| 12 Aug 2026
Analytical improvements and a machine-learning assessment of age dispersion in laser-ablation (U-Th)/He zircon dating
Abstract. Laser-ablation (U-Th)/He dating offers major advantages over conventional whole-grain dissolution analysis, including spatially resolved sampling, minimal sample preparation, and much faster data acquisition. Despite these advantages, the method’s full potential is held back by issues including the lack of standardized, fully documented analytical protocols, and age dispersion that commonly exceeds the propagated analytical uncertainty. In this study, we document a complete LA-(U-Th)/He workflow for zircon, combining ultra-high-vacuum laser extraction, white-light-interferometric pit-volume measurement, and time-resolved LA-ICP-MS parent-nuclide quantification, and introducing two methodological refinements: a dynamic multi-phase oxide normalization for the parent concentration calculation, and parent-source depth weighting to account for alpha redistribution. Applied to 811 spots in Fish Canyon Tuff (FCT) zircon, the workflow yields a median age of 28.32 Ma (σMAD = 3.85 Ma; 13.6 %), in agreement with the published reference age, and reduces dispersion by a factor of ~1.7 relative to conventional 29Si- or 91Zr-based reductions. To identify the drivers of the remaining dispersion, we train a gradient-boosted regression model on 33 simultaneously measured per-spot parameters. He pit volumes emerge as the strongest predictor of age residuals, highlighting volume-correlated analytical and geometric effects as the largest identifiable and potentially correctable source, while elevated 208Pb concentrations flag inclusion-related disturbance and provide a practical spot-screening criterion. The measured degree of zonation within the ICP-MS profile on the other hand carries no independent predictive information. Instead, a simple two-layer zonation model shows that U-Th zonation in the grain volume removed during polishing can alone generate age offsets exceeding the entire residual dispersion, a mechanism confirmed by the contrast in dispersion between zoned FCT zircon (13.6 %) and the compositionally homogeneous LGC megacryst (1.5 %). The residual dispersion is therefore not an instrumental limitation, and further gains in precision will require a new generation of methods that characterize or circumvent the polished-away parent-source volume.
General comments:
This contribution describes an analytical workflow for laser-ablation-based zircon (U-Th)/He dating, introducing several modifications in data reduction compared to existing procedures, and tests these on a large dataset acquired on Fish Canyon Tuff (FCT) and LGC zircon. It investigates the factors that contribute to age dispersion by (1) a comparison of ablation-pit-volume measurements by white-light-interferometry and µCT; (2) a machine-learning model of age dispersion depending on a series of analytical parameters; and (3) an α-redistribution model capturing effects of parent zoning and implantation from grain parts that were polished away.
The authors present the main workflow and results in the main manuscript, while the technical details of sample preparation, data reduction and modeling are presented in several Appendices. The accompanying code allows to carry out the full data reduction and machine-learning model as described in the text. The authors demonstrate how taking into account non-stoichiometric zircon composition and the effect of α-redistribution during data reduction improve the age variation and accuracy of the FCT reference material and how the observed age dispersion depends strongly on measured pit volumes and inclusion-related parameters.
This is a highly welcome publication dealing with the problem of age “overdispersion” in laser-ablation-based (U-Th)/He dating involving theoretical, analytical, and data-reduction aspects. I expect that the large (U-Th)/He dataset, detailed technical Appendix, and the related software will help the community to improve existing procedures and to compare results across laboratories.
My personal highlights are the innovative data-reduction approach using oxide-sum normalization and the correction of LA-ICP-MS-pit contributions to the measured helium, which fits well into the ongoing research on optimal pit-size configurations and dealing with zoned grains (e.g., Maier et al., 2006; Hoemann et al., in review)
While this manuscript is clearly suitable for Gchron, there are three issues that I’d like the authors to address:
First, the documentation of the analytical data is insufficient. The manuscript presents a large set of FCT and LGC in-situ (U-Th)/He data, but there is no dedicated supplementary table besides the data table for testing the machine-learning code, which, as far as I can tell, only contains 800 of the FCT analyses and little description. I suggest documenting all FCT and LGC data together with the analytical parameters for each session, e.g., following the community guidelines for (U-Th)/He dating (Flowers et al., 2023). The methods section also needs to be more specific, e.g., regarding the laser-ablation pit sizes and depths for the acquisition of the FCT and LGC data (see comments on L. 88, 127, 128), and the elements analyzed by LA-ICP-MS (comment on L. 133).
Second, the authors have acquired a large dataset on the zircon reference material FCT to test their improvements for the laser-ablation (U-Th)/He workflow. However, while they provide a thorough explanation with respect to the improved fit of the FCT data to the reference age, other results are also worth exploring and will likely make a stronger case for the proposed modifications. In particular, the LGC dataset could provide an important distinction between improvements that address heterogeneous zircon from more general effects (see comments on L. 62, Fig. 6). Another aspect is the effect of the oxide-sum-based data reduction on the U and Th concentrations of the secondary reference materials, and a comparison of the present FCT dataset to existing laser-ablation (U-Th)/He data from the same material (see comment on L. 283–295).
Third, the rationale for the machine-learning dispersion model needs to be clarified. I understand why the authors deliberately excluded the He, U, and Th concentrations because of their direct appearance in the age equation. However, I wonder whether including the pit volume, which also affects the (U-Th)/He age directly, introduces a similar correlation with the model’s target. In particular, could the direct relationship between pit volume and age obscure correlations with other variables (see comments on L. 234, L. 371)? At least, I would appreciate a discussion of the practical implications, e.g., if a more conservative treatment of pit-volume uncertainties is necessary.
With these issues resolved, the manuscript will be an excellent contribution to Geochronology.
Birk Härtel
Please find specific suggestions below.
Specific comments:
L. 43: I encourage the authors to develop these factors a little more for LA-based (U-Th)/He dating and include the recent literature dealing with these issues. For example, we can expect the effects of α-ejection bias mentioned for whole-grain dating to be mostly negligible. However, other factors such as inclusions or zoning would hit the small, mis-matched ablation volumes harder. Please include references to the articles that discuss the problems related to zoning in laser-ablation (U-Th)/He dating such as Danišik et al. (2017), Léger et al. (2023), and Vermeesch et al. (2023). The recent preprint by Hoemann et al. (in review) discussed and even quantified the effect of lateral zoning data with respect to different ablation pits.
L. 62: I agree with the authors that LGC seems to be a nearly-ideal material to test accuracy and precision of the presented workflow. I thus wonder how (besides the final result for LGC presented in L. 301) the step-wise modifications of the analytical procedure affected the LGC data relative to the heterogeneous and inclusion-bearing FCT. I would predict that the oxide normalization probably would have improved accuracy, while parent-source weighting would leave the data from the homogeneous crystals mostly unaffected. I suggest presenting these data in parallel to the FCT data in Fig. 6 and section 3.2.
L. 70: Consider citing Dunkl et al. (2024) here, who presented an alternative way of mounting in epoxy resin.
L. 86: Is this the nominal fluence corresponding to the 7 mJ? Please clarify.
L. 88: The variation in the number of pulses is not explained in the subsequent text. Do these different numbers of pulses refer to different acquisition sessions for the FCT data? This broad range of pulses is surprising to me because it is not reflected in the pit depth given in the supplementary material.
L. 127: The description of the LA-ICP-MS analysis requires some more detail. Where the main analyses of this study on the FCT grains done with one specific diameter or with different ones? If these varied between sessions, this needs to be documented properly in the Supplementary Material.
L. 128: I assume the 25 µm pit depth for the LA-ICP-MS refers to the added depth on top of the He-extraction pit? Were the reference materials (AusZ2, LGC-3, NIST SRM 610) ablated prior to LA-ICP-MS to obtain the same ablation geometry as for the unknowns?
L. 133: I suggest listing 235U and 147Sm with the other parent nuclides, although 235U was not directly used for age calculation. Also, list the other trace elements of interest that were analyzed (i.e., 27Al, 57Fe, etc.).
L. 185: Vermeesch et al. (2012) and Evans et al. (2015) describe small-diameter LA-ICP-MS pits placed within their He-extraction pits.
L. 222: Add the primary U/Pb reference material GJ-1 to the methods section.
L. 234: I understand the decision of excluding parent and daughter nuclides to avoid the dominance of inherent correlations, but what about the pit volume, which is also part of the age calculation (see L. 209)? If the measured He concentration (ncc) co-varies more strongly with U and Th than with the pit volume, I would expect the pit volume to strongly affect the apparent (U-Th)/He age and thereby mask the effect of other variables.
L. 270: I wonder what the implications of the pit-volume comparison are for the uncertainties of the FCT analyses in the subsequent discussion. Looking at the variation Fig. 5, I wonder whether applying the 5.5–6% from the 40/8 µm pits may be more appropriate for the FCT data.
L. 271: Please cite an overall range of single-grain uncertainties for the (U-Th)/He ages to compare to the age range in Fig. 6 and the uncertainties mentioned in the text.
L. 283–295: This paragraph adds interpretation to the results presented in this section. While interpreting part of the results here is not a bad idea, I would like to see a dedicated discussion section on how the analytical modifications affected the results for the FCT dataset. This improvement represents one of the main goals of the manuscript and is strongly supported by the bulk of the data, so I see the risk that focusing only on FCT age dispersion undersells the strongest aspect of the manuscript. I also suggest using this discussion section to further explore: (1) the bias in U and Th concentrations due to substitution and extrapolation, e.g., by discussing concentrations in the secondary reference materials; (2) the effect of the modifications on the well-behaved reference material LGC; and (3) comparing the results to existing laser-ablation (U-Th)/He datasets from FCT zircon (e.g., Horne et al., 2016; Léger et al., 2023), which used different pit-size combinations and did not reduce the LA-ICP-MS data by oxide normalization.
L. 289: Replace 15.3 wt% by 50 wt%, the approximate stoichiometric Zr content in zircon. Please check.
L. 371: I like the discussion about the influence of pit volume beyond its obvious contribution to the measured age. I think the mismatch between variation in pit volume and pit-volume uncertainty is another important observation (L. 393). As mentioned in the general comments, I wonder if its inclusion in the machine-learning models may either mask effects of other variables and what the direct implications of this analysis are for treating measurement uncertainties on the pit volume that include both the instrument capacities and pit-specific factors.
L. 394: Which exact analysis does this 10x10 grid refer to, the LGC data presented in section 3.2 or measurements for volume benchmarking? Please specify.
L. 415: There are two interesting aspects here that would merit a short note each (2–3 sentences). First, is there a mechanism that would explain why low-208Pb grains have a positive PDP (Fig. 9)? Is this just thorium’s contribution to the age equation? Second, what are the practical implications besides using U/Pb age and 208Pb as proxies for disturbance? I think we can safely assume that the likelihood of hitting inclusions that disturb the isotopic systems increases with the LA-ICP-MS pit volume. I’d then expect that large spot diameters compromise the main advantages of laser-based zircon (U-Th)/He dating.
L. 419: Please consider restructuring the discussion here by combining sections 4.3 and 4.4. From L. 435 on, the discussion here already includes the “ghost grain” volume, and section 4.4 is rather the quantitative extension.
L. 424: Again, I would add a reference to Léger et al. (2023) here.
L. 431: I appreciate the critical discussion of the 1D vs. 3D characterization. This is especially important for the “over-coring” geometry of laser-ablation pits as material from below the He pit is mixed with the annular region around the He pit (see discussion in Härtel et al., 2026). This geometry further complicates the relationship between vertical zoning and time-resolved ICP-MS signal.
L. 433: I would expect sub-stopping-distance zoning to be substantially averaged out if it is not super-imposed on a larger-scale U-Th enrichment or depletion trend.
L. 445: I think it would be helpful for the reader to also include the He pit depths here.
L. 451: Please indicate the diameters of the He and ICP-MS pit assumed for the model.
L. 480: I agree that without U-Th zonation from the polished-away part of the grain, we cannot be sure how much our eU estimates are in error. I’d appreciate if references to our discussion in Härtel et al. (2026) and that in Hoemann et al. (in review) on robust pit-size configurations could be included here. I also suggest adding a reference to Evans et al. (2015) regarding pit geometry and accurate volume measurement.
L. 498: The statement about the one-dimensional U and Th zoning needs to be rephrased because it omits the fact that the parent-source weighting largely takes care of the vertical zoning. This is an important positive result that should be emphasized.
L. 594: Please indicate the interval of instrumental-blank measurements; are these repeated after every ablation?
L. 660: Was the σ-scaled or the raw 5x MAD used for filtering?
L. 742: I see that the x and y coordinates are ignored for the calculation of the parent-source weights. What does that mean for the accuracy of the correction for different ICP-MS pit sizes?
Figure 4: I suggest adapting the figure to the actual ablation-pit parameters used for the FCT analyses.
Figure 6: Consider adding another row with boxplots for samples LGC to have a comparison of the improvements for a zoned and an unzoned reference material.
Figure B2: Please indicate that the lower left plot corresponds to the He-pit configuration used for the data acquired in this study.
Code: I test-ran the code provided with the manuscript and did not encounter any obvious errors, apart from some open3d-compatibility issues related to Python version.
Technical corrections
L. 7: I would rephrase the second part of the sentence to make it more understandable that the depth-weighting applies to LA-ICP-MS.
L. 9: Age variation – “dispersion” may make people think of σ in the radial plot.
L. 11: He pit volume emerges (singular as in the rest of the phrase).
L. 40: FT with a subscript.
L. 49: I suggest to rephrase: “The goal of this study is to (1) present an improved routine workflow for LA-(U-Th)/He dating and (2) identify key factors contributing to the analytical uncertainties using a machine-learning approach.”
L. 99 and throughout (optional): Consider using mass-concentration units such as µg/g instead of ppm.
L. 129: I think the minimum grain size (diameter) for a He pit of 30 µm and a 30–40 µm broader LA-ICPMS pit should rather be in the 70–80 µm range. Please check.
L. 220: Rephrase: This represents an advantage over conventional dissolution-based (U-Th)/He dating.
L. 225: I only counted 800 analyses in the Supplementary data sheet, please double-check.
L. 255: Please briefly introduce the mean bias.
L. 275: Consider using “variation” instead of “precision” here; I think most readers would link “precision” with single-grain uncertainties.
L. 277: Replace “reducing” by “processing”.
L. 307: Please indicate if the variations are given as σMAD.
L. 404: Superscript on 208Pb.
L. 406: Missing period.
L. 420: I am not sure why this result is unexpected, since the downhole variation in U and Th was corrected for (see L. 425). Please rephrase.
L. 464: Consider “5 µm pit depth” in the parentheses, as the reader may think of a 5-µm-diameter pit.
L. 497: Superscript on 208Pb.
L. 697: Missing “to” after “assigned”.
Figure 2: I suggest to rescale the axes to the typical spot sizes used in zircon (U-Th)/He studies and to eU values more relevant to typical zircon (U-Th)/He applications.
Figure 5: Introduce the abbreviation WLI in the caption.
Figure 7: Specify the meaning of the annotations below each boxplot, I assume these are again medians and σMAD.
Figure 10: Please add to the caption or axis labels that the axes refer to the He pit. Also rescale the axes to depth and diameter ranges that are typically used (see Forte et al., 2022 or Zawacki et al., 2022).
Figure 11: The annotations in the grey sketches at the top are difficult to read and should have a larger font size. Consider using a legend for the eU concentrations and the pits to reduce the clutter with larger fonts. Please also indicate the diameters of the He and ICP-MS pits.
In B, I suggest using the same scales for the y axis in B to show the differences in eU error more intuitively.
Figure A2: Period missing at the end of the caption.
Figure B2: The legends are difficult to read, consider increasing the font sizes. Please also declutter the overlapping x-axis labels in the upper plots. I suggest adding index letters for easier reference to the figure.
Supplement Table zFCT_AGES_RESULTS.xlsx: Zr concentrations are labeled as %, but the numbers suggest they are in µg/g.
References
Dunkl., I., Mallis, F., Lünsdorf, N.K., Schönig, J., and von Eynatten, H.: Zircon U-Pb-He double dating of modern sands from the Inn River catchment: assessing resolution and potential in a complex orogenic setting, J. Geophys. Res. Earth Surf., 129, e2023JF007360, https://doi.org/10.1029/2023JF007360, 2024.
Härtel, B., Enkelmann, E., and Whelan, E.: Combined zircon U/Pb-(U-Th)/He-Raman dating, Chem. Geol., 123435, https://doi.org/10.1016/j.chemgeo.2026.123435, 2026.
Hoemann, H., Hueck, M., and Dunkl, I.: Technical Note: Reducing age dispersion in in-situ (U-Th)/He dating: How to optimize ablation pit geometries with OptiPit, EGUsphere [preprint], https://doi.org/10.5194/egusphere-2026-2590, 2026.