Elsevier

Geomorphology

Volume 94, Issues 3–4, 15 February 2008, Pages 353-378
Geomorphology

Comparing models of debris-flow susceptibility in the alpine environment

https://doi.org/10.1016/j.geomorph.2006.10.033Get rights and content

Abstract

Debris-flows are widespread in Val di Fassa (Trento Province, Eastern Italian Alps) where they constitute one of the most dangerous gravity-induced surface processes. From a large set of environmental characteristics and a detailed inventory of debris flows, we developed five models to predict location of debris-flow source areas. The models differ in approach (statistical vs. physically-based) and type of terrain unit of reference (slope unit vs. grid cell). In the statistical models, a mix of several environmental factors classified areas with different debris-flow susceptibility; however, the factors that exert a strong discriminant power reduce to conditions of high slope-gradient, pasture or no vegetation cover, availability of detrital material, and active erosional processes. Since slope and land use are also used in the physically-based approach, all model results are largely controlled by the same leading variables.

Overlaying susceptibility maps produced by the different methods (statistical vs. physically-based) for the same terrain unit of reference (grid cell) reveals a large difference, nearly 25% spatial mismatch. The spatial discrepancy exceeds 30% for susceptibility maps generated by the same method (discriminant analysis) but different terrain units (slope unit vs. grid cell). The size of the terrain unit also led to different susceptibility maps (almost 20% spatial mismatch). Maps based on different statistical tools (discriminant analysis vs. logistic regression) differed least (less than 10%). Hence, method and terrain unit proved to be equally important in mapping susceptibility.

Model performance was evaluated from the percentages of terrain units that each model correctly classifies, the number of debris-flow falling within the area classified as unstable by each model, and through the metric of ROC curves. Although all techniques implemented yielded results essentially comparable; the discriminant model based on the partition of the study area into small slope units may constitute the most suitable approach to regional debris-flow assessment in the Alpine environment.

Introduction

In the Alpine mountain belt debris flows are widespread and constitute one of the most dangerous gravity-induced surface processes that cause severe damage to dwellings, roads and other lifelines. Evaluation of the susceptibility requires the identification and the quantitative assessment of the factors leading to the initiation, propagation and deposition of debris-flows. The task is not easy and is complicated by the fact that the term “debris-flow” is commonly used to indicate a wide spectrum of slope-instability phenomena that may significantly differ in mechanical properties of the material (fine particles vs. rock boulders, etc.), geomorphological setting of the process (open slope, channel, etc.), and hydrological conditions (high or low water content, etc.).

We confined the investigation to develop models to predict debris-flow source areas, neglecting the runout of the phenomena. Indeed, most of the predicting geo-environmental variables, which can be cost-effectively acquired at regional scale (rock type, land-use, structural setting, etc.), are poorly suited to model the propagation of debris-flows. The latter task requires the collection of site-specific data generally not available over large areas. Despite the few attempts to model debris flow runout along a 3D topography with a spatially distributed approach at the regional scale (Iverson et al., 1998, O'Brien et al., 1993, Chau and Lo, 2004), we believe that the uncertainties relative to the parameters controlling runout and the complexity of the processes involved still hamper reliable and physically sound modelling of the runout susceptibility.

Confining the analysis to the problems related to model the initiation of debris-flows, we point out that local micro-topography and mechanical and hydraulic soil properties greatly control the processes responsible for triggering debris flows. As a consequence, “physically-based” models are widely used to assess debris-flow susceptibility (Montgomery and Dietrich, 1994, Wu and Sidle, 1995, Burton and Bathurst, 1998, Crosta and Frattini, 2003). Indeed, these models are more frequently applied than empirical and statistical approaches (Irigaray et al., 1999, Dai and Lee, 2001, Baeza and Corominas, 2001, Lee et al., 2002) that are commonly and successfully used for evaluating the occurrence of deep-seated landslides (e.g., Carrara, 1983, Yin and Yan, 1988, Carrara et al., 1995, Chung et al., 1995).

Although physically-based models may be suitable for modelling the hydrological conditions leading to debris-flow initiation, they have several limitations when applied to predict the spatial distribution of debris flows, i.e. to map debris-flow susceptibility. With the exception of slope morphology, in fact, the physical variables that control the spatial distribution of landslides within such models (i.e., physical and mechanical parameters of the slope and the failed material) cannot be acquired over large areas at reasonable cost.

Conversely, data-driven, statistical models are intrinsically based on the analysis of relations between the spatial distribution of “mappable” instability factors (i.e., the environmental factors that are directly or indirectly correlated with slope instability) and the observed distribution of landslides. Under the assumption that the factors that caused slope failures in a specific region are similar to those that will generate landslides in the future, statistical techniques may provide information on the future spatial distribution of landslides.

In the domain of geomorphology, land evaluation (Dymond et al., 1995, Gallant and Wilson, 2000, Pike, 2000) and mineral exploration (Isaaks and Srivastava, 1989), investigators have highlighted the relevance of the terrain unit (or mapping unit or homogeneous domain) in analysing and modelling environmental or geological phenomena. With a few exceptions, the topic has not received adequate attention in landslide literature. This is quite surprising since, when a given unit is selected, all subsequent works will refer to and treat each terrain unit as a spatially homogeneous domain in terms of both instability characteristics and the degree of susceptibility. In landslide susceptibility assessment, geomorphological units (Hansen, 1984, Meijerink, 1988), grid-cells, unique condition units (resulting from the overlay of categorical map; Bonham-Carter et al., 1989), stream tubes (Gallant and Wilson, 2000), terrain facets (Dymond et al., 1995), and slope units (Carrara et al., 1991) all have been used with different degrees of frequency and success (Carrara et al., 1995).

The Val di Fassa (Trento Province, Eastern Italian Alps) is famous for its landscape of very high cliffs carved in thick dolostone or limestone sequences that frequently rise to over 1000 m. The steep slopes and channel networks of these spectacular mountains are, however, a major source of debris flows that seriously threaten the inhabitants of the valley, their dwellings, local roads, and the other lifelines.

As part of a long-term project aimed at assessing the susceptibility from both shallow fast-moving (debris flows, rockfalls, etc), and deep-seated (slides, rock and earth flows, etc.) landslides in Val di Fassa, we carried out a systematic inventory of debris flows through aerial photo interpretation and field work (Carrara et al., 2004). Owing to the wealth of data acquired, and taking advantage of previous experience on susceptibility assessment (Carrara, 1983, Carrara et al., 1995, Carrara et al., 2003a), we developed different predictive models of debris-flow source areas by applying statistical and physically-based approaches.

In this paper we illustrate and compare statistical and physically based models designed to assess debris-flow occurrence. We discuss the application of statistical models to debris flows using different terrain units, and critically evaluate the results in terms of data quality and potential application of the resulting susceptibility maps in land management and planning. Lastly, we examine and discuss the verification and evaluation of these models.

Section snippets

Study area

The study area, which is almost 300 km2 in size and includes the upper drainage basin of the Avisio river (Val di Fassa, Trento Province), lies in the Dolomites that are part of the Southern Calcareous Alpine belt (Fig. 1). Bedrock geology is made up by different rock types that witness the complex history of the region (Bosellini et al., 2003). A basal effusive igneous (rhyolite) complex, which crops out in the southern part of the area, underlies a sedimentary succession that ranges from

Debris-flow inventory and thematic maps

Debris flows are common in Val di Fassa, especially on the upper slopes, where glacial talus and colluvial deposits supply large amounts of detrital material for the initiation and propagation of the process (Fig. 1). Historically, debris flows have reached the alluvial fans along the main valley where villages are located, causing damage to human life and property. The last significant event dates from 1989, when three small channels were invaded by debris flows (Dona and Udai creek) or

Models of debris-flow susceptibility

To predict location of the debris-flow source areas in Val di Fassa, we designed and implemented five models that differ in method and type of terrain unit of reference (Table 1). Among the various indirect methods and techniques applied in the domain of landslide assessment (empirical, statistical, neuronal networks, deterministic, physically-based, etc.), we selected two multivariate procedures (discriminant and logistic regression) and a physically-based (SHALSTAB) approach.

Over the last

Model comparison

Having developed different models of debris-flow source areas, we faced the issue to compare them to both highlight their respective limitations/potentials, and establish which would be most suitable for planning soil conservation measures and setting forth land use regulations by local administrators. Comparison was accomplished through traditional, GIS-based, map overlay operations, and by calculating receiver operating characteristic (ROC) curves of each model.

For discriminant (fSU_DIS) and

Factors leading to debris-flow initiation

The modelling strategies for debris-flow susceptibility mapping implemented and tested here for the Val di Fassa area differ in terms of method (statistical vs. deterministic and type of terrain-unit of reference (slope unit vs. grid cell; Table 2). However, all share the same input dataset: the debris-flow inventory map and the thematic layers describing the geo-environmental characteristics of the study area. This uniformity makes it possible to compare the different methods and identify the

Conclusions

Developing and comparing five different models for predicting the occurrence of debris-flow source areas in Val di Fassa study area led to the following results:

  • Models based on multivariate statistical techniques and a physically-based approach proved to differ significantly, as expressed by spatial mismatch, close to 25%, between the maps displaying areas predicted by each model as stable and unstable.

  • A larger spatial discrepancy (over 30%) is observed between susceptibility maps generated by

Acknowledgements

We are grateful to Richard Pike, U.S Geological Survey-Menlo Park, for the critical review of the manuscript. We thank Paolo Campedel, Geological Survey of the Provincia Autonoma di Trento, and Marco Stefani, University of Ferrara, for having made available geological and land-use data for the Val di Fassa area. Elena Valbuzzi, Samuel Cucchiaro, and Sabrina Iannaccone are acknowledged for the contribution in the preparation of the landslide inventory map by aerial photographic interpretation.

References (82)

  • TerlienM.T.J.

    Hydrological landslide triggering in ash-covered slopes of Manizales (Colombia)

    Geomorphology

    (1997)
  • WarrickA.W. et al.

    Spatial variability of soil physical properties in the field

  • AgterbergF.P.

    Geomathematics – Mathematical Background and Geo-Science Applications

    (1974)
  • ArdizzoneF. et al.

    Impact of mapping errors on the reliability of landslide hazard models

    Natural Hazard and Earth Systems Science

    (2002)
  • BaecherG.B. et al.

    Critical parameters for tailings embankments

    (1983)
  • BaezaC. et al.

    Assessment of shallow landslide susceptibility by means of multivariate statistical techniques

    Earth Surface Processes and Landforms

    (2001)
  • BarlingR.D. et al.

    A quasi-dynamic wetness index for characterizing the spatial distribution of zones of surface saturation and soil water content

    Water Resources Research

    (1994)
  • BaumR.L. et al.

    TRIGRS - A Fortran Program for Transient Rainfall Infiltration and Grid-Based Regional Slope-Stability Analysis. U.S. Geological Survey Open-File Report 02–424

    (2002)
  • BevenK.J. et al.

    A physically based, variable contributing area model of basin hydrology

    Hydrological Science Bulletin

    (1979)
  • Bonham-CarterG.F. et al.

    Weights of evidence modelling: a new approach to mapping mineral potential

  • BoselliniA. et al.

    Geology of the dolomites

    Episodes

    (2003)
  • BurtonA. et al.

    Physically based modelling of shallow landslide sediment yield at a catchment scale

    Environmental Geology

    (1998)
  • CarraraA.

    Multivariate models for landslide hazard evaluation

    Mathematical Geology

    (1983)
  • CarraraA. et al.

    GIS techniques and statistical models in evaluating landslide hazard

    Earth Surface Processes and Landforms

    (1991)
  • CarraraA. et al.

    Uncertainty in assessing landslide hazard and risk

    ITC Journal

    (1992)
  • CarraraA. et al.

    GIS technology in mapping landslide hazard

  • CarraraA. et al.

    Geomorphological and historical data in assessing landslide hazard

    Earth Surface Processes and Landforms

    (2003)
  • CarraraA. et al.

    A multivariate model for predicting debris-flow source areas in Valsassina (Central Alps)

  • Carrara, A., Campedel, P., Crosta, G.B., Stefani, M., Bozzolan, M., Frattini, P., 2004. Valutazione della pericolosità...
  • CartonA. et al.

    Geomorfologia dell’alpe di Lusia tra Moena e il Monte Vezzena

    Studi Trentini di Scienze Naturali, Acta Geologica

    (1983)
  • CasadeiM. et al.

    Testing a model for predicting the timing and location of shallow landslide initiation in soil-mantled landscapes

    Earth Surface Processes and Landforms

    (2003)
  • ChauK.T. et al.

    Hazard assessment of debris flows for Leung King Estate of Hong Kong by incorporating GIS with numerical simulations

    Natural Hazards and Earth System Sciences

    (2004)
  • ChungC.J.F. et al.

    Validation of spatial prediction models for landslide hazard mapping

    Natural Hazards

    (2003)
  • ChungC.J.F. et al.

    Multivariate regression analysis for landslide hazard zonation

  • CrostaG.B. et al.

    Distributed modelling of shallow landslides triggered by intense rainfall

    Natural Hazards and Environmental Systems Science

    (2003)
  • DaiF.C. et al.

    Terrain-based mapping of landslide susceptibility using a geographical information system: a case study

    Canadian Geotechnical Journal

    (2001)
  • DhakalA.S. et al.

    Distributed simulations of landslides for different rainfall conditions

    Hydrological Processes

    (2004)
  • DhakalA.S. et al.

    Landslide hazard mapping and its evaluation using GIS: an investigation of sampling schemes for a grid-cell based quantitative method

    Photogrammetric Engineering and Remote Sensing

    (2000)
  • DietrichW.E. et al.

    Geoscience and geotechnical engineering aspects of debris-flow hazard assessment

  • DietrichW.E. et al.

    A validation study of the shallow slope stability model, SHALSTAB, in forested lands of Northern California

    (1998)
  • DymondJ.R. et al.

    Automated mapping of land components from digital elevation data

    Earth Surface Processes and Landforms

    (1995)
  • Cited by (220)

    View all citing articles on Scopus
    View full text