**3. Results**

This section is organized as follows. The first and second subsections show the spatial framework of, respectively, landslide hazard and the LEAC groups across the study area. The following subsection presents the outcomes of the estimate of regression model (1) implemented into the spatial context identified in Section 2.2.

#### *3.1. Landslide Hazard in the Study Area*

As Table 3 and Figure 4, panel A, show, in the vast majority of CMT (i.e., 4476.42 km2, or 80.29% of the CMT land mass) the hazard level was assessed as null by the PAI, while around a fifth of the subdistrict is prone to landslides, mostly of medium (580.01 km2, or 10.40% of the CMT surface) or high severity (371.84 km2, or 6.67%); a very small share of the CMT features moderate landslide hazard (107.15 km2, or 1.92%) and a negligible one is characterized by very high hazard levels (39.80 km2, i.e., 0.71%). As for the 300 m grid, LH is greater than zero in 30,775 out of the total 62,231 300 m grid cells (Figure 4, panel B); hence, in nearly half of the cells, landslide hazard, of whichever level, affects a certain share of the cell.

**Table 3.** Landslide hazard levels in the Coghinas-Mannu-Temo (CMT) subdistrict.


**Figure 4.** Landslide hazard levels as assessed by the regional PAI in the Coghinas-Mannu-Temo subdistrict (panel **A**), and spatial distribution of the LH variable in the 300 m grid used in this study (panel **B**).

#### *3.2. The Spatial Framework of the LEAC Groups*

Three LEAC groups prevail in the CMT subdistrict, as shown in Table 4 and Figure 5, panel A: arable land and permanent crops (32.09%); pastures and mosaic farmland (26.68%); natural grasslands, sclerophyllous vegetation and heathlands (23.19%). Together, they make up 81.96% of the study area. Next come standing forests (14.98%), while artificialized land amounts to 2.37% of the study area, and a negligible share (0.69%) is that of waters, which are not listed in Table 4 because they were not relevant for this study.

**Table 4.** LEAC groups as share of the Coghinas-Mannu-Temo subdistrict.


**Figure 5.** Spatial distribution of land covers in CMT classed through the LEAC groups (panel **A**), and share of each LEAC group within the 300 m grid used in this study (panels **B**–**F**).

Panels B-F in Figure 5 show the spatial layout of the share of each LEAC group in the 300 m grid cells used within this study. Cells having non-null values of L\_TAKE form small and spatially disarticulated bundles. Cells where a share of arable land and permanent crops (ARA) is present cluster especially along the main plains; however, they are spread across the subdistrict, except for the Asinara Island to the north and the mountain areas that delineate the borders of the watersheds. In the latter, clusters of cells hosting standing forests (FOR) are clearly visible in the map, while the Asinara Island is a hotspot for natural grasslands, sclerophyllous vegetation, and heathlands (GRSH), which also feature along the rugged western coast and are scattered across CMT. Finally, cells hosting pasture and mosaic farmland (PMF) are diffuse across the subdistrict, with the larger assemblage along the Marghine mountain chain to the southern border.

#### *3.3. The Outcomes of the Regression Model*

The strength and significance of correlations between the explanatory variables in model (1) were preliminarily assessed through the Pearson product–moment correlation coefficient; the outcomes of this assessment, which was carried out on the attribute table of the shapefile containing the 30,775 cells having non-null values of LH, are provided in Appendix A, Table A1. The strongest correlation is that between PMF and GRSH (−0.4033, *p* < 0.01), while |r| < 0.4 for the remaining couples of variables. The lack of strong correlations between the explanatory variables highlights the absence of issues of multicollinearity in model (1).

The estimates of the coefficients of DEPOL and VOLSE are significant and show the expected signs, since comparatively higher values of LH are associated with the incoherent and loose substrates that characterize quaternary deposits, and comparatively lower LH values are correlated with the solid and resistant substrates that feature volcanic sedimentary rocks.

Moreover, lower altitudes are associated with higher landslide hazard, and a decrease of 100 m is correlated with an increase of 1.8% in landslide hazard. This outcome may seem rather counterintuitive, since, in general, it is expected that landslide hazard increases with elevation, or the higher the altitude, the higher the probability that landslides may occur. The reason of this finding can be detected from the peculiar spatial taxonomy of landslide hazard in the study area, mapped in Figure 6, which shows the most relevant concentration of high-landslide-hazard cells in locations characterized by comparatively low and medium elevation.

**Figure 6.** Spatial overlay between historically recorded landslides, landslide hazard areas mapped in the PAI, and regional DTM.

Moreover, the spatially lagged variable shows a positive and significant sign, in terms of *p*-values, which shows an effective control of the spatial autocorrelation of the dependent variable.

That being so, since the estimates related to the control variables DEPOL, VOLSE, and ELEV are statistically significant and consistent with the expectations in terms of the expected signs, whereas the model offers an adequate control for spatial autocorrelation, the impacts of the LEAC covariates and, in particular, the influence of the land take variable on landslide hazard, identified by their estimated coefficients, are reliable and consequential.

The estimated coefficients of the five explanatory variables are significant at 1% and entail the following results, provided that everything else is equal.

Agricultural land, whether characterized by extensive or intensive production, is negatively associated with landslide hazard, showing comparatively low correlations, since, on average, a 10% increase in pastures and mosaic farmland or in arable land corresponds to a 0.7% decrease or to a 1% decrease in landslide hazard.

Positive correlations are shown by the coefficient of FOR, since 10% increases in FOR and GRSH are associated with 1.4% and 0.9% increases in LH, respectively. Increases in forests, natural grasslands, sclerophyllous vegetation, and heathlands are associated with higher values of LH, which entails that such land covers are likely to identify buffer zones with respect to areas characterized by relevant landslide hazard. The spatial contexts featured by these land covers are usually almost totally devoid of human settlements, which highlights a virtuous spatial organization, which aims at protecting urbanized areas from the negative impacts generated by landslides, by preserving natural forests and grasslands from land-taking processes.

All in all, crop production is not associated with increases in LH. On the other hand, forests are the LEAC group that reveals the most relevant positive correlation with landslide hazard, whereas natural grasslands, sclerophyllous vegetation, and heathlands are less relevant in terms of association with LH.

Finally, the regression model identifies the association of LH and L\_TAKE as a relevant positive correlation; namely, a 10% increase in L\_TAKE is associated with a 0.8% increase in LH. In other words, the higher the size of the land take-related covariate, the higher the size of areas characterized by relevant landslide hazard. This finding highlights that the spatial structure of the study area is characterized by artificialized areas intertwined with areas featured by relevant landslide hazard, or that land-taking processes have taken place in locations that should have been preserved free from urbanization processes due to the magnitude of landslide hazard. Table 5 reports the results of the estimate and relevant statistics of regression model (1).


**Table 5.** Estimate of regression model (1).

## **4. Discussion**

The mapping of landslide hazard in the study area is quite consistent with the taxonomies of similar spatial contexts described and discussed in the current literature. As described in Section 2, the CMT subdistrict features a hilly ground orography (the Marghine-

Goceano Chain and the Mount Limbara), with widespread uphill and downhill stretches, and by a limited coastal plain (the Nurra). As described in Section 3.1, just about onefifth of the study area is characterized by a more-or-less relevant landslide hazard which, nevertheless, has generated a relevant geological instability, demonstrated by nearly 400 events. Hilly spatial contexts intertwined with plain areas are often associated with limited zones characterized by relevant landslide hazard and by diffused geological instability, as demonstrated by the regional screening of landslide phenomena in the lowlands of Calabria, Southern Italy [60]. The European screening study by Jaedicke et al. [61] identifies the European hotspots concerning landslide hazard, based on the implementation of the models defined by the International Center for Geohazards (ICG) and the Joint Research Center of the European Commission (JRC), which are often located in hilly and plain areas, i.e., with morphological characteristics similar to the CTM subdistrict, sometimes featured by high levels of precipitation and seismic activity. The European screening is quite consistent with what was detected in the case of the landslide inventory implemented by Solís-Castillo et al. [62] as regards the Mexican tropical region of Sierra Costa, characterized by low precipitation rates and landslide hazard diffused over mountainous and plain zones. Analogous findings are shown, among many, in recent studies concerning the Freetown region in Sierra Leone [63], and the Whitsunday Region, located in North Queensland, Australia [64].

The mapping of the LEAC groups in the study area, featuring pastures and mosaic farmland, natural grasslands, sclerophyllous vegetation and heathlands, and arable land, is consistent with the spatial taxonomies reported in other studies concerning landslide hazard in hilly spatial contexts intertwined with plain areas. As in the case of the CTM subdistrict, important relations are identified between landslide hazard and farming production in hilly and plain zones by Rendon et al. [65], which are addressed by several policy tools, aimed at improving the quality of degraded ground and agrosystems, such as the Common agricultural policy [66], the strategy "Farm to Fork" [67] and the Biodiversity Strategy of the EU [68]. According to Borrelli et al. [69] and Panagos et al. [70], landslide hazard and related events in hilly spatial contexts intertwined with plain areas are mainly related to soil erosion phenomena, which should be addressed by increasing soil retention capacity [60], and the endowment of ecosystem services such as ground and superficial water resources quality and recharge, ground and underground biodiversity, and soil resilience to the impacts of climate change and of landslide events [71].

The outcomes of the regression model can be straightforwardly discussed in the theoretical and technical context of the current literature.

Negative correlations are associated with agricultural land, whether it is characterized by intensive or extensive crop farming. This is consistent with the results of several studies which relate the effectiveness of soil conservation practices based on agriculture. For example, Suci et al. [72] highlight the importance of crop farming and crop rotation in improving soil conservation conditions and landslide hazard mitigation in the rural area planning in the Indonesian Cidadap Subdistrict located in Western Java. Extensive and, wherever it is suitable, intensive crop farming are identified as effective approaches to recovering from scars generated by landslide-related events in Mount Elgon, Uganda [73], where such practices are implemented through the direct cooperation of local communities. The extensive mapping of rural areas' exposure to landslide hazard in Central Italy developed by Santangelo et al. [74] shows the association of extensive and intensive crop farming to low-hazard areas as well.

The mapping of the quoted study by Santangelo et al. is consistent with the regression outcomes related to the covariates that identify forests (FOR), and natural grasslands, sclerophyllous vegetation, and heathlands (GRSH). Since the latter two LEAC groups characterize non-urbanized areas, it has to be put in evidence that a virtuous approach to land use planning brings together Central Italy and the Sardinian CMT subdistrict, since areas with relevant landslide hazards have been kept almost totally settlement-free.

The outcomes related to the DEPOQ covariate are consistent with the studies by Sasaky and Sugai concerning the Hachimantai region located in Northeastern Japan [75], and by Akumu et al. [76], where significant landslide hazard is correlated to inland wetlands, whereas coastal wetlands are associated with low LH, which brings together the CMT and the Central Italy coastal wetlands, as characterized by Santangelo et al. [74].

As for forests, woodlands, and shrubs, not only do they almost totally feature the non-urbanized areas of the CMT district, but also they act as spatial contexts whose managemen<sup>t</sup> is crucial to implement planning policies aimed at decreasing the environmental risk associated with landslide hazard. The association of these LEAC groups with areas characterized by landslide hazard is consistent with their environmental protection function. This issue is widely addressed in the current literature. The enhancement and strengthening of forests and woodlands is basically related to the protection of primary forests, to forest recovery activities, to sustainable managemen<sup>t</sup> of forests and woodlands, and to tree planting in spatial contexts characterized by different prevailing ecosystems, such as urban and agricultural areas, where these LEAC groups play a decisive role in mitigating the impact of landslide hazard [77]. FAO identifies forest sustainable managemen<sup>t</sup> as the most important operational category to enhance economic, social, and environmental quality of rural areas, mainly because of its impact on improvement of crop production and productivity connected to protection from flood and landslide effects [78]. Forest and woodlands' recovery and new arboreal plantations are particularly relevant for the definition and implementation of spatial planning policies since the assessment of their economic impact in terms of mitigation and adaptation to climate change is generally recognized as particularly effective in the medium and long runs, especially due to decrease in flood and landslide risk [79,80], as well as for biodiversity protection and enhancement [81–83].

Finally, the regression model shows that landslide hazard is associated with land take in significant and quantitatively relevant terms in the CMT subdistrict. This finding is supported by the fact that the other results of the regression are consistent with the outcomes of several studies available in the current literature, which implies that the definition and implementation of spatial planning policies aimed at addressing landslide hazard in the study area are almost entirely an issue of countering the ongoing land-taking processes and of deurbanizing at least a part of the areas located in landslide-prone zones. This is a key issue in the current scientific and technical debate (among many, [84–87]), and it is widely discussed in the fifth section.

#### **5. Policy Implications**

The results show a positive correlation between coefficient of FOR and landslide hazard due to virtuous spatial organization aimed at protecting urbanized areas and at preserving natural forests and grassland from land-taking processes. Forests have positive effects on reducing impacts of landslide. In shallow soils, deep-rooted trees and shrubs may reduce the occurrence of rapid landslide [88] by anchoring and stabilizing superficial soil layer to more sound substrates [89]. Moreover, trees may represent a physical barrier to contrast rocks and debris falls [90]. Therefore, spatial planning policies concerning forest and woodland recovery and plantations of deep-rooted trees and shrubs are particularly significant in order to decrease landslide hazard. From this perspective, two main policy implications can be identified as follows. First, forest managemen<sup>t</sup> should consider the potential of forests and woodlands for landslide protection by restoring and protecting natural forests [91] and by maintaining forest cover. Health and vitality of forests are two key factors to reduce landslide hazard by strengthening rooting systems of tree in relation to climate change [89]. Secondly, spatial planning should localize forest in high-risk areas in order to support a virtuous spatial organization that locates human settlements and activities in zones characterized by low levels of landslide hazard [89].

The most prominent result is, however, the significant positive correlation concerning L\_TAKE; therefore, as far as land covers and their effects on landslide hazard are concerned, controlling land-taking processes is the main road to mitigating the hazard. On this premise, three main groups of policy implications, respectively concerning land densification, land recycling, and strategic environmental assessment, can be identified as follows.

At the international level, governments are using different measures to reduce landtaking processes, such as policy targets [3], financial or fiscal incentives, and environmental assessment of spatial plan and projects [92]. On the other hand, as shown in the introduction, land-taking processes are steadily increasing; therefore, further measures are necessary in order to achieve the EU goal of "no net land take by 2050". According to the EU Soil Strategy for 2030, Member States should integrate the actions defined in the "land take hierarchy"; that is, avoid, reuse, minimize, and compensate, into urban greening plans, and promote the reuse and the recycling of land and high-quality urban soil [3]. In particular, land recycling is defined as "the reuse of abandoned, vacant or underused land for redevelopment. It includes 'grey recycling' and 'green recycling'. Grey recycling is when 'grey' urban objects, such as buildings or transport infrastructures, are built under redevelopment. Green recycling is when 'green' urban objects, such as green urban areas or sport facilities, are built" [93]. Moreover, land recycling includes three components: gray land recycling, green land recycling, and land densification.

As for land densification, it implies that land is developed within existing settlements so as to take advantage of existing infrastructure without using undeveloped land [93]. Therefore, national and regional administrations should promote land recycling strategies within regional plans, to be further downscaled at the local level through municipal masterplans where new development should be allowed only if its impacts on land take are negligible. Moreover, regional strategies should promote a compact urban model based on the land densification concept to reduce demand for undeveloped areas. However, this should not be regarded as a "one size fits all" solution, as local specificities need to be taken into account. Indeed, such measures have been found to be particularly effective in developed countries, whereas in Latin America and Sub-Saharan Africa, as well as in already hyper-dense Asian megacities, further urban densification can bring about negative effects in terms of inequality of spatial distribution and social conditions of the local communities [94].

With regard to land recycling, this is mainly supported through financial and regulatory tools. Concerning the former, financial support through publicly funded programs [95] and subsidies generated through impact fees, soil sealing fees, or improvement levies [96] are among the most common tools to promote land recycling. However, national and regional governments should diversify the set of fiscal tools that usher in creative and innovative ways to manage land uses effectively and efficiently. For example, the transfer of development rights can be used to direct development towards already taken up and well-serviced areas, rather than towards greenfield areas that are poorly connected in terms of transport, infrastructures, facilities, and services. As for regulatory tools, such as zoning schemes and land use regulations, these could successfully promote the participation of the private sector within land recycling projects. Furthermore, flexible and performance-based zoning regulations could be adopted within municipal masterplans [97]: these should pursue strict limits and constraints on land take, while allowing land uses that do not result in artificial land, hence in turn promoting mixed land uses where different functions coexist. However, because in Italy land use plans are drafted and approved by local municipalities independently of each other, monitoring and evaluating the provisions of existing municipal land use plans is necessary in order to understand the cumulative effects of land use regulations in adjacent cities and towns, and their consequences on land-taking processes. In this regard, in Sardinia, the regional administration can play a key role, since the regional planning office actively participates in the approval processes of regional and local plans with a view to ensuring their compliance with both regional planning laws and the regional landscape plan [98]. Due to the complexity of interests at stake, measures concerning land-take prevention and limitation, be they finance-based or regulation-based, call for active involvement of local communities and municipalities [99], and for effective vertical and horizontal cooperation between governments and other public bodies [100].

A third important policy implication concerns strategic environmental assessment (SEA), an appraisal planning tool that is mandatory in EU countries, which ensures that environmental considerations and sustainable-oriented goals are integrated into planmaking processes by assessing their likely effects on the environment, by considering reasonable and more sustainable alternatives, and by taking into account the mutual relations between the environment and the economic and social sectors [100,101]. Through the comparison of alternative land uses, the identification of areas that are more suitable for certain uses, and the evaluation of cumulative, direct, and indirect impacts of land-based investments, SEA can pave the way for the integration, within spatial plans, of measures aimed at preventing or minimizing land take.

In areas prone to landslide hazard (as well as in areas prone to flood hazard, which are not the object of this study), in Sardinia the PAI maps serve as a spatial reference for the PAI regulations, which restrict land uses and prevent land transformations depending on the magnitude of the hazard. In this way, the PAI provides a legally binding framework for municipal masterplans, whose zoning choices must comply with the PAI regulations, contrary to what has been reported in other countries [102], where new development in landslide hazard areas is not prohibited [103]. Hence, this higher-level regional planning tool contributes to limiting land-taking processes in fragile areas, while also providing relevant spatial information to planners in charge of drafting land use plans and appraising them through the SEA. Moreover, because the PAI maps are publicly available through the regional geoportal, they also contribute to raising local authorities' and local communities' awareness of landslide hazard and, by doing so, to granting transparency and legitimacy to restrictions that otherwise would be, in principle, quite conflictual.
