*2.4. Methodology*

Italian regulation (D. P. C. M. 27/12/1998) dictates that environmental assessment should be performed by subdividing the environment into environmental components, each of them described and characterized by indicators, which are parameters used to describe a given phenomenon and that should have the following characteristics: being concise, easy to understand and easy to measure and update.

In this framework, the objective of this study is to propose a set of indictors at a national scale to characterize landslide risk by depicting how urban expansion interferes with geomorphological slope processes. To this end, we started with some input data that consists of the outputs of bigger ongoing or concluded research activities, and we combined them by means of GIS analyses.

Input data are:


In short, the procedure consists of identifying a landslide risk following a revised and simplified version of Equation (1). For our purposes, hazard is considered equal to the spatial probability of occurrence (thus, equal to susceptibility). Over the susceptibility we superimpose the spatial distribution of anthropic elements (depicted by the soil sealing map), in order to consider elements at risk only on a presence/absence basis. Vulnerability is neglected (mathematically it is considered equal to 1 in Equation (1)) for different reasons: first, it would be nearly impossible to assess separately the physical vulnerability of each element (e.g., buildings) at national scale (and, to our knowledge this is a still unattempted task); and second, the soil sealing map does not effectively allow for distinguishing between different typologies of buildings or infrastructure. Moreover, in national scale studies, the approach of considering vulnerability as equal to 1 (the maximum possible degree) is

considered a viable and cautionary approach [12]. The resulting index is then aggregated at the municipality basis.

The first step of the proposed procedure consists of blending the three susceptibility maps into a single information. It can be considered quite unlikely that, in a single spatial unit of the susceptibility map (pixel with 50 m size), two or more landslides of different typologies could be contemporarily present. Indeed, every predictive landslide model should first make a typological prediction, trying to predict what kind of landslide will take place [55]. As a consequence, the three susceptibility maps were imported into ArcGis software, and the "cell statistics" operation was performed to assess the "maximum" value. In this way, the output is a raster map in which the susceptibility index associated to each cell is the highest value found in the three input maps. This is equivalent to considering the landslide type with the highest susceptible value as the most probable to occur in a given location, and surmising that this landslide typology is the one that will be most likely affecting that area, controlling the related hazard. The resulting raster will be called "hazard index map" henceforth (Figure 2).

Before overlying the soil sealing map to the hazard index map, a procedure of homogenization is needed as the two raster maps have different cell sizes. Using ArcGis "block statistics" function, the resolution of the soil sealing raster was changed from 10 m to 50 m. Despite the loss of spatial resolution, this operation was necessary for the perfect match of soil sealing map with landslide hazard map, and some authors demonstrated that this spatial resolution is a good compromise in wide-area landslide hazard assessment studies [49]. The "minimum" statistics type was used: in this way the resulting raster obtains the value 1 (sealed soil) if at least one 10 m cell of sealed soil is present in each 50 m block. This choice determines a small expansion of the sealed soil that could be considered precautionary, and that is more desirable than alternate approaches. For instance, we verified that using the "majority" operator, many small infrastructure are completely neglected (e.g., roads cutting rural or mountain areas usually represent a small fraction of the 10 m pixels inside the 50 m block, and a relevant source of landslide risk would be completely ignored). In addition, it should be noted that the original soil sealing map represents the presence of sealed soil, but it is widely acknowledged [19] that the effects of the sealing may extend also to the surrounding areas (e.g., concerning hillslope hydrology, small surficial drainage systems connected to infrastructure could have discharge outlets a few meters away from the sealed area).

The resulting raster was reclassified, assuming a value of 1 in soil-sealed 50 m pixels and "no data" elsewhere. From a mathematical point of view, the reclassified soil sealing map and the hazard index raster were combined with a multiplication by means of the "raster calculator" tool of ArcMap. From the point of view of spatial information, the values "1" and "no data" in a multiplication act as a filter that maintains unaltered the input value of spatial probability of occurrence only in correspondence of anthropic elements, while far from them the index is not defined (conceptually, it is similar to assuming a risk equal to zero). This output raster was named Landslide Risk Index (LRI), because it accounts for the interaction between hazard and anthropic elements, giving a spatially distributed picture of how much they are exposed to landslide risk (Figure 3). It should be observed that a thorough assessment of the interaction between landslides and elements at risk would require accounting for the propagation of mass movements (for which run-out models would be necessary). This element is rarely encompassed in landslide susceptibility assessments, especially in wide-area applications; this shortcoming will be further investigated in the discussion of the results.

**Figure 3.** (**a**) Landslide Risk Index (LRI) map for the whole Italian territory; (**b**) Focus on hazard index map; (**c**) Focus on LRI map. Roads and buildings are from OpenStreetMap dataset.

LRI ranges from 0 to 100 and represents a spatially distributed indicator, which can be considered a basic element to be aggregated over larger spatial units in order to characterize them with respect to landslide risk. In this work, we derived from LRI two more indexes at municipal scale. The LRI raster and the shapefile of the borders of the Italian municipalities were overlaid in ArcMap and a "zonal statistics" was performed twice, using "mean" and "sum" to characterize each municipality with respect to two indexes named Average Landslide Risk (ALR) and Total Landslide Risk (TLR), respectively. The outcome of this operation represents the last step of the proposed procedure: the resulting indexes and a discussion about their interpretation are contained in the next section.

#### **3. Results and Discussion**

The TLR index (Figure 4) expresses for each mapping unit (municipalities in this study) the sum of the susceptibility values of all the cells with urbanized soil. Basically, this index cumulates for each administration the situations of interaction between spatial hazard and urbanized areas, expressing how much the development of the municipality has let hazardous areas to be "invaded" by constructions, infrastructure and services. In this regard, TLR could be used to describe the attention of an administration to harmonize the urban development with the main geomorphological hazard affecting its territory. Figure 4 shows that the Italian areas characterized by the highest TLR values are the Apennines (mainly the northern and central sectors), the isle of Sicily and, to a lesser extent, the eastern Alps. The drawback of this index is that it is sensitive to the extension of each aggregation unit: large municipalities have a greater chance than small ones to have a high TLR value, because of the higher number of pixels. For this reason, the value of the index does not have a fixed upper limit, and the value could theoretically tend to infinite, requiring particular attention for a correct interpretation. Indeed, when comparing different municipalities, a similarly high value of the index could be determined by many pixels with mid LRI values or by fewer pixels with higher LRI values. For this reason, the municipalities with the higher TLR index are large and densely urbanized municipalities. This result is not an artifact or a bias: the index effectively describes a recurring situation in some of the largest and most densely urbanized municipalities, which are exposed to a very high landslide risk in their territory because, during their urban expansion, they have had to cope with more hazardous areas than small municipalities. The highest values are found in the cities of Rome and Genova (both characterized by a very wide territory, densely populated and almost completely urbanized), and in the municipalities of Perugia, Gubbio and Messina, which are less populated but still have large portions of territory urbanized in hazardous areas (Figure 4). Nevertheless, TLR seems effective in highlighting the municipalities most affected by landslide risk, as the aforementioned territories correspond to areas where news about landslides continuously appear in newspapers and online blogs, as reported by [32]. In the last ten years, 4% of the landslide news catalogued and geotagged by their semantic engine is located in the aforementioned five municipalities with higher TLR values: in particular, 600 online news providers talked about landslides in Genova, 533 in Rome, and 235 in Messina.

Our results are further supported by the governmental data coming from ItaliaSicura web platform (http://mappa.italiasicura.gov.it/ last accessed on 31 May 2021), which collects the number of interventions and the economic resources allocated to mitigate hydrogeological risk in Italy. Rome is the Italian municipality with the highest number of interventions (64), likewise Genova has the highest total cost (about 378 m €) (however, it should be noted that data also include interventions for flood risk mitigation).

**Figure 4.** Characterization of the Italian municipalities with the Total Landslide Risk (TLR) index.

ALR index characterizes each municipality with the mean value of hazard found in correspondence of anthropic elements (Figure 5a). This index expresses, for each municipality, how hazardous is the portion of the territory where buildings, infrastructure and other services have been located. The values of the index range from 0 (minimum value) to 100 (maximum value): low values mean that the local administration has been cautious in planning urban development avoiding landslide risk, while high values are associated with municipalities where a consistent percentage of the urban structure has been built in hazardous areas, resulting in a relevant level of risk. It should be stressed that this does not necessarily mean that urban expansion has been recklessly planned: landslide hazard is so widespread in Italy that sometimes a municipality could be almost entirely interested by a relevant level of hazard posed by landslides or other geohazards (e.g., flood or volcanic activity). Nevertheless, also in such cases, ALR is an indicator that can be used to highlight situations where landslide risk is a very serious issue and should be carefully evaluated before further planning activities, or in the perspective of considering mitigation strategies. From a mathematical point of view, the value of ALR is independent from the areal extension of each municipality. However, a close investigation on the distribution of the values (Figure 5b) reveals that the highest values are found in small municipalities, most of them renowned international holiday destinations located by the sea, in rocky coasts (Positano, Amalfi, Capri, and Portofino, to name a few). We do not consider this outcome as a bias, and we explain it with a concurrence of factors of different nature. Firstly, in correspondence of many rocky and high-cliff coasts, the susceptibility to rockfalls presents very high values. Secondly, the territory of these municipalities is very steep and traditionally managed with the terracing method. This could be an effective method to cope with landslide hazard, but several studies highlighted that currently the loss of farmed land and the lack of maintenance seem to have recently increased the landslide hazard in

these areas [56–58]. Thirdly, in the touristic locations with very high real estate value, the building of houses, accommodation facilities, infrastructure and services has been more intense than elsewhere. It has been driven mainly by market law and, especially in the last decades of the last century, not adequately counterbalanced by countermeasures concerning landslide hazard or environmental protection. This effect is particularly exacerbated in small municipalities, because the territory that can be used for urban expansion is limited and causes a severe competition between economic interests (urban expansion to support tourism and investments on the real estate market) and geomorphological processes. This is particularly alarming because small municipalities usually have scarce resources (both in terms of funds and manpower) to effectively face emergencies or to manage in-house risk mitigation strategies.

**Figure 5.** (**a**) Characterization of the Italian municipalities with the Average Landslide Risk (ALR) index; (**b**) Focus on the Amalfi Coast, where seven municipalities are ranked among the 10 Italian municipalities with the highest ALR value.

Our findings are in accordance with the evidence resulting from the governmental WebGIS platform presented by [12]: in most of the high-ALR municipalities highlighted in Figure 5 (especially Positano, Amalfi, Conca dei Marini), a high number of buildings are built in areas classified as landslide hazard areas according to Italian laws. For example, at least 90% of Amalfi buildings are located in hazardous areas.

The combined use of ALR and TLR indexes can be useful in gaining preliminary insights on the landslide risk of municipalities. Starting from the LRI index, which is defined at the pixel level, the same principle could be applied to other spatial units and ALR and TLR could be calculated for administrative subdivision of different level (e.g., provinces or districts) or for geographical areas (e.g., basins). It should be stressed that the proposed indexes are environmental indicators and, by definition, are conceived to simplify a complex phenomenon to aid an easy understanding also for non-experts. As a consequence, we acknowledge that the proposed indexes are an oversimplification of reality and cannot substitute a thorough quantitative risk assessment. The main utility of the indexes lies in the fact that a nation-wide quantitative landslide risk assessment is still far from being

accomplished for Italy; thus, the proposed indexes can be used to explain, at scales ranging from the local to the national, the severity of the phenomenon, and to evaluate how the administrations have dealt with landslide hazards when planning urban expansion and associated services.

One of the most important requirements for indicators is the possibility to be easily updated. Concerning LRI, the updating procedure can be accomplished in GIS environment whenever updated input data (susceptibility and soil sealing maps) are available. Soil sealing is a dynamic anthropogenic process, and an updated nation-wide map is officially released every year, thus allowing for a yearly update of LRI to account for variations in urban expansion. Conversely, susceptibility is traditionally considered a quasi-static element and a constant update of this element is not expected. However, the index could be updated if a nation-wide susceptibility map is released and deemed more accurate than the one used in this work. E.g., a susceptibility map considering also the runout of landslides would be particularly indicated to thoroughly consider the interactions between hillslope processes and elements at risk. Indeed, we acknowledge that one of the main limitations of the present work is the absence of a method to explicitly include the landslide runout in the model. Unfortunately, complex modeling techniques are required to assess the post-failure displacement of landslides [59,60] and the travel distance is correlated to lithological and morphological factors [61]. For these reasons, a model accounting for landslide runout at the scale of this work (3\*10<sup>6</sup> km2) has not been proposed yet; even the latest attempts to include landslide runout in susceptibility assessments are limited to few case studies with limited extension [62,63].

Once LRI is updated, the derivation of TLR and ALR at municipal level can be also accomplished easily in a GIS system. This procedure could be carried out using the last update of the shapefile representing the Italian municipalities, which is also updated every year to account for small variations mainly consisting of the merging of very small and scarcely populated municipalities.
