03 Feb 2021
03 Feb 2021
Optimizing and validating the Gravitational Process Path model for regional debrisflow runout modelling
 ^{1}Department of Geography, Friedrich Schiller University Jena, German
 ^{2}Institute of Geosciences, University of Potsdam, Germany
 ^{3}Instituto de Geografía, Pontificia Universidad Católica de Chile, Chile. Centre for Sustainable Urban Development, CEDEUS. Centro de Cambio Global UC
 ^{1}Department of Geography, Friedrich Schiller University Jena, German
 ^{2}Institute of Geosciences, University of Potsdam, Germany
 ^{3}Instituto de Geografía, Pontificia Universidad Católica de Chile, Chile. Centre for Sustainable Urban Development, CEDEUS. Centro de Cambio Global UC
Abstract. Knowing the source and runout of debrisflows can help in planning strategies aimed at mitigating these hazards. Our research in this paper focuses on developing a novel approach for optimizing runout models for regional susceptibility modelling, with a case study in the upper Maipo river basin in the Andes of Santiago, Chile. We propose a twostage optimization approach for automatically selecting parameters for estimating runout path and distance. This approach optimizes the random walk and Perla's twoparameter modelling components of the opensource Gravitational Process Path (GPP) modelling framework. To validate model performance, we assess the spatial transferability of the optimized runout model using spatial crossvalidation, including exploring the model's sensitivity to sample size. We also present diagnostic tools for visualizing uncertainties in parameter selection and model performance. Although there was considerable variation in optimal parameters for individual events, we found our runout modelling approach performed well at regional prediction of potential runout areas. We also found that although a relatively small sample size was sufficient to achieve generally good runout modelling performance; larger samples sizes (i.e. ≥ 80) had higher model performances and lower uncertainties for estimating runout distances at unknown locations. We anticipate that this automated approach using opensource software R and SAGAGIS will make processbased debrisflow models more readily accessible and thus enable researchers and spatial planners to improve regionalscale hazard assessments.
Jason Goetz et al.
Status: final response (author comments only)

RC1: 'Comment on nhess202122', Anonymous Referee #1, 14 Feb 2021
General Comments
Dear Editor, dear Authors,
this a wellwritten and interesting paper on the automatic calibration and validation of a framework for regional debrisflow modelling. Besides the modelling of debrisflow initiation sites with a GAM, the GPP model is used for debrisflow path and runout modelling in the upper Maipo river basin, Andes of central Chile. The authors develop and present a novel approach for model optimization and validation, including several aspects like uncertainty in parameter selection, spatial transferability, and the models's sensitivity to sample size. The results are well presented and discussed, including very nice and informative figures to illustrate the findings. Most parts of section 2 (material and methods) are also well written, but I think this is the section which could be improved most by adding some more detail on some of the aspects (see specific comments below). Apart from that I think the paper is well suited for publication in NHESS. It is also really nice to see that the tools developed for this paper (as well as the data) are also made available to the public.
With best regards.
Specific CommentsSection 2.1.2
Please use a different color for debris flows and roads in Fig. 1, they are both grey and can't be distinguished very well.Section 2.1.3
Regarding the sampling of presence and absence of source points: how do you exactly determine the nonsource points? Do you somehow guarantee that the samples are not "too close" to mapped source points? There are much more nonsource than source points in your study area, how does this influence the results? This affects training as well as validation, please elaborate.After denoising, you apply a sink filling algorithm to the DEM, which one?
Section 2.2.1
Regarding the rating of the random walk performance (line 160): performance was rated higher if observed debrisflow tracks were within the modelled paths. Please provide more details on how this was done exactly, e.g. did you also take the number of cells into consideration that were outside the mapped track? Otherwise you might get optimized parameters that overestimate the process area.Regarding the random walk parameter optimization before the runout optimization (twostage approach): in order to optimize the random walk parameters, wouldn't you also require to use some kind of friction model to limit the runout distance? This overlaps with the previous question, please explain.
Regarding runout distance optimization: here, you use a minimum area bounding box to measure length. What impact has the character of the derbis flow path on this concept? For example, take (1) a quite short, more or less straight debrisflow path versus a (2) very long path, which runs from a hillslope into a channel with a distinct change of direction, let's say 90°? Then you get (1) a bounding box matching the real length quite well and (2) a bounding box which is almost square, strongly underestimating the runout length.
Regarding the optimization of the 2 parameters of the PCM model (sliding friction coefficent "my" and masstodrag ratio "M/D"): a general problem with the PCM model calibration is, that there is some mathematical redundancy between the parameters. I.e., you can achieve the same runout length with different parameter combinations of my and M/D. How does your calibration approach handle this? Please add some information on this, because this may also have some impact on other sections of the paper, e.g. section 3.2 ("low sensitivity for a large range of parameter combinations"), section 3.5 ("no clear spatial pattern in optimal my and M/D parameter combinations across the study area"), section 4.1.2 ("we observed high variability in optimal PCM parameters").
Section 2.2.2
You assessed the transferability of optimized model parameters by 5fold spatial crossvalidation. In section 2.2.1 you state that you are using a random sample of 100 debrisflow tracks for optimization. Is this the sample size you use here too? Or how is this related?Section 2.2.4
To calculate the AUROC, you used 1000 samples of both debrisflow and nondebrisflow locations. How did you sample the nondebrisflow locations? Thematically similar to my question on the nonsource point sampling.Section 3.1
You write that areas with slightly concave profile curvature were modelled as more likely being source areas. So far plan (not profile) curvature was used, and it is also plan curvature that is shown in Fig. 5.Section 3.2
I think it would improve the reading of Table 2 if you would name the "third" model component "Runout distance (spatially varying friction)" instead of only "Runout distance" (like the "second" model component).Section 3.4
In line 294 you write "... the modelled runout paths failed to follow the flow direction ...": is this due to a general problem of the flow path model or is this caused by errors in the DEM?In line 299 you write that "these cases were related to missclassifying stream erosion ...": was the runout lemgth over or underestimated in these cases?
Section 4.1.2
This section (mostly) discusses the runout distance model, please also add a few sentences on the runout path model.
Technical correctionsp1, l5: fix typo in "Germany"
p1, l29: remove additional blank after "learning"; "source" instead of "sources"
p2, l30: remove additional blank after "and"
p2, l46/47: not sure if a comma should be used instead of a semicolon in the enumeration
p2, l55/56: add missing periods after "al" in three citations
p2, l56: remove additional blanks after "be"
p3, l73: "our" instead of "out"
p3, l77: add missing periods after "al" in three citations; Moreiras et al. 2012 and Serey et al. 2019 are missing in the references, please add
p3, l80: add missing period after "al" in the citation; Sepulveda et al. 2006 is missing in the references, please add
p3, l81: add missing period after "al" in the citation
p3, l83: add missing period after "al" in the citation
p4, l95: add period at the end of the table description
p5, l115: add missing "the" in "with ___ remaining set"
p6, l136: the PCM model was developed by three authors, so it isn't "Perla's" model, please rephrase
p9, l208: throughout the text you use a hypen in "debrisflow", here you write "nondebris flow"; should this be changed?
p9, l214: add the missing "a", the package is called "Rsagacmd"
p11, l243: "masstodrag ratio" not "masstodragratio"
p14, l278: "towards a threshold of 0.5" instead of "thresholds"
p14, l279: There's quite a break between the two sentences, I had to read it twice to realize that "The resulting runout prediction map ..." was meant to be that with a threshold of 0.7. Maybe it would help to start a new paragraph here or to reformulate the sentence to something like "The runout prediction map resulting from the best threshold ..."
p18, l314, Figure 11: "... runout path (a), ..." "... relative error (b), actual runout length error (c), and ..."
p22, l264: "source conditions to spatially" instead of "source conditions spatially"
p22, l373: "parameters of the PCM model" instead of "parameters the PCM model"
p24, References: please add the missing references and also have a look at the formatting  there are many references in which the author's first names are not shortened to the initials

AC1: 'Reply on RC1', Jason Goetz, 27 Apr 2021
General Comments
Dear Editor, dear Authors,this a wellwritten and interesting paper on the automatic calibration and validation of a framework for regional debrisflow modelling. Besides the modelling of debrisflow initiation sites with a GAM, the GPP model is used for debrisflow path and runout modelling in the upper Maipo river basin, Andes of central Chile. The authors develop and present a novel approach for model optimization and validation, including several aspects like uncertainty in parameter selection, spatial transferability, and the models's sensitivity to sample size. The results are well presented and discussed, including very nice and informative figures to illustrate the findings. Most parts of section 2 (material and methods) are also well written, but I think this is the section which could be improved most by adding some more detail on some of the aspects (see specific comments below). Apart from that I think the paper is well suited for publication in NHESS. It is also really nice to see that the tools developed for this paper (as well as the data) are also made available to the public.
With best regards.We thank this reviewer for their highly constructive comments. As shown in our following response, we will make improvements to the methods section to help clarify our approach on the use of the AUROC as a metric for runout path modelling, sampling debrisflow and nondebrisflow source areas, and how the the GPP implementation of the random walk model limits runout distance. Additionally, we believe by addressing their comments on the potential multiple optimal PCM (runout distance) model solutions, we enhance our discussion and further demonstrate the suitability of our approach for regional debris flow runout modelling.
Specific Comments
Section 2.1.2
Please use a different color for debris flows and roads in Fig. 1, they are both grey and can't be distinguished very well.We will update the colours in the figures.
Section 2.1.3
Regarding the sampling of presence and absence of source points: how do you exactly determine the nonsource points? Do you somehow guarantee that the samples are not "too close" to mapped source points? There are much more nonsource than source points in your study area, how does this influence the results? This affects training as well as validation, please elaborate.We will rephrase some sentences in this section to help clarify how nonsource points were sampled and how many.
“The nonsource (i.e. absence) points were determined by random sampling locations within the mapped subbasins outside of the debris flow polygons. The resulting training and test data contained 541 source points and 541 nonsource points.”
We guaranteed that the samples were not too close to source points by sampling outside of the mapped polygons. As mentioned in L.102. We used the commonly applied 1:1 sampling ratio of source and nonsource points.
After denoising, you apply a sink filling algorithm to the DEM, which one?
We used the sink filling algorithm from Planchon and Darboux (2001). This citation will be added.
Section 2.2.1
Regarding the rating of the random walk performance (line 160): performance was rated higher if observed debrisflow tracks were within the modelled paths. Please provide more details on how this was done exactly, e.g. did you also take the number of cells into consideration that were outside the mapped track? Otherwise you might get optimized parameters that overestimate the process area.We accounted for the cells outside of the mapped track. The ROC curve plots the true positive rate (TPR) vs the false positive rate (FPR; Zweig and Campbell 1993). Therefore the AUROC does consider cells inside and outside of the mapped tracks. We will add a brief description of the AUROC to the paper to help clarify this, as well as the Zweig and Campbell 1993 citation.
Regarding the random walk parameter optimization before the runout optimization (twostage approach): in order to optimize the random walk parameters, wouldn't you also require to use some kind of friction model to limit the runout distance? This overlaps with the previous question, please explain.
The Gamma (2000) random walk model implemented in the GPP model (Wichman 2017) does not have controls for runout distance. The flow paths will continue downslope until neighboring cells have a higher or equal elevation compared to the central cell being processed. We will add this detail to the paper.
Regarding runout distance optimization: here, you use a minimum area bounding box to measure length. What impact has the character of the derbis flow path on this concept? For example, take (1) a quite short, more or less straight debrisflow path versus a (2) very long path, which runs from a hillslope into a channel with a distinct change of direction, let's say 90°? Then you get (1) a bounding box matching the real length quite well and (2) a bounding box which is almost square, strongly underestimating the runout length.
This is an excellent question. It was also brought up by Reviewer #2.
It is possible that the runout length of the minimum area bounding box can be underestimated when a debris flow makes an abrupt 90 degree change in direction. This may occur for some iterations of decreasing sliding friction coefficient (or increasing masstodrag ratio) past the actual optimal value. However, to mitigate this issue in optimal parameter selection, we use the AUROC to break any ties in performance. Longer runout paths should have a lower AUROC. We will add the following paragraph to the discussion to better explain this issue:
“Although we obtained a unique regional model solution, runout distance relative errors were only slightly higher than the best performer for pairs of sliding friction coefficients and masstodrag ratios across a band in grid search space of lower sliding friction coefficients. This insensitivity of performance to different combinations of PCM model parameters may be due to the uniqueness problem. Our approach using the minimum areabounding box could also contribute to this observed parameter insensitivity. Abrupt changes in flow perpendicular to the initial flow direction, such as a flow meeting a channel, may only slightly increase the length of the bounding box for several iterations of decreasing sliding coefficient (or increasing masstodrag ratio). However, our approach of breaking parameter ties using the AUROC ensures that we select the parameter set that best fits the runout extent in addition to distance – slides with longer runout distances should also have a lower AUROC performance”
Regarding the optimization of the 2 parameters of the PCM model (sliding friction coefficent "my" and masstodrag ratio "M/D"): a general problem with the PCM model calibration is, that there is some mathematical redundancy between the parameters. I.e., you can achieve the same runout length with different parameter combinations of my and M/D. How does your calibration approach handle this? Please add some information on this, because this may also have some impact on other sections of the paper, e.g. section 3.2 ("low sensitivity for a large range of parameter combinations"), section 3.5 ("no clear spatial pattern in optimal my and M/D parameter combinations across the study area"), section 4.1.2 ("we observed high variability in optimal PCM parameters").
If there were ties in the PCM model, we select the parameter set that resulted in the highest AUROC of the runout path performance. We did explore this, and found that this was not a major issue for regional optimization. Only 5% of the repeated spatial crossvalidation iterations (n = 5000) had multiple optimal solutions. For individual events this occurrence rate was much higher (56%). However, after using the AUROC to break ties, the vast majority of individual events (97%) had a unique solution  we will add these results to sections 3.2 and 3.5. For the remaining cases, which still had ties, we simply select the first record. Additionally, in the case where relative error may seem insensitive to perpendicular changes in flow directions, the AUROC enables us to select the optimized flow distance that best matches the flow extent of the mapped debris flow tracks.
We will add the following to the discussion to better highlight this issue with optimizing the PCM model.
“The twoparameter PCM model has a uniqueness problem (Perla et al 1980). There are possibly infinitely many pairs of the sliding friction coefficient and masstodrag ratio that result in the same runout distances. When optimizing individual events, we did observe this phenomenon. The majority of individual events had more than one optimal combination of parameters. Obtaining a unique solution was not an issue for the regional optimization in this case study for the given grid search space. Likely this is due to having to satisfy the runout distances for a variety of hillslope conditions and lengths across the study area. Through our investigation of sample size, we observed a reduced variability in PCM model optimal solutions for larger sample sizes (Figure 15).”
Section 2.2.2
You assessed the transferability of optimized model parameters by 5fold spatial crossvalidation. In section 2.2.1 you state that you are using a random sample of 100 debrisflow tracks for optimization. Is this the sample size you use here too? Or how is this related?It is the same sample. We will add, “Based on our random sample of 100 debrisflow tracks, ...” to help clarify this.
Section 2.2.4
To calculate the AUROC, you used 1000 samples of both debrisflow and nondebrisflow locations. How did you sample the nondebrisflow locations? Thematically similar to my question on the nonsource point sampling.We randomly sampled locations outside of debris flow polygons. We will rephrase this to, “The AUROC was calculated using a sample of 1,000 debrisflow runout locations and 1,000 nondebris flow locations outside of the debrisflow polygons”.
Section 3.1
You write that areas with slightly concave profile curvature were modelled as more likely being source areas. So far plan (not profile) curvature was used, and it is also plan curvature that is shown in Fig. 5.Thanks, this was a typo. We mean plan curvature.
Section 3.2
I think it would improve the reading of Table 2 if you would name the "third" model component "Runout distance (spatially varying friction)" instead of only "Runout distance" (like the "second" model component).Good point! We will make this change.
Section 3.4
In line 294 you write "... the modelled runout paths failed to follow the flow direction ...": is this due to a general problem of the flow path model or is this caused by errors in the DEM?This is likely a problem of the errors in the DEM than the flow path model. We previously mentioned this in the discussion  however, we will add references to works that cover these issues in more detail,
“Poorly individually optimized events could be attributed to locally poor DEM quality (Horton et al., 2013) and mapping uncertainties (Ardizzone et al., 2002)”.
In line 299 you write that "these cases were related to missclassifying stream erosion ...": was the runout lemgth over or underestimated in these cases?
Runout was underestimated for these cases (Figure 11c), likely due to the relatively gentle slope of these stream channels.
Section 4.1.2
This section (mostly) discusses the runout distance model, please also add a few sentences on the runout path model.Thanks, we will add the following interpretation of the runout path model results to the discussion,
“The bestperforming regional randomwalk parameters allowed for maximum lateral spreading of the runout path given the range of parameters for optimization. Individual events tended to also optimize for high lateral spreading, but not as strongly as the regional model. We believe this high lateral spreading may be due to the location of the observed debris flows relative to simulated paths and the quality of the DEM. A large proportion of the observed debrisflow tracks were located at the fringe of the most frequent simulated paths. Thus, a higher slope threshold and exponent of divergence are required to capture these fringe debris flows. Additionally, the surface of DEMs with resolutions greater than 20 m can be too general to capture minor gullies that may have high flow accumulation (Blahut et al 2010b). The 12.5 m resolution ALOS DEM used in this study is derived from downsampled SRTM data, and would likely contain some of the topographic generalizations of the original DEM (~ 30 m spatial resolution). Despite potential issues with DEM quality, similarly to Horton et al. (2013), we illustrated valuable results can still be achieved.”
Technical corrections
p1, l5: fix typo in "Germany"Corrected.
p1, l29: remove additional blank after "learning"; "source" instead of "sources"
Corrected.
p2, l30: remove additional blank after "and"
Corrected.
p2, l46/47: not sure if a comma should be used instead of a semicolon in the enumeration
Corrected.
p2, l55/56: add missing periods after "al" in three citations
Corrected.
p2, l56: remove additional blanks after "be"
Corrected.
p3, l73: "our" instead of "out"
Corrected.
p3, l77: add missing periods after "al" in three citations; Moreiras et al. 2012 and Serey et al. 2019 are missing in the references, please add
Corrected.
p3, l80: add missing period after "al" in the citation; Sepulveda et al. 2006 is missing in the references, please add
Corrected.
p3, l81: add missing period after "al" in the citation
Corrected.
p3, l83: add missing period after "al" in the citation
Corrected.
p4, l95: add period at the end of the table description
Corrected.
p5, l115: add missing "the" in "with ___ remaining set"
Corrected.
p6, l136: the PCM model was developed by three authors, so it isn't "Perla's" model, please rephrase
We will make this change.
p9, l208: throughout the text you use a hypen in "debrisflow", here you write "nondebris flow"; should this be changed?
Will rephrase to, “locations outside of the debrisflow polygons"
p9, l214: add the missing "a", the package is called "Rsagacmd"
Corrected.
p11, l243: "masstodrag ratio" not "masstodragratio"
Corrected.
p14, l278: "towards a threshold of 0.5" instead of "thresholds"
Corrected.
p14, l279: There's quite a break between the two sentences, I had to read it twice to realize that "The resulting runout prediction map ..." was meant to be that with a threshold of 0.7. Maybe it would help to start a new paragraph here or to reformulate the sentence to something like "The runout prediction map resulting from the best threshold ..."
Thanks for the recommendation. We will rephrase it to, “The runout prediction map resulting from the best threshold”.
p18, l314, Figure 11: "... runout path (a), ..." "... relative error (b), actual runout length error (c), and ..."
Corrected.
p22, l264: "source conditions to spatially" instead of "source conditions spatially"
Corrected.
p22, l373: "parameters of the PCM model" instead of "parameters the PCM model"
Corrected.
p24, References: please add the missing references and also have a look at the formatting  there are many references in which the author's first names are not shortened to the initials
Corrected.
References
Ardizzone, F., Cardinali, M., Carrara, A., Guzzetti, F., & Reichenbach, P. (2002). Impact of mapping errors on the reliability of landslide hazard maps. Natural hazards and earth system sciences, 2(1/2), 314.Gamma, P.: Dfwalk – MurgangSimulationsmodell zur Gefahrenzonierung, Geographica
Bernensia, G66, 2000.Horton, P., Jaboyedoff, M., Rudaz, B. E. A., & Zimmermann, M. (2013). FlowR, a model for susceptibility mapping of debris flows and other gravitational hazards at a regional scale. Natural hazards and earth system sciences, 13(4), 869885.
Planchon, O., & Darboux, F. (2001). A fast, simple and versatile algorithm to fill the depressions of digital elevation models. Catena, 46(23), 159176.
Wichmann, V. (2017). The Gravitational Process Path (GPP) model (v1. 0)–a GISbased simulation framework for gravitational processes. Geoscientific Model Development, 10(9), 33093327.
Zweig, M. H., & Campbell, G. (1993). Receiveroperating characteristic (ROC) plots: a fundamental evaluation tool in clinical medicine. Clinical chemistry, 39(4), 561577.

AC1: 'Reply on RC1', Jason Goetz, 27 Apr 2021

RC2: 'Comment on nhess202122', Anonymous Referee #2, 02 Mar 2021
General comments
The paper presents an approach to optimize the parameters of the Gravitational Process Path model for regional debrisflow runout modelling. It addresses the evaluation of the source areas as well as of the runout path and its length. The approach is illustrated with a case study in the upper Maipo river basin in the Andes of Santiago, Chile. The method and the sensitivity analyses are interesting and add value to the field of regional debris flow modelling. The paper is well written, and the figures are of high quality. I recommend publishing it after consideration of two main concerns I have about performance metrics that may require additional work.
Main concerns
I have two main concerns about the performance metrics of the runout distance and runout path:
AUROC for the path: you process the AUROC as defined by: “Model performance was rated higher if the random walk model contained observed debrisflow tracks within its simulated paths” (2.2.1). The problem here is that there is no “false positive” in your approach, and thus the model is not penalized for overpredicting. The approach is correct for the source areas but not for the runout path. As we can see in Fig. 2, the extent of the modelled debris flow is much larger than the observed one, but the AUROC is almost = 1. It means that your model needs to spread as widely as possible to have a good score. I get the difficulty of comparing potential events to a single observed event, but you might then use another contingency table score that does not have false positives. Using a ROCtype score is misleading here if there is no false positive.
Relative error for the runout distance: your approach of using a bounding box on the median frequency (2.2.1) to quantify the runout distance is interesting, but I have an issue with it. Most observed debris flows will likely propagate to the valleybottom, where they might meet the main river. My problem is that when you model the debris flow propagation with small friction values, it is likely to reach the main river and continue perpendicularly, thus not increasing the bounding box for some iterations of the parameter values. We can see such behaviour in your Fig. 2. There is, therefore, a discontinuity as too long propagations are less penalized than too short ones. I believe this might play a role in the results of Fig. 6, where the runout length error remains low for a large range of sliding friction coefficients. It might provide a misleading impression of insensitivity. Or is it the case that most propagations reach a flatter area where they quickly stop anyway? Although an approach based on actual length (for example, defined by a D8) might better represent the difference in runout distance, it might not be trivial to use the median frequency criteria. What about using the median length of all random walk runs for one setting, provided it’s a piece of information you can get? This problem should be at least discussed and considered in the interpretation of the sensitivity analyses. Then, interpretation such as in l. 380381 (“This may indicate that the combination of random walk and the process based PCM model dictates a general runout pattern that is insensitive to values within a broad and nearly optimal range of physically reasonable parameters”) might not be stated this way. Same for l. 407 (“we observed a general insensitivity in runout distance performance of the PCM model to a range of parameters”).
Specific comments
Figure 1: The caption should be a bit more comprehensive, explaining, for example, the random sample.
Section 2.1.3: It might be useful to describe the fundamental principles of the AUROC in 1 sentence.
Section 2.2: Please provide more details about the models and their parameters. For example, mention the random component in the iterative simulations of the random walk and give more information about the persistence factor and the exponent of divergence. As they are key parameters for the rest of the paper, adding a few sentences to describe them and 12 equations would be beneficial for the readers.
Section 2.2.1: Please mention that you do an exhaustive grid search.
Section 2.2.4: You have chosen 1000 “nondebris flow locations”. However, could these be excluded to be potential source areas for future events? Could they become source areas under certain triggering conditions?
Figure 5: It would deserve some more interpretation. For example, what can explain the role of elevation in debris flow conditioning? Why is the slope angle contribution decreasing after a certain threshold? What about the plan curvature?
Section 3.3 & Figure 9: Is the runout frequency relative to a single source? How are they combined when different propagations overlap? Please add some clarifications.
Section 3.4, l. 299: “these cases were related to misclassifying stream erosion”: can you identify such information from satellite imagery?
Section 3.4, l. 309311: not so clear; please clarify.
Figure 12a: You do not mention plot 12a in the text, i.e., the slope threshold values in the grid of other parameters.
Section 3.6 & Figure 14: You might mention again here that these scores are processed on the test data.
Section 4.2: The ability to optimize the runout path and the runout distance separately is related to the fact that the random walk mainly controls the path/spreading, and the PCM controls the runout distance. The influences of these algorithms are quite distinct.
Conclusion: Should contain some more results of your study.
Technical corrections
l. 5: “y” is missing in Germany
l. 73: “our” instead of “out”
l.186: “We explored *for* such spatial…” ?
l. 378: what do you mean by “ambiguous events”?
l. 386: “very *a* specific problem”

AC2: 'Reply on RC2', Jason Goetz, 27 Apr 2021
General comments
The paper presents an approach to optimize the parameters of the Gravitational Process Path model for regional debrisflow runout modelling. It addresses the evaluation of the source areas as well as of the runout path and its length. The approach is illustrated with a case study in the upper Maipo river basin in the Andes of Santiago, Chile. The method and the sensitivity analyses are interesting and add value to the field of regional debris flow modelling. The paper is well written, and the figures are of high quality. I recommend publishing it after consideration of two main concerns I have about performance metrics that may require additional work.We would also like to thank this reviewer for providing highly constructive comments. We believe the first concern regarding the use of the AUROC as a metric for runout path can be addressed by providing a more explicit description of how the AUROC is computed, as well as by providing a more detailed interpretation of the random walk optimization in the discussion. We addressed the second issue regarding runout distance based on the minimum area bounding box length by providing a discussion of how this metric may impact the results. Overall, we believe by addressing these concerns and the specific comments, we are able to provide an even more valuable contribution to the debris flow modelling community.
Main concerns
I have two main concerns about the performance metrics of the runout distance and runout path:
AUROC for the path: you process the AUROC as defined by: “Model performance was rated higher if the random walk model contained observed debrisflow tracks within its simulated paths” (2.2.1). The problem here is that there is no “false positive” in your approach, and thus the model is not penalized for overpredicting. The approach is correct for the source areas but not for the runout path. As we can see in Fig. 2, the extent of the modelled debris flow is much larger than the observed one, but the AUROC is almost = 1. It means that your model needs to spread as widely as possible to have a good score. I get the difficulty of comparing potential events to a single observed event, but you might then use another contingency table score that does not have false positives. Using a ROCtype score is misleading here if there is no false positive.Thanks for bringing up this concern. The AUROC does account for false positives. The receiver operating characteristic curve (ROC), from which we calculate the area under the curve (AUROC), is a plot of true positive vs. false positive rates (Zweig and Campbell 1993). We did not explicitly state this in the original manuscript, so we will put a brief description of the AUROC into the methods section. Our results also indicated that the path optimization does not always favour maximum lateral spread. This is illustrated by the variety of exponentofdivergence values in the parameter selection frequency plot (Figure 12a). We didn’t make this point clear in the paper, so we will add it, as well as the following interpretation of the optimization of the random walk model:
“The bestperforming regional randomwalk parameters allowed for maximum lateral spreading of the runout path given the range of parameters for optimization. Individual events tended to also optimize for high lateral spreading, but not as strongly as the regional model. We believe this high lateral spreading may be due to the location of the observed debris flows relative to simulated paths and the quality of the DEM. A large proportion of the observed debrisflow tracks were located at the fringe of the most frequent simulated paths. Thus, a higher slope threshold and exponent of divergence are required to capture these fringe debris flows. Additionally, the surface of DEMs with resolutions greater than 20 m can be too general to capture minor gullies that may have high flow accumulation (Blahut et al 2010b). The 12.5 m resolution ALOS DEM used in this study is derived from downsampled SRTM data, and would likely contain some of the topographic generalizations of the original DEM (~ 30 m spatial resolution). Despite potential issues with DEM quality, similarly to Horton et al. (2013), we illustrated valuable results can still be achieved.”
Relative error for the runout distance: your approach of using a bounding box on the median frequency (2.2.1) to quantify the runout distance is interesting, but I have an issue with it. Most observed debris flows will likely propagate to the valleybottom, where they might meet the main river. My problem is that when you model the debris flow propagation with small friction values, it is likely to reach the main river and continue perpendicularly, thus not increasing the bounding box for some iterations of the parameter values. We can see such behaviour in your Fig. 2. There is, therefore, a discontinuity as too long propagations are less penalized than too short ones. I believe this might play a role in the results of Fig. 6, where the runout length error remains low for a large range of sliding friction coefficients. It might provide a misleading impression of insensitivity. Or is it the case that most propagations reach a flatter area where they quickly stop anyway? Although an approach based on actual length (for example, defined by a D8) might better represent the difference in runout distance, it might not be trivial to use the median frequency criteria. What about using the median length of all random walk runs for one setting, provided it’s a piece of information you can get? This problem should be at least discussed and considered in the interpretation of the sensitivity analyses. Then, interpretation such as in l. 380381 (“This may indicate that the combination of random walk and the process based PCM model dictates a general runout pattern that is insensitive to values within a broad and nearly optimal range of physically reasonable parameters”) might not be stated this way. Same for l. 407 (“we observed a general insensitivity in runout distance performance of the PCM model to a range of parameters”).
Thanks. Excellent questions.
Regarding parameter insensitivity. We will add the following paragraph to the discussion highlighting that the bounding box approach may have some influence on the sensitivity of parameter performance, and remove our previous statements (l. 380381 and l. 407) on this issue.
“Although we obtained a unique regional model solution, runoutdistance relative errors were only slightly higher than the best performer for pairs of slidingfriction coefficients and masstodrag ratios across a band in gridsearch space of lower slidingfriction coefficients. This insensitivity of performance to different combinations of PCM model parameters may be due to the uniqueness problem. Our approach using the minimum areabounding box could also contribute to this observed parameter insensitivity. Abrupt changes in flow perpendicular to the initial flow direction, such as a flow meeting a channel, may only slightly increase the length of the bounding box for several iterations of decreasing slidingfriction coefficient (or increasing masstodrag ratio). However, our approach of breaking parameter ties using the AUROC ensures that we select the parameter set that best fits the runout extent in addition to distance – slides with longer runout distances should also have a lower AUROC performance”
We don’t believe there is a general issue of longer slides being less penalized than short ones. This seems to only occur when the flow path is nearly perpendicular to the initial flow direction, which is not the case for all the mapped debrisflows tracks used for regionalmodel training.
Thank you for suggesting other approaches to estimating runout distance. We are generally satisfied with our approach to quickly estimate runout length for our study area. Much of the work in this paper was developing an opensource framework to optimize processbased models for runout simulation that can be adapted by others. We highly encourage and look forward to seeing future applications of this approach modify this framework, such as trying different performance metrics, to best suit particular applications.
Specific comments
Figure 1: The caption should be a bit more comprehensive, explaining, for example, the random sample.Thanks, we will add, “The selection of these debris flow polygons was based on a random sample” to the caption.
Section 2.1.3: It might be useful to describe the fundamental principles of the AUROC in 1 sentence.
Here, we will add the following, “The receiver operating characteristic (ROC) is a plot of the true positive rate versus the false positive rate. AUROC values range from 0.5 (random discrimination between classes) and 1.0 (a perfect classifier).”
Section 2.2: Please provide more details about the models and their parameters. For example, mention the random component in the iterative simulations of the random walk and give more information about the persistence factor and the exponent of divergence. As they are key parameters for the rest of the paper, adding a few sentences to describe them and 12 equations would be beneficial for the readers.
We agree that a better description of the random walk model (Gamma 2000) can help improve the reader’s interpretation of the results. We therefore will add the following to Section 2.2:
“Flow path is determined using a 3×3 window that first controls the path of a central cell by considering only neighboring cells with lower elevation. If the neighbouring cells are below the slope threshold, the neighboring cell with the steepest descent is selected; otherwise, neighbours are assigned transition probabilities based on slope. These probabilities are adjusted using the exponent of divergence and the persistence factor. A higher exponent of divergence will result in more even probabilities across the neighbouring cells, allowing for a higher likelihood of not selecting the steepest descent path. The persistence factor considers the previous flow direction in weighting the probabilities. A higher persistence factor increases the probability that the selected neighbor will follow the direction of the previous cell. Based on these transition probabilities, a pseudorandom number generator selects a cell to define the flow path (see Gamma 2000; and Wichman 2017 for a more detailed description). With this randomwalk implementation, the flow path stops when the neighboring cells have a higher or equal elevation compared to the central cell.”
Section 2.2.1: Please mention that you do an exhaustive grid search.
Added.
Section 2.2.4: You have chosen 1000 “nondebris flow locations”. However, could these be excluded to be potential source areas for future events? Could they become source areas under certain triggering conditions?
This is a general challenge in selecting nondebris flow locations. Future work could focus on improving methods for identifying these locations.
Figure 5: It would deserve some more interpretation. For example, what can explain the role of elevation in debris flow conditioning? Why is the slope angle contribution decreasing after a certain threshold? What about the plan curvature?
The relationship between elevation and debris flow activity is complex. In the upper Maipo river basin elevation can be a proxy for vegetation, snow cover duration, terrain ruggedness, permafrost and glacial bodies, and geology. It is therefore difficult to discern any direct relationships between elevation and likelihood of being debris source areas. However, we suspect that lower elevations are predicted to be less prone to be source areas due to increased vegetation cover and less rugged terrain. The decrease observed at the highest elevations may relate to permafrost and glacial bodies holding potentially mobilized sediment (e.g. Sattler et al. 2011).
We observed a decrease in likelihood of source areas occurring at high slope angles (e.g. ~ >45°). These steep slopes can be associated with steep rock faces that are more likely sources of rock falls than debris flows (Loye et al 2009).
Slightly concave plan curvature of the slopes (relative to the DEM) are associated with being more likely source areas.
We will add this interpretation to the results.
Section 3.3 & Figure 9: Is the runout frequency relative to a single source? How are they combined when different propagations overlap? Please add some clarifications.
As computed from the GPP model, the runout frequencies are the total times a cell is traversed from all source areas (Wichmann 2017). We will add the following to section 2.2.
“This is a cumulative frequency based on simulations from all source areas”
Section 3.4, l. 299: “these cases were related to misclassifying stream erosion”: can you identify such information from satellite imagery?
Through expert interpretation of DEM derived hillslope angles and very high resolution satellite imagery (0.50 m) we are confident in our ability to identify such information. We didn’t make it clear in the paper that hillslope angle was used to help with the interpretation, so we will add this to the paper.
Section 3.4, l. 309311: not so clear; please clarify.
Thanks. We will clarify this section by rephrasing it to:
“The optimization of the runout model avoided overfitting to debrisflow tracks of a certain magnitude and general terrain conditions. That is, we did not observe a strong correlation between runout distance performance to length of observed debris flow (ρ = 0.36), starting elevations (0.21), catchment area (0.11) or hillslope angle (0.29) of source points used for model training.”
Figure 12a: You do not mention plot 12a in the text, i.e., the slope threshold values in the grid of other parameters.
We will add the following to the results to describe the simulated path behaviour of the individual events:
“Most individual events optimized runout paths with parameter sets leading to high lateral spreading. The optimalpath parameters for most of the individual events had a 40° slope threshold, high exponent of divergence and low persistence values (Figure 12a). By individually examining the optimal simulated paths for each training event, we observed that 60% of the observed debrisflow tracks did occur within the most frequent simulated paths. The other 40% of events were typically located on the fringes of the most frequent paths.”
Section 3.6 & Figure 14: You might mention again here that these scores are processed on the test data.
We will add that we used spatial crossvalidation to assess the performance in the figure caption.
“Figure 14. Comparison of runout path (a) and distance (b) performances for different model training samples sizes assessed using spatial crossvalidation. The error bars indicate the standard deviation in performances.”
We will also add some clarification of this in the methods (Section 2.2.3).
“Spatial crossvalidation was applied to data sets of varying training sample sizes using the random sample of 100 debris flows used for model optimization”
Section 4.2: The ability to optimize the runout path and the runout distance separately is related to the fact that the random walk mainly controls the path/spreading, and the PCM controls the runout distance. The influences of these algorithms are quite distinct.
Thank you. This is a really valid point to remind the readers in the discussion. We will add the following to Section 4.2.:
“The modular framework of the GPP model provides the ability to optimize two distinct runout components, the runout path including lateral spreading and the runout distance. In our study, we used the random walk and PCM components of the GPP model to simulate spatial extent of runout.”
Conclusion: Should contain some more results of your study.
We will add the following points to our conclusion.
 The combination of the statistical learning for source area prediction and regional optimization of the random walk and PCM model components of the GPP runout model performed well at generalizing runout patterns across the upper Maipo river basin.
 In addition to its strong performance, the transparency and interpretability of the GAM provided further user confidence in predicting debris flow source areas.
 Unique regionaloptimal PCM model solutions were more prone with larger sample sizes, as well as higher model performances and lower uncertainties.
Technical corrections
l. 5: “y” is missing in GermanyCorrected.
l. 73: “our” instead of “out”
Corrected.
l.186: “We explored *for* such spatial…” ?
Corrected.
l. 378: what do you mean by “ambiguous events”?
We meant to refer to uncertainties in mapping debris flows.
We will change this sentence to, “Poorly individually optimized events could be attributed to locally poor DEM quality (Horton et al 2013) and mapping uncertainties (Ardizzone et al 2002).”
l. 386: “very *a* specific problem”
Corrected.
References
Gamma, P.: Dfwalk – MurgangSimulationsmodell zur Gefahrenzonierung, Geographica
Bernensia, G66, 2000.Horton, P., Jaboyedoff, M., Rudaz, B. E. A., & Zimmermann, M. (2013). FlowR, a model for susceptibility mapping of debris flows and other gravitational hazards at a regional scale. Natural hazards and earth system sciences, 13(4), 869885.
Loye, A., Jaboyedoff, M., & Pedrazzini, A. (2009). Identification of potential rockfall source areas at a regional scale using a DEMbased geomorphometric analysis. Natural Hazards and Earth System Sciences, 9(5), 16431653.
Sattler, K., Keiler, M., Zischg, A., & Schrott, L. (2011). On the connection between debris flow activity and permafrost degradation: a case study from the Schnalstal, South Tyrolean Alps, Italy. Permafrost and Periglacial Processes, 22(3), 254265.
Wichmann, V. (2017). The Gravitational Process Path (GPP) model (v1. 0)–a GISbased simulation framework for gravitational processes. Geoscientific Model Development, 10(9), 33093327.
Zweig, M. H., & Campbell, G. (1993). Receiveroperating characteristic (ROC) plots: a fundamental evaluation tool in clinical medicine. Clinical chemistry, 39(4), 561577.

AC2: 'Reply on RC2', Jason Goetz, 27 Apr 2021
Jason Goetz et al.
Data sets
Debris flow inventory and data for regionally modelling runout in the upper Maipo river basin, Chile Eric Parra Hormazábal, Jazmine Calabrese Fernández, Manuel Bustos Morales, María Belén Araneda Riquelme, Jason Goetz, Robin Kohrs, Alex Brenning, and Cristián Henríquez https://doi.org/10.5281/zenodo.4428080
Model code and software
runoptGPP: An R package for optimizing mass movement runout models Jason Goetz https://doi.org/10.5281/zenodo.4428050
Jason Goetz et al.
Viewed
HTML  XML  Total  BibTeX  EndNote  

410  149  16  575  3  8 
 HTML: 410
 PDF: 149
 XML: 16
 Total: 575
 BibTeX: 3
 EndNote: 8
Viewed (geographical distribution)
Country  #  Views  % 

Total:  0 
HTML:  0 
PDF:  0 
XML:  0 
 1