| 18 Aug 2026
An extensible, algorithmic framework to enumerate, ascertain, and compare mechanical properties from common snowpit observations
Abstract. Mechanical properties of snow, such as Young’s modulus and Poisson's ratio, are used in avalanche release models but rarely measured directly in snowpits. Researchers chain published parameterizations to estimate these properties, but the combinations and their practical coverage have not been systematically compared. We present SnowPyt-MechParams, an extensible, graph-based framework that encodes parameterizations as a directed acyclic graph, enumerates valid pathways, and propagates measurement uncertainty through each one. For an example dataset, we developed SnowPylot to parse CAAML files from the SnowPilot database, structuring 50,278 snowpits (371,429 layers, 2015–2025 water years) into 14,776 slabs from Extended Column Test results with propagation. As an example application, we implemented pathways for the shear component of slab weight per unit area with and without elastic properties. The highest-coverage shear-weight pathway succeeds for 5,470 slabs (37.0 %), while the highest-coverage shear-weight-with-elasticity pathway succeeds for only 687 slabs (4.6 %), reflecting elasticity’s longer calculation chains. Slab bending stiffness, D11, carries a median relative uncertainty of 78 % (IQR: 56–85 %). Because multiple pathways can reach the same target, estimates vary substantially, with a median max/min ratio for D11 of 32 (up to 63) per slab. These results demonstrate the framework's utility to systematically calculate coverage, uncertainty, and range in mechanical properties from snowpit observations using different parameterizations.
Summary:
This paper presents SnowPyt-MechParams. It is an open-source, graph-based framework that identifies all the possible ways of estimating the mechanical properties of a snow slab from common snowpit observations by chaining published parameterizations. For each resulting calculation pathway, the framework reports how many slabs can be processed, the resulting values and their propagated measurement uncertainty. The authors also present SnowPylot, a tool to read SnowPilot data, and apply the framework to slabs derived from a decade of snowpit observations. They show that coverage decreases sharply when elastic properties are required and that different pathways can yield slab bending stiffness values that differ by more than an order of magnitude for the same slab.
The paper addresses a relevant and timely problem. Mechanical stability models are increasingly used in avalanche research and are being considered for operational applications, but they require input parameters that are rarely measured in the field. The approach proposed by the authors has several strengths. The framework is extensible and openly available. The authors also point out that it could be driven by snow cover model outputs (L450-456), which opens interesting perspectives. The application to a decade of SnowPilot data (50,278 snowpits, 14,776 slabs constructed from ECT results) demonstrates that the framework can be used at population scale and SnowPylot will likely be useful to other researchers working with SnowPilot data. The explicit handling of the validity domain of each parameterization, the propagation of measurement uncertainty and the transparent discussion of limitations are also appreciated.
My main concerns are detailed below, followed by minor comments.
Major comments:
Minor comments:
L11-12 and L481: The value "up to 63" corresponds to the slab at the 90th percentile of the spread-ratio distribution (L377–378, Figure 10) rather than to the maximum. Could the maximum be reported, or the wording adjusted?
L59: "a failure criteria" to "a failure criterion"
L65-76: The generalized Hooke's law in tensor form (Equations 4 and 5) does not appear to be used later in the paper. Could this part be shortened or moved to an appendix?
L85: κ is defined here but does not appear in Equation 10. Should it be included?
L93: Missing full stop after "framework".
L106: "ECT's" to "ECTs"
L109: Could the authors clarify whether layer thicknesses in SnowPilot are measured vertically or perpendicular to the slope, and whether this is accounted for in Equation 1?
L144: "figner" to "finger"
L145-146: Since hand hardness is a categorical observation, it would be helpful to explain how the uncertainty of two subclasses is converted into a numerical uncertainty on density.
L188-189: Could the authors confirm that these ice properties are consistent with those used in the original studies, since Equations 13 and 15–17 are normalized by them?
L221-222: "a target a parameter", remove the second "a"
L302-311: Using a single density method for all layers explains the low coverage of direct density measurements. It may be worth reporting the coverage of a combined strategy, using measured density where available and a regression elsewhere, as practitioners would likely do.
L313-316: The target "shear weight with elasticity" is not defined by an equation and seems to be a check that E and ν are available for all layers. Could this be clarified or the target renamed?
L358-360: This sentence could be clarified, as it suggests that Köchle is the bottleneck, whereas the Poisson's ratio step seems to limit coverage.
L361: "in in"
L367: Please explain why the 16 pathways retained are those combining the Kim T2 and Geldsetzer density methods, and why Kim T6 and direct density were excluded.
L377-382: The statement that the spread between pathways exceeds the propagated measurement uncertainty is illustrated with only three example slabs (Figure 10), and in these examples the error bars of the Köchle pathways overlap those of the Bergfeld and Schöttner pathways. Could this comparison be quantified over all 317 slabs? These error bars may also be affected by the linearization discussed in comment 8.
L380-381: Please specify which quantity and which pathways and slabs, the median uncertainty of 78 % refers to.
L442: "calcuation" to "calculation"
L459: "mechancial" and "have have"
L460: "develeoped" to "developed"
L479: The statement that the range of estimates "grows as pathways lengthen" does not seem to be supported, since the spread between pathways is computed only for D11.
L486-489: It would be helpful to archive the list of snowpits used, since the SnowPilot database keeps growing. It would also be useful to report the computational time required to process all slabs and to discuss how the framework would scale to larger datasets or real-time operational applications.
Throughout: Use "snowpit" or "snow pit" consistently and define ECTP at first use.
Figures:
Some captions include interpretation that should be mentioned in the text, for example Figure 1 with the restatement of the objective.
Figure 2: The caption mentions Rutschblock results, whereas the figure shows PST results.
Figure 3: The thin coloured lanes and the number of labels make this figure difficult to read. One panel per method, or a table, may be clearer.
Figure 5: "slab weigh" to "slab weight"
Figure 6: Since all arrows appear in every panel and the active pathway is only shown in bold, the four panels look very similar at first glance. Showing only the arrows used by each pathway would make the differences clearer.
Figure 7b: The three-line axis labels overlap.
Figure 10: The labels repeated on the vertical axis of each panel make this figure dense and difficult to read. I suggest increasing the font size and the sixe of the panels. See also major comment 8.