Rainfall-Induced Shallow Landslide Susceptibility for Risk Management of Underground Services in a Mediterranean Metropolitan City

Abstract

Shallow landslide susceptibility assessment is an essential research activity for land management and risk assessment. In this study, a GIS-based approach was developed to assess rain-induced landslide susceptibility in the Municipality of Genoa, a Mediterranean anthropized area historically characterized by intense rainfall events that frequently trigger shallow landslides with high destructive power. Based on a detailed inventory of historical landslides, a semi-quantitative method was applied to assess the influence of seven causal factors of natural and anthropogenic landslides. The areas were categorized into five classes of rain-induced shallow landslide susceptibility, indicating slopes where newly triggered landslides may occur. The landslide susceptibility map was subsequently integrated with the map of gas and water utilities, whose features were used to assess their vulnerability. Finally, an early-stage risk assessment of the two utility networks was developed to serve as a decision support tool for strategic planning and integrated asset management in the context of climate change. The results show that about 9.8% and 6.8% of the total length of water and gas pipelines are exposed to higher risk classes 4 and 5.

1. Introduction

Landslides cause damage every year and significant socio-economic losses in mountain areas [1,2,3]. Shallow landslides and debris/mudflows are the most common processes damaging buildings and infrastructures, often inducing a loss of human life [4,5,6]. Landslide triggers include intense rainfall, which often generates small-volume landslides that are characterized by a high speed of propagation [7,8,9]. The high kinetic energy of small but quickly moving masses can cause serious damage to buildings and infrastructure.

Landslide hazards can be defined as the probability of a potentially damaging event occurring in a specific area over a determined return period [10,11]. Risk assessment requires estimation of landslide volume, frequency, and location [12]. The extent and recurrence time of the event are difficult parameters to predict, as they are linked to the natural processes that control gravity dynamics. Moreover, landslide inventories are a relatively recent creation [13,14,15].

The spatial distribution of potentially unstable areas can be determined by the probability of event occurrence based on local geological and physical–geographical features [16]. The definition of landslide susceptibility represents the first step in risk assessment [17]. Identifying and mapping areas of historical slope instability processes with similar geotechnical characteristics provides useful information for predicting the location of newly initiated shallow landslides [18]. Landslide susceptibility assessment, therefore, is an essential tool for risk management, supporting authorities and decision makers in sustainable spatial planning and the definition of mitigation measures, including monitoring and warning activities [19,20,21].

Over the past decades, several methods have been developed to assess landslide susceptibility, combining parameters in a GIS (Geographic Information System) environment and collecting data with precision remote sensing techniques such as ALS (Airborne Laser Scanning) surveys [22,23,24,25,26]. The methods used to assess landslide susceptibility can be distinguished into [27,28,29,30]: (1) a geomorphological survey and inventory map of landslide phenomena; (2) a semi-quantitative approach based on the combination of parameters, linked to geomorphological processes and landslide factors; and (3) physically based methods [31], linked to the constitutive laws of the geomaterials that characterize the stratigraphic sequence of the investigated slopes.

The definition of rainfall-induced shallow landslide susceptibility is thus the first key step in risk assessment, which is performed through subsequent cross-referencing with hazard-exposed elements. Much research has been conducted on rainfall-induced shallow landslide risk assessment with reference to buildings and large infrastructures.

Rainfall-induced shallow landslides are also an important element of risk for vulnerable elements such as underground services. Scientific research over the last 20 years has been conducted on large oil and gas pipelines, with interesting case studies in Albania [32,33], China [34], and Turkey [35] related to transcontinental oil and gas pipelines affected by landslides.

Research has focused on different topics: (i) the strains of underground gas pipelines subjected to landslide movement [32]; (ii) an expert determination approach to define a consensus for estimating the risk of a gas pipeline [33]; (iii) a quantitative model for regional risk assessment involving an analysis of the patterns of historical landslide disasters along oil and gas pipelines, with an interesting dual approach to both susceptibility and vulnerability assessment [34]; (iv) appropriate methodology for using a Tunnel Boring Machine to construct a micro-tunnel to bury a natural gas pipeline beneath an active landslide [33]; (v) the assessment of landslide susceptibility at a damaged gas pipeline by applying both statistical index and weightings factor in a GIS environment [35]; (vi) a risk assessment model for a buried pipeline geohazard-like landslide, sinkhole phenomena, and running water processes [36]; (vii) a geotechnical finite element approach to a landslide hazard case for an underground gas pipeline [37]; (viii) the maximum susceptibility conditions under which an underground pipeline is subject to leakage and rupture as a result of a landslide [38].

An interesting scientific contribution lies in the susceptibility assessment of soil erosion for shallow underground pipelines. This method is used to analyze geotechnical soil parameters and, through an AHP (Analytic Hierarchy Process) procedure, the susceptibility values of potential erosion [39,40]. On the other hand, there are few studies on risk assessment for underground services, such as gas and water networks in urban areas located in hilly and mountainous environments, which represent an important criticality in terms of land management [41,42,43,44,45].

In this research, a semi-quantitative method was adopted to assess shallow landslide susceptibility in the Municipality of Genoa, which is internationally known for its cultural and landscape heritage and historically affected by natural hazards such as landslides, debris flows, floods, and flash floods [46]. The method was modified from previous research [19,47], improving the quantitative component, and then applied to the risk assessment of water and gas distribution pipelines.

The aim of the present work is to improve upon previous research in assessing susceptibility to shallow landslides, increase the quantitative component, and apply the results to the risk assessment of water and gas distribution pipelines in the Municipality. The results provide a preliminary evaluation of the more critical stretches and a possible DSS (decision support system) to plan and adopt a risk mitigation strategy and consequent measures.

2. Materials and Methods

2.1. Geographical and Geological Outline

This subsection summarizes the geographic, climatic, and geological controls underlying the selection of factors for susceptibility mapping and subsequent overlap with gas/water networks. There is a strong link between the landscape features and the intense soil sealing and population density that affect the area, underlining the importance of developing a reliable methodology to assess the network’s risk.

Genoa is a coastal city and the capital of the Liguria region. It is located in the north-western Italian macro-region (Figure 1) and extends over approximately 240 km², mostly distributed along a coastal strip stretching for 30 km from east to west (Figure 1). Subordinately, the city also extends along the two main drainage axes orthogonal to the coastline, the Bisagno and Polcevera streams (Figure 1), which limit the morphological amphitheater on which the historic city center has been built [46].

The resident population of Genoa is about 560,000 [48], but in the 1970s, when the industrial sector (especially related to the port) was at its peak, it reached 800,000 [49]. In that condition, urban planning had already foreseen a city of 1 million inhabitants. The population density is the highest in the central municipal area, particularly in the final stretch of the alluvial plain of the Bisagno stream, in the historical amphitheater behind the port area, and near the mouth of the Polcevera stream. The population reaches 25,000 inhabitants/km² in Marassi (Lower Bisagno Valley), east of the amphitheater, in an area of high flood hazard, and west of Sampierdarena district.

The physical–geographical features of Genoa are peculiar both in terms of meteorological–climatic aspects and landscape. Due to its location at the northern apex of the Western Mediterranean, in the late summer–autumn period, there is frequent depression over the Gulf of Genoa, bringing intense and high-intensity rainfall, and the metropolitan city of Genoa has rainfall records at a Mediterranean scale. The highest values have been exceeded during major flooding events: on 7–8 October 1970 (948 mm/24 h), 4 November 2011 (181 mm/h), and 4 October 2021 (883.8 mm/24 h, 740.6/12 h, 496 mm/6 h, and 377.8 mm/3 h).

This last event, which was particularly intense, affected the hinterland to the west of the city but avoided the municipal territory, triggering diffuse shallow landslides [50]. During this event, the maximum values recorded over 12, 6, and 3 h exceeded the previous maximums that were measured during 1970 and 2011. The peculiarities of the ground effects and the landscape features of the study area suggested the causal factors used in the present research.

The average annual rainfall is 1250 mm in the city center and varies significantly with altitude. Figure 2 shows the average monthly values recorded at the Genoa University station, which has been operating since 1833. The higher values of annual rainfall correspond to some catastrophic events that occurred in the past, while higher monthly values mainly occurred in the fall and, secondarily, in the spring.

Climate change in Genoa is demonstrated by an increase in the average annual air temperature and a decrease in the number of rainy days, resulting in an increase in the intensity of rainfall, leading to more frequent floods and landslides [51,52]. Moreover, proxies related to extreme rain events show a statistically significant increase over 1979–2019 [53].

The area includes a large number of small catchment basins, mainly oriented perpendicular to the coastline. Not all of the watersheds belong entirely to the Municipality of Genoa: both those in the upper Bisagno basin and the upper Polcevera are part of other municipalities.

Based on the distance of the watershed from the coastline and the nature of the rock masses, slope gradients are generally high, especially in the eastern and western sectors of the municipal area, where values above 50% prevail. The morphometric and elevation features of the basins (Figure 1) represent one of the most significant landslide factors, which results in a widespread hazard: during heavy rainfall, the reduced time of concentration, which may be even less than 30’, causes flash floods in the small floodplain. The result is a high risk due to intense urbanization.

During flood events, often a flash flood, shallow landslides are triggered, along with, depending on the morphological features of the territory, debris mudflows. These quickly moving masses cause a significant hazard for directly exposed elements, such as buildings and infrastructure, and an indirect hazard related to the saturation of the culverts, which are diffusively present in the urban area. The effects on the ground can be very serious and include a loss of life as well as substantial economic damage.

Table 1. Rainfall events triggering geo-hydrological phenomena with ground effects in the City of Genoa in the 1920–present day period, organized in decades (database AVI project available at https://sici.irpi.cnr.it/storici.htm, accessed on 20 October 2025). For each year, the number in round brackets represents the number of rainfall events that trigger geo-hydrological phenomena.

Period	Date and Number of Events
1920–1929	1920 (1), 1923 (1), 1926 (1), 1927 (5), 1929 (5)
1930–1939	1930 (5), 1931 (3), 1932 (7), 1934 (6), 1935 (1), 1936 (3), 1937 (3), 1938 (4), 1939 (4)
1940–1949	1940 (1), 1942 (3), 1945 (13), 1948 (1), 1949 (1)
1950–1959	1951 (32), 1952 (1), 1953 (3), 1954 (2), 1955 (9), 1956 (5), 1957 (6), 1958 (4), 1959 (12)
1960–1969	1960 (5), 1961 (4), 1962 (2), 1963 (12), 1964 (7), 1965 (4), 1966 (13), 1967 (8), 1968 (9), 1969 (5)
1970–1979	1970 (22), 1971 (2), 1972 (3), 1976 (3), 1977 (19), 1978 (1), 1979 (1)
1980–1989	1980 (1), 1981 (1), 1982 (2), 1983 (1), 1984 (4), 1985 (7), 1987 (2), 1989 (3)
1990–1999	1990 (2), 1991 (13), 1992 (19), 1993 (23), 1994 (21), 1995 (18), 1996 (8), 1997 (28), 1998 (17), 1999 (4)
2000–2009	2000 (9), 2002 (9), 2003 (2), 2005 (2), 2006 (6), 2007 (3), 2008 (3), 2009 (2)
2010–2019	2010 (12), 2011 (10), 2012 (1), 2014 (24), 2016 (2), 2018 (2), 2019 (4)
2020–2024	2020 (2), 2021 (1), 2022 (3), 2023 (3)

and Figure 3 show the geo-hydrological hazard events that have characterized the Genoese territory over the last 100 years based on the properly updated AVI project database [54].

The city of Genoa’s geologic features contribute substantially to slope instability, inducing a natural hazard: Figure 4 schematically shows the rock masses located in the Genoa area, predominantly sedimentary in the East (limestone–marly flysch), and ophiolitic in the West (serpentinites and calcareous schist). The central sector is particularly complex and characterized by argillitic and siltitic flysch, basalts, and dolomites [55]. In the present study, we focused on shallow landslides, which involve soil over the rock substratum. The lithology and the morphometry of the study area are considered crucial factors: in particular, slope steepness is related to the lithology and tectonic deformation that occurred after the Alpine and Apennine orogenic events. Simultaneously, lithology affects the development of different soil types, which may trigger a shallow landslide. The altitude ranges from sea level to 1183 m asl, with a mean value of 279 m asl, while the mean slope gradient is 45.6%. The morphology is related to the hydrographical networks, which are characterized by good hierarchical development with many first-order Strahler sensu streams and with a prevailing north–south and east–west orientation due to ductile deformations that affected the rocks. These features affect the running water concentration.

The geological history of rock formations is an additional element causing slope instability and triggering landslides characterized by different kinematic processes. Many phases of ductile and brittle deformation affect the rocks, which slightly differ in strength and deformability; consequently, the various deformation patterns influence the rocks in terms of weathering resistance and geotechnical behavior.

2.2. Research Methods

The susceptibility assessment of shallow landslides in the Municipality of Genoa was conducted by employing a semi-quantitative approach, which means combining a quantitative and a qualitative method according to previous research in a nearby territory [19,47,50]. This approach was partially modified, including a quantitative component to evaluate the causative factors. The workflow is as follows: (i) shallow landslide inventory extraction from national databases and random subdivision in 80% and 20% subsets; (ii) the selection of seven causal factors related to the area features, as described in Section 2.1; (iii) statistics of the phenomena per causal factor to be used as weights for every factor; (iv) weighting process for every causal factor; (v) application of AHP methodology using five experts’ evaluation of the causal factors, to be used as factor weights; (vi) weighting process and linear sum of the seven factors; (vii) normalization of the result and subdivision into five susceptibility classes.

The method, schematized in Figure 5, is based on identifying a set of descriptive parameters of the area as factors triggering shallow landslides, followed by a related descriptive statistical evaluation. The identification of factors follows the area features described in Section 2.1 and the numerous triggering events of shallow landslides that have been observed in the last twenty years. Then, the results of the descriptive statistics for every factor are assigned as weights to the different classes into which every causing factor has been divided; every factor is consequently reclassified [56,57].

Then, the seven factors are weighted and linearly summed, and the result is normalized. These weights are identified by applying the AHP heuristic method (Analytic Hierarchy Process) [58,59,60,61,62] by combining subjective evaluations of five experts [61,62]. This methodology is a modified version of the one already proposed and tested in the Mediterranean environment [19,47] in previous research. The AHP is a semi-quantitative multi-criteria decision support technique used to compare heterogeneous physical quantities, and it is widely applied in natural hazard management [63,64,65,66,67,68] and landslide susceptibility analysis [8,9,19,63,64,65,66,67,68]. Further, the susceptibility analysis results are applied to the risk assessment of the gas and water distribution pipelines in the Municipality of Genoa, which is shown in Figure 5.

The reference database used is the inventory of landslide phenomena in Italy from the IFFI project, integrated with data collected through the Copernicus system after the most recent intense events and from data acquired by the governing authority of the basin plans involving the municipal area [54]. The combination of these geodatabases resulted in a total of 485 fast-developing shallow landslides that occurred in the municipal area. The inventory was used for statistical weighting and validation.

The case history of such phenomena in Liguria and beyond is expansive and has increased since the year 2000 [46]: the events that occurred in Western Liguria in 2000; the hundreds of shallow landslides triggered during the flash flood that affected the Cinque Terre/Vara Valley in 2011; the events of October/November 2014 in Genoa Province; and the landslide that in Leivi, in the hinterland 50 km east from Genoa, claimed two lives; the 2016 Lavina di Rezzo (Imperia) debris flow; and the 2019 landslides in Cava Lupara (Genoa), Cenova di Rezzo (Imperia), and Madonna del Monte (Savona) that caused the collapse of the viaduct along the A6 highway.

The set of census events at the municipal scale was divided via random extraction into two subsets: a training set of 388 phenomena accounting for 80 percent of the total and a calibration set of 97 phenomena accounting for the remaining 20 percent (Figure 6).

The first set was used to obtain the descriptive statistics of phenomena with respect to the seven causative factors (Figure 7): slope gradient, slope aspect, lithology, land use, planar curvature and tangential surface curvature, and presence of slope debris cover.

These data, whose characteristics are summarized in

Table 2. List of data, in raster and vector formats, used in the assessment of shallow landslide susceptibility.

Name	Source	Scala/Pixel	Date	Type
Administrative Unit	Liguria Region	1:5000	2018	V
Aspect	Liguria Region	1:10,000	2007	V
CORINE Land Use (II Level)	Liguria Region	1:10,000	2018–2019	V
CTR (Regional Technical Map)	Liguria Region	1:5000	1990–2006	R
DEM (Digital Elevation Model)	Liguria Region	5 m	2016	R
Hydrographic Network and Catchments	Liguria Region	1:10,000	2019	V
Landslide—Project IFFI	Liguria Region	1:10,000	2014	V
Lithology	Liguria Region	1:10,000	2017	V
Slope	Liguria Region	1:10,000	2016	V
DEM	Genoa Municipality	1 m	2018	R
Geomorphological map	Genoa Municipality	1:5000	2015	V

, were partly obtained from the available data of regional databases at a scale of 1:10,000, and partly obtained by the application of processing algorithms to the 5 m mesh DTM created by the Region of Liguria. Debris covers are also considered to play a significant role in terms of the probability of triggering shallow landslides. Therefore, based on evidences of recent phenomena in the area, the slope cover factor was included in the analysis of shallow landslide susceptibility: these geomorphological data were obtained by combining the informative layer included in the geological map made by the Municipality of Genoa with original field surveys and remote sensing analysis, which enabled more detailed recognition and mapping of the anthropogenic terraces widely spread in the Ligurian territory.

The presence of anthropogenic terraces is often crucial in triggering the phenomena that affected the region over the past 20 years. They were identified through a dedicated analysis according to the methods identified in the same morpho-climatic environment [69,70,71] in other research. The analysis was completed using the DTM obtained from the ALS survey acquired by the Municipality of Genoa in 2018 with a 1 m resolution. The analysis is based on the computation of the SVF—Sky View Factor (SAGA GIS algorithm), whose reliability and accuracy were deemed satisfactory for the purposes at hand [72]. The high resolution of the ALS DTM was crucial to perfectly identify the terraces’ presence, but in the final combination of factors, the deposit layer was resampled to 5 m, according to the other layers’ resolution.

The causal factors were examined according to the extensive bibliography available, which enables the evaluation of both geological–morphological–morphometric characteristics and anthropogenic influence through land use. Descriptive statistics, which are presented in the Results section, made it possible to objectively assign different weights to the relevant classes into which each factor was divided. The causing factors are as follows:

• Lithology: This drives the soil type, evolution, and the presence of debris after weathering effects, along with geotechnical features. The study area includes flysch and ophiolitic rocks, conglomerate, cherts, carbonate, and shale rocks.
• Aspect: This drives weathering effects on the rock mass. Moreover, the high-intensity rainfalls are mainly related to winds from the south.
• Slope gradient: This influences soil/debris stability along the slopes. The study area is largely defined by high values.
• Land use: in the study area, there is quite a significant spatial variability, with many small villages and related infrastructure along the slopes.
• Slope deposits: apart from natural deposits, the area is characterized by a diffuse presence of man-made terraces, which often play the role of source areas for shallow landslides.
• Tangential curvature: morphometry drives the running water concentration and general stability of the slope.
• Profile curvature: morphometry drives the running water concentration and general stability of the slope.

The AHP technique was applied to obtain the seven factors’ weights after five experts’ evaluation, and according to the methodology conducted in previous research [19,47]; the weights of factors are shown in

Table 3. Weights assigned to the seven causal factors for shallow landslide susceptibility assessment.

Landslide Factor	Weight
Lithology	8.4
Aspect	5.6
Slope gradient	27.6
Land use	13.3
Slope deposits	29.3
Tangential curvature	8.5
Profile curvature	7.3

.

In Appendix A, the pairwise matrix is shown. It is created from the five experts’ factor evaluation. In order to assess the reliability of the evaluation, the consistency ratio C.R. and AHP consensus indicator S* were computed according to the method described in [19]. The computed indicator figures are as follows:

$$C.R.=6.5\%$$

(1)

$$S^{*}=96.8\%$$

(2)

A C.R. value smaller than 10% is considered to be acceptable, indicating the consistency of the matrix, while the S* value indicates high agreement between the participants.

In order to quantitatively assess the predictive accuracy of the susceptibility map that arises from the calculation process, the Receiver Operating Characteristic curve (ROC) and the Area Under Curve (AUC) indicators were computed [73,74,75]. Finally, the reliability of the result was subsequently verified via the calibration process, i.e., by checking the susceptibility classes corresponding to the 97 landslides in the calibration set.

A risk assessment of the water and gas distribution networks was conducted by evaluating the vulnerability of the two through the local diameter of the pipes and then calculating the related risk [76,77,78]. No further data, like burial depth, pipeline material, or age, are available at this stage. This assessment is to be considered an early-stage evaluation of vulnerability. Since further data should be available in the future, a more precise vulnerability evaluation could be implemented in a future calculation.

Table 4. The water and gas networks’ vulnerability classes according to the pipe’s diameters.

Pipe Diameter (mm)	Vulnerability Class
d ≤ 50	1
50 < d ≤ 100	2
100 < d ≤ 200	3
200 < d ≤ 300	4
d > 300	5

identifies the five vulnerability classes defined according to the pipe’s diameters, which range from 10 to 900 mm for the water network and from 15 to 800 mm for the gas network. Then, the five risk classes were determined by combining the vulnerability classes with the hazard classes identified in the susceptibility computation: figures in the risk matrix were obtained after the product of the hazard score and the vulnerability score. Finally, the results were conservatively classified into the five risk classes according to the risk matrices in Figure 8.

In sealed urban soil areas, where the shallow landslide hazard is absent (class 0), in order to highlight the potential presence of risk factors caused by falls or wall collapses, pipe risk was defined as follows: (i) where the slope gradient is under 5°, the hazard is considered to be absent, and the associated risk is zero; (ii) in areas steeper than 5°, the fall hazard is considered as belonging to only one class, and the associated risk is obtained considering the vulnerability identified in Table 4. This evaluation has to be considered preliminary and can be improved in future research.

3. Results

The statistics of shallow landslides belonging to the training set for every causal factor are shown in

Table 5. The occurrence of shallow landslides for the 7 identified causal factors grouped into classes, along with corresponding weights used to assess landslide susceptibility.

Slope (%)	Number of Shallow Landslides	Weight
0–10	2	0.0052
11–20	36	0.0935
21–35	153	0.3974
36–50	139	0.3610
51–75	54	0.1403
>75	1	0.0026
Aspect	Number of Shallow Landslides	Weight
South	67	0.1740
East	44	0.1143
West	59	0.1532
North	22	0.0571
North-East	34	0.0883
South-East	69	0.1792
South-West	68	0.1766
North-West	22	0.0571
Lithology	Number of Shallow Landslides	Weight
Schists and shales	103	0.2675
Carbonate rocks	5	0.0130
Alluvial deposits	18	0.0468
Conglomerates	2	0.0052
Flysch	72	0.1870
Quartzite and cherts	91	0.2364
Ophiolitic rocks	94	0.2442
Land Use	Number of Shallow Landslides	Weight
Artificial areas	8	0.0208
Crop and meadows	119	0.3091
Woods	195	0.5065
Scrub and herbaceous areas	61	0.1584
Peri-fluvial areas	2	0.0052
Tangential Curvature	Number of Shallow Landslides	Weight
<−0.025	0	0.0000
−0.025/−0.01	55	0.429
−0.01/−0.0025	92	0.2390
−0.0025/0.0025	98	0.2545
0.0025/0.01	85	0.2208
0.01/0.025	47	0.1221
>0.025	8	0.0208
Profile Curvature	Number of Shallow Landslides	Weight
<−0.025	0	0.0000
−0.025/−0.01	44	0.1143
−0.01/−0.0025	97	0.2519
−0.0025/0.0025	134	0.3481
0.0025/0.01	86	0.2234
0.01/0.025	23	0.0597
>0.025	1	0.0026
Slope Deposits	Number of Shallow Landslides	Weight
Presence	359	0.9325
Without slope deposits	26	0.0675

; the figures were used to compute the weights for the reclassification process of every causal factor. As shown in Table 5, the highest frequency of shallow landslides for Genoa Municipality is related to exposures to the southern quadrants; slopes of intermediate classes; cultivated land uses, i.e., the presence of terraces, meadows, and forests; and lithotypes of schist, quartzitic, flysch, and ophiolitic rock masses. In addition, the possibility of triggering even in the absence of cover, as its presence was defined, i.e., for thicknesses greater than 1 m or in the presence of terraces, although it may seem incongruent, is to be attributed to the activation of modest-sized phenomena or conditions where the presence of cover was not precisely recognized. The shallow landslide susceptibility map (SLSM) was obtained as the AHP weight linear sum of the causal factor reclassification maps, which were weighted using the statistics in Table 5. The final result was normalized in the interval 0–1 and classified into five equal-length intervals. The combination of factors and the normalization process is as follows:

$$SLSM={\displaystyle \raisebox{1ex}{$\left(\sum _{i=1}^7{w}_i{F}_i\right)$}\!\left/ \!\raisebox{-1ex}{$\left(\sum _{i=1}^7{w}_i{F}_i\right)max$}\right.}$$

where w is the AHP weight of the factor i, and F is the factor value after reclassification with the weights obtained from statistics.

The network of underground utilities shown in Figure 5 was cross-referenced with the susceptibility map obtained according to the aforementioned procedure. Areas made impermeable by urbanization were excluded from the assessment because they lack potential source material for triggering shallow landslides. This type of phenomenon shows the highest capacity to generate damage on structures and infrastructures [79,80,81] because of the high speed of movement that, even in the presence of relatively moderate masses, results in high kinetic energy. Moreover, their evolution, whose triggering is driven by high rainfall intensities, occurs rapidly, preventing any kind of intervention during the event. Sealed surfaces can concentrate runoff and thus possibly contribute to triggering a downstream runoff phenomenon, as may occur through a “gutter effect” due to a road with inadequate or possibly occluded surface drainage.

The generated susceptibility map (Figure 9) shows a large concentration of areas in the highest class in Polcevera and Bisagno Valleys and in the small catchment areas of the western portion of the Municipality. The latter were developed in the proximity of the Municipality’s highest elevations, which exceed 1000 m above sea level at a very short distance from the coastline and thus in the presence of extremely high slope gradients. In addition, due to these morphometric characteristics, they have historically been affected by higher and more intense rainfall values than the eastern parts of the municipal territory.

More generally, the susceptibility classes and their spatialization can also be ascribed to the different lithologies’ differing sensitivities to fast-developing landslides. These lithologies make up the geological bedrock and have less cover, including terraces. The high slope gradient also results in high erosive power, preventing or significantly reducing debris and soil accumulations.

The predictive accuracy of the map was quantitatively computed using the Receiver Operating Characteristic curve (ROC) and the Area Under the Curve (AUC) [36,73,74]. Then, the true positive rate (TPR), i.e., the correctly predicted events, was plotted opposite to the false positive rate (FPR), i.e., falsely predicted events, by varying the cut-off value. The AUC value of the prediction curve was 0.79 (Figure 10), which is categorized in the good class, according to Sewts [75]. The computed map has good prediction accuracy for shallow landslides in the study area when considering the factors. The calibration process also made it possible to further verify the reliability of the susceptibility assessment model. The results, obtained by cross-referencing the events in the calibration set (Figure 6) with landslide susceptibility mapping, are shown in Table 6: 92 percent of the landslides in the calibration set fall in landslide susceptibility zones classes 4 and 5, and 73 percent fall in class 5. These results testify to the reliability of the developed model.

Figure 10. The susceptibility map ROC curve (dark blue).

Figure 10. The susceptibility map ROC curve (dark blue).

Table 6. A majority of the landslides in the calibration set, 92 per cent of the total of 97 landslides, fall into the two maximum susceptibility categories, testifying to the model’s reliability.

Susceptibility Class	Shallow Landslides—Calibration Set
1	0
2	3
3	5
4	18
5	71
Total	97

Table 6. A majority of the landslides in the calibration set, 92 per cent of the total of 97 landslides, fall into the two maximum susceptibility categories, testifying to the model’s reliability.

Susceptibility Class	Shallow Landslides—Calibration Set
1	0
2	3
3	5
4	18
5	71
Total	97

The obtained landslide susceptibility map was used to assess the hazard (H) to which the underground utilities are exposed in relation to the development of shallow landslides. Considering the association of vulnerability (V) to the pipe’s diameters defined in Table 4 and the following definition of risk classes in Figure 8, the network’s risk maps are presented in Figure 11 according to

$$R=H\ast V$$

(3)

Figure 11 presents the classification of the two networks in terms of risk for shallow landslide occurrence in classes from 1 to 5 and in terms of fall risk from 1f to 5f. The 0 class denotes the absence of the two risk factors. Network segmentation is based on homogeneous classes, and the results were topologically cleaned with dedicated QGIS algorithms. The detailed maps in Figure 12 indicate steep slopes and highlight the mixing and co-penetration between natural cover areas and scattered built-up areas: high-risk zones occur at the edge of areas of intermediate or even low risk. As already stated in Section 2.2, the present evaluation is to be considered preliminary as it is based solely on pipe diameters. A more thorough assessment will be possible if further data are available in the future, i.e., burial depth, pipe material, and age.

In addition, and only for illustrative purposes, runoff accumulation was added to Figure 12. Its evaluation was computed through the flow accumulation algorithm (SAGA GIS) applied to the DTM: this parameter represents the potential triggering factor for fast-developing landslides, and its spatial analysis in relation to the landslide’s susceptibility enables a further assessment of shallow landslides’ potential risk. The combination of the two analyses helps to identify both a potential shallow landslide and its development path, allowing us to point out, through the subsequent spatial analysis, the potentially exposed elements.

Finally, in Figure 13, the percentage of gas and water networks in Genoa Municipality exposed to the risk is shown. The performed calculation illustrates that 12% of the total length of the water network and 11% of the total length of the gas network are classified into risk classes 4 and 5, and thus exposed to the highest risk. Considering Figure 11, these high-risk stretches are widely spread along the slopes where the networks expand from the urban sealed soil area to the sparse building area.

4. Discussion

The computed susceptibility map for rainfall-induced shallow landslides in the Municipality of Genoa, following a qualitative–quantitative approach, enabled us to perform a risk assessment for the water and gas networks. The descriptive statistics method, based on the evaluation and analysis of 485 rapidly developing surface landslide events recognized in the municipal territory, was combined with the AHP method, which is part of the heuristic techniques for evaluation and decision support. The methodology was applied by employing seven causal factors for the initiation of rain-induced surface landslides and using baseline and remote sensing survey data combined and analyzed in a GIS environment. The model’s predictive accuracy was validated thoroughly using ROC and AUC methods: the results confirm the reliability of the obtained results [73,74,75]. Moreover, a calibration process confirmed the consistency: 92% of the calibration set, which includes 97 landslides, belong to the two highest susceptibility classes.

The result was subsequently cross-referenced with the underground utility network vulnerability, as defined in terms of the pipe diameter, in order to evaluate the risk and provide an early prevention tool. This first assessment can be further refined and developed by adding the analysis of runoff water flow accumulation. Including the triggering factor of water concentration would enable periodic on-site checks in the most critical areas and then taking preventive measures. The approach is presented in detail in Figure 12. At the present research stage, flow accumulation data are shown with an illustrative aim and do not conform to the overall methodology.

The availability of detailed information on the characteristics of the networks, and especially on how they are laid, may enable subsequent refinement of the risk assessment by assigning further differentiated vulnerability values. The potential risk related to the wall’s collapse or rockfall in urban areas, where shallow landslide hazard is considered absent, was included in the risk evaluation in order to provide a more complete evaluation of the risk the two networks are exposed to.

Furthermore, it is necessary to highlight how the susceptibility map for the fast-developing shallow landslides was obtained based on the events that occurred in the municipal territory over the past 2–3 decades [46,52,53]. Although their number represents a robust statistical basis, it should be emphasized that their occurrence is determined by intense rainfall, which in the area and for the short-duration intervals presents the maximum at the national and European scale, with 180 mm/1 h recorded in 2011 [82]. Such high-intensity phenomena exhibit strong spatial localization; it is possible that the areas where the landslide phenomena have developed less frequently (Figure 6) have not yet been affected by extremely high rainfall intensities, but they could be in the future. Moreover, some areas in the western sector and along both Polcevera and Bisagno Basins are more prone to developing shallow landslides because of the peculiar geological and geomorphological features.

The presented approach to the vulnerability assessment and then the following risk evaluation represents an evolution of previous research activities dedicated to hazard assessment [7,8,19,26]: in particular, the weights of all causal factors were estimated through statistics of recent phenomena, thanks to the robust available inventory. Further research developments may additionally include investigating both these aspects and including more descriptive parameters of the utility network’s vulnerability to risk.

The performed analysis enables planning and designing a possible mitigation strategy. First, the distribution of the two networks in risk classes constitutes a priority scale of attention along very long networks. Then, depending on the local situation, and considering the shallow nature of the considered landslides, the necessary prevention measures may include NBS—Nature Based Solution—techniques [83,84,85,86,87] and bioengineering measures to stabilize slopes and reinforce terraces. These measures are successfully used to improve slopes’ stability, particularly in the presence of abandoned man-made terraces, which represent a possible source of risk [78,79,80,88,89]. NBS prevention measures also present other co-benefits: (i) the small and low-cost measures may be widely adopted throughout the Municipality with proper planning of resources and interventions; (ii) the short extension of the measures allows overcoming access issues that may arise in some steep or critical areas.

Finally, the importance of a prevention approach is increasingly necessary and urgent, considering the rainfall trend in the study area, which sees a reduction in rainy days in the face of annual averages that remain substantially unchanged [90]. Figure 14 shows the annual maximum rainfall intensities for 1 h, 3 h, and 6 h registered between 1960 and 2024 by three rain stations, whose localization is shown in Figure 1. The absolute values are relevant for all the time intervals and for all the stations, underlining the high hazard for pluvial-induced phenomena in the area.

Table 7. Descriptive statistics of hourly rainfall at Fiorino (FIO 1960–2024), Ponte Carrega (PCO 1960–2024), and Crocetta d’Orero (CRO 1961–2024) rain gauges.

Statistics	FIO			PCA			CRO
	1 h	3 h	6 h	1 h	3 h	6 h	1 h	3 h	6 h
Mean	59.1	95.0	116.5	48.9	75.9	96.8	42.1	66.3	86.9
Median	54.2	80.5	94.7	43.0	66.4	83.0	38.4	55.1	73.7
Maximum	126.0	269.0	332.0	124.2	230.0	283.6	88.0	214.0	267.0
Minimum	14.8	31	38.8	17.0	24.6	35.0	15.2	29.6	42.0

shows the statistics for the intervals and for the stations: the maximum exceeds 120 mm/1 h, 260 mm/3 h, and 330 mm/6 h. The trends seem to increase in the 64-year period:

Table 8. Mann–Kendall trend test results at 95% confidence for hourly rainfall series at Fiorino (FIO 1960–2024), Ponte Carrega (PCO 1960–2024), and Crocetta d’Orero (CRO 1961–2024) rain gauges.

Statistics	FIO			PCA			CRO
	1 h	3 h	6 h	1 h	3 h	6 h	1 h	3 h	6 h
z-stat	2.40470	1.73823	1.54695	−0.12457	−0.283092	−0.192505	1.92901	−0.04445	1.29781
p-value	0.01619	0.08217	0.12188	0.90086	0.777101	0.847347	0.05373	0.96455	0.19435
Trend	yes	no	no	no	no	no	no	no	no
Sen’ slope	0.44721	0.42311	0.39500	−0.01847	−0.051355	−0.03333	0.31623	−0.01456	0.34170
Lower	0.09268	−0.07143	−0.13600	−0.31111	−0.426229	−0.50345	−0.01053	−0.45000	−0.23125
Upper	0.82857	0.98571	1.05625	0.28	0.333333	0.51429	0.70769	0.50625	0.94595

shows the Mann–Kendall trend test performed at 95% significance and Sen’s slope trend estimation with the relative confidence intervals. The analysis shows that the 1 h yearly maximum intensity increase is statistically significant for the Fiorino station, which is located in the western portion of the Municipality (Figure 1). All the other diagrams present an increasing trend, even if not statistically significant. Finally, these results are coherent with other studies conducted on extreme pluvial events’ meteorological proxies [91]. These results, considering that the impact of intense rainfalls is the triggering factor, suggest that the western areas could experience an increase in shallow landslides in the future.

5. Conclusions

This research assessed shallow landslide risk for the two water and gas distribution networks in Genoa Municipality. The assessment was conducted by evaluating the hazard according to a methodology previously developed in a contiguous area [17,55], which was modified and adapted to the territory’s features.

It must be underlined that, compared to previous research, the conditioning factor set was modified, reduced in number, and expanded by introducing new factors related to the morphometric features. The conditioning factor number reduction, obtained by including all anthropogenic features within the land use factor, enables a more agile model application without losing reliability, as indicated by the AUC prediction rate (0.79) and the calibration process results. Moreover, a crucial factor, related to the presence of man-made terraces that indicate human modification of the landscape, was maintained, as their presence was precisely identified using methods from previous research and included in the slope deposit factor. The availability of a recent and precise LiDAR-derived 1 m resolution DTM enabled accurate evaluation of the presence of terraces and landscape morphometric features.

The increasing instability of the slopes due to shallow landslides and the effects on the peri-urban areas cause increasing damage every year and often involve water and gas distribution networks. This effect is related to the increasing intensity of rainfalls in the study area, as underlined by the 95% statistically significant increase in the 1 h duration maximum intensity at one meteorological station in the studied area.

These results underline the necessity of adopting prevention measures to mitigate the risk of ground effects induced by high-intensity rainfall for the two vulnerable networks, which represent critical exposed elements. About 9.8% and 6.8% of the total length of water and gas pipelines are exposed to class 4 and 5 risk. Further on-site investigations should be performed in these areas, and subsequent mitigation measures should be adopted. The developed risk assessment tool represents the basis of a municipal-scale decision support tool for planning and adopting a series of dedicated prevention measures. Further development of the obtained results may include the integration of other networks’ technical data for additional vulnerability assessment. Finally, the methodology may be adapted to different geographic/geologic contexts by changing or adding the causative factors in relation to the local peculiarities. Finally, a preliminary assessment of the risk related to wall collapse in urban areas was included for both investigated pipelines.