# Preparing first-time slope failures hazard maps: from pixel-based to slope unit-based

## Abstract

In this work, we present a novel quantitative geographical information system-based procedure to obtain the magnitude (area) and frequency of medium to large first-time shallow slope failures. The procedure has been set up at the Barcedana Valley, in the Tremp Basin (Eastern Pyrenees). First, pixel-based susceptibility classes were defined using a slope stability index obtained with the physically based model SINMAP. The frequency calculated from the number of first-time failures recorded during the last 60 years was then assigned to each susceptibility class. We devised a procedure to estimate the size of potential failures by means of the aggregation of pixels within the boundaries of morphological slope units, optimized for the purpose. Finally, the landslide hazard was prepared using the magnitude-frequency matrix. Results show that a proper pixel clustering has been carried which avoids the generation of small groups of pixels with different susceptibility degrees within the same slope unit. For a given hill slope, the area of the cluster of pixels depends on the size of the slope unit, which is not unique as it depends on the criterion used to delineate them. Therefore, the latter is a key factor in the final results. In this study, we validated our results with the size distribution of the observed landslides. The methodology presented in this work can be applied using any susceptibility assessment method with a pixel-based output.

## Keywords

Landslide susceptibility mapping Landslide hazard mapping Slope units SINMAP Physically-based## Introduction

Magnitude and frequency are the two key factors for a quantitative assessment of landslide hazard (Dai et al. 2002; Fell et al. 2005; Picarelli et al. 2005; Van Den Eeckhaut et al. 2007; Corominas et al. 2014; Lari et al. 2014; Vranken et al. 2014). Thus, knowledge of the probability of occurrence of landslides of a given magnitude is mandatory to obtain landslide hazard maps. In this work, the landslide magnitude is expressed as the area (number of pixels) affected by the slope failure.

The concept of intact slopes is not straightforward as strictly speaking, they do not exist in nature: evidences of slope processes acting in the past are not present after a few years due to the erosive cycle. From a geotechnical point of view, a slope can be considered as intact if it mobilizes peak strength conditions. Geomorphologically speaking, intact slopes display undisturbed strata and show consistent dip angles in contrast with existing landslide bodies. The latter can be characterised by undulated or hummocky ground surface, presence of scarps, open cracks, tilted trees and disturbed soils and rocks. The stability of intact slopes can be assessed by means of physically based, statistical, analytical or heuristic approaches (Aleotti and Chowdhury 1999; Dai and Lee 2002; Das et al. 2011; Ghosh et al. 2012). Some physically based models calculate the safety factor (SF) that can be integrated in the landslide hazard assessment (Wu and Abdel-Latif 2000; Frattini et al. 2004; Haneberg 2004 Baum et al. 2005). Advanced approaches couple the hydrological model and the infinite slope stability model and both can be implemented with various degrees of sophistication, including either steady-state or transient conditions (Montgomery and Dietrich 1994; Savage et al. 2004; Baum et al. 2008; Godt et al. 2008; Salciarini et al. 2008; Rossi et al. 2013; Alvioli et al. 2018). The first physically based models were two-dimensional (2D), but nowadays, three-dimensional (3D) slope stability models exist (Marchesini et al. 2009; Jia et al. 2012; Mergili et al. 2014; Raia et al. 2014; Alvioli and Baum 2016). Many physically based models are implemented within a GIS, or are supported by GIS processing, and calculate the SF for single pixels. Meaningful combinations of the spatially distributed SFs allow to define 3D geometries of instable portions of the terrain (e.g. Marchesini et al. 2009) or evaluate the SF for the most critical slip surfaces (e.g. Jia et al. 2012; Mergili et al. 2014). More sophisticated models incorporate a probabilistic package that calculates the probability of failure (probability of SF being smaller than unity) by evaluating the different slip surfaces as well as the uncertainty in the input values (e.g. Pack et al., 1998; Rossi et al., 2013). Another example, for the case of shallow landslides, is the integration of the TRIGRS model (Baum et al., 2008; Alvioli and Baum, 2016) with SCOOPS3D (Reid et al. 2015), where the latter uses the time-dependent infiltration model results as calculated by the former and performs an additional 3D analysis (Tran et al. 2018), providing an overall 3D, time-dependent analysis. Physically based models work at pixel level by design, and the interpretation of the results at slope level needs to be addressed in a meaningful way (e.g. Alvioli et al. 2014; Bellugi et al. 2015). Physically based models also require higher resolution data, with respect to statistical ones, typically geotechnical parameters of the soil. Such data is often difficult to obtain on large areas, and for this reason, probabilistic approaches have been introduced, for example using TRIGRS in Raia et al. (2014) and Salciarini et al. (2017). Another limitation posed by the physically based models accounting for water infiltration is the dependence on the initial water table depth (Baum et al. 2005), which is difficult to estimate and needs to be measured on the field. An evaluation of the different options for data gathering for physically based models, specifically for TRIGRS, was recently described in Yatheendradas et al. (2019).

The SF output of physically based models can be transformed into a stability index (SI), for example by means of the Monte Carlo method, providing multiple SF analysed statistically. The SI, which can be used as a proxy for susceptibility (spatial probability of landslide occurrence) due to its statistic component, can be empirically converted into annual probability by correlating it with the temporal frequency of observed failures: the larger the temporal window of the inventory, the better the conversion will be. However, the SI resulting from 2D physical models is pixel-based, and if one does not resort it to explicit 3D methods, it is difficult to go beyond the pixel-based scale. In other words, a main restriction of the pixel-based analysis is the lack of correspondence between the landslide size and the pixel size. In many landslide susceptibility maps carried out in clayey formations, the size of the landslide often does not match to the size of the pixel (landslides are either bigger or smaller than the pixel size) (e.g. Baum et al. 2005; Salciarini et al. 2008; Pradhan et al. 2018) and this may not reflect the actual occurrence of the slope failures.

- 1.
Assessing the susceptibility of intact slopes.

- 2.
Assessing the magnitude of failures bigger than the pixel size.

A procedure to overcome this restriction is by defining the physical boundaries of the hill slope where the potential failure can occur (e.g. Guzzetti et al. 1999, 2006). In this case, any analysis carried out affects the whole slope unit in terms of uniform spatial probability of landslide occurrence within the same slope unit, and therefore, this approach is difficult to carry out in large slopes showing variable steepness as the failure may only affects a small portion. In this work, we have assumed that failures cannot be bigger than the physical boundaries of the hill slope. We have devised a procedure to obtain the area of failures bigger than the pixel size and constrained by physiographic boundaries. The latter are commonly referred to as terrain units (TUs).

### Terrain units

TUs are portions of terrain with similar geological and geomorphological features which divide a region in portions that have a set of common properties, different from the adjacent ones, across definable boundaries (e.g. Hansen 1984; Lee and Evangelista 2006; Pavel et al. 2008). In other words, they maximise the internal homogeneity and the between-unit heterogeneity (Jia et al. 2012; Alvioli et al. 2016). TUs embody the geomorphologist assumption that both geological and geomorphological features properly describe the boundaries resulting from the relationship between materials, landforms and processes. Since landslides occur in sites with a certain slope gradient, the failure is constrained between the highest and the lowest part, i.e. between the divide and the drainage line of a given slope, respectively.

Slope units (SUs) are a particular type of TUs causally related to the hydro- and geomorphological conditions and processes that shape natural landscapes, including landslides. Each SU is encompassed by a drainage (or valley) and a divide line (Carrara 1988; Carrara et al. 1991, 1995, 2003; Giles 1998b; Guzzetti et al. 1999; Xie et al. 2004; Komac 2006; Guzzetti et al. 2006; Copons and Vilaplana 2008; Reichenbach et al. 2014). SUs are a formalization of what geomorphologists describe as slopes, and failures take place within their boundaries (Guzzetti et al. 2000).

SUs are known to be the most appropriate terrain partitioning for hydrological and geomorphological studies, including landslide susceptibility modelling and zonation (Carrara et al. 1991, 1995; Guzzetti et al. 1999, 2006) and landslide hazard assessment (Xie et al. 2003; Giles 1998a, b; Van Den Eeckhaut et al. 2009; Jia et al. 2012). In the literature, the use of grid cells is often preferred to other types of mapping units, including slope units. This is largely due to the simplicity of using regular grid cells, in conjunction with a gridded digital elevation model, with respect to irregular polygons. Most of the existing GIS software can perform complex operations on a pixel basis, while calculating quantities within slope units requires some additional effort. Moreover, the very delineation of slope units is not straightforward, especially on large areas. Nevertheless, using a meaningful slope unit delineation has several advantages for a number of applications. Signatures of geomorphological processes are uniform within slope units, and maps produced on the basis of slope units are inherently more suitable for land use planning and hazard management, since their boundaries are recognizable in the field. Moreover, for the specific purpose of this study, pixels are not suitable to define the magnitude of potential landslides, simply because the stability model does not provide a means to go beyond the pixel-based description. Hence, SUs are a valuable tool to assess the magnitude of failures larger than the pixel size.

### Beyond pixel-based susceptibility using slope units

A rough way to account for failures larger than a pixel is by assigning a unique susceptibility value to the whole SU. This can be done, for instance, by averaging the susceptibility values within the SU polygons (Fig. 1b) and assuming that the size of the SU defines the size of the potential failure. In the example of Fig. 1b, however, two main inconsistencies are observed: the average susceptibility obtained is low (yellow) and hides the fact that part of the slope is susceptible to fail. Moreover, if the spatial extent of the potential failure is smaller than the SU (orange in Fig. 1a), the area of the latter is not a proper proxy of the possible slope failure magnitude.

In order to overcome the disadvantages of averaging, we intend to work directly with the area of the pixel clusters defined within the same SU, which will be used as a proxy for the magnitude of the potential failure. The size of the clusters of pixels with the highest-susceptibility values are expected to collectively furnish a proxy of the estimation of the magnitude of a potential landslide in the SU: we expect that failures of different magnitude will take place within the clusters according to their probability of occurrence. To this end, a process of pixel clustering has to be developed.

### Landslide hazard mapping

Landslide hazard can be expressed as the frequency of landslides of given magnitude within a certain area and a given time period; it is evaluated through different methods depending on the scale of analysis (Fell et al. 2008). According to the suggestions given by Fell et al. (2008), landslide hazard zoning assigns an estimated frequency to the potential landslides resulting from susceptibility mapping. Frequency is combined with the magnitude of potential instabilities to map landslide hazard by means of magnitude-frequency (MF) relations expressed as an MF matrix. Usually, these matrices are prepared for each landslide mechanism (Strozzi et al. 2013) and the hazard degree results from the combination of both input parameters. For instance, frequent small-size landslides are not ranked as highly hazardous because their intensity and the subsequent damage are low and vice-versa. This topic has been widely discussed in countries such as Switzerland, Italy as well as Australia, among others, where custom methodologies have been developed (AGS 2007; Kranitz and Bensi 2009; Raetzo and Loup 2009) and implemented elsewhere (e.g. Cardinali et al. 2002, 2006; Raetzo et al. 2002; Sterlacchini et al. 2007; Posch-Trözmüller 2010; Strozzi et al. 2013).

Landslide hazard can be expressed in qualitative or quantitative terms. Usually, qualitative assessment includes subjective criteria of experts and, therefore, a certain subjectivity in the interpretation of the threat posed by different landslides. MF matrices are usually presented in qualitative terms as low, medium, high and very high. However, if frequency and magnitude can be quantified, the different degrees of hazard can be quantified as well.

The objective of this work is to evaluate the potential for large slope failures (translational slides) of the intact slopes by applying a deterministic model in the Barcedana Valley (Eastern Pyrenees, Spain) and using the area of the most susceptible pixel clusters as a proxy of the magnitude. The details of the methodology and the study area are given in the following sections.

## Study area and data

### General setting

The study area is located in the Barcedana Valley (or Basin), in the southern part of the Tremp-Graus Basin, which is an erosive depression within the South Central Structural Unit (SCU) (Seguret 1972). The SCU includes, from north to south the Boixols, Montsec, and Sierras Marginales thrusts. The Tremp-Graus basin is located north from the Montsec thrust sheet and forms a wide syncline. Materials filling the Tremp-Graus basin are mostly claystones and sandstones from the Upper Cretaceous/Palaeocene (Garumnian facies) and limestones and conglomerates from Eocene-Oligocene ages.

^{2}and an altitude ranging from 400 to 650 m asl. It bounds the Montsec Range to the south and the Llimiana Range to the north. Surrounding summits have a maximum elevation of 1700 m asl (Fig. 2a).

Typical Mediterranean climatic conditions prevail in the area with intense storms at the end of spring and in the beginning of autumn and dry periods in January–February and early summer. The mean annual precipitation is about 600–700 mm mostly in the form of high-intensity storms (Novoa 1984), some of them reaching up to 100 mm in 10 h (approx.), and the mean annual temperature is 12.5 °C with a mean maximum temperature of 27.6 °C and a mean minimum temperature of 0.4 °C.

### Geological description

The Llimiana Range is formed by well-cemented materials from Oligocene: sandstones, conglomerates, reddish and grey lutites, limestones with alveolines and marls and lutites. To the south, the Montsec Range exhibits also well-cemented materials from Garumnian Facies: Micritic limestones, calcarenites and lutites of Fm La Posa (Maastrichtian); Sandstones and calcarenites of Fm Gresos d’Areny (Maastrichtian) and limestones, calcarenites and sandstones of Campanian–Maastrichtian age (Fig. 2b). Between both ranges, within the valley, a big extension of colluvium can be found. It is mainly composed of clays with angular sandy and silty clasts of Holocene age. Garumnian deposits are also present: red lutites, paleosoils, sandstones and gypsum of Selandian–Thanetian age, as well as micritic limestones, calcarenites and lutites of Fm La Posa (Maastrichtian).

- 1.
Grey Garumnian: from the diagenetic point of view, it is completely different of the other typical Garumnian materials. It is made of grey lutite, lignite, sandstone and interbedded limestone layers. It is of Maastrichtian age.

- 2.
Lower Red Garumnian (red claystones): it is made of red silty materials with interbedded levels of lenticular fining upward sandstones. Between silty deposits, there exist thin layers of fine-grain sandstone with cross lamination. These silty deposits are often affected by joints and present calcareous nodules. The roof of the Lower Red Garumnian level has been defined as the limit between Mesozoic and Cenozoic.

- 3.
Vallcebre Limestones (intermediate limestone): in the Tremp Basin, they exhibit several micritic calcareous sections, frequently recrystallised, of Maastrichtian-Danian age.

- 4.
Upper Red Garumnian (red claystones): it is the most expansive level of Garumnian facies. It is of Thanetian age and may contain gypsum and anhydrite nodules. Well-developed evaporitic lentils can be found as well.

- 1-
PPlg, which in some parts is covered by Qco, corresponds to the Upper Red Garumnian

- 2-
KMlg corresponds to the Lower Red Garumnian

- 3-
KMcp corresponds to the Gray Garumnian.

Vallcebre limestones do not outcrop in Barcedana Valley as they are either covered by quaternary colluvium or they are not present in this part of the basin. Materials from the Red Garumnian Facies (PPlg and KMlg) are the main responsible for the several numbers of landslides occurred, especially, during rainy events.

^{2}but up to more than 20,000 m

^{2}), they are relatively shallow because the sliding surface develops at the contact between the disturbed (weathered) and the intact claystone. Most of the failures show depths ranging between 1 and 5 m. Locally, in case of slopes next to entrenched streams, the steep slopes may also generate rotational failures (Fig. 3c). Therefore, in terms of area, they can be considered medium to large landslides but not in terms of thickness. These landslides are mostly self-contained and short runout (e.g. Carrara et al. 2003; Guzzetti et al. 2005). The debris sometimes accumulate in hollows and drainage ways and may eventually turn into slow-moving earthflow. Contrary to the typically clay formations, slope failures in the Tremp basin may be triggered due to short-lasting very intense rainfall events. The reason is that the most superficial part of the claystones has been unloaded by the erosion and presents cracks and macropores (Fig. 3b) (Corominas 2001, 2006). Furthermore, volumetric strain and cracks usually appear within the first meters below the ground surface due to moisture changes, thus favouring rainfall infiltration.

### Data

Data used in this research. The aerial images are true-colour georeferenced orthophotos. Shadows or clouds masking the study area are not present

Document | Format | Year | Scale | Pixel resolution | Source |
---|---|---|---|---|---|

Digital elevation model | Digital | 2014 | – | 5 m/pixel | ICGC |

Orthophoto | |||||

1956 | Digital | 1956 | 1:33,000 | 50 cm | ICGC |

1990 | Digital | 1990 | 1:25,000 | 50 cm | ICGC |

1993 | Digital | 1993 | 1:25,000 | 50 cm | ICGC |

1997 | Digital | 1997 | 1:5000 | 50 cm | ICGC |

2003 | Digital | 2003 | 1:5000 | 50 cm | ICGC |

2005 | Digital | 2005 | 1:5000 | 50 cm | ICGC |

2007 | Digital | 2007 | 1:5000 | 50 cm | ICGC |

2008 | Digital | 2008 | 1:5000 | 50 cm | ICGC |

2009 | Digital | 2009 | 1:5000 | 50 cm | ICGC |

2011 | Digital | 2011 | 1:5000 | 50 cm | ICGC |

2013 | Digital | 2013 | 1:5000 | 50 cm | ICGC |

Geological map | Digital | 2007 | 1:50,000 | – | ICGC |

Geomechanical units map | Digital | 2009/2013 | 1:5000 | – | Georisc S.L. |

Landslide inventory map | Digital | 2009/2013 | 1:5000 | – | Georisc S.L. |

Simplified geomechanical units defined as the basis for the landslide hazard assessment

Unit code | 1 | 2 | 3 | 4 | 5 | 6 |
---|---|---|---|---|---|---|

Geomechanical unit | Hard rock | Interbedded weak and hard rocks | Weak rock | Coarse colluvium | Fine colluvium | Slide material |

Lithologies (simplified) | Limestone, conglomerate and massive sandstone | Sandstone, limestone, conglomerate interbedded with argillaceous rocks | Claystone, siltstone, marl and either thin or poorly cemented sandstone | Coarse colluvium with clayey and silty matrix | Fine colluvium with clayey and silty matrix | Fine colluvium with clayey and silty matrix |

^{2}, and they mainly occurred on fine colluvium (53%), followed by coarse colluvium (29%) and weak rock unit (18%). Such landslides have occurred in quite gentle slopes, ranging from 8° to 37° with a mean slope angle of 17° (Canales 2011).

Main characteristics of the first-time slope failures in the study area during 1956–2013

Landslide | Year | Area [m | Geomechanical unit | Landslide | Year | Area [m | Geomechanical unit |
---|---|---|---|---|---|---|---|

1 | 2005 | 816 | Weak rock | 10 | 2013 | 11,550 | Fine colluvium |

2 | 1990 | 1053 | Weak rock | 11 | 2003 | 15,110 | Coarse colluvium |

3 | 2013 | 1091 | Weak rock | 12 | 2003 | 15,120 | Fine colluvium |

4 | 2013 | 1229 | Fine colluvium | 13 | 1990 | 19,590 | Coarse colluvium |

5 | 2005 | 2961 | Fine colluvium | 14 | 2011 | 23,700 | Fine colluvium |

6 | 1990 | 7002 | Coarse colluvium | 15 | 1997 | 27,700 | Fine colluvium |

7 | 1990 | 7153 | Fine colluvium | 16 | 1997 | 28,450 | Coarse colluvium |

8 | 1993 | 8849 | Fine colluvium | 17 | 1990 | 52,010 | Coarse colluvium |

9 | 2009 | 11,540 | Fine colluvium |

## Methodology

Establishing a relation between the outcomes of physically based models, as the one adopted in this work, and our ultimate aim of preparing hazard maps for large first-time failures that exceed the pixel size of 5 m (DEM resolution) consists of several steps. The first step of our method is the susceptibility assessment through the calculation of the SI for each pixel using a slope stability model coupled with a hydrological model (SINMAP) (Pack et al. 1998). Then, to constrain and to calculate the size of the potential failures, larger than the pixel size, we delineated SUs using an existing software and re-classifying the SI map within each SU polygon, using a novel pixel clustering procedure. The ultimate goal of our approach is to consider the sizes of the high-susceptibility pixel clusters as a measure, or a proxy, of the magnitude of potential landslides in intact slopes. The temporal component of landslide hazard (i.e. frequency) was estimated using the probability of the first-time failures from the inventoried instabilities. Eventually, hazard was calculated using a MF matrix considering the mentioned magnitude and frequency.

### Susceptibility assessment using a physically based model

*w*for each pixel of a digital elevation model (DEM). It considers the Mohr–Coulomb criterion to calculate the SF as follows:

*C*

_{dl}is the adimensional cohesion,

*ρ*

_{w}[kg/m

^{3}] is the water density,

*ρ*

_{s}[kg/m

^{3}] is the soil density,

*φ*[°] is the friction angle, and

*θ*[°] is the slope angle. The adimensional cohesion is given by

*C*[N/m

^{2}] is the soil cohesion,

*g*= 9.81 [m/s

^{2}] is the acceleration of gravity, and

*h*is the height [m] of the soil column in the given pixel.

*w*starting from a few assumptions:

- 1.
Surface runoff follows topographic gradient. This assumption allows calculating the specific catchment area at each pixel as:

*a*is the specific catchment area [m],

*A*is the contributing area [m

^{2}], and

*b*is the unit contour length (pixel size) [m].

- 2.
Lateral discharge (

*q*) at each point is in equilibrium with a steady-state recharge,*R*[m/h]. - 3.
The capacity for lateral flux at each point is given by

*T*[m

^{2}/h] is the soil transmissivity defined as the product of the permeability,

*K*[m/h], and the soil thickness

*h*[m].

*q*[m

^{2}/h] reads as follows:

*w*can be calculated as follows:

_{min}> 1, the SI is the FS

_{min}(SI = FS

_{min}). For areas where FS

_{min}< 1, the SI = Prob (FS > 1). In case FS

_{max}< 1, then SI = Prob (FS > 1) = 0. The numerical values of input parameters used for applying SINMAP in this research (

*C*

_{dl},

*φ*and

*T*/

*R*) are listed in Table 4 and discussed in more detail in Domènech (2015). They have been obtained from different field campaigns carried out in the study area (Canales 2011; Montero 2011; Oliveras 2011; Domènech 2015) and from the literature (Coduto 1999; González de Vallejo et al. 2002). For each unit, the same

*h soil*value has been assigned to all the pixels where it outcrops. Pixels where units 1 and 2 outcrop correspond to hard rock and interbedded weak and hard rock (Table 2), and they require another specific analysis for rockfalls.

Parameters used in the SINMAP slope stability simulation for each geomechanical unit

Unit code | 1 | 2 | 3 | 4 | 5 | 6 |
---|---|---|---|---|---|---|

| 19.62 | 2.45 | 0 | 0 | 0 | 0 |

| 39.24 | 39.24 | 0.2 | 0 | 0.058 | 0 |

| 0.01 | 0.01 | 5 | 5 | 5 | 5 |

| 2600 | 2600 | 2500 | 2300 | 2200 | 2200 |

| 9.81 | 9.81 | 9.81 | 9.81 | 9.81 | 9.81 |

| 1 * 10 | 1 * 10 | 1 * 10 | 1 * 10 | 1 * 10 | 1 * 10 |

| 1 * 10 | 1 * 10 | 1 * 10 | 1 * 10 | 1 * 10 | 1 * 10 |

| 1 * 10 | 1 * 10 | 5 * 10 | 5 * 10 | 5 * 10 | 5 * 10 |

| 1 * 10 | 1 * 10 | 5 * 10 | 5 * 10 | 5 * 10 | 5 * 10 |

| 6 * 10 | 6 * 10 | 6 * 10 | 6 * 10 | 6 * 10 | 6 * 10 |

| 3 * 10 | 3 * 10 | 3 * 10 | 3 * 10 | 3 * 10 | 3 * 10 |

| 79,623 | 9615 | 0 | 0 | 0 | 0 |

| 153,846 | 153,846 | 1.5 | 0 | 0.5 | 0 |

| 40 | 30 | 30 | 18 | 18 | 9 |

| 50 | 50 | 35 | 30 | 25 | 13 |

| 2 * 10 | 2 * 10 | 0.0865 | 0.865 | 0.087 | 0.087 |

| 0.004 | 0.037 | 18.5 | 185.2 | 18.52 | 185.2 |

Finally, the SI is reclassified into four susceptibility classes of increasing probability of failure according to the observations made in the field. To this aim, the classification of the SI values has been modified from that proposed by Pack et al. (2005), using the following classes: 0 < SI < 0.25 (high susceptibility); 0.25 < SI < 1.0 (medium susceptibility); 1.0 < SI < 1.5 (low susceptibility); 1.5 < SI < 10 (very low susceptibility). Results of SINMAP have been validated by a receiving operating characteristics (ROC) analysis (Fawcett 2006).

### Parametric delineation of slope units

Although a unique SU size for a given hill slope does not exist, in mountainous areas, SUs can be easily identified, as compared to more gentle slopes areas. In the latter, the topography has not been sufficiently shaped by the erosive processes and, therefore, the drainage lines and the subsequent catchments to be later partitioned into SUs are more difficult to delineate and several possibilities exist. In the latter, it might happen that the morphodynamic processes trespass the SU limits. In such cases, it has to be decided whether the landslide extension prevails over the SU or vice versa.

Our starting point is that the observed first-time slope failures in the field should not exceed the size of the SU and conversely, they should not be too small in relation to the size of the SU. To this end, we have first carried out a preliminary delineation of SU to achieve a correspondence between the size of the SU generated and the size of the observed slope failures.

In this work, we used the software r.slopeunits of Alvioli et al. (2016) for SU delineation. They assumed that no unique SU delineation exists, since there is an inherent dependence on the scale and nature of the process under investigation. Non-univocity implies no definite SU size and different degrees of homogeneity of aspect, and other morphological properties, within each SU. The r.slopeunits software accepts a number of input parameters, which control the size and shape of the output SU map. The numerical values of the input parameters must be provided by the user based on his/her expertise or on some objective parameter selection method. The SUs boundaries delineation also depends on the DEM resolution (Mashimbye et al. 2014; Schlögel et al. 2018). The algorithm of Alvioli et al. (2016) is implemented in the GRASS GIS (Neteler and Mitasova 2008) module r.slopeunits freely available from http://geomorphology.irpi.cnr.it/tools/slope-units. In addition to the DEM, relevant inputs of the method implemented by the software are two parameters that drive the SU delineation process: (i) the circular variance, *c*, representing the degree of slope aspect homogeneity required within each SU; (ii) the minimum area, *a* [m^{2}], i.e. the threshold size under which a polygon is considered as an SU. Other input parameters are required for using the software; their role in the SU delineation is more technical and it is described in detail in Alvioli et al. (2016); thus, we focus here only on the determination of suitable *c* and *a* parameter values, which are directly related to the local terrain morphology.

The r.slopeunits software has been run for many combinations of the values of the input parameters *a* and *c*, namely *c* = (0.01, 0.05, 0.1, 0.15, 0.2, 0.4) for the circular variance threshold and *a* = (5000, 10,000, 25,000, 50,000, 100,000, 200,000, 500,000) m^{2}, which determined 42 sets of SU, obtained from all the possible (*a*, *c*) combinations. The choice of the ranges of variation of input parameters was not dictated by specific criteria, but only on heuristic judgement based on a brief preliminary analysis, and they are similar to the ones used in previous work (Alvioli et al. 2016; Bornaetxea et al. 2018; Schlögel et al. 2018). Each SU delineation is a cover of the study area, with polygons of different shapes and sizes, but all of them represent valid hydrological subdivisions of the terrain. For example, it is intuitive that the size of SU is larger for sets produced using larger values of the parameter *a*. Similarly, larger values of the parameter *c* require SU with large terrain variability that is likely found within larger polygons. A finer control of the output of the r.slopeunits software as a function of the whole set of input parameters is virtually impossible, since it depends from the complex interplay of individual parameter values and local terrain characteristics.

### Re-assessment of susceptibility values within the SU: magnitude of the potential failure

The SI map produced by SINMAP was treated as follows. The first step is to avoid the formation of pixel mosaics that are difficult to interpret. The isolated pixels located within a given SU are reclassified to the predominant class of the adjacent pixels. Furthermore, clusters of pixels involving an area smaller than 800 m^{2} have also been reclassified to the neighbouring predominant class (we neglected the cluster sizes smaller than a given value and replaced their values with the class with which the removed cluster shared the longest linear boundary). The 800-m^{2} threshold was defined heuristically and corresponds to the smaller landslide size observed in the study area. On the other hand, an upper limit to the size of potential failures has been imposed by constraining them within SU boundaries (a failure cannot be bigger than the slope in which it is contained). Therefore, clusters of pixels exceeding SU boundaries are not accepted.

*a*,

*c*) combination. For the purpose of this work, the cumulated frequency of the mapped first-time slope failures and the predicted ones (high-susceptibility cluster of pixels) are used to optimize SU parameters for better agreement with the distribution of the inventoried landslides. In such case, we are assuming that the landslide frequency size distribution will not change over time, which is not completely true, especially in areas affected by a strong disturbance such as large earthquakes (Fan et al. 2018). Therefore, it is advisable to update the model according to the local conditions. In this study, a custom metric has been defined to establish which SU delineation produces the best result. It is a positive calibration that calculates the difference between measured and modelled values and is defined as follows:

*D*

_{i}is the

*i*th point of the observed distribution and

*P*

_{i}(

*a*,

*c*) is the corresponding

*i*th point of the predicted distribution obtained with the specified values of (

*a*,

*c*), among the 42 different SU delineations.

It is worth noting that an optimal SU map with respect to aspect segmentation can also be selected by finding the maximum of the general-purpose objective function proposed by Espindola et al. (2006) and adapted for dealing with aspect segmentation in Alvioli et al. (2016).

Eventually, the magnitude of the potential failures was evaluated considering the size of the pixel clusters belonging to the highest susceptibility class and located within the same SU, which are the most likely to fail.

### Assessing the probability of the first-time failures

*N*

_{i}in each susceptibility class over the total number of first-time slope failures

*N*

_{T}. Then, the ratio has been divided by the observed period of time

*P*and normalised by the area occupied by each susceptibility class

*A*

_{i}. The frequency is expressed in terms of the number of first-time slope failures per year and per square kilometre as follows:

The hazard has been evaluated by combining the frequency and magnitude, obtained in the previous sections, for each cluster of pixels located within a given SU. For each combination, the degree of hazard has been established and expressed in qualitative terms, too. The number of classes and the definition of each one have been established according to the Swiss Federal Guidelines (Raetzo et al. 2002): very low, low, medium and high hazard. We assume that the failure affects only the cluster of pixels.

## Results and discussion

Validation of SINMAP by a receiving operating characteristics (ROC) analysis comparing the model results versus observed landslide presence. We converted the probabilistic result to 0/1 using a threshold (column Th.). We show true positives, false negatives, true negatives and false positives, along with hit rate and false alarm rate, as a function of the threshold. Plotting HR versus FAR one gets the ROC curve. A value of area under the curve (AUC) of 0.87 has been obtained by means of the R function (https://www.rdocumentation.org/packages/verification/versions/1.42/topics/roc.area)

Th. | TP | FN | TN | FP | HR | FAR |
---|---|---|---|---|---|---|

0.0 | 3918 | 5818 | 341,606 | 45,009 | 1.000 | 1.000 |

0.1 | 9154 | 582 | 282,811 | 103,804 | 0.991 | 0.629 |

0.2 | 9224 | 512 | 272,700 | 113,915 | 0.986 | 0.585 |

0.3 | 9294 | 442 | 259,869 | 126,746 | 0.978 | 0.526 |

0.4 | 9335 | 401 | 244,834 | 141,781 | 0.971 | 0.461 |

0.5 | 9389 | 347 | 227,757 | 158,858 | 0.964 | 0.411 |

0.6 | 9458 | 278 | 208,275 | 178,340 | 0.959 | 0.367 |

0.7 | 9525 | 211 | 183,373 | 203,242 | 0.955 | 0.328 |

0.8 | 9603 | 133 | 160,538 | 226,077 | 0.947 | 0.295 |

0.9 | 9644 | 92 | 143,546 | 243,069 | 0.940 | 0.268 |

1.0 | 9736 | 0 | 0 | 386,615 | 0.402 | 0.116 |

Validation of SINMAP by a receiving operating characteristics (ROC) analysis for high and medium susceptibility classes. Number of pixels classified as true positive (TP), false positive (FP), false negative (FN), true negative (TN) and the resulting true positive rate (TPr) and false positive rate (FPr) are shown

Susceptibility class | TP | FP | FN | TN | TPr | FPr |
---|---|---|---|---|---|---|

High | 9304 | 131,686 | 432 | 254,929 | 0.96 | 0.34 |

Medium | 9483 | 194,495 | 253 | 192,120 | 0.98 | 0.5 |

^{2}), the frequency of the remaining ones increases considerably. Another important factor is the size of the SUs. For clarity, a few SU sets covering the whole range of frequency size distributions have been selected (Fig. 7b). The SU sets denoted by “set 2”, “set 5”, “set 10”, “set 20”, “set 35” and “set 42” have been obtained respectively with the (

*a*,

*c*) combinations: (10,000 m

^{2}, 0.01), (100,000 m

^{2}, 0.01), (25,000 m

^{2}, 0.05), (200,000 m

^{2}, 0.1), (500,000 m

^{2}, 0.2) and (500,000 m

^{2}, 0.4). The slopes of the six distributions clearly indicate that the smaller the SU polygons (roughly proportional to the values of both the

*a*and

*c*parameters), the smaller the sizes of the high-susceptibility pixel clusters and vice versa. It is worth nothing that most of the curves nearly fit with the pure clustering distribution (grey line) for areas between 1000 and 10,000 m

^{2}, i.e. the smaller clusters of pixels given in the pure clustering process (grey line) are not affected by the SUs. From these results, we can already state that the SUs and their size (i.e. the type of relief of a given area) are a key factor that controls landslide size.

*F*(

*a*,

*c*) is shown in Fig. 9 with a yellow solid curve, denoted by “set 12”, and it has been obtained by (

*a*,

*c*) = (100,000 m

^{2}, 0.05).

^{2}over a maximum of 8.59) and the final frequency fluctuation is about 4%. Hence, the influence of different SUs on the time dependence can be neglected. The results of Table 7 are consistent with the susceptibility class of the clusters of pixels. The frequency of the first-time slope failures in the highest susceptible class (SC4) is one order of magnitude bigger than the frequency in the mid susceptibility class (SC3). Additionally, the null values of frequency obtained for the two lowest susceptibility classes (1 and 2) and the increasing for the susceptibility class 3 and 4 have confirmed the reliability of SINMAP and the reclassified SI values. The fact that the susceptibility classes 1 and 2 show the same frequency (0) is a matter of the small number of first-time slope failures present in our study area. However, the distinction between both classes is important as in the stability analysis they showed a different susceptibility.

Normalized frequency of first-time failures (no. of failures/year/km^{2}) per susceptibility class, for the selected SU. *SC* susceptibility class, *No*. number of first-time slope failures, *Failure ratio* number of first-time slope failures of each susceptibility class/overall, *T* observed period of time (1956 to 2013)

SC | No. | Failure ratio | | Area of intact slopes in 1956 (km | Frequency of first-time slope failures (no./year/km | ||||||
---|---|---|---|---|---|---|---|---|---|---|---|

SU no. 2 | SU no. 20 | SU no. 35 | SU no. 42 | SU no. 2 | SU no. 20 | SU no. 35 | SU no. 42 | ||||

1 | 0 | 0 | 57 | 8.56 | 8.55 | 8.58 | 8.59 | 0 | 0 | 0 | 0 |

2 | 0 | 0 | 57 | 2.21 | 2.15 | 2.24 | 2.25 | 0 | 0 | 0 | 0 |

3 | 1 | 0.059 | 57 | 3.26 | 3.25 | 3.26 | 3.26 | 0.005 | 0.005 | 0.005 | 0.005 |

4 | 16 | 0.941 | 57 | 5.60 | 5.68 | 5.56 | 5.54 | 0.05 | 0.049 | 0.051 | 0.051 |

^{2}). However, for events characterized by a medium frequency and high magnitude (index 33, Fig. 10), the large size and thickness of the movements, and its high recurrence, suggests that the damage produced to the buildings may be more severe.

^{2}) showing susceptibility values smaller than the surrounding pixels will not be considered as less susceptible, but as an artefact due to the pixel-based nature of the SF. Therefore, the susceptibility has been homogenised along the slope by means of the SU-dependent clustering procedure devised in this work, and the area of the adjacent clusters, located within the same SU, with the same SI has been calculated (Fig. 11c). Thereafter, the MF matrix has been applied (Fig. 11d) obtaining different degrees of hazard according to the size of the pixel clusters.

It is important to notice that within the same SU, pixel clusters of different susceptibility are possible. In terms of probability of failure, we expect a frequency of failures of each susceptibility class according to the results of Table 7. The hazard classes of the clusters of pixels (Fig. 11d) are defined based on the pairs of frequency of failures-size of the pixel cluster. Thus, despite some pixel clusters being classified as high susceptibility in Fig. 11c, they may become medium hazard if their size do not exceed the established threshold (Fig. 11d).

## Conclusions

A supervised procedure has been developed and applied in the Barcedana Valley (Eastern Pyrenees, Spain) to evaluate the frequency and magnitude of first-time slope failures. A deterministic model is used to obtain susceptibility classes, which is applied to the SUs to generate the magnitude-frequency matrix. In particular, the aim was to map predicted potential failures of intact slopes with size considerably larger than the pixel size, but smaller or equal than the slope unit in which they fall in. Compared to other hazard analysis approaches using the entire SU (e.g. Guzzetti et al. 2006), the procedure developed here is able to identify and define the area mostly likely to fail within the SU, i.e. most susceptible part of the slope only.

The susceptibility assessment has been performed using the physically based SINMAP model (Pack et al. 2005). The probability of failure has been reclassified into a classified pixel-based SI map.

The magnitude of the potential failures has been obtained performing an automatic aggregation of groups of adjacent pixels with the same susceptibility class, located within the same SU and having an area smaller than a given threshold. In that case, SUs were used as aggregation domains. To this end, we have developed a procedure to split the clusters of pixels of the same SI among different adjacent SU. Since a predefined SU delineation for a given study area does not exist, several predictions using different SU delineations, including SUs of different sizes and shape, were carried out using the parametric r.slopeunits software developed by Alvioli et al. (2016). We have checked whether the number of pixels of a given susceptibility class depends on the SU size. The magnitude of the potential failure is calculated as the size of the pixel cluster of the highest susceptibility class within the SU. The latter has been chosen by maximizing the agreement with the observed landslide data. The landslide size distribution modelled in this way is considered as our best estimate of the magnitude component of landslide hazard.

The first-time slope failures, mapped by analysing stereoscopically a set of orthophotos covering a 57-year period, have been used to calculate the frequency of each susceptibility class. The frequency was calculated by dividing the percentage of observed first-time failures, falling in each susceptibility class, by the time span covered by the orthophotos, and normalised by the area occupied by each susceptibility class. In this way, we have obtained an estimate of the temporal frequency component of landslide hazard. The procedure of landslide frequency obtained matches consistently with the landslide susceptibility classes generated (Table 7). We have checked that the influence of SU size on the determination of the frequency of the potential failures is negligible.

A magnitude-frequency matrix has been prepared considering that the frequency prevails over the magnitude for low and very low frequency classes. The landslide hazard maps have been obtained considering the optimal SU delineation and applying the MF matrix. Our procedure shows that not all the most susceptible pixel clusters are the most hazardous as the latter also depends on the size of the potential failure (Fig. 11): for small SUs, the final hazard degree is typically smaller, and vice versa. Hence, the SU size and the mapping units in general play an important role to obtain the final results (Guzzetti et al. 2006; Pavel et al. 2008; Catani et al. 2013; Erener and Düzgün 2013) and the best SU delineation should be chosen after checking their consistency with the observed slope failures.

The general framework can be applied using any physically based model with a pixel-based output, which must be properly classified. The presented methodology allows assessing the hazard in a semi-automatic and objective way, and considering the landslide size instead of individual pixel values or the whole slope unit, thus using both pixel and mapping unit information to the maximum extent.

## Notes

### Acknowledgments

The authors thank two anonymous reviewers and the editor for their helpful comments and improvement of the English text. M. Alvioli acknowledges the Division of Geotechnical Engineering and Geosciences, Department of Civil and Environmental Engineering, Universitat Politècnica de Catalunya-BarcelonaTech for warm hospitality, and the CINECA award Gmorph3 under the ISCRA initiative for the availability of high performance computing resources and support.

### Funding information

The FPU-UPC 2009 Research Fellowship by the Universitat Politècnica de Catalunya-BarcelonaTech provided funding to the first author, and the Spanish Economy and Competitiveness Ministry to the Rockrisk research project (BIA2013-42582-P) and the Interreg Project SUDOE IV B/DO-SMS, ref. SOE1/P2/F157 of the European Union provided financial support. M. Alvioli received “Short Mobility Grant”, 2016, from CNR during which part of this work was completed, and the Division of Geotechnical Engineering and Geosciences, Department of Civil and Environmental Engineering, Universitat Politècnica de Catalunya-BarcelonaTech for the warm hospitality. Massimiliano Alvioli acknowledges partial support within the SARF-LIGURIA project of CNR IRPI.

## References

- AGS (2007) Guidelines for landslide susceptibility, Hazard and Risk Zoning for Land Use Planning Australian Geomechanics Society Landslide Taskforce Landslide Zoning Working Group. Australian Geomechanics Society 42(1):13–36Google Scholar
- Aleotti P, Chowdhury R (1999) Landslide hazard assessment: summary review and new perspectives. Bull Eng Geol Environ 58:21–44. https://doi.org/10.1007/s100640050066 CrossRefGoogle Scholar
- Alvioli M, Baum RL (2016) Parallelization of the TRIGRS model for rainfall-induced landslides using the message passing interface. Environ Model Softw 81:122–135. https://doi.org/10.1016/j.envsoft.2016.04.002 CrossRefGoogle Scholar
- Alvioli M, Guzzetti F, Rossi M (2014) Scaling properties of rainfall induced landslides predicted by a physically based model. Geomorphology 213:38–47. https://doi.org/10.1016/j.geomorph.2013.12.039 CrossRefGoogle Scholar
- Alvioli M, Marchesini I, Reichenbach P, Rossi M, Ardizzone F, Fiorucci F, Guzzetti F (2016) Automatic delineation of geomorphological slope units with r.slopeunits v1.0 and their optimization for landslide susceptibility modeling. Geosci Model Dev 9:3975–3991. https://doi.org/10.5194/gmd-9-3975-2016 CrossRefGoogle Scholar
- Alvioli M, Melillo M, Guzzetti F, Rossi M, Palazzi E, von Hardenberg J, Brunetti MT, Peruccacci S (2018) Implications of climate change on landslide hazard in Central Italy. Sci Total Environ 630:1528–1543. https://doi.org/10.1016/j.scitotenv.2018.02.315 CrossRefGoogle Scholar
- Baum RL, Coe JA, Godt JW, Harp EL, Reid ME, Savage WZ, Schulz WH, Brien DL, Chleborad AF, McKenna JP, Michael JA (2005) Regional landslide-hazard assessment for Seattle, Washington, USA. Landslides 2:266–279. https://doi.org/10.1007/s10346-005-0023-y CrossRefGoogle Scholar
- Baum RL, Savage WZ, Godt JW (2008) TRIGRS — a Fortran program for transient rainfall infiltration and grid based regional slope-stability analysis, version 2.0. USGS Open File Report 2008 1159Google Scholar
- Bellugi D, Milledge DG, Dietrich WE, McKean JA, Perron JT, Sudderth EB, Kazian B (2015) A spectral clustering search algorithmfor predicting shallow landslide size and location. J Geophys Res Earth Surf 120:300–324. https://doi.org/10.1002/2014JF003137 CrossRefGoogle Scholar
- Beven KJ, Kirkby MJ (1979) A physically based, variable contributing area model of basin hydrology. Hydrol Sci Bull 24:43–69. https://doi.org/10.1080/02626667909491834 CrossRefGoogle Scholar
- Bornaetxea T, Rossi M, Marchesini I, Alvioli M (2018) Effective surveyed area and its role in statistical landslide susceptibility assessments. Nat Hazards Earth Syst Sci Discuss 1–22. https://doi.org/10.5194/nhess-2018-88
- Canales CJ (2011) Soil strength parameters and stability conditions of the slopes of the Tremp basin (eastern Pyrenees, Spain) derived from back-analysis and field reconnaissance. Master thesis. Technical Univesity of Catalonia. Barcelona. Technical University of CataloniaGoogle Scholar
- Cardinali M, Reichenbach P, Guzzetti F, Ardizzone F, Antonini G, Galli M, Cacciano M, Castellani M, Salvati P (2002) A geomorphological approach to the estimation of landslide hazards and risks in Umbria , Central Italy. Nat Hazards Earth Syst Sci 2:57–72. https://doi.org/10.5194/nhess-2-57-2002 CrossRefGoogle Scholar
- Cardinali M, Galli M, Guzzetti F, Ardizzone F, Reichenbach P, Bartoccini P (2006) Rainfall induced landslides in December 2004 in South-Western Umbria, Central Italy: types, extent, damage and risk assessment. Nat Hazards Earth Syst Sci 6:237–260. https://doi.org/10.5194/nhess-6-237-2006 CrossRefGoogle Scholar
- Carrara A (1988) Drainage and divide networks derived from high-fidelity digital terrain models. In: Chung CF, Fabbri AG, Sinding-Larsen R (eds) Quantitative Analysis of Mineral and Energy Resources. NATO ASI Series (Series C: Mathematical and Physical Sciences). Springer, DordrechtGoogle Scholar
- Carrara A, Cardinali M, Detti R, Guzzetti F, Pasqui V, Reichenbach P (1991) GIS techniques and statistical models in evaluating landslides hazard. Earth Surf Process Landf 16:427–445. https://doi.org/10.1002/esp.3290160505 CrossRefGoogle Scholar
- Carrara A, Cardinali M, Guzzetti F, Reichenbach P (1995) GIS technology in mapping landslide hazard. In: Carrara A, Guzzetti F (eds) Geographical information systems in assessing natural hazards. Kluwer Academic Publishers, Dordrecht, pp 135–175CrossRefGoogle Scholar
- Carrara A, Crosta G, Frattini P (2003) Geomorphological and historical data in assessing landslide hazard. Earth Surf Process Landf 28:1125–1142. https://doi.org/10.1002/esp.545 CrossRefGoogle Scholar
- Catani F, Lagomarsino D, Segoni S, Tofani V (2013) Landslide susceptibility estimation by random forests technique: sensitivity and scaling issues. Nat Hazards Earth Syst Sci 13:2815–2831. https://doi.org/10.5194/nhess-13-2815-2013 CrossRefGoogle Scholar
- Coduto D (1999) Geotechnical engineering: principles and practices. Prentice-Hall, New JerseyGoogle Scholar
- Copons R, Vilaplana J (2008) Rockfall susceptibility zoning at a large scale: from geomorphological inventory to preliminary land use planning. Eng Geol 102:142–151. https://doi.org/10.1016/j.enggeo.2008.03.020 CrossRefGoogle Scholar
- Corominas J (2001) Landslides and climate. In: 8th International Symposium on Landslides. Keynote Lecturers.CD-Rom, Cardiff. 33 ppGoogle Scholar
- Corominas J (2006) El clima y sus consecuencias sobre la actividad de los movimientos de ladera en España. Cuaternario Geomorfol 20:89–113Google Scholar
- Corominas J, van Westen C, Frattini P, Cascini L, Malet JP, Fotopoulou S, Catani F, van den Eeckhaut M, Mavrouli O, Agliardi F, Pitilakis K, Winter MG, Pastor M, Ferlisi S, Tofani V, Hervás J, Smith JT (2014) Recommendations for the quantitative analysis of landslide risk. Bull Eng Geol Environ 73:209–263. https://doi.org/10.1007/s10064-013-0538-8 CrossRefGoogle Scholar
- Dai FC, Lee CF (2002) Landslide characteristics and slope instability modeling using GIS, Lantau Island, Hong Kong. Geomorphology 42:213–228. https://doi.org/10.1016/S0169-555X(01)00087-3 CrossRefGoogle Scholar
- Dai FC, Lee CF, Ngai YY (2002) Landslide risk assessment and management: an overview. Eng Geol 64:65–87. https://doi.org/10.1016/S0013-7952(01)00093-X CrossRefGoogle Scholar
- Das I, Stein A, Kerle N, Dadhwal VK (2011) Probabilistic landslide hazard assessment using homogeneous susceptible units (HSU) along a national highway corridor in the northern Himalayas. India Landslides 8:293–308. https://doi.org/10.1007/s10346-011-0257-9 CrossRefGoogle Scholar
- David RM (2002) Arc hydro: GIS for water resources. ESRI Press, RedlandsGoogle Scholar
- Domènech G (2015) Assessment of the magnitude-frequency relationship of landslides and rockfalls: application to hazard mapping - Ph. D thesis. Universitat Politècnica de Catalunya-BarcelonaTechGoogle Scholar
- Erener A, Düzgün HBS (2013) A regional scale quantitative risk assessment for landslides: case of Kumluca watershed in Bartin, Turkey. Landslides 10:55–73. https://doi.org/10.1007/s10346-012-0317-9 CrossRefGoogle Scholar
- Espindola GM, Camara G, Reis IA et al (2006) Parameter selection for region-growing image segmentation algorithms using spatial autocorrelation. Int J Remote Sens 27:3035–3040. https://doi.org/10.1080/01431160600617194 CrossRefGoogle Scholar
- Fan X, Domènech G, Scaringi G, et al (2018) Spatio-temporal evolution of mass wasting after the 2008 Mw 7 . 9 Wenchuan Earthquake revealed by a detailed multi-temporal inventory. Landslides 1–17. https://doi.org/10.1007/s10346-018-1054-5 CrossRefGoogle Scholar
- Fawcett T (2006) An introduction to ROC analysis. Pattern Recogn Lett 27:861–874. https://doi.org/10.1016/j.patrec.2005.10.010 CrossRefGoogle Scholar
- Fell R, Ho KKS, Lacasse S, Leroi E (2005) A framework for landslide risk assessment and management. In: Hungr O, Fell R, Couture R, Eberhardt E (eds) Landslide risk management. Taylor and Francis, London, pp 3–19Google Scholar
- Fell R, Corominas J, Bonnard C, Cascini L, Leroi E, Savage WZ (2008) Guidelines for landslide susceptibility, hazard and risk zoning for land use planning. Eng Geol 102:85–98. https://doi.org/10.1016/j.enggeo.2008.03.022 CrossRefGoogle Scholar
- Frattini P, Crosta G, Fusi N, Dalnegro P (2004) Shallow landslides in pyroclastic soils: a distributed modelling approach for hazard assessment. Eng Geol 73:277–295. https://doi.org/10.1016/j.enggeo.2004.01.009 CrossRefGoogle Scholar
- Ghosh S, van Westen CJ, Carranza EJM, Jetten VG, Cardinali M, Rossi M, Guzzetti F (2012) Generating event-based landslide maps in a data-scarce Himalayan environment for estimating temporal and magnitude probabilities. Eng Geol 128:49–62. https://doi.org/10.1016/j.enggeo.2011.03.016 CrossRefGoogle Scholar
- Giles PT (1998a) Geomorphological signatures: classification of aggregated slope unit objects from digital elevation and remote sensing data. Earth Surf Process Landf 23:581–594. https://doi.org/10.1002/(SICI)1096-9837(199807)23:7<581::AID-ESP863>3.0.CO;2-S CrossRefGoogle Scholar
- Giles PT (1998b) An automated approach to the classification of the slope units using digital data. Geomorphology 21:251–264. https://doi.org/10.1016/S0169-555X(97)00064-0 CrossRefGoogle Scholar
- Godt JW, Baum RL, Savage WZ, Salciarini D, Schulz WH, Harp EL (2008) Transient deterministic shallow landslide modeling: requirements for susceptibility and hazard assessments in a GIS framework. Eng Geol 102:214–226. https://doi.org/10.1016/j.enggeo.2008.03.019 CrossRefGoogle Scholar
- González de Vallejo LI, Ferrer M, Ortuño L, Oteo C (2002) Ingeniería Geológica. PEARSON EDUCACIÓN, Madrid (in Spanish)Google Scholar
- Guzzetti F, Carrara A, Cardinali M, Reichenbach P (1999) Landslide hazard evaluation: a review of current techniques and their application in a multi-scale study, Central Italy. Geomorphology 31:181–216. https://doi.org/10.1016/S0169-555X(99)00078-1 CrossRefGoogle Scholar
- Guzzetti F, Cardinali M, Reichenbach P, Carrara A (2000) Comparing landslide maps: a case study in the upper Tiber River basin, Central Italy. Environ Manag 25:247–263. https://doi.org/10.1007/s002679910020 CrossRefGoogle Scholar
- Guzzetti F, Reichenbach P, Cardinali M, Galli M, Ardizzone F (2005) Probabilistic landslide hazard assessment at the basin scale. Geomorphology 72:272–299. https://doi.org/10.1016/j.geomorph.2005.06.002 CrossRefGoogle Scholar
- Guzzetti F, Galli M, Reichenbach P, Ardizzone F, Cardinali M (2006) Landslide hazard assessment in the Collazzone area, Umbria, Central Italy. Nat Hazards Earth Syst Sci 6:115–131. https://doi.org/10.5194/nhess-6-115-2006 CrossRefGoogle Scholar
- Haneberg WC (2004) A rational probabilistic method for spatially distributed landslide hazard assessment. Environ Eng Geosci 10:27–43. https://doi.org/10.2113/10.1.27 CrossRefGoogle Scholar
- Hansen A (1984) Landslide hazard analysis. In: Brunsden D, Prior DB (eds) Slope instability. John Wiley and Sons, New York, pp 523–602Google Scholar
- Jia N, Mitani Y, Xie M, Djamaluddin I (2012) Shallow landslide hazard assessment using a three-dimensional deterministic model in a mountainous area. Comput Geotech 45:1–10. https://doi.org/10.1016/j.compgeo.2012.04.007 CrossRefGoogle Scholar
- Komac M (2006) A landslide susceptibility model using the analytical hierarchy process method and multivariate statistics in perialpine Slovenia. Geomorphology 74:17–28. https://doi.org/10.1016/j.geomorph.2005.07.005 CrossRefGoogle Scholar
- Kranitz F, Bensi S (2009) In: Posch-Trözmüller G (ed) Second Scientific Report to the INTERREG IV A project MASSMOVE - Minimal standards for compilation of danger maps like landslides and rock fall as a tool for disaster prevention. Attachment 4 to the second progress reportGoogle Scholar
- Lari S, Frattini P, Crosta GB (2014) A probabilistic approach for landslide hazard analysis. Eng Geol 182:3–14. https://doi.org/10.1016/j.enggeo.2014.07.015 CrossRefGoogle Scholar
- Lee S, Evangelista DG (2006) Earthquake-induced landslide-susceptibility mapping using an artificial neural network. Nat Hazards Earth Syst Sci 6:687–695. https://doi.org/10.5194/nhess-6-687-2006 CrossRefGoogle Scholar
- Marchesini I, Cencetti C, De Rosa P (2009) A preliminary method for the evaluation of the landslides volume at a regional scale. Geoinformatica 13:277–289. https://doi.org/10.1007/s10707-008-0060-5 CrossRefGoogle Scholar
- Mashimbye ZE, De Clercq WP, Van Niekerk A (2014) An evaluation of digital elevation models (DEMs) for delineating land components. Geoderma 213:312–319. https://doi.org/10.1016/j.geoderma.2013.08.023 CrossRefGoogle Scholar
- Mergili M, Marchesini I, Rossi M, Guzzetti F, Fellin W (2014) Spatially distributed three-dimensional slope stability modelling in a raster GIS. Geomorphology 206:178–195. https://doi.org/10.1016/j.geomorph.2013.10.008 CrossRefGoogle Scholar
- Metz M, Mitasova H, Harmon RS (2011) Efficient extraction of drainage networks from massive, radar-based elevation models with least cost path search. Hydrol Earth Syst Sci 15:667–678. https://doi.org/10.5194/hess-15-667-2011 CrossRefGoogle Scholar
- Montero I (2011) Determinación y análisis de la actividad del deslizamiento de Les Pales (Cuenca de Tremp, Lleida). Master thesis. Technical Univesity of Catalonia. Barcelona. 1–125 (in Spanish)Google Scholar
- Montgomery DR, Dietrich WE (1994) A physically based model for the topographical control on shallow landsliding. Water Resour Res 30:1153–1171. https://doi.org/10.1029/93WR02979 CrossRefGoogle Scholar
- Neteler M, Mitasova H (2008) Open source GIS: a GRASS GIS approach. Third edition. The International Series in Engineering and Computer Science: Volume 773 406 pages, 80 illus., Springer, New York. ISBN: 038735767XGoogle Scholar
- Novoa M (1984) Precipitaciones y avenidas extraordinarias en Catalunya. J Trab Inestabilidades Laderas Pirineo Barcelona, pp I.1.1–I.1.15 (in Spanish)Google Scholar
- Oliveras G (2011) Determinació i analisi de l’activitat de l’esllavissada de Clot del Oms (Conca de Tremp, Lleida). Master thesis. Technical Univesity of Catalonia Barcelona 1–118 (in Spanish)Google Scholar
- Pack RT, Tarboton DG, Goodwin CN (1998) The SINMAP approach to terrain stability mapping. In: Eighth Congres of the International Association of Engineering GeologyGoogle Scholar
- Pack RT, Tarboton DG, Goodwin CN, A. Prasad (2005) SINMAP. A stability index approach to terrain stability hazard mapping. Version for ArcGIS 9.x and Higher. Utah State UniversityGoogle Scholar
- Pavel M, Fannin R, Nelson J (2008) Replication of a terrain stability mapping using an artificial neural network. Geomorphology 97:356–373. https://doi.org/10.1016/j.geomorph.2007.08.012 CrossRefGoogle Scholar
- Picarelli L, Oboni F, Evans SG et al (2005) Hazard characterization and quantification. In: Hungr O, Fell R, Couture R, Eberthardt E (eds) Landslide risk management. Taylor and Francis, London, pp 27–61Google Scholar
- Posch-Trözmüller G (2010) AdaptAlp WP 5.1 Hazard mapping - geological hazards. Literature survey regarding methods of hazard mapping and evaluation of danger by landslides and rock fall. Final Report, Geologische Bundesanstalt, WienGoogle Scholar
- Pradhan AMS, Lee S-R, Kim Y-T (2018) A shallow slide prediction model combining rainfall threshold warnings and shallow slide susceptibility in Busan, Korea. Landslides 16:647–659. https://doi.org/10.1007/s10346-018-1112-z CrossRefGoogle Scholar
- Raetzo H, Loup B (2009) Bafu Schutz vor Massenbewegungen. Technische Richtlinie als Vollzugshilfe. Entwurf 9. Sept.(in German)Google Scholar
- Raetzo H, Lateltin O, Bollinger D, Tripet J (2002) Hazard assessment in Switzerland - codes of practice for mass movements. Bull Eng Geol Environ 61:263–268. https://doi.org/10.1007/s10064-002-0163-4 CrossRefGoogle Scholar
- Raia S, Alvioli M, Rossi M, Baum RL, Godt JW, Guzzetti F (2014) Improving predictive power of physically based rainfall-induced shallow landslide models: a probabilistic approach. Geosci Model Dev 7:495–514. https://doi.org/10.5194/gmd-7-495-2014 CrossRefGoogle Scholar
- Reichenbach P, Busca C, Mondini AC, Rossi M (2014) The influence of land use change on landslide susceptibility zonation: the Briga catchment test site (Messina, Italy). Environ Manag 54:1372–1384. https://doi.org/10.1007/s00267-014-0357-0 CrossRefGoogle Scholar
- Reid ME, Christian SB, Brien DL, Henderson S. (2015) Scoops3D — software to analyze three-dimensional slope stability throughout a digital landscape. In: U.S. Geological Survey Techniques and Methods, book 14. p 218Google Scholar
- Rosell J, Llompart R, Linares C (2001) El “garumniense” prepirenaico. Rev Soc Geol España 14:47–56Google Scholar
- Rossi G, Catani F, Leoni L, Segoni S, Tofani V (2013) HIRESSS: a physically based slope stability simulator for HPC applications. Nat Hazards Earth Syst Sci 13:151–166. https://doi.org/10.5194/nhess-13-151-2013 CrossRefGoogle Scholar
- Salciarini D, Godt JW, Savage WZ, Baum RL, Conversini P (2008) Modeling landslide recurrence in Seattle, Washington, USA. Eng Geol 102:227–237. https://doi.org/10.1016/j.enggeo.2008.03.013 CrossRefGoogle Scholar
- Salciarini D, Fanelli G, Tamagnini C (2017) A probabilistic model for rainfall—induced shallow landslide prediction at the regional scale. Landslides 14:1731–1746. https://doi.org/10.1007/s10346-017-0812-0 CrossRefGoogle Scholar
- Savage WZ, God JW, Baum RL (2004) Modelling time-dependent areal slope instability. In: Lacerda W, Ehrlich M, Fountoura S, Sayao A (eds) Landslide evaluation and stabilization, IX International Symposium on Landslides. AA Balkem, Rio de Janeiro, pp 23–38Google Scholar
- Schlögel R, Marchesini I, Alvioli M, Reichenbach P, Rossi M, Malet JP (2018) Optimizing landslide susceptibility zonation: effects of DEM spatial resolution and slope unit delineation for logistic regression models. Geomorphology 31:10–20. https://doi.org/10.1016/j.geomorph.2017.10.018 CrossRefGoogle Scholar
- Seguret M (1972) Etude tectonique des nappes et séries décollées de la partie centrale du vessant du des Pyrénées - Ph.D thesis. University of Montpellier (in French)Google Scholar
- Sterlacchini S, Frigerio S, Giacomelli P, Brambilla M (2007) Landslide risk analysis: a multi-disciplinary methodological approach. Nat Hazards Earth Syst Sci 7:657–675. https://doi.org/10.5194/nhess-7-657-2007 CrossRefGoogle Scholar
- Strozzi T, Ambrosi C, Raetzo H (2013) Interpretation of aerial photographs and satellite SAR interferometry for the inventory of landslides. Remote Sens 5:2554–2570. https://doi.org/10.3390/rs5052554 CrossRefGoogle Scholar
- Tran TV, Alvioli M, Lee G, An HU (2018) Three-dimensional, time-dependent modeling of rainfall-induced landslides over a digital landscape: a case study. Landslides 1–14. https://doi.org/10.1007/s10346-017-0931-7 CrossRefGoogle Scholar
- Van Den Eeckhaut M, Poesen J, Govers G et al (2007) Characteristics of the size distribution of recent and historical landslides in a populated hilly region. Earth Planet Sci Lett 256:588–603. https://doi.org/10.1016/j.epsl.2007.01.040 CrossRefGoogle Scholar
- Van Den Eeckhaut M, Reichenbach P, Guzzetti F et al (2009) Combined landslide inventory and susceptibility assessment based on different mapping units: an example from the Flemish Ardennes, Belgium. Nat Hazards Earth Syst Sci 9:507–521. https://doi.org/10.5194/nhess-9-507-2009 CrossRefGoogle Scholar
- Vranken L, Vantilt G, Van Den Eeckhaut M et al (2014) Landslide risk assessment in a densely populated hilly area. Landslides. 12:787–798. https://doi.org/10.1007/s10346-014-0506-9 CrossRefGoogle Scholar
- Wu TH, Abdel-Latif MA (2000) Prediction and mapping of landslide hazard. Can Geotech J 37:781–795. https://doi.org/10.1139/t00-006 CrossRefGoogle Scholar
- Xie M, Esaki T, Zhou G, Mitani Y (2003) Geographic information systems-based three-dimensional critical slope stability analysis and landslide hazard assessment. J Geotech Geoenvironmental Eng 129:1109–1118. https://doi.org/10.1061/(ASCE)1090-0241(2003)129:12(1109) CrossRefGoogle Scholar
- Xie M, Esaki T, Zhou G (2004) GIS-based probabilistic mapping of landslide hazard using a three-dimensional deterministic model. Nat Hazards 33:265–282. https://doi.org/10.1023/B:NHAZ.0000037036.01850.0d CrossRefGoogle Scholar
- Yatheendradas S, Kirschbaum D, Nearing G, et al (2019) Bayesian analysis of the impact of rainfall data product on simulated slope failure for North Carolina locationsGoogle Scholar

## Copyright information

**Open Access** This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.