 |
| Titel |
Landslide Hazard Assessment and Mapping in the Guil Catchment (Queyras, Southern French Alps): From Landslide Inventory to Susceptibility Modelling |
| VerfasserIn |
Louise Roulleau, François Bétard, Benoit Carlier, Candide Lissak, Monique Fort |
| Konferenz |
EGU General Assembly 2016
|
| Medientyp |
Artikel
|
| Sprache |
en
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 18 (2016) |
| Datensatznummer |
250128337
|
| Publikation (Nr.) |
 EGU/EGU2016-8320.pdf |
|
|
|
|
|
| Zusammenfassung |
| Landslides are common natural hazards in the Southern French Alps, where they may affect
human lives and cause severe damages to infrastructures. As a part of the SAMCO
research project dedicated to risk evaluation in mountain areas, this study focuses on
the Guil river catchment (317 km2), Queyras, to assess landslide hazard poorly
studied until now. In that area, landslides are mainly occasional, low amplitude
phenomena, with limited direct impacts when compared to other hazards such as floods or
snow avalanches. However, when interacting with floods during extreme rainfall
events, landslides may have indirect consequences of greater importance because of
strong hillslope-channel connectivity along the Guil River and its tributaries (i.e.
positive feedbacks). This specific morphodynamic functioning reinforces the need to
have a better understanding of landslide hazards and their spatial distribution at
the catchment scale to prevent local population from disasters with multi-hazard
origin.
The aim of this study is to produce a landslide susceptibility mapping at 1:50 000 scale as
a first step towards global estimation of landslide hazard and risk. The three main
methodologies used for assessing landslide susceptibility are qualitative (i.e. expert opinion),
deterministic (i.e. physics-based models) and statistical methods (i.e. probabilistic models).
Due to the rapid development of geographical information systems (GIS) during the last two
decades, statistical methods are today widely used because they offer a greater objectivity and
reproducibility at large scales. Among them, multivariate analyses are considered as the
most robust techniques, especially the logistic regression method commonly used
in landslide susceptibility mapping. However, this method like others is strongly
dependent on the accuracy of the input data to avoid significant errors in the final results.
In particular, a complete and accurate landslide inventory is required before the
modelling.
The methodology used in our study includes five main steps: (i) a landslide inventory was
compiled through extraction of landslide occurrences in existing national databases (BDMvt,
RTM), photointerpretation of aerial photographs and extensive field surveys; (ii) the main
predisposing factors were identified and implemented as digital layers into a GIS together
with the landslide inventory map, thus constituting the predictive variables to introduce into
the model; (iii) a logistic regression model was applied to analyze the spatial and
mathematical relationships between the response variable (i.e. absence/presence of
landslides) and the set of predictive variables (i.e. predisposing factors), after a selection
procedure based on statistical tests (χ2-test and Cramer’s V coefficient); (iv) an evaluation of
the model performance and quality results was conducted using a validation strategy based on
ROC curve and AUC analyses; (v) a final susceptibility map in four classes was proposed
using a discretization method based on success/prediction rate curves. The results of the
susceptibility modelling were finally interpreted and discussed in the light of what was
previously known about landslide occurrence and triggering in the study area. The
major influence of the distance-to-streams variable on the model confirms the strong
hillslope-channel coupling observed empirically during rainfall-induced landslide events. |
|
|
|
|
|
|