2 on a log-log scale. The data split into two distinct sets: blue points represent samples dated using the MICADAS system and red points correspond to measurements using the 5 MV NEC Pelletron. For both accelerators, the distributions are rather narrow and follow straight lines of the similar slope which transition into plateaus at higher masses. For masses below approximately 1 μg C, the 13C currents are systematically higher than predicted by the linear fit. This observation can be rationalized from a relative increase of the contribution from potential contamination at very small sample sizes. Despite this, excellent linearity is maintained across a wide range of sample masses for both accelerators. The Pelletron measurements exhibit a consistently higher ion current which can be explained by a different source geometry and sputter efficiency, resulting in a higher number of carbon ions entering the accelerator per unit of time. The plateau at higher masses can be interpreted similarly: The fixed opening of the source cathode limits the sputtering area, such that adding more graphite into the cathode only increases the sample thickness and thus the available measurement time, but not the maximum achievable ion current. This interpretation is also in agreement with the different masses at which saturation is observed for the two systems, reflecting different cathode designs. Data points that fall significantly below the slopes represent samples with drastically reduced current, typically due to failure of the graphitization process or (partial) loss of graphite during cathode preparation.
Dependency of measured 13C current on carbon mass for small samples dated using the MICADAS (blue dots, 2014–2025) in comparison to using the 5 MV NEC Pelletron (red dots, 2001–2014), for 6679 samples each, on a log-log scale.
As small samples with low carbon mass produce low 13C currents of only a few nA as shown in Figure 2, the dating of very small samples becomes highly sensitive to 1) contamination even at the nanogram level during sample handling, 2) contributions from the sample surroundings during the measurement, and 3) the need of a highly accurate current measurement down to the picoamp regime.
The graphite obtained from each sample is pressed into an aluminium cathode which is sealed at the back with a small aluminium sphere. As the sample material is progressively sputtered away during the measurement process, the surroundings can start contributing increasingly to the ion currents. For example, upon observing irregularities in the 12C LE-current, Time-of-Flight Elastic Recoil Detection (ToF-ERDA) measurements were performed on the spheres, revealing a carbon content of roughly one percent. For different aluminium spheres, the absence of carbon could be confirmed by TOF-ERDA measurements within the detection limit (< 0.1%), and the 12C LE-current for blank samples decreased by a factor of ten.
Since the F14C value of a sample is calculated from the ratio of the measured 14C and 12C isotope currents and corrected using a 13C-dependent factor, any offset in current will introduce a mass-dependant bias in the calculated age. A thorough determination and continuous monitoring of the dark currents in the Faraday cups used to measure the 12C and 13C isotope currents, as well as a precise correction of these currents in the pA regime is therefore critical for obtaining reliable ages for ultra-small samples.
The effect of poorly vs. well-adjusted Faraday cups is illustrated in Figure 3, which shows F14C as a function of carbon mass for two OxII mass series from exemplary test measurements. The measurement data for both series were current corrected in BATS using the 500 µg OxII-standards and are plotted as open symbols. For the largest samples (around 500 μg C), both series converge to the standard value of 1.341. However, for poorly adjusted Faraday cups, the F14C increases artificially to higher values for small masses (red open circles), whereas it decreases for well-adjusted Faraday cups (blue open squares). This behavior can be understood by recalling that, when Faraday cups currents are measured correctly, the remaining mass-dependent effect should be dominated by contamination, which can be expected to typically feature lower F14C than OxII.
Comparison of F14C for two mass-series of OxII samples from two test magazines, with (filled symbols) and without (open symbols) mass correction according to the contamination model (see Eq. 2). Red and blue curves correspond to badly vs. well-adjusted Faraday cups. Error bars are purely statistical and were scaled from measured to corrected data proportional to the change in F14C.
Utilizing the contamination model as given in Equation (5) and the fact that OxII is a standard material of known F14C, series of OxII-standards can be fitted to determine the mass and age of the contamination. The resulting fits for both series are shown as solid lines together with their respective 95% confidence intervals. Applying these fit results to the measurement data yields contamination-corrected values, shown as filled symbols in Figure 3, which scatter around the expected standard value.
Knowledge of the mass and apparent age of the contamination for a given measurement allows the age of unknown samples from the same magazine to be corrected accordingly. However, because both the contamination mass and its F14C value exhibit a significant uncertainty, and as small samples may have different F14C values, resulting in a different impact from the correction, we routinely run other small sample standards as mass series, as described in chapter 2.2. Small samples. Figure 4 shows the mass-corrected F14C values for various standard materials over a wide range of carbon masses. In all cases, the corrected data align well along the respective reference value down to masses of around 20 µg, providing a robust validation of the contamination-model correction derived from OxII size-series measurements.
F14C for mass series of standard sample materials used for calibration and control of small sample dating: C6, OxII, salmon DNA, C7 and C2. The data is current-corrected using BATS and mass-corrected using the contamination model (see section 2.2).
3.3. Mass dependency of stable isotopes
Ratios of stable isotopes obtained by IRMS are expected to be mass independent. To investigate a potential size dependence of the deduced values for δ13C and δ15N, weight-series of a protein standard (casein, B2155, Elemental Microanalysis Ltd.) have been measured weekly over the past two years. The combined data are shown as red dots in panels A and B of Figure 5. The peak area of the respective isotope in the EA which is proportional to the sample mass is plotted on the x-axis, while the corresponding stable isotope ratios are shown on the y-axis. For clarity, data points belonging to single weekly mass series have been enlarged. In the ideal case, the measured data should scatter around the reference δ13C value of –26.98‰ and the reference δ15N value of 5.94‰, respectively. However, both plots show a clear dependence on sample size while no temporal drift is observed over the investigated period. From linear fits to each weekly series, mass-dependent corrections are calculated and applied to all samples measured during the corresponding week. The resulting corrected datapoints are shown in panels A and B as blue dots. To quantify the improvement, histograms of the corrected and uncorrected data are shown in Figures 5C and 5D. For both, δ13C and δ15N, the distributions become clearly narrower, and the overall uncertainty is reduced to approximately one third of the original spread, resulting in an uncertainty of 0.18‰ for δ13C and 0.16‰ for δ15N, respectively.
Panels A–D: IRMS measurements on protein standard casein (B2155, Elemental Microanalysis Ltd). Data for stable isotope δ13C are shown in panel A and the corresponding histograms in panel C, data for δ15N in panel B and histograms in panel D. Uncorrected measurement data is shown in red, corrected data is shown in blue. Data sets of one example week are highlighted as large dots.
We conclude that reliable stable isotope measurements require a mass correction in addition to standard isotope calibration.
4. Summary and conclusion
This work reports on ten years of operation of the MICADAS compact tandem accelerator for radiocarbon analysis at the Tandem Laboratory, Uppsala University, covering over 40,000 measured samples. The long operational record enables a detailed evaluation of measurement performance, especially for target scatter, small-sample behavior, and stable-isotope mass dependence.
Our long-term dataset shows that target scatter derived from single magazines runs routinely underestimate true variability, often yielding values close to zero. In contrast, long-term C7 statistics reveal a consistent scatter of 0.24–0.30%, driven largely by the δ1 3C uncertainty of MICADAS. Consequently, we suggest the use of the long‑term target scatter for accurate uncertainty estimation, as it yields dating uncertainties consistent with the observed long‑term behavior.
For small and ultra‑small samples, low ion currents make measurements highly sensitive to contamination and current offsets. When these effects are properly corrected, using contamination models and carefully monitored Faraday‑cup dark currents, reliable F1⁴C values can be obtained down to approximately 20 µg C. Mass‑series of multiple standards confirm the robustness of these corrections across a wide range of activities.
We also find that stable isotope ratios measured by IRMS exhibit a clear mass dependence, requiring weekly mass‑dependent corrections to achieve high precision (0.16–0.18‰). Without such corrections, δ1 3C‑dependent fractionation adjustments could bias radiocarbon ages, especially for small samples.
Overall, the ten‑year dataset highlights that MICADAS delivers precise and reliable radiocarbon measurements, but also that robust correction procedures for small samples and stable‑isotope measurements are essential for maintaining accuracy. Building on these findings, we aim to extend this work with a more comprehensive analysis of potential error sources, comparable to the approach of Aerts‑Bijma et al. (Reference Aerts-Bijma, Paul, Dee, Palstra and Meijer2021), with the goal of further improving dating precision.
Acknowledgments
This work has been supported by the Swedish Research Council (VR) under grant numbers 2019-00191 and 2023-00155. The Radiocarbon Laboratory is part of ArchLab, funded by VR grant number 2023-00174, as well as of SwedigArch, supported by VR grant number 2021-00161.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
δ13C=(−14.48±0.21)
Open access
