Topographic controls on landslide mobility: Modeling hurricane-induced landslide runout and debris-flow inundation in Puerto Rico
Abstract. In 2017, Hurricane Maria triggered more than 70,000 landslides in Puerto Rico. After initiation, these predominantly shallow landslides mobilized to varying degrees – some landslides only traveled partway downslope, whereas others reached drainages and mobilized into long-traveled debris flows that could severely impact roads and infrastructure. Thus, forecasting potential landslide runout and inundation zones is critical for estimating landslide and debris flow hazards. Here we conduct an in-depth topographic analysis of landslide-affected areas from nine study areas and apply a linked modeling technique to estimate locations susceptible to varying degrees of landslide runout in Lares, Utuado and Naranjito municipalities.
We find that longer runout length is observed on high-relief escarpments, although highly mobile long-runout debris flows also occurred in lower-relief dissected uplands. These topographic differences indicate that landslides initiating under similar conditions and possessing equal potential to mobilize as debris flows may not travel the same distances or affect the same areal extent. Our modeling approach allows the local topography to automatically control the implementation of two runout methods: 1) H/L runout zones are assigned directly downslope of landslide source zones, and 2) debris-flow inundation zones are estimated in the presence of a channel network. Debris-flow volumes are calculated as a function of area-integrated growth factors, estimated as a function of the upstream areas susceptible to shallow landslides. Applying our empirical modeling scheme over an area of 560 km2 , our results highlight the efficacy of our methods for assessment of the potential for landslide runout and debris-flow inundation over diverse terrains with varied susceptibility.
- Preprint
(3976 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 29 Nov 2024)
-
RC1: 'Reviewer comment on nhess-2024-141', Martin Mergili, 28 Oct 2024
reply
The authors analyze landslide mobility related to the 2017 hurricane Maria for three municipalities in Puerto Rico. Based on published landslide inventories and a published slope stability and landslide susceptibility analysis, terrain and geological information, they analyze landslide runout based on empirical models, distinguishing between rather short-runout landslides on slopes (H/L model) and longer runout debris flows in channels. A growth function is applied to account for erosion and landslide input to debris flows, based on topographic and empirical parameters. The results of the debris flow runout model are evaluated against mapped runout zones using an ROC analysis.
The topic investigated is highly relevant from both scientific and practical perspectives. It is certainly suitable for NHESS. The manuscript is very well written, structured, and illustrated. I would certainly like to see this work published in NHESS. I just have two specific comments and suggestions for improvement (see below), so that I recommend some minor revisions. All in all, I congratulate the authors for their excellent work.
Specific comments:
- The work flow of the study is, in principle, clearly presented and nicely illustrated through Fig. 5. There is just one point which is not clear to me: are the datasets used to develop the models and those used to evaluate their predictive success completely independent, or do they overlap? For H/L, it is written in L339 that, besides global datasets, also datasets from Hurricane Maria are used. Are those the same as used for evaluation (L373)? Are there similar issues for the debris flows? If yes, I suggest to address and justify such overlap in one or two sentences.
- The ROC analysis is, again, very well illustrated (Fig. 15) and explained. I am rather familiar with another type of ROC plot and was looking for values of the area under the curve (AUROC), which is very commonly used to evaluate the success of landslide simulations, but I found none. Would it be possible to use AUROC values, or is this just not useful for the type of ROC analysis performed here?
Citation: https://doi.org/10.5194/nhess-2024-141-RC1 -
AC1: 'Reply on RC1', Dianne Brien, 08 Nov 2024
reply
Thank you for the thoughtful comments and suggestions. Here is a reply:
Input datasets for selection of model parameters
For debris-flow inundation modeling, Maria's most mobile landslides (MMM) provided statistics to constrain the range of values for stream order, planform curvature, and Psrc (Table 7). Scenarios defining the location of debris flow growth were built using the extracted statistics in combination with stream slopes measured in the field. Debris-flow growth factors and maximum debris-flow volumes were not available for all MMM. The growth factors and volumes were obtained from Coe et al., 2021. The MMM dataset was used to test the predictive success of the debris flow inundation modeling
For the landslide runout modeling, we explored a variety of published references with H/L values: 1) Hurricane Maria landslides (Bessette-Kirton, 2020), 2) worldwide datasets (Corominas, 1996), and 3) data from flume experiments (Iverson, 1997). Although one might expect that the ideal dataset is Hurricane Maria landslides, we found that median values presented in Bessette-Kirton, 2020 (27 to 34 degrees) would have resulted in an underestimation of runout and a gap between non-channelized runout zones and the channel. Maximum values from Hurricane Maria were from channelized flows – these values would be extremely low (14 degrees). It was apparent that H/L values extracted from regional datasets, such as the Hurricane Maria inventories, are influenced by the local slope and may not be a good metric for landslide mobility when used in this context. H/L is valuable for comparison of mobility when the local slope and hillslope lengths are equivalent. With much of the topography in Puerto Rico consisting of mountainous areas with greater than 40 degree slopes, landslides occurring on these steep slopes that do not make it to the bottom of the hill are common in the inventories and result in high values of α.The inventories included only a small number of landslides in non-channelized topography, hence we resorted to an approach that assesses how much the predicted area of runout is affected by changes in α. We ultimately conclude that 1) H/L values extracted from landslide inventories are strongly influenced by the local slope, and 2) for much of the terrane in Puerto Rico, there is very little difference in the area affected regardless of selected H/L value (Fig. 13). Using the landslide source areas published in the landslide inventories, figure 13 illustrates the minimal change in area affected (added yellow area) with a decrease of α from 25 to 20 degrees.
We attempt to explain the biases in H/L values extracted from regional landslide inventories from widespread landslide-inducing events in the results (6.1) and discussion (7.3.1). Ultimately, we did not quantitively assess predictive success given that in areas with predominantly steep slopes the statistics would have looked good, but this is really a bias in the distribution of topographic slopes rather than an indication of predictive success.
We will work to provide additional clarification in the revisions.
ROC analysis
AUROC values are typically used for comparison of different methods. In this case, one method is applied with different parameters rather than comparing different methods. The consideration of TPR allows direct comparison with the metric used for the source area modeling (Baum et al., 2024). For the intended purpose it is not as useful to extract AUROC values, since we need to identify a point in ROC space where TPR is less than 1.0 (TPR=1 is unobtainable without unrealistic over-prediction). With the goal of selecting parameters for two scenarios of regional susceptibility maps, PLR allows us to determine the balance between TPR and FPR for one method with different input parameters.
Citation: https://doi.org/10.5194/nhess-2024-141-AC1
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 160 | 52 | 61 | 273 | 6 | 6 |
- HTML: 160
- PDF: 52
- XML: 61
- Total: 273
- BibTeX: 6
- EndNote: 6
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1