the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
A new tool to automatise the characterisation of fracture networks from 3D point cloud data
Abstract. Fracture networks linked to the brittle deformation of rocks are often hosts of fluid flows or geomechanical discontinuities which are important to model for rock mass stability analyses, reservoir assessment or storage capacity evaluation. However, the fractures can be mapped only poorly by subsurface geophysics or borehole imagery. 1D scan lines or 2D maps on outcrops analogues are thus one of the current methods used for the assessemet of the networks, but are very limited regarding the 3D distribution of the fractures. This paper shows the first tests and results for a new automated workflow of fracture properties characterization, using high resolution LiDAR and Photogrammetry data on outcrops. From the obtained 3d point clouds on different objects such as quarries, cliffs or road sides, we reconstruct the outcrop surface in a 3D meshed surface with the help of a Simple and Scalable Surface Reconstruction (SSSR). The fracture planes are automatically detected based on a region-growing segmentation approach based on the spatial distribution of the points. It allows to quickly and effectively extract fracture properties such as orientation, geometry and position in the 3d space, with a possibility for direct 3D density computation without using proxies from 1D or 2D data. This paper present the workflow methodology and the tests on 5 basement rock outcrops of different outcropping quality. We also compare the method with a manual fracture picking tool and a classical 1D scan line methodology on the field. It is a first step to get more precise, automatic, easier and faster modelling workflows on fractured rocks from outcrop analogues data.
- Preprint
(68943 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on egusphere-2023-1316', Adam Cawood, 08 Jul 2023
GENERAL COMMENTS:
The manuscript provides a new method for automatic fracture detection and analysis from 3D point-cloud data. I like the general approach taken here and I believe that this method could be of great benefit to practitioners who want to perform fracture characterization more efficiently. The manuscript is interesting, and once concerns have been addressed, should be of interest to a broad readership. The manuscript suffers from some major issues, however, and I therefore recommend Major Revision.
My greatest concern with the manuscript is that in its current form, not enough attention is given to ground-truthing of the proposed method. Given that this method is supposed to provide an accurate, efficient, and reliable approach for fracture detection, the authors need to prove that the technique actually works. The first time I saw any comparisons with field data was in the Discussion section – given the importance of ground-truthing, this data should be in the results sections, and probably very early on. Further, more detailed comparisons between digital and field data are needed.
My second concern is that this code or workflow is not available for testing and implementation by reviewers and/or the wider community. If I am to critically assess this code, I need access to it.
A third concern is that the paper is poorly organized. The geologic setting section is too long, confusing, and a mix of geologic setting and acquisition workflows. The discussion section is almost entirely composed of primary results with not enough consideration of wider implications, reference to other work, potential pitfalls of the approach, or scope for future developments. Major reorganization of the results and discussion sections are needed.
Finally, there are a number of typos throughout the paper, and the discussion has a number of places where the authors intended to cite references but left them blank. This gives the overall impression of the manuscript being a bit rushed, with a lack of attention to detail.
SPECIFIC COMMENTS:
- Section 2.1 is currently too long and a bit confusing. This paper is not focused on geologic histories, deformation conditions, fracture genesis etc. and as such I think this section could be shortened substantially. I’m not convinced the reader needs the level of detail provided in sentences such as “magmatic association of acidic to ultramafic rocks, high pressure rocks and migmatites formed by partial melting of pelitic and quartzofeldspathic rocks”, given that this is in essence a methods paper.
- Acquisition tools and workflows are also currently included in Section 2.1 which is confusing. I suggest that the geologic setting be shortened and the data acquisition text be put into a separate subsection (in Materials and Methods). It might be good to have a table with the outcrop name, rock type, location, scanner used, point-cloud density, outcrop area etc. This would make this information much more easily accessible to the reader without the need for a long section about the geologic setting.
- Section 4.1.1 (Comparison with 1D scan lines) should be in the Results section of the paper. Primary data is presented here and therefore it is not really appropriate to have this in the Discussion section. Further, given my concerns about ground-truthing of your data, I think it is very important to present any comparisons with field data more prominently.
- Section 4.1.2 (Comparison with 2D maps). This section is confusing. Are you comparing your data to your own 2D maps or those already published? It is difficult to tell from your text.
- Section 4.2 (Implications for DFN models). This section is full of primary results related to calculated P21 and P32 values. According to your abstract, this is one of the main objectives of your approach so I don’t understand why this is not included in the results section?
Specific comments and edits in the attached PDF.
Adam Cawood,
San Antonio, Texas
8 July 2023
- AC1: 'Reply on RC1', Lionel Bertrand, 02 Oct 2023
-
RC2: 'Comment on egusphere-2023-1316', Thomas Seers, 05 Sep 2023
The submitted work presents a method for the automated detection of fracture planes from 3D remote sensing data (i.e., terrestrial lidar, digital photogrammetry) based upon region growing using mesh topology to define the search kernel and angular deviation of neighboring mesh elements to define the stop criteria. The authors also present a semi-automated workflow to process and analyze the extracted planar facet dataset to computed key fracture network properties, such as orientation and set distribution, fracture size and fracture intensity measures (i.e., linear, areal and volumetric intensity: P10, P21, P32). The authors are to be commended on undertaking an ambitious objective in the development of automated fracture feature extraction and analysis tools, which can be of major benefit to the community, potentially negating the need for time consuming manual surveys. However, the presented manuscript contains several major flaws which should be addressed in order for the work to be suitable for publication. I therefore recommend that the authors undertake major revisions to their manuscript addressing the following concerns:
- The way the work is presented in terms of prior arts is disingenuous to the existing literature. The earliest automated fracture characterization studies typically focused upon planar extraction of fracture facets (i.e., the current work) yet the authors present the study as though it is groundbreaking in this regard. Some of the earliest approaches are ostensibly similar to the one presented here (i.e., using mesh region growing based upon angular deviation): see Turner, A.K., Kemeny, J.O.H.N., Slob, S.E.I.F.K.O. and Hack, R.O.B.E.R.T., 2006. Evaluation, and management of unstable rock slopes by 3-D laser scanning. IAEG, 404, pp.1-11. The authors need to give the literature its due, framing the novelty of their work in this context. Omitting pertinent literature, some of which has been published for nearly two decades and is very well known within the community is completely unacceptable (I have placed some suggested references in the edited .pdf uploaded with this review).
- The description of the authors’ methodology in terms of their automated fracture analysis is quite confusing and difficult to justify or replicate in its current state. I find some of the calculations to be overly complex (for example the authors’ calculation of mesh area using vectors in convoluted without justification). P21 is described in the text as fractures per unit area which is incorrect. I find the computation of P32 to be particularly problematic: why do you use smoothed and unsmoothed surfaces to define the investigative volume for P32? Surely this results in a thin and uneven volume of interest? How do you use single integrals in two dimensions to approximate the volume between the two surfaces? Surely for this to be valid you would need to evaluate the separate integrals between the two curves of intersection at regular intervals (this is not clear from the text)? Even if we play devil’s advocate and accept this analysis there is a fundamental flaw in using apparent fracture surfaces to compute absolute volumetric intensity measures such as P32, as the apparent values obtained are curtailed due to censoring related to both the orientation of the outcrop vis-a-vis the fracture system and the preference of smaller fractures to be censored by the finite window of observation provided by the outcrop. These are first order considerations in fracture analysis which have been ignored. See Priest (1993) and Wang (2005) for discussions (references in the attached .pdf). Thus, you are only calculating ‘apparent’ P32 at best which is inappropriate for fractured rock mass modelling (i.e., via DFNs).
- Leading on from Point 2, in the results section the authors are overly optimistic about the suitability of their fracture properties for DFN model conditioning due to an overly simplistic view of the derivation of volumetric fracture properties for quasi-2D data (i.e., outcrops). For example, it is conceptually broken to believe that their obtained length distributions are appropriate for DFN model conditioning, as the trace distribution does not fully reflect the underlying distribution of fracture SIZE: as (1) Traces may represent the true diameter of a fracture but are more commonly a chordal length of a larger disk, and (2) trace lengths are right skewed due to the low P of intersection with smaller fractures. This issue can be dealt with using analytical solutions (see Priest 1993) or forward modelling (for example Golder Associates FracMan software has tools for this). In short, you have not fully appreciated the scale of the problem of estimating 3D fracture size from outcrop: a problem that has been dealt with for four decades in the rock engineering literature. This issue is mirrored in the authors’ belief that their P32 estimates constitute appropriate input for DFN modelling. As alluded to in Point 2, this is not conceptually valid. The reason why Wang's (2005) work which the authors dismiss on P32 correction factors exists is because you cannot estimate P32 from the exposed network without considering the occluded component of the network related to stereological effects.
- There are numerous issues with the organization and presentation of the study, including informal or incomprehensible English, missing references or inaccurate technical language / descriptions. I have detailed these issues specifically within the attached .pdf file.
- AC2: 'Reply on RC2', Lionel Bertrand, 02 Oct 2023
Status: closed
-
RC1: 'Comment on egusphere-2023-1316', Adam Cawood, 08 Jul 2023
GENERAL COMMENTS:
The manuscript provides a new method for automatic fracture detection and analysis from 3D point-cloud data. I like the general approach taken here and I believe that this method could be of great benefit to practitioners who want to perform fracture characterization more efficiently. The manuscript is interesting, and once concerns have been addressed, should be of interest to a broad readership. The manuscript suffers from some major issues, however, and I therefore recommend Major Revision.
My greatest concern with the manuscript is that in its current form, not enough attention is given to ground-truthing of the proposed method. Given that this method is supposed to provide an accurate, efficient, and reliable approach for fracture detection, the authors need to prove that the technique actually works. The first time I saw any comparisons with field data was in the Discussion section – given the importance of ground-truthing, this data should be in the results sections, and probably very early on. Further, more detailed comparisons between digital and field data are needed.
My second concern is that this code or workflow is not available for testing and implementation by reviewers and/or the wider community. If I am to critically assess this code, I need access to it.
A third concern is that the paper is poorly organized. The geologic setting section is too long, confusing, and a mix of geologic setting and acquisition workflows. The discussion section is almost entirely composed of primary results with not enough consideration of wider implications, reference to other work, potential pitfalls of the approach, or scope for future developments. Major reorganization of the results and discussion sections are needed.
Finally, there are a number of typos throughout the paper, and the discussion has a number of places where the authors intended to cite references but left them blank. This gives the overall impression of the manuscript being a bit rushed, with a lack of attention to detail.
SPECIFIC COMMENTS:
- Section 2.1 is currently too long and a bit confusing. This paper is not focused on geologic histories, deformation conditions, fracture genesis etc. and as such I think this section could be shortened substantially. I’m not convinced the reader needs the level of detail provided in sentences such as “magmatic association of acidic to ultramafic rocks, high pressure rocks and migmatites formed by partial melting of pelitic and quartzofeldspathic rocks”, given that this is in essence a methods paper.
- Acquisition tools and workflows are also currently included in Section 2.1 which is confusing. I suggest that the geologic setting be shortened and the data acquisition text be put into a separate subsection (in Materials and Methods). It might be good to have a table with the outcrop name, rock type, location, scanner used, point-cloud density, outcrop area etc. This would make this information much more easily accessible to the reader without the need for a long section about the geologic setting.
- Section 4.1.1 (Comparison with 1D scan lines) should be in the Results section of the paper. Primary data is presented here and therefore it is not really appropriate to have this in the Discussion section. Further, given my concerns about ground-truthing of your data, I think it is very important to present any comparisons with field data more prominently.
- Section 4.1.2 (Comparison with 2D maps). This section is confusing. Are you comparing your data to your own 2D maps or those already published? It is difficult to tell from your text.
- Section 4.2 (Implications for DFN models). This section is full of primary results related to calculated P21 and P32 values. According to your abstract, this is one of the main objectives of your approach so I don’t understand why this is not included in the results section?
Specific comments and edits in the attached PDF.
Adam Cawood,
San Antonio, Texas
8 July 2023
- AC1: 'Reply on RC1', Lionel Bertrand, 02 Oct 2023
-
RC2: 'Comment on egusphere-2023-1316', Thomas Seers, 05 Sep 2023
The submitted work presents a method for the automated detection of fracture planes from 3D remote sensing data (i.e., terrestrial lidar, digital photogrammetry) based upon region growing using mesh topology to define the search kernel and angular deviation of neighboring mesh elements to define the stop criteria. The authors also present a semi-automated workflow to process and analyze the extracted planar facet dataset to computed key fracture network properties, such as orientation and set distribution, fracture size and fracture intensity measures (i.e., linear, areal and volumetric intensity: P10, P21, P32). The authors are to be commended on undertaking an ambitious objective in the development of automated fracture feature extraction and analysis tools, which can be of major benefit to the community, potentially negating the need for time consuming manual surveys. However, the presented manuscript contains several major flaws which should be addressed in order for the work to be suitable for publication. I therefore recommend that the authors undertake major revisions to their manuscript addressing the following concerns:
- The way the work is presented in terms of prior arts is disingenuous to the existing literature. The earliest automated fracture characterization studies typically focused upon planar extraction of fracture facets (i.e., the current work) yet the authors present the study as though it is groundbreaking in this regard. Some of the earliest approaches are ostensibly similar to the one presented here (i.e., using mesh region growing based upon angular deviation): see Turner, A.K., Kemeny, J.O.H.N., Slob, S.E.I.F.K.O. and Hack, R.O.B.E.R.T., 2006. Evaluation, and management of unstable rock slopes by 3-D laser scanning. IAEG, 404, pp.1-11. The authors need to give the literature its due, framing the novelty of their work in this context. Omitting pertinent literature, some of which has been published for nearly two decades and is very well known within the community is completely unacceptable (I have placed some suggested references in the edited .pdf uploaded with this review).
- The description of the authors’ methodology in terms of their automated fracture analysis is quite confusing and difficult to justify or replicate in its current state. I find some of the calculations to be overly complex (for example the authors’ calculation of mesh area using vectors in convoluted without justification). P21 is described in the text as fractures per unit area which is incorrect. I find the computation of P32 to be particularly problematic: why do you use smoothed and unsmoothed surfaces to define the investigative volume for P32? Surely this results in a thin and uneven volume of interest? How do you use single integrals in two dimensions to approximate the volume between the two surfaces? Surely for this to be valid you would need to evaluate the separate integrals between the two curves of intersection at regular intervals (this is not clear from the text)? Even if we play devil’s advocate and accept this analysis there is a fundamental flaw in using apparent fracture surfaces to compute absolute volumetric intensity measures such as P32, as the apparent values obtained are curtailed due to censoring related to both the orientation of the outcrop vis-a-vis the fracture system and the preference of smaller fractures to be censored by the finite window of observation provided by the outcrop. These are first order considerations in fracture analysis which have been ignored. See Priest (1993) and Wang (2005) for discussions (references in the attached .pdf). Thus, you are only calculating ‘apparent’ P32 at best which is inappropriate for fractured rock mass modelling (i.e., via DFNs).
- Leading on from Point 2, in the results section the authors are overly optimistic about the suitability of their fracture properties for DFN model conditioning due to an overly simplistic view of the derivation of volumetric fracture properties for quasi-2D data (i.e., outcrops). For example, it is conceptually broken to believe that their obtained length distributions are appropriate for DFN model conditioning, as the trace distribution does not fully reflect the underlying distribution of fracture SIZE: as (1) Traces may represent the true diameter of a fracture but are more commonly a chordal length of a larger disk, and (2) trace lengths are right skewed due to the low P of intersection with smaller fractures. This issue can be dealt with using analytical solutions (see Priest 1993) or forward modelling (for example Golder Associates FracMan software has tools for this). In short, you have not fully appreciated the scale of the problem of estimating 3D fracture size from outcrop: a problem that has been dealt with for four decades in the rock engineering literature. This issue is mirrored in the authors’ belief that their P32 estimates constitute appropriate input for DFN modelling. As alluded to in Point 2, this is not conceptually valid. The reason why Wang's (2005) work which the authors dismiss on P32 correction factors exists is because you cannot estimate P32 from the exposed network without considering the occluded component of the network related to stereological effects.
- There are numerous issues with the organization and presentation of the study, including informal or incomprehensible English, missing references or inaccurate technical language / descriptions. I have detailed these issues specifically within the attached .pdf file.
- AC2: 'Reply on RC2', Lionel Bertrand, 02 Oct 2023
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
473 | 200 | 56 | 729 | 32 | 37 |
- HTML: 473
- PDF: 200
- XML: 56
- Total: 729
- BibTeX: 32
- EndNote: 37
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1