the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 15 Dec 2021
Impact of spatial data uncertainty in debris flow susceptibility analysis
Abstract. In a study of debris flow susceptibility on the European continent, an analysis of the impact between known location and a location accuracy offset for 99 debris flows, demonstrates the impact of uncertainty in defining appropriate predisposing factors, and consequent analysis for areas of susceptibility.
The dominant predisposing environmental factors, as determined through Maximum Entropy modeling, are presented, and analyzed with respect to the values found at debris flow event points versus a buffered distance of locational uncertainty around each point.
Five Maximum Entropy susceptibility models are developed utilizing the original debris flow inventory of points, randomly generated points, and two models utilizing a subset of points with an uncertainty of 5 km, 1 km, and a model utilizing only points with a known location of “exact”. The AUCs are 0.891, 0.893, 0.896, 0.921, and 0.93, respectively. The “exact” model, with the highest AUC, is ignored in final analyses due to the small number of points, and localized distribution, and hence susceptibility results likely non-representational of the continent.
Each model is analyzed with respect to the AUC, highest contributing factors, factor classes, susceptibility impact, and comparisons of the susceptibility distributions and susceptibility value differences.
Based on model comparisons, geographic extent and context of this study, the models utilizing points with a location uncertainty of less than or equal to 5 km best represent debris flow susceptibility of the continent of Europe. A novel representation of the uncertainty is expressed, and included in a final susceptibility map, as an overlay of standard deviation and mean of susceptibility values for the two best models, providing additional insight for subsequent action.
This preprint has been withdrawn.
-
Withdrawal notice
This preprint has been withdrawn.
-
Preprint
(1705 KB)
Interactive discussion
Status: closed
-
RC1: 'Comment on nhess-2021-364', Anonymous Referee #1, 17 Jan 2022
Dear authors,
first of all thank you for the nice reading, I enjoyed going through your manuscript.
I have suggested the editor for minor revisions and below what I will do is to initially summarize what I understood of the work you proposed and then report my suggestions.
What you present is a susceptibility model at the continental scale. The phenomena you model are debris flows, which you access from the NASA repository. The mapping unit you chose are grid-cells (did you mention their size in the text?) while the covariates you chose span from climatic to terrain ones, to land use and more. You run this experiment by making use of the locational uncertainty provided in the debris flow metadata. As a result you can select debris flows with different level of certainty of their positional accuracy, then run a susceptibility model for each group respectively.
The modeling framework is solid. I only have one suggestion on this. You can remove the term presence-only from the text because it is true that MaxEnt is often referred to as presence only model in the ecological literature. But, in a landslide context, all the model implementations we run do exactly the same thing that maxent does. For instance, even a logistic regression does exactly what you did here. It starts from a set of locations where you consider your presences, then it extract absences or pseudo-absences at random, with a number equal to what you set here to be your background. So, shall we call all the other models presence-only? I think it is more of a phylosofical definition but in the daily life of every susceptibility paper out there what happens is that the two framework coincide.
Also, your exact model, with only 5 debris flows is quite difficult to justify. There I would stress the limitations even more in the text.
One thing I have noticed is that you use the natural break method to classify your susceptibility. This is something that Lombardo et al. 2020 stress in their work. Often authors use one method or not to justify the classification they opt for. I would suggest to write a couple of lines on why you chose this over any other criterion.
Ref: Lombardo, L., Opitz, T., Ardizzone, F., Guzzetti, F. and Huser, R., 2020. Space-time landslide predictive modelling. Earth-Science Reviews, p.103318.
As for the last comments, in all figures you use the acronym for kilometer as Km. This is incorrect as the symbol for kilometer in the international system is km. I would suggest to change it across all figures.
Good luck with the progress of your paper.
Kind regards,
Rev
Citation: https://doi.org/10.5194/nhess-2021-364-RC1 -
AC1: 'Reply on RC1', Laurie Kurilla, 17 Jan 2022
We greatly appreciate the detailed suggestions provided by the reviewer, and their time spent in providing the feedback. We agree with and will implement recommended modifications for the final submission. We regret that we did not "catch" some of these issues before submission.
Regarding the discussion on MaxEnt background points constituting "absence" data, it is noted that “presence” is unknown at the MaxEnt background locations (Merow et al 2013). When using the default setting (as was done in this study), the MaxEnt software uniformly at random selects background locations which may include the “known” debris flow sites, as well. MaxEnt uses background data primarily to characterize environments in the study region rather than to act as “absence” data. (Phillips et al. 2009). Perhaps as further evidence, “A simple strategy to remove sample selection bias is to replace the uniform background data by a random sample of background data drawn from the sampling distribution” (Phillips and Dudek 2008).
We agree that logistic regression is another statistical method not requiring the input of "absence" data. Pointing out that MaxEnt is a “presence-only” model is just one of the justifications for utilizing this methodology. The discussion and methodologies of “presence-only” vs “presence-absence” will continue to be an important topic for this researcher and your insights are much appreciated.
Merow, M. et al A practical guide to MaxEnt for modeling species’ distributions: what it does, and why inputs and settings matter 10.1111/j.1600-0587.2013.07872.x
Phillips, S. et al. 2009. Sample selection bias and presence-only distribution models: implications for background and pseudo-absence data. – Ecol. Appl. 19: 181–197.
Phillips, S. and Dudik, M. 2008. Modeling of species distributions with MaxEnt: new extensions and a comprehensive evaluation. – Ecography 31: 161.
Respectfully,
Laurie J. Kurilla
Citation: https://doi.org/10.5194/nhess-2021-364-AC1
-
AC1: 'Reply on RC1', Laurie Kurilla, 17 Jan 2022
Interactive discussion
Status: closed
-
RC1: 'Comment on nhess-2021-364', Anonymous Referee #1, 17 Jan 2022
Dear authors,
first of all thank you for the nice reading, I enjoyed going through your manuscript.
I have suggested the editor for minor revisions and below what I will do is to initially summarize what I understood of the work you proposed and then report my suggestions.
What you present is a susceptibility model at the continental scale. The phenomena you model are debris flows, which you access from the NASA repository. The mapping unit you chose are grid-cells (did you mention their size in the text?) while the covariates you chose span from climatic to terrain ones, to land use and more. You run this experiment by making use of the locational uncertainty provided in the debris flow metadata. As a result you can select debris flows with different level of certainty of their positional accuracy, then run a susceptibility model for each group respectively.
The modeling framework is solid. I only have one suggestion on this. You can remove the term presence-only from the text because it is true that MaxEnt is often referred to as presence only model in the ecological literature. But, in a landslide context, all the model implementations we run do exactly the same thing that maxent does. For instance, even a logistic regression does exactly what you did here. It starts from a set of locations where you consider your presences, then it extract absences or pseudo-absences at random, with a number equal to what you set here to be your background. So, shall we call all the other models presence-only? I think it is more of a phylosofical definition but in the daily life of every susceptibility paper out there what happens is that the two framework coincide.
Also, your exact model, with only 5 debris flows is quite difficult to justify. There I would stress the limitations even more in the text.
One thing I have noticed is that you use the natural break method to classify your susceptibility. This is something that Lombardo et al. 2020 stress in their work. Often authors use one method or not to justify the classification they opt for. I would suggest to write a couple of lines on why you chose this over any other criterion.
Ref: Lombardo, L., Opitz, T., Ardizzone, F., Guzzetti, F. and Huser, R., 2020. Space-time landslide predictive modelling. Earth-Science Reviews, p.103318.
As for the last comments, in all figures you use the acronym for kilometer as Km. This is incorrect as the symbol for kilometer in the international system is km. I would suggest to change it across all figures.
Good luck with the progress of your paper.
Kind regards,
Rev
Citation: https://doi.org/10.5194/nhess-2021-364-RC1 -
AC1: 'Reply on RC1', Laurie Kurilla, 17 Jan 2022
We greatly appreciate the detailed suggestions provided by the reviewer, and their time spent in providing the feedback. We agree with and will implement recommended modifications for the final submission. We regret that we did not "catch" some of these issues before submission.
Regarding the discussion on MaxEnt background points constituting "absence" data, it is noted that “presence” is unknown at the MaxEnt background locations (Merow et al 2013). When using the default setting (as was done in this study), the MaxEnt software uniformly at random selects background locations which may include the “known” debris flow sites, as well. MaxEnt uses background data primarily to characterize environments in the study region rather than to act as “absence” data. (Phillips et al. 2009). Perhaps as further evidence, “A simple strategy to remove sample selection bias is to replace the uniform background data by a random sample of background data drawn from the sampling distribution” (Phillips and Dudek 2008).
We agree that logistic regression is another statistical method not requiring the input of "absence" data. Pointing out that MaxEnt is a “presence-only” model is just one of the justifications for utilizing this methodology. The discussion and methodologies of “presence-only” vs “presence-absence” will continue to be an important topic for this researcher and your insights are much appreciated.
Merow, M. et al A practical guide to MaxEnt for modeling species’ distributions: what it does, and why inputs and settings matter 10.1111/j.1600-0587.2013.07872.x
Phillips, S. et al. 2009. Sample selection bias and presence-only distribution models: implications for background and pseudo-absence data. – Ecol. Appl. 19: 181–197.
Phillips, S. and Dudik, M. 2008. Modeling of species distributions with MaxEnt: new extensions and a comprehensive evaluation. – Ecography 31: 161.
Respectfully,
Laurie J. Kurilla
Citation: https://doi.org/10.5194/nhess-2021-364-AC1
-
AC1: 'Reply on RC1', Laurie Kurilla, 17 Jan 2022
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 591 | 196 | 43 | 830 | 43 | 46 |
- HTML: 591
- PDF: 196
- XML: 43
- Total: 830
- BibTeX: 43
- EndNote: 46
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1