the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 11 Jul 2022
Potential of satellite-derived hydro-meteorological information for landslide hazard assessment thresholds in Rwanda
Abstract. Satellite and hydrological model-based technologies provide estimates of rainfall and soil moisture over larger spatial scales and now cover multiple decades, sufficient to explore their value for the development of landslide early warning system in data scarce regions. In this study, we used statistical metrics to compare gauge-based to satellite-based precipitation products and assess their performance in landslide hazard assessment and warning in Rwanda. Similarly, the value of high resolution satellite and hydrological model-derived soil moisture was compared to in situ soil moisture observations at Rwanda weather station sites. Based on statistical indicators, the NASA GPM-based IMERG rainfall product showed the highest skill to reproduce the main spatiotemporal precipitation patterns at the studies sites in Rwanda. Similarly, the satellite and model-derived soil moisture time series broadly reproduce the most important trends of in situ soil moisture observations. We evaluated two categories of landslide meteorological triggering conditions from IMERG satellite precipitation. First, the maximum rainfall amount during a multiple day rainfall event. Second, the cumulative rainfall over the past few day(s). For each category, the antecedent soil moisture recorded at three levels of soil depth, top 5 cm by satellite-based technologies as well as top 50 cm and 2 m through modelling approaches, was included in the statistical models to assess its potential for landslide hazard assessment and warning capabilities. The results reveal the cumulative 3 day rainfall RD3 as the most effective predictor for landslide triggering. This was indicated not only by its highest discriminatory power to distinguish landslide from no landslide conditions (AUC ~0.72) but also the resulting true positive alarms TPR of ~80 %. The modelled antecedent soil moisture in the 50 cm root zone Seroot(t-3) was the most informative hydrological variable for landslide hazard assessment (AUC ~0.74 and TPR of 84 %). The hydro-meteorological threshold models that incorporate the Seroot(t-3) and RD3 following the cause–trigger concept in a bilinear framework reveal promising results with improved landslide warning capabilities in terms of reduced rate of false alarms by ~20 % at the expense of a minor reduction of true alarms by ~8 %.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(3284 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(3284 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Review to egushere-2022-596 by Uwihirwe et al.', Anonymous Referee #1, 10 Aug 2022
General comment
The manuscript deals with the investigation of the use satellite-derived rainfall and soil moisture information to derive thresholds useful for hazard assessment (in the sense of early warning) in Rwanda. In particular, the authors compare different satellite-derived precipitation and soil moisture products with observations. They also use soil moisture derived from a simple hydrological model. Then they use several rainfall variables to analyse their predictive power for landslides (single rainfall variables and rainfall+soil moisture 2D analysis).
The manuscript is well written and the investigation is sound. I think that just a few points need clarification and some additional comments, as detailed below. For this reason I suggest minor revisions for this manuscript.
Specific comments
Section 3.3.3 This section describes the hydrological model-derived soil moisture. I feel that maybe the manuscript can benefit from a few more words about the model and its calibration. Perhaps just 5-10 lines may be sufficient, as I understand that you do not want to break too much the flow of the manuscript. Otherwise go for an appendix/supplementary material.
Equation 7. I really appreciate the approach of normalizing the soil water content to make comparisons between models and observations. However more details should be given on this: which values of theta_max and theta_min have been found for the various soil moisture products?
Section 3.1 The landslide inventory is made of 32 useful landslides. These are a bit few (see analyses in https://link.springer.com/article/10.1007/s10346-021-01704-7). A comment on this may be added. However, for the manuscript this is not a big issue as it focuses on Rwanda which is an area for which only a few studies exist.
Section 4.2.3 : A comment on the limitations of the analysis related to the constraint of using a bilinear threshold form may be added (see e.g., https://www.mdpi.com/2073-4441/13/13/1752/htm, where other forms are suggested).
Minor comments/technical corrections
LL 364-365 This is unclear: I immagine that the critical level for landslide occurrence is sort of fixed and then it is reached more or less easily based on the prior rainfall and the time lag.
LL 500 The authors apply a threshold of 10 mm on satellite products to make them better agree with observations. This is a sort of a “bias correction”, about which a lot of literature exist. Perhaps make a fast literature review and add some references. (This could be useful also for future work)
Fig. 1 is perhaps a little bit messy (especially in B/W).
LL 512 the authors write “inter-event time” as the minimum dry interval between rainfall events. Perhaps add “minimum”, even if I understand that IET is aligned with previous literature in the field of landslides.
L120 a “)” is missing after Mukungwa.
L586 thus “can be” (?) very useful (something is missing in the sentence).
Citation: https://doi.org/10.5194/egusphere-2022-596-RC1 -
AC1: 'Reply on RC1', Judith Uwihirwe, 07 Sep 2022
Dear Referee,
We are so thankful for your overall positive feedback on the manuscript and for the important comments and suggestions. We have therefore addressed them as follow:
Specific comments
Comment1: Section 3.3.3 This section describes the hydrological model-derived soil moisture. I feel that maybe the manuscript can benefit from a few more words about the model and its calibration. Perhaps just 5-10 lines may be sufficient, as I understand that you do not want to break too much the flow of the manuscript. Otherwise go for an appendix/supplementary material.
Response: An appendix about the Wflow model and its calibration will be added to the final version of the manuscript
Comment 2: Equation 7. I really appreciate the approach of normalizing the soil water content to make comparisons between models and observations. However more details should be given on this: which values of theta_max and theta_min have been found for the various soil moisture products?
Response: We agree that the normalization of soil water content (theta) was made for easy comparison of the observed, model-derived and satellite-based soil moisture products. However, for all compared soil moisture products, the and were 1 and 0 respectively which led to almost similar values of Se (effective soil moisture) and (actual soil water content). We will add such information in Section 3.4.1 of the manuscript for clarification.
Comment 3: Section 3.1 The landslide inventory is made of 32 useful landslides. These are a bit few (see analyses in https://link.springer.com/article/10.1007/s10346-021-01704-7). A comment on this may be added. However, for the manuscript this is not a big issue as it focuses on Rwanda which is an area for which only a few studies exist.
Response: This short paragraph on the constraints related to the used small sized sample (32 hazardous landslides events) will be added to Section 4.2.4. ”Ideally one would have a landslide inventory of about 200 landslides events in order to have a precise estimation of threshold parameters (Peres and Cancelliere, 2021). However, the landslide inventory used for this study counts for only 32 hazardous landslides. Although, the reliance on this limited sample size is likely to lead to a bias towards the larger landslide events and those with impact to society, this landslide inventory is the most comprehensive currently available in the study area”.
Comment 4: Section 4.2.3 : A comment on the limitations of the analysis related to the constraint of using a bilinear threshold form may be added (see e.g., https://www.mdpi.com/2073-4441/13/13/1752/htm, where other forms are suggested).
Response: The limitations and constraints of using the bilinear format have been shortly presented in Section 4.2.4. However, additional comment on the constraint of using a bilinear threshold will be added in Section 4.2.4 referring to Conrad et al., (2021).
Minor comments/technical corrections
Comment 1: LL 364-365 This is unclear: I imagine that the critical level for landslide occurrence is sort of fixed and then it is reached more or less easily based on the prior rainfall and the time lag.
Response: It is true that the critical level for landslide occurrence is more or less fixed when other geological and geomorphological condition are kept constant and it is reached more or less easily depending on the prior rainfall expressed in terms of antecedent soil moisture and the time lag between the landslide triggering rainfall and the soil hydrological response. LL 364-365 will be paraphrased accordingly.
Comment 2: LL 500 The authors apply a threshold of 10 mm on satellite products to make them better agree with observations. This is a sort of a “bias correction”, about which a lot of literature exist. Perhaps make a fast literature review and add some references. (This could be useful also for future work)
Response: The threshold definition of a rainy day (10 mm) improved the similarities between the satellite-based and gauge-based landslide thresholds and thus considered as a bias correction between the two sources of rainfall data. Similarly, bias correction methods were adopted by other researchers to ensure for the high accuracy between ground- and satellite-based rainfall data (Bhatti et al., 2016; Vernimmen et al., 2012). This paragraph and references will be added to LL 500
Comment 3: Fig. 1 is perhaps a little bit messy (especially in B/W).
Response: Fig 1 will be improved to ensure for a better visibility and readability.
Comment 4: LL 512 the authors write “inter-event time” as the minimum dry interval between rainfall events. Perhaps add “minimum”, even if I understand that IET is aligned with previous literature in the field of landslides.
Response: The word “minimum” will be added and thus “minimum inter-event time”
Comment: L120 a “)” is missing after Mukungwa.
Response: a “)” will be added after Mukungwa
Comment: L586 thus “can be” (?) very useful (something is missing in the sentence).
Response: The word “can be “ will be added “thus can be very useful for landslide hazard….”
Citation: https://doi.org/10.5194/egusphere-2022-596-AC1
-
AC1: 'Reply on RC1', Judith Uwihirwe, 07 Sep 2022
-
RC2: 'Comment on egusphere-2022-596', Anonymous Referee #2, 22 Aug 2022
The work presents a comprehensive analysis of the possible combination of rainfall variables and soil moisture data (both measured and modelled) from ground and satellite for predicting rainfall-induced landslides. The outcomes of the analysis are based on ROC analysis, and skill scores, especially TPR and FPR.
General remarks
The title contains the term “landslide hazard”, which is misleading if referred to the manuscript contents. Hazard is generally intended to be an “off line” property of a territory, which is a function of susceptibility, temporal and magnitude (size) probability.
Regarding the discussion of threshold results, in my experience, false negatives (i.e., missed alarms) are more important than false positives (i.e., false alarms), because the consequences of missed alarms (e.g., deaths and injuries) are certainly more severe than those caused by false alarms (e.g., the unnecessary evacuation of a school). Therefore, looking at the number of missed alarms in Fig. 8b and 8e, I would not rely too heavily on these thresholds in a LEWS (Landslide Early Warning System). I suggest you to review the Discussion and Conclusions.
Minor revisions
- Figs. 1, 2, 3, and 4 are similar and repetitive, and the 5-km buffers locally obscure the information from high-granularity maps in Figs. 1 and 2. For a better readability, my suggestion is to merge Figures (perhaps Fig. 1 with Fig. 3, and Fig. 2 with Fig.4) selecting two maps (perhaps elevation and geomorphology), and grouping sensor and landslide information, in Fig.2-4. By the way, I think that the year of occurrence in not that important for this analysis given the relatively low number of failures. Other information on the landslide sites (mean terrain slope in ROIs, aquifer type) could be provided in a Table with the list of failures.
- I suggest using “cumulated event rainfall” for E, “event duration” for D, and “rainfall mean intensity” for I.
- A quantitative error estimation on your findings is lacking.
- Tables 4 and 5 are a bit confusing. If I understand correctly, the first group of 5 columns refers to the whole landslide area, while the second group only refers to the modeled catchments. If so, please amend the tables accordingly.
- In Section 4.2.3 on hydro-meteorological thresholds, I would suggest to calculate also classical ED thresholds, which could provide competitive skill scores.
- Figs. 8 and 9. You should improve the quality, especially that of 8d and 9d. Please, also avoid too small characters in the legenda, and instead use the caption to explain symbols and colors.
Additional minor revisions are in the attached PDF file.
-
AC2: 'Reply on RC2', Judith Uwihirwe, 07 Sep 2022
Dear Referee,
Thank you for your feedback on our manuscript. The raised comments and suggestions are of great value for the improvement of the manuscript. We have considered all as follow:
Comment: The title contains the term “landslide hazard”, which is misleading if referred to the manuscript contents. Hazard is generally intended to be an “off line” property of a territory, which is a function of susceptibility, temporal and magnitude (size) probability.
Response: The word “Hazard assessment “ will be removed from the title to fit with the manuscript content as suggested. The Title will be “Potential of satellite-derived hydro-meteorological information for landslide initiation thresholds in Rwanda”.
Comment: Regarding the discussion of threshold results, in my experience, false negatives (i.e., missed alarms) are more important than false positives (i.e., false alarms), because the consequences of missed alarms (e.g. deaths and injuries) are certainly more severe than those caused by false alarms (e.g., the unnecessary evacuation of a school). Therefore, looking at the number of missed alarms in Fig. 8b and 8e, I would not rely too heavily on these thresholds in a LEWS (Landslide Early Warning System). I suggest you to review the Discussion and Conclusions.
Response: We agree that the consequences of offering false alarms (FPR) are less harmful on the short-term than missed alarms (FNR) which implies that the best threshold should maximize the rate of true positives TPR (true alarms) while minimizing the FNR. However, the thresholds in Fig. 8b and 8e are classical thresholds ED relying exclusively on rainfall (Trigger), leading to the high rate of missed alarms and thus less important for a robust LEWS development. Similar to this study, previous studies (Bogaard and Greco, 2018; Peres et al., 2018, Marino et al., 2020; Thomas et al., 2020; Zhuo et al., 2019, Mirus et al., 2018a; Thomas et al., 2019; Uwihirwe et al., 2020, 2021) indicated that the consideration of the prior subsurface hydrological conditions reduce the number of missed alarms FNR as well as the number of false alarms FPR relative to the exclusive use of rainfall-only thresholds. In Figure 8a, 8c and 8d, 9a, 9b and 9c, we integrated the hydrological information (i.e. antecedent soil moisture) in landslide thresholds to improve the rate of TPR and reduce the rate of FNR and FPR. The main goal of hydro-meteorological thresholds (Cause-trigger) is to maximize the rate of true positives TPR (true alarms) i.e. minimize the FNR but at the same time reducing the rate of false positives FPR (False alarms). The used statistical metrics (TSS and RAD) are also in line with this concept aiming at maximizing the rate of true positives TPR while minimizing the rate of false positives FPR. Once TPR is maximized, the FNR is also minimized though difficult and or impossible to have a perfect threshold model with zero FNR and FPR. We will add a discussion point about this information in Section 4.2.4 and in conclusion part. We will also correct the number of false negative (FNR) in Figure 8f and 9d.
Minor revisions
Comment: Figs. 1, 2, 3, and 4 are similar and repetitive, and the 5-km buffers locally obscure the information from high-granularity maps in Figs. 1 and 2. For a better readability, my suggestion is to merge Figures (perhaps Fig. 1 with Fig. 3, and Fig. 2 with Fig.4) selecting two maps (perhaps elevation and geomorphology), and grouping sensor and landslide information, in Fig.2-4. By the way, I think that the year of occurrence in not that important for this analysis given the relatively low number of failures. Other information on the landslide sites (mean terrain slope in ROIs, aquifer type) could be provided in a Table with the list of failures.
Response: We will merge Fig.1&2 containing quite similar information and keep Fig. 3&4 showing different extent of our study area (ROIs) to keep the flow of the methodology. The mean terrain slope (Map) will only be kept in text and be removed in Fig 2.
Comment: I suggest using “cumulated event rainfall” for E, “event duration” for D, and “rainfall mean intensity” for I.
Response: We will replace “ rainfall event volume E” by “cumulated event rainfall E” ; and “event intensity E/D” by “rainfall mean intensity E/D” as Suggested. The “event duration” for D is same as suggested and will be kept.
Comment: Tables 4 and 5 are a bit confusing. If I understand correctly, the first group of 5 columns refers to the whole landslide area, while the second group only refers to the modeled catchments. If so, please amend the tables accordingly.
Response: We will amend the Tables’ captions accordingly even though same explanation have been provided in the footnotes of Table 4 and Table 5.
Comment: In Section 4.2.3 on hydro-meteorological thresholds, I would suggest to calculate also classical ED thresholds, which could provide competitive skill scores.
Response: In Figure 8b, 8e and 8f the classical ED thresholds are presented despite the weak prediction capability (TPR=50%) and weak skill score.
Comment: Figs. 8 and 9. You should improve the quality, especially that of 8d and 9d. Please, also avoid too small characters in the legend a, and instead use the caption to explain symbols and colours.
Response: We will improve the quality of Figure 8 and 9 by increasing the font size of the legend and explaining some of symbols in the Figures’ captions.
Citation: https://doi.org/10.5194/egusphere-2022-596-AC2
-
AC2: 'Reply on RC2', Judith Uwihirwe, 07 Sep 2022
Interactive discussion
Status: closed
-
RC1: 'Review to egushere-2022-596 by Uwihirwe et al.', Anonymous Referee #1, 10 Aug 2022
General comment
The manuscript deals with the investigation of the use satellite-derived rainfall and soil moisture information to derive thresholds useful for hazard assessment (in the sense of early warning) in Rwanda. In particular, the authors compare different satellite-derived precipitation and soil moisture products with observations. They also use soil moisture derived from a simple hydrological model. Then they use several rainfall variables to analyse their predictive power for landslides (single rainfall variables and rainfall+soil moisture 2D analysis).
The manuscript is well written and the investigation is sound. I think that just a few points need clarification and some additional comments, as detailed below. For this reason I suggest minor revisions for this manuscript.
Specific comments
Section 3.3.3 This section describes the hydrological model-derived soil moisture. I feel that maybe the manuscript can benefit from a few more words about the model and its calibration. Perhaps just 5-10 lines may be sufficient, as I understand that you do not want to break too much the flow of the manuscript. Otherwise go for an appendix/supplementary material.
Equation 7. I really appreciate the approach of normalizing the soil water content to make comparisons between models and observations. However more details should be given on this: which values of theta_max and theta_min have been found for the various soil moisture products?
Section 3.1 The landslide inventory is made of 32 useful landslides. These are a bit few (see analyses in https://link.springer.com/article/10.1007/s10346-021-01704-7). A comment on this may be added. However, for the manuscript this is not a big issue as it focuses on Rwanda which is an area for which only a few studies exist.
Section 4.2.3 : A comment on the limitations of the analysis related to the constraint of using a bilinear threshold form may be added (see e.g., https://www.mdpi.com/2073-4441/13/13/1752/htm, where other forms are suggested).
Minor comments/technical corrections
LL 364-365 This is unclear: I immagine that the critical level for landslide occurrence is sort of fixed and then it is reached more or less easily based on the prior rainfall and the time lag.
LL 500 The authors apply a threshold of 10 mm on satellite products to make them better agree with observations. This is a sort of a “bias correction”, about which a lot of literature exist. Perhaps make a fast literature review and add some references. (This could be useful also for future work)
Fig. 1 is perhaps a little bit messy (especially in B/W).
LL 512 the authors write “inter-event time” as the minimum dry interval between rainfall events. Perhaps add “minimum”, even if I understand that IET is aligned with previous literature in the field of landslides.
L120 a “)” is missing after Mukungwa.
L586 thus “can be” (?) very useful (something is missing in the sentence).
Citation: https://doi.org/10.5194/egusphere-2022-596-RC1 -
AC1: 'Reply on RC1', Judith Uwihirwe, 07 Sep 2022
Dear Referee,
We are so thankful for your overall positive feedback on the manuscript and for the important comments and suggestions. We have therefore addressed them as follow:
Specific comments
Comment1: Section 3.3.3 This section describes the hydrological model-derived soil moisture. I feel that maybe the manuscript can benefit from a few more words about the model and its calibration. Perhaps just 5-10 lines may be sufficient, as I understand that you do not want to break too much the flow of the manuscript. Otherwise go for an appendix/supplementary material.
Response: An appendix about the Wflow model and its calibration will be added to the final version of the manuscript
Comment 2: Equation 7. I really appreciate the approach of normalizing the soil water content to make comparisons between models and observations. However more details should be given on this: which values of theta_max and theta_min have been found for the various soil moisture products?
Response: We agree that the normalization of soil water content (theta) was made for easy comparison of the observed, model-derived and satellite-based soil moisture products. However, for all compared soil moisture products, the and were 1 and 0 respectively which led to almost similar values of Se (effective soil moisture) and (actual soil water content). We will add such information in Section 3.4.1 of the manuscript for clarification.
Comment 3: Section 3.1 The landslide inventory is made of 32 useful landslides. These are a bit few (see analyses in https://link.springer.com/article/10.1007/s10346-021-01704-7). A comment on this may be added. However, for the manuscript this is not a big issue as it focuses on Rwanda which is an area for which only a few studies exist.
Response: This short paragraph on the constraints related to the used small sized sample (32 hazardous landslides events) will be added to Section 4.2.4. ”Ideally one would have a landslide inventory of about 200 landslides events in order to have a precise estimation of threshold parameters (Peres and Cancelliere, 2021). However, the landslide inventory used for this study counts for only 32 hazardous landslides. Although, the reliance on this limited sample size is likely to lead to a bias towards the larger landslide events and those with impact to society, this landslide inventory is the most comprehensive currently available in the study area”.
Comment 4: Section 4.2.3 : A comment on the limitations of the analysis related to the constraint of using a bilinear threshold form may be added (see e.g., https://www.mdpi.com/2073-4441/13/13/1752/htm, where other forms are suggested).
Response: The limitations and constraints of using the bilinear format have been shortly presented in Section 4.2.4. However, additional comment on the constraint of using a bilinear threshold will be added in Section 4.2.4 referring to Conrad et al., (2021).
Minor comments/technical corrections
Comment 1: LL 364-365 This is unclear: I imagine that the critical level for landslide occurrence is sort of fixed and then it is reached more or less easily based on the prior rainfall and the time lag.
Response: It is true that the critical level for landslide occurrence is more or less fixed when other geological and geomorphological condition are kept constant and it is reached more or less easily depending on the prior rainfall expressed in terms of antecedent soil moisture and the time lag between the landslide triggering rainfall and the soil hydrological response. LL 364-365 will be paraphrased accordingly.
Comment 2: LL 500 The authors apply a threshold of 10 mm on satellite products to make them better agree with observations. This is a sort of a “bias correction”, about which a lot of literature exist. Perhaps make a fast literature review and add some references. (This could be useful also for future work)
Response: The threshold definition of a rainy day (10 mm) improved the similarities between the satellite-based and gauge-based landslide thresholds and thus considered as a bias correction between the two sources of rainfall data. Similarly, bias correction methods were adopted by other researchers to ensure for the high accuracy between ground- and satellite-based rainfall data (Bhatti et al., 2016; Vernimmen et al., 2012). This paragraph and references will be added to LL 500
Comment 3: Fig. 1 is perhaps a little bit messy (especially in B/W).
Response: Fig 1 will be improved to ensure for a better visibility and readability.
Comment 4: LL 512 the authors write “inter-event time” as the minimum dry interval between rainfall events. Perhaps add “minimum”, even if I understand that IET is aligned with previous literature in the field of landslides.
Response: The word “minimum” will be added and thus “minimum inter-event time”
Comment: L120 a “)” is missing after Mukungwa.
Response: a “)” will be added after Mukungwa
Comment: L586 thus “can be” (?) very useful (something is missing in the sentence).
Response: The word “can be “ will be added “thus can be very useful for landslide hazard….”
Citation: https://doi.org/10.5194/egusphere-2022-596-AC1
-
AC1: 'Reply on RC1', Judith Uwihirwe, 07 Sep 2022
-
RC2: 'Comment on egusphere-2022-596', Anonymous Referee #2, 22 Aug 2022
The work presents a comprehensive analysis of the possible combination of rainfall variables and soil moisture data (both measured and modelled) from ground and satellite for predicting rainfall-induced landslides. The outcomes of the analysis are based on ROC analysis, and skill scores, especially TPR and FPR.
General remarks
The title contains the term “landslide hazard”, which is misleading if referred to the manuscript contents. Hazard is generally intended to be an “off line” property of a territory, which is a function of susceptibility, temporal and magnitude (size) probability.
Regarding the discussion of threshold results, in my experience, false negatives (i.e., missed alarms) are more important than false positives (i.e., false alarms), because the consequences of missed alarms (e.g., deaths and injuries) are certainly more severe than those caused by false alarms (e.g., the unnecessary evacuation of a school). Therefore, looking at the number of missed alarms in Fig. 8b and 8e, I would not rely too heavily on these thresholds in a LEWS (Landslide Early Warning System). I suggest you to review the Discussion and Conclusions.
Minor revisions
- Figs. 1, 2, 3, and 4 are similar and repetitive, and the 5-km buffers locally obscure the information from high-granularity maps in Figs. 1 and 2. For a better readability, my suggestion is to merge Figures (perhaps Fig. 1 with Fig. 3, and Fig. 2 with Fig.4) selecting two maps (perhaps elevation and geomorphology), and grouping sensor and landslide information, in Fig.2-4. By the way, I think that the year of occurrence in not that important for this analysis given the relatively low number of failures. Other information on the landslide sites (mean terrain slope in ROIs, aquifer type) could be provided in a Table with the list of failures.
- I suggest using “cumulated event rainfall” for E, “event duration” for D, and “rainfall mean intensity” for I.
- A quantitative error estimation on your findings is lacking.
- Tables 4 and 5 are a bit confusing. If I understand correctly, the first group of 5 columns refers to the whole landslide area, while the second group only refers to the modeled catchments. If so, please amend the tables accordingly.
- In Section 4.2.3 on hydro-meteorological thresholds, I would suggest to calculate also classical ED thresholds, which could provide competitive skill scores.
- Figs. 8 and 9. You should improve the quality, especially that of 8d and 9d. Please, also avoid too small characters in the legenda, and instead use the caption to explain symbols and colors.
Additional minor revisions are in the attached PDF file.
-
AC2: 'Reply on RC2', Judith Uwihirwe, 07 Sep 2022
Dear Referee,
Thank you for your feedback on our manuscript. The raised comments and suggestions are of great value for the improvement of the manuscript. We have considered all as follow:
Comment: The title contains the term “landslide hazard”, which is misleading if referred to the manuscript contents. Hazard is generally intended to be an “off line” property of a territory, which is a function of susceptibility, temporal and magnitude (size) probability.
Response: The word “Hazard assessment “ will be removed from the title to fit with the manuscript content as suggested. The Title will be “Potential of satellite-derived hydro-meteorological information for landslide initiation thresholds in Rwanda”.
Comment: Regarding the discussion of threshold results, in my experience, false negatives (i.e., missed alarms) are more important than false positives (i.e., false alarms), because the consequences of missed alarms (e.g. deaths and injuries) are certainly more severe than those caused by false alarms (e.g., the unnecessary evacuation of a school). Therefore, looking at the number of missed alarms in Fig. 8b and 8e, I would not rely too heavily on these thresholds in a LEWS (Landslide Early Warning System). I suggest you to review the Discussion and Conclusions.
Response: We agree that the consequences of offering false alarms (FPR) are less harmful on the short-term than missed alarms (FNR) which implies that the best threshold should maximize the rate of true positives TPR (true alarms) while minimizing the FNR. However, the thresholds in Fig. 8b and 8e are classical thresholds ED relying exclusively on rainfall (Trigger), leading to the high rate of missed alarms and thus less important for a robust LEWS development. Similar to this study, previous studies (Bogaard and Greco, 2018; Peres et al., 2018, Marino et al., 2020; Thomas et al., 2020; Zhuo et al., 2019, Mirus et al., 2018a; Thomas et al., 2019; Uwihirwe et al., 2020, 2021) indicated that the consideration of the prior subsurface hydrological conditions reduce the number of missed alarms FNR as well as the number of false alarms FPR relative to the exclusive use of rainfall-only thresholds. In Figure 8a, 8c and 8d, 9a, 9b and 9c, we integrated the hydrological information (i.e. antecedent soil moisture) in landslide thresholds to improve the rate of TPR and reduce the rate of FNR and FPR. The main goal of hydro-meteorological thresholds (Cause-trigger) is to maximize the rate of true positives TPR (true alarms) i.e. minimize the FNR but at the same time reducing the rate of false positives FPR (False alarms). The used statistical metrics (TSS and RAD) are also in line with this concept aiming at maximizing the rate of true positives TPR while minimizing the rate of false positives FPR. Once TPR is maximized, the FNR is also minimized though difficult and or impossible to have a perfect threshold model with zero FNR and FPR. We will add a discussion point about this information in Section 4.2.4 and in conclusion part. We will also correct the number of false negative (FNR) in Figure 8f and 9d.
Minor revisions
Comment: Figs. 1, 2, 3, and 4 are similar and repetitive, and the 5-km buffers locally obscure the information from high-granularity maps in Figs. 1 and 2. For a better readability, my suggestion is to merge Figures (perhaps Fig. 1 with Fig. 3, and Fig. 2 with Fig.4) selecting two maps (perhaps elevation and geomorphology), and grouping sensor and landslide information, in Fig.2-4. By the way, I think that the year of occurrence in not that important for this analysis given the relatively low number of failures. Other information on the landslide sites (mean terrain slope in ROIs, aquifer type) could be provided in a Table with the list of failures.
Response: We will merge Fig.1&2 containing quite similar information and keep Fig. 3&4 showing different extent of our study area (ROIs) to keep the flow of the methodology. The mean terrain slope (Map) will only be kept in text and be removed in Fig 2.
Comment: I suggest using “cumulated event rainfall” for E, “event duration” for D, and “rainfall mean intensity” for I.
Response: We will replace “ rainfall event volume E” by “cumulated event rainfall E” ; and “event intensity E/D” by “rainfall mean intensity E/D” as Suggested. The “event duration” for D is same as suggested and will be kept.
Comment: Tables 4 and 5 are a bit confusing. If I understand correctly, the first group of 5 columns refers to the whole landslide area, while the second group only refers to the modeled catchments. If so, please amend the tables accordingly.
Response: We will amend the Tables’ captions accordingly even though same explanation have been provided in the footnotes of Table 4 and Table 5.
Comment: In Section 4.2.3 on hydro-meteorological thresholds, I would suggest to calculate also classical ED thresholds, which could provide competitive skill scores.
Response: In Figure 8b, 8e and 8f the classical ED thresholds are presented despite the weak prediction capability (TPR=50%) and weak skill score.
Comment: Figs. 8 and 9. You should improve the quality, especially that of 8d and 9d. Please, also avoid too small characters in the legend a, and instead use the caption to explain symbols and colours.
Response: We will improve the quality of Figure 8 and 9 by increasing the font size of the legend and explaining some of symbols in the Figures’ captions.
Citation: https://doi.org/10.5194/egusphere-2022-596-AC2
-
AC2: 'Reply on RC2', Judith Uwihirwe, 07 Sep 2022
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 268 | 106 | 14 | 388 | 5 | 5 |
- HTML: 268
- PDF: 106
- XML: 14
- Total: 388
- BibTeX: 5
- EndNote: 5
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Cited
Alessia Riveros
Hellen Wanjala
Jaap Schellekens
Frederiek Sperna Weiland
Markus Hrachowitz
Thom A. Bogaard
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(3284 KB) - Metadata XML