the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 11 Feb 2026
Assessing the reliability of firn microstructural measurements from micro-CT data
Abstract. X-ray micro-computed tomography (micro-CT) has become a common technique used to characterize the microstructure of snow and firn, yet the sensitivity of micro-CT-derived microstructural parameters to image processing choices remains poorly understood. In particular, the selection of a binarization threshold can influence the reconstructed representation of the microstructure and, thus, any quantitative properties computed from it. Here, we systematically evaluate the sensitivity of six firn microstructural parameters to binarization threshold choice using micro-CT data from three samples of the NEEM 2009 S2 firn core that were extracted from shallow (7 m), intermediate (26 m), and deep (70 m) depths of the core. We generated reconstructions of the microstructure of each sample at every threshold value across the grayscale threshold range, and compare three thresholding approaches representing subjective, statistical, and topological strategies. Microstructural parameters describing bulk volume-fraction properties (Percent Object Volume, Percent Open Porosity, and Surface Area to Volume ratio) and microstructural complexity (Structural Model Index, Surface Convexity, and Euler Number) were computed across the full grayscale threshold range and evaluated using a normalized sensitivity metric. We find that bulk volume-fraction parameters are robust to threshold choice across all firn depths, while parameters describing microstructural complexity and connectivity exhibit strong threshold and depth-dependent sensitivity. Modeled estimates of the intrinsic permeability of the reconstructed microstructures generated at threshold values between 60–120 for each sample underscore the impact of threshold choice on the microstructural complexity and connectivity. These results demonstrate that binarization threshold choice can substantially influence interpretations of firn microstructural complexity and, therefore, transport properties, highlighting the need for careful selection of image processing steps, including binarization, in firn micro-CT studies.
- Preprint
(9651 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed (peer review stopped)
- RC1: 'Comment on egusphere-2026-1', Anonymous Referee #1, 10 Mar 2026
-
RC2: 'Comment on egusphere-2026-1', Anonymous Referee #2, 27 Apr 2026
This manuscript deals with a very significant and urgent challenge for everybody quantitatively analysing CT images: How do I binarise the original gray value images correctly to ensure comparability? Which impact do changes in this choice have on the subsequent geometric analysis or simulated materials properties? The data chosen for exploring these questions is very appropriate and interesting, too.
However, there are two major methodological flaws that are hard to overcome: First, the authors start off with reconstructed CT images with grey values in the 8bit range. Tomographic reconstructions have float valued voxels that are typically mapped to 16bit integer values. Sometimes, just 12 out of these 16bit are used but 8bit is definitely a considerable reduction that can seriously affect all following steps. It is however never discussed why and how the grey value range is reduced. Second, global grey value thresholding can work only if there are no global grey value fluctuations. It should at least be checked, if this prerequisite is met for the image data investigated.
Finally, the conclusion is weak: Topological features are more affected than volume related ones. Please take care when choosing a threshold. In which way does this help a user?
Minor remarks:
Please cite the CTAn software properly.
p2, lines 35: The wording here is somewhat misleading. The tomographic reconstruction uses 2D projections from various angles and reconstructs a volume from them. Please do not mix up 2D X-ray projection images and 2D slices trough the reconstructed 3D image. Same remark for Figure 2. (b) is not what the CT device generates but the result of the reconstruction.
Table 1: The differences can only be assessed based on the grey value histogramms of the images. Please provide them.
SMI = Structure Model Index
lines 329: binarized
lines 390: threshold
Citation: https://doi.org/10.5194/egusphere-2026-1-RC2
Status: closed (peer review stopped)
-
RC1: 'Comment on egusphere-2026-1', Anonymous Referee #1, 10 Mar 2026
The study addresses a well-known but often under-quantified issue in micro-CT analysis. The contribution is timely and practical, especially since many studies report topological measurements without clearly accounting for segmentation uncertainty. The systematic threshold sweep and the clear distinction between robust bulk parameters and more sensitive connectivity-based parameters are particular strengths of the manuscript.
Minor Revisions:
lines 51 - 53: Threshold sensitivity is widely studied in general image processing literature. The novelty could be more clearly framed by emphasizing the domain-specific contribution to firn microstructure analysis.
lines 170-172: The boundary conditions and flow direction in the permeability simulations would benefit from clearer explanation Or a simple schematic illustrating inlet, outlet, and no-flow boundaries would improve clarity
lines 265-266: The reasoning behind defining the threshold range as “reasonable” should be introduced earlier, ideally when the range is first mentioned (lines 160–162), to maintain logical flow
lines 270-272: Since the sensitivity metric relies on median-based normalization, a brief explanation of why the median was chosen would improve transparency.
"reliability" in Title: The manuscript refers to assessing the “reliability” of firn microstructural measurements, but the analyses mainly examine sensitivity to threshold choice. It does not clearly identify which binarization method is more physically representative, and the framing may benefit from clarification. -
RC2: 'Comment on egusphere-2026-1', Anonymous Referee #2, 27 Apr 2026
This manuscript deals with a very significant and urgent challenge for everybody quantitatively analysing CT images: How do I binarise the original gray value images correctly to ensure comparability? Which impact do changes in this choice have on the subsequent geometric analysis or simulated materials properties? The data chosen for exploring these questions is very appropriate and interesting, too.
However, there are two major methodological flaws that are hard to overcome: First, the authors start off with reconstructed CT images with grey values in the 8bit range. Tomographic reconstructions have float valued voxels that are typically mapped to 16bit integer values. Sometimes, just 12 out of these 16bit are used but 8bit is definitely a considerable reduction that can seriously affect all following steps. It is however never discussed why and how the grey value range is reduced. Second, global grey value thresholding can work only if there are no global grey value fluctuations. It should at least be checked, if this prerequisite is met for the image data investigated.
Finally, the conclusion is weak: Topological features are more affected than volume related ones. Please take care when choosing a threshold. In which way does this help a user?
Minor remarks:
Please cite the CTAn software properly.
p2, lines 35: The wording here is somewhat misleading. The tomographic reconstruction uses 2D projections from various angles and reconstructs a volume from them. Please do not mix up 2D X-ray projection images and 2D slices trough the reconstructed 3D image. Same remark for Figure 2. (b) is not what the CT device generates but the result of the reconstruction.
Table 1: The differences can only be assessed based on the grey value histogramms of the images. Please provide them.
SMI = Structure Model Index
lines 329: binarized
lines 390: threshold
Citation: https://doi.org/10.5194/egusphere-2026-1-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 754 | 451 | 113 | 1,318 | 186 | 243 |
- HTML: 754
- PDF: 451
- XML: 113
- Total: 1,318
- BibTeX: 186
- EndNote: 243
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The study addresses a well-known but often under-quantified issue in micro-CT analysis. The contribution is timely and practical, especially since many studies report topological measurements without clearly accounting for segmentation uncertainty. The systematic threshold sweep and the clear distinction between robust bulk parameters and more sensitive connectivity-based parameters are particular strengths of the manuscript.
Minor Revisions:
lines 51 - 53: Threshold sensitivity is widely studied in general image processing literature. The novelty could be more clearly framed by emphasizing the domain-specific contribution to firn microstructure analysis.
lines 170-172: The boundary conditions and flow direction in the permeability simulations would benefit from clearer explanation Or a simple schematic illustrating inlet, outlet, and no-flow boundaries would improve clarity
lines 265-266: The reasoning behind defining the threshold range as “reasonable” should be introduced earlier, ideally when the range is first mentioned (lines 160–162), to maintain logical flow
lines 270-272: Since the sensitivity metric relies on median-based normalization, a brief explanation of why the median was chosen would improve transparency.
"reliability" in Title: The manuscript refers to assessing the “reliability” of firn microstructural measurements, but the analyses mainly examine sensitivity to threshold choice. It does not clearly identify which binarization method is more physically representative, and the framing may benefit from clarification.