Sediment Connectivity in Human-Impacted vs. Natural Conditions: A Case Study in a Landslide-Affected Catchment

Abstract

This research aims to characterize sediment dynamics in the Rupinaro catchment, a uniquely terraced and human-shaped basin in Italy’s Liguria region, employing geomorphometric methods to unravel sediment connectivity in a landscape vulnerable to shallow landslides. Within a scenario-based approach, we utilized high-resolution LiDAR-derived digital terrain models (DTMs) to calculate the Connectivity Index, comparing sediment dynamics between the original terraced landscape and a virtual natural scenario. To reconstruct a pristine slope morphology, we applied a topographic roughness-based skeletonization algorithm that simplifies terraces into linear features to simulate natural hillslope conditions and remove anthropogenic structures. The analysis was carried out considering diverse targets (e.g., hydrographic networks, road networks) and the effect of land use. The results reveal significant differences in sediment connectivity between the anthropogenic and natural morphologies, with implications for erosion and landslide susceptibility. The findings reveal that sediment connectivity is moderately higher in the scenario without terraces, indicating that terraces function as effective barriers to sediment transfer. This highlights their potential role in mitigating landslide susceptibility on steep slopes. Additionally, the results show that roads exert a stronger influence on the Connectivity Index, significantly altering flow paths. These modifications appear to contribute to increased landslide susceptibility in adjacent areas, as reflected by the higher observed landslide density within the study region.

1. Introduction

Understanding the spatial variability of sediment dynamics is crucial for identifying unstable areas that act as active sediment sources across diverse environments [1,2,3]. These sediment sources often provide the material mobilized during debris flows or shallow landslides [4,5].

Sediment connectivity represents a recognized system property, describing the ability of sediments to move through a geomorphic system itself [6,7,8]. Analyzing connectivity helps establish relationships between sediment sources and potentially impacted areas downstream within the system [9,10].

A fundamental distinction is often drawn between structural connectivity, which describes the physical arrangement of landscape features (e.g., topography, land cover, terraces), and functional connectivity, which pertains to the actual movement of materials governed by hydrological processes [7].

Several factors influence sediment connectivity. Key drivers include the following: (i) the morphological complexity of the catchment, such as its relief, stream network density, shape, soil type, and surface roughness [9,11,12]; (ii) the spatial distribution of vegetation, which affects erosion and transport processes [13,14]; and (iii) human-induced changes to the landscape, such as modifications in land use, drainage systems, and road networks [15]. The authors of [16] investigated the effect of anthropogenic modification on the landscape, including changes in land use and road presence, on sediment connectivity, as well as its relationship with shallow landslides triggered by a severe rainfall event.

It is important to study the spatial variability of sediment dynamics, which aids in identifying unstable areas that act as active sediment sources for potential processes, such as debris flow and shallow landslides, in diverse environmental settings [2,3,17].

One of the main factors that drives the temporal variability of sediment dynamics is anthropogenic modification, such as terraces, often constructed for agricultural purposes, which are common in many regions worldwide [18].

In the Liguria region in northwestern Italy, the areas with terraced landscapes are often affected by severe storm events that trigger widespread shallow landslides, such as in the Cinque Terre area in October 2011 [18], in the central-eastern sector in 2014 [19], and in the western sectors in 2016 [19].

Terrain reconstruction techniques have gained increasing attention for their value in simulating pre-anthropogenic surface conditions, enabling a more accurate analysis of the topographic processes uninfluenced by human modifications. Llena et al. [20] proposed a terrain smoothing approach to eliminate terraces from DTMs, allowing for the reconstruction of natural surface morphology and improving the understanding of sediment transfer in pristine settings. Similarly, Crema et al. [21] compared several interpolation strategies, including robust and heat-diffusion-based inpainting, to reconstruct terrain in areas with missing or altered elevation data. Their work demonstrates the importance of restoring natural surface features to enhance geomorphic interpretations and assess structural connectivity more realistically. Building upon foundational studies by [20,21], our approach introduces specific improvements, particularly refined threshold criteria and an adjusted skeletonization workflow, to accurately represent the unique geomorphological conditions of the Rupinaro basin and better understand terraces’ impacts on sediment connectivity.

Identifying terraces using high-resolution Digital Terrain Models (DTMs) [22] is a crucial first step in understanding their impacts on sediment connectivity and geomorphic processes. Godone et al. [23] proposed a methodology for detecting terraces in high-resolution digital terrain models (DTMs), which has provided valuable insights into their role in landscape stability and sediment dynamics.

The main aim of this paper is to investigate the influence of terraces on sediment connectivity modeled through the Connectivity Index (CI), developed by [4] and based on the original approach by [12], in a small catchment in the Liguria region (Italy), an area frequently affected by shallow landslides. Our specific objective is to develop a methodology to remove anthropogenic features, such as terraces and roads, from high-resolution LiDAR-derived DTMs, and analyze the resulting connectivity under multiple scenarios.

2. Study Area

2.1. Geological and Geomorphological Setting

The Rupinaro catchment, a small coastal basin of 11 km² in the Liguria region (NW Italy, Chiavari and Leivi municipalities), spans elevations from sea level up to 544 m of M. Anchetta (Figure 1A). The terrain is predominantly characterized by steep hills and mountains, with limited flat areas, reflecting the tectonic structure of the Ligurian Apennines. This setting leads to short, low-order channels with high erosive potential [24]. Slopes rarely exceed 45–50°, with a prevailing south–southwest orientation. In our case, we excluded the coastal and alluvial plain of the Rupinaro from the analysis; the AOI is limited to 9.4 km².

Geologically (Figure 1B), the study area lies entirely within the Northern Apennines flysch sequence [25,26]. It spans the Mt. Antola Unit, primarily composed of Mt. Antola Limestones (Late Cretaceous-Paleocene), and the Mt. Gottero Unit, which includes the Slates of Mt. Verzi, Manganese-rich Shales (Late Cretaceous-Paleocene), and Palombini Clays (Low Cretaceous). The Mt. Antola Limestones (MAL) (Figure 1B) are located in the western region of the basin, along the right bank of the Rio Campodonico tributary, and consist mainly of a turbiditic sequence of marly limestone with minor shales and slates [27]; this formation is overthrust by the Mt. Gottero Unit and is intersected by fault systems. The Manganese-rich Shales (MNS) and Slates of Mt. Verzi (VES) dominate much of the catchment; the former outcrops along the central stream axis and consists of dark shales with thick siltstone and calcarenite interlayers, while the latter occupies the left bank of the Rio Campodonico and includes dark slates and marl [28]. Meanwhile, the Palombini Clays (PAC) outcrops are present in a small sector of the eastern area of the study. Finally, the central valley of Rupinaro is characterized by a thin layer of Quaternary alluvial deposits. The structural settings show a general plunge of the Manganese-rich Shales formation toward south–southwest, which drives the morphology of the Rupinaro valley, with the slope facing SW being more gentle and the steep slope to NW. The role of stratification is less evident in the Rio Campodonico basin.

Figure 1. (A) Location of the study area; (B) lithological map of the study area, modified from [29]; (C) land use map of the study area.

Figure 1. (A) Location of the study area; (B) lithological map of the study area, modified from [29]; (C) land use map of the study area.

2.2. Land Use Patterns

Land use (Figure 1B) is heavily influenced by centuries-old anthropogenic terraces, comprising dry-stone walls that cover approximately 40% of the catchment, primarily for agricultural purposes, especially for olive cultivators [23]. Moreover, the southwest slope also exhibits a high degree of urbanization, characterized by buildings and roads, especially between the Leivi and San Bartolomeo hamlets. In contrast, the steep slope facing northeast retains natural vegetation, including woods and Mediterranean shrubland, and is less terraced.

2.3. Shallow Landslide Distribution

Like other sectors of the Liguria Region, the Rupinaro basin was affected in the autumn of 2014 by severe rainfall events [30] that triggered hundreds of shallow landslides, as well as some debris flows and rotational landslides. Across the entire surveyed area of 400 km², 1641 landslides were mapped, resulting in an average density of 4 landslides per km² [30].

We used the shallow landslide inventory form [30]. The landslide distribution and density are driven by lithology, slope, aspect, and land use, and, according to [31], density correlates well with susceptibility models. In particular, higher landslide density is found in shale–flysch formations, and this is particularly pronounced on 30–40 degree south-facing slopes that have a similar strike and dip of the formation bedding. Olive tree land and sparse urban areas are also area factors that drive landslide density. These areas often exhibit reduced vegetation cover and altered surface drainage patterns that can enhance soil erosion and slope instability.

Additionally, anthropogenic features, such as roads and terraces, significantly influence the spatial distribution of landslides [32]. Roads, in particular, can act as pathways for concentrated runoff, increasing the likelihood of slope failures along their edges [5]. While they are constructed to stabilize slopes, terraces can also create localized sediment accumulation zones that may become unstable during intense rainfall.

We extracted shallow landside trigger points (or crowns) as polygon geometry by semi-automatically detecting the source area of shallow landslides. Using DTM, we selected the highest points of the landslide polygon and created a small buffer to allow for a natural variability range of the actual trigger location.

Further, we classified shallow landslide trigger points based on their spatial relationships with terrain features to better understand the interactions between sediment connectivity and landslide occurrence. Three primary categories were identified: (i) terraced slopes, where landslides occurred within or near agricultural terraces; (ii) abandoned terraces, where the terraces show high natural vegetation regrowth; (iii) road upslope failures, where landslides originated above road structures related to slope cut; (iv) road downslope failures, where slopes below roads experienced failure due to runoff concentration and slope modification; and (v) artificial cut, which include all of the slope cuts that are related to road or terraces.

Figure 2A presents the classified trigger points from the 2014 landslide inventory for the Rupinaro Basin, highlighting the distribution of failures across terraces, roads, and other slope conditions. Most failures (66 out of 94) were observed on the terraced slopes. Additionally, 14 failures were associated with road downslope conditions, while 6 failures occurred upslope of roads. Other types are limited, and are not significant enough to support a statistical analysis.

Figure 2B,C shows a more detailed view of the classification of shallow landslide triggering points, which overlapped with post-event orthophoto and hillshade derived from DTM LiDAR. In particular, it is possible to detect the four main types of triggering points: two are located downslope of the road (blue points), another is related to the slope cut above the road (yellow points), and the highest one is on the terraced landscape (red points).

3. Materials and Methods

3.1. Connectivity Index with and Without Anthropogenic Disturbances

To contrast sediment connectivity patterns and highlight variations in sediment dynamics between anthropogenically modified and natural conditions, we employed a multi-step methodology to simulate pre-anthropogenic surface morphology and assess its impact on sediment transfer efficiency.

Our methodology follows four main steps: (1) data collection, (2) DTM preprocessing and improvement, (3) terrace removal, and (4) sediment connectivity analysis. The flowchart in Figure 3 below summarizes the methodological workflow, illustrating the key processing steps from raw data to final CI calculations.

We first collected high-resolution LiDAR-derived DTMs along with supporting datasets, including land use and landslide inventories. The DTM was then processed to remove artificial features, such as bridges and pits, ensuring accurate flow modeling. Additionally, we worked with a pre-processed DTM where terraces and roads had been removed, allowing us to analyze sediment connectivity in both anthropogenic and natural landscape conditions. In the final step, we calculated the Connectivity Index (CI) [9] for both DTMs using three different target scenarios: catchment outlet (without a specific target), a hydrographic network, and roads. We also used land use as weight instead of DTM-derived roughness.

Using spatial and statistical analyses, we also investigated how changes in connectivity patterns are related to the spatial distribution of shallow landslides.

3.1.1. Lidar Data–DTM Data

In early 2015, CNR IRPI organized a LiDAR survey of the area, including the Rupinaro Basin, one of the regions most impacted by the 2014 floods in the Province of Genoa [30,33]. The survey’s primary objective was to obtain high-resolution DTMs and orthophotos to identify geohydrological instabilities triggered by the flood and to provide a detailed description of the basin morphology under investigation.

The raw point cloud, stored in LAS format, was georeferenced using the ETRF2000 coordinate system in UTM zone 32N, with EGM2008 as the vertical datum. A TIN interpolation was then applied to the scanned data to generate a Digital Terrain Model (DTM) at a 0.25 × 0.25 m resolution. For the Rupinario AOI, we resampled the DTM resolution to 0.5 × 0.5 m to have enough computer RAM for the Connectivity Index calculation. Additionally, we utilized the associated high-resolution terrestrial orthoimages, with a ground resolution of 15 cm per pixel, to classify landslide trigger points.

3.1.2. DTM Pre-Processing

A rigorous preprocessing workflow was applied to prepare a high-resolution Digital Terrain Model (DTM) of the Rupinaro catchment for sediment connectivity analysis [34]. This included removing anthropogenic structures, such as bridges, which could disrupt flow path calculations. The initial DTM_T, obtained at 0.5 m resolution, was processed using the several geospatial tools outlined below.

DTM preprocessing involved removing artificial structures, such as bridges, and filling depressions to ensure realistic flow path modeling. Subsequently, the DTMs were processed using our adapted skeletonization algorithm, tailored specifically for the Rupinaro basin, to simulate pre-anthropogenic landscape conditions. The Connectivity Index was calculated across multiple scenarios (catchment outlet, hydrographic network, and road network), and statistical significance was assessed using Kolmogorov–Smirnov and Mann–Whitney U tests.

These non-parametric tests were selected due to the non-normal distribution of Connectivity Index values and the need to compare differences in CI across diverse geomorphic scenarios. This statistical approach enables us to assess whether changes in terrain configuration (e.g., the presence or absence of terraces and roads) result in significant shifts in sediment connectivity.

Next, the Raster Extraction by Mask Layer tool was applied to the original DTM. The AOI with “holes” served as the mask layer, ensuring that the resulting DTM excluded the bridge areas. This process created a DTM with gaps corresponding to the removed bridges.

The Fill No Data tool from the GDAL Raster Analysis toolbox in QGIS was used to generate a seamless terrain surface. This tool uses inverse distance weighted (IDW) interpolation to estimate missing elevation values based on the surrounding cells. Although relatively simple, this method was sufficient for our case, where the missing areas were small and well-distributed.

This type of reconstruction fits within a growing interest in contrasting altered and pristine topographies, as discussed in studies like [20,21]. These works support the value of reconstructing pre-disturbance surface morphology for geomorphological analysis and justify the approach used here.

Further corrections were performed to address terrain depressions, which could cause incorrect flow routing. The TauDEM “http://hydrology.usu.edu/taudem/taudem5/index.html (accessed on 10 May 2025)” [35] Pit Removal tool was employed to adjust elevation values and ensure uninterrupted flow paths. This tool elevates depressions to the surrounding terrain level, eliminating spurious pits and ensuring accurate hydrological modeling.

The same preprocessing steps were repeated for the DTM with terraces and roads removed. After generating this version—where anthropogenic features had been removed using a dedicated workflow—we applied the same sequence of operations: bridge removal, raster extraction, no-data filling, and depression correction. This ensured uniformity across all of the datasets used in the sediment connectivity analysis.

This preprocessing pipeline yielded two fully corrected DTMs: one representing the original terraced landscape and another depicting a terrain with anthropogenic features removed. Both DTMs formed the foundation for the subsequent sediment connectivity analyses.

3.1.3. Terrace Removal and DTM Smoothing

The terrace-removed DTM was generated using an adapted roughness-based skeletonization procedure. While inspired by previous methods [20,21], our procedure incorporates tailored adjustments, such as customized threshold values and morphological refinements specifically suited to the Rupinaro basin’s densely terraced landscape. The proposed procedure transforms a digital terrain model that contained terraces (DTM_T) into a digital terrain model with the same properties, simulating the original state of the area before human intervention (DTM_NoT). The terrace smoothing procedure consists of four main steps:

• From DTM_T, the Terrain Ruggedness Index (TRI) is derived and classified using two thresholds, one for the terraces and another for the terrace walls, as determined via QGIS.
• The newly created binary masks and the original DTM are used in Python 1.2.1 to generate central/skeleton lines and skeleton line points for the terraces and terrace walls.
• The points for the terraces and terrace walls are input in QGIS and merged to create the points used in the interpolation of the new raster.
• The smoothed raster (DTM_NoT) is created by interpolating the merged set of points.

Spotting the location of the terraces relies on first enhancing the contrast of the DTM. The Terrain Ruggedness Index (TRI) was used as a terrace detection tool due to its sensitivity to local elevation differences. As defined by [36], TRI is a geomorphometric parameter that quantifies surface roughness by considering the absolute elevations of neighboring cells in a Digital Elevation Model (DEM). It is calculated using the following Equation (1):

TRI = (Σ(zc − zi)²)²

(1)

where:

• zc represents the elevation of the central cell.
• zi represents the elevation of one of the eight neighboring cells (i = 1, 2, …, 8).

After localizing the terraces and terrace walls using TRI (Figure 4A), two binary masks are created to extract them (Figure 4B,C) by leveraging the TRI radiometric variance.

After creating binary masks for the terraces and terrace walls using empirical thresholds, these masks served as input for a developed Python code. The entire concept of the terrace smoothing algorithm is tied to the creation of skeleton lines.

Skeletonization is a digital image processing technique used to reduce the representation of a shape to a simplified structure. In the context of this study, skeletonization is a process that simplifies the shape of the terraces and terrace walls into lines while capturing the essential topology.

In other words, it resembles the “backbone” of a shape. According to [37], skeletonization algorithms are primarily utilized to extract feature parameters from an image. They are often a key component of thinning algorithms, which work by iteratively removing pixels from the boundary of an object while ensuring that the object’s connectivity remains intact. This process gradually shrinks the object down to a thin representation that resembles a skeleton.

The developed Python code [38] aims to convert raster data into a .csv file that can be interpolated to produce the final raster. Firstly, several Python libraries were imported. The following process in the code was to create a skeleton mask for the terraces and the terrace walls. The skeleton mask is a binary raster (0–1) that creates central lines in the active pixels of the previous binary raster. To create the skeleton mask, the code reads the binary mask to define the location of the skeleton lines. The skeleton lines were placed in the middle of the terraces or terrace walls by incorporating the activated binary pixels (pixel value = 1). After creating the skeleton raster, the DTM values are read, allowing for the points to register the coordinates and elevation data at the locations of the skeleton lines and convert them to points. The result is a CSV file containing the coordinates and elevation data for the skeleton lines converted into points (Figure 4D). This process was repeated twice: once for the terraces and once for the walls. Then, the points were merged in QGIS to create the final step in the terrace smoothing process. A Triangulated Irregular Network (TIN) was used, with the parameters reported in

Table 1. Parameters used for interpolation.

Parameter	Attribute
Type	TIN
Method	Linear
Vector layer	Merged points
Attribute	Elevation values
Resolution	0.5 m

, to create an artificial raster. This raster was generated using the merged points from the terraces and terrace walls to recreate the original morphology (DTM_NoT) (Figure 4E).

3.1.4. Connectivity Index Calculations

The adopted index of connectivity represents the potential connection between the slopes and features selected as targets for the analysis, with a particular focus on evaluating the possible connection between the hillslopes and features that act as targets or storage areas (sinks) for transported sediment [9]. In this context, targets refer to specific landscape features, such as hydrographic networks or road networks, that direct sediment transport. At the same time, weight factors account for the surface characteristics that influence sediment movement efficiency, such as surface roughness or land use types. The Connectivity Index (CI) [9] was calculated using the stand-alone software SedInConnect 2.3 [39,40]. This application, available as a single executable file for Windows under a GPL v2.0 license, is accompanied by its Python source code. The software relies on the installation of TauDEM Version 5 tools [35] (http://hydrology.usu.edu/taudem/taudem5/index.html, (accessed on 10 May 2025)) for several hydrological functions critical to the analysis [41].

All input raster datasets must adhere to the GeoTIFF format, as this is the only format supported by TauDEM. They must have identical extents, origins, and grid dimensions to ensure compatibility. The primary input required is a Digital Terrain Model (DTM), typically a depressionless version (with pits removed), to produce connectivity maps at the catchment scale.

In SedInConnect 2.3, we configured key processing parameters:

• Cell size: Set to 0.5 map units, matching the DTM spatial resolution of 0.5 m.
• Moving window size: Defined as 5 pixels, which is the optimal setting for capturing local terrain variations.

SedInConnect 2.3 allows for users to enhance standard analyses by incorporating target and/or sink features to refine sediment transport pathways. Additionally, the software provides the option to use the surface roughness-based weighting factor proposed by [9], which can be automatically calculated by the tool.

3.1.5. Connectivity Index Scenarios

To analyze the sediment connectivity in the study area, we computed four different CI scenarios, each designed to represent different sediment transport dynamics.

• The catchment outlet (without a specific target) scenario represents the baseline sediment connectivity, where no specific transport pathway is defined. It allows for the observation of general connectivity trends based solely on terrain characteristics.
• Hydrographic network target scenario: This scenario models sediment connectivity toward the hydrographic networks, assessing how sediment is likely to be transported toward drainage systems under different conditions.
• Road network target scenario: In this case, sediment transport is modeled toward roads, reflecting how infrastructure influences connectivity by either acting as a sediment sink or a conduit for sediment displacement.
• Land use weighting scenario: Unlike the other scenarios, which use DTM-derived weights [4] as impedance factors, this scenario applies land use as a weighting factor, using Manning’s flow resistance [16] table (

  Table 2. Manning’s flow resistance table used for the land use weighting scenario.

  Land Use	Manning Roughness	Weight Factor
  abandoned olive	0.05	0.95
  Arable land	0.2	0.8
  Grasslands	0.25	0.75
  Low Urban areas	0.15	0.85
  Olive/vineyards	0.2	0.8
  Sparse vegetation	0.05	0.95
  Uncultivated areas	0.35	0.65
  Urban areas	0.02	0.98
  Wood	0.4	0.6

  ). This approach provides a more process-based evaluation of sediment transport based on different land cover types.

For the first three scenarios—catchment outlet (without a specific target), hydrographic network, and road—we applied roughness-based weighting factors as impedance factors.

For the fourth scenario, instead of using a DTM-based weighting factor, we used land use categories as weighting factors, derived from Manning’s flow resistance (Table 2).

Each of these four CI calculations was performed for both of the following:

• The original DTM, which includes terraces and roads.
• The modified DTM, where terraces were removed.

This dual approach enabled a comparative analysis of sediment connectivity between anthropogenic and natural landscapes across various transport scenarios.

In addition to the standard CI values, we calculated the normalized Connectivity Index (CI_z) following the equation proposed by [16]. This normalization accounts for variations in landscape characteristics, allowing for a comparative analysis between different sub-catchments and across various sediment connectivity scenarios. CI_z enables a scaled interpretation of connectivity values, ensuring that connectivity differences between terraced and non-terraced landscapes are not influenced solely by terrain complexity, but also by relative sediment transfer potential.

3.2. Statistical Analysis of Sediment Connectivity and Correlations with Shallow Landslides

To further investigate the role of sediment connectivity in relation to shallow landslide occurrences, we applied statistical tests across the entire catchment and within specific polygons identified as landslide triggers.

We created a Grid in QGIS to sample the values of CI within trigger points and exported it to a CSV file. Then, we used the Mann–Whitney U and KS_2SAMP functions of SciPy.stats library of Python to perform the Kolmogorov–Smirnov (KS) and Mann–Whitney tests.

The Kolmogorov–Smirnov (KS) test [42] was used to compare the distribution of CI values between terraced and non-terraced DTMs for each of the four target scenarios. This non-parametric test assesses whether two datasets originate from the same distribution by measuring the maximum difference between their cumulative distribution functions (CDFs).

The KS test was particularly useful in evaluating whether the removal of terraces caused significant shifts in sediment connectivity patterns, providing insights into how human-induced modifications influence sediment transfer at the landscape scale.

The Mann–Whitney U test [43] was used to assess differences in CI values between the original and terrace-free DTMs. The test was performed in two ways:

• Across the entire basin to capture overall differences in connectivity.
• Within landslide trigger point polygons to evaluate how terrain modifications influence connectivity in unstable areas.

These statistical analyses provided quantitative insights into the impact of terraces on sediment connectivity, helping to establish links between connectivity dynamics and slope instability.

4. Results

Following the procedure described in the Materials and Methods section, we calculated the CI for each scenario. Each scenario was computed for both the original DTM (DTM_T), which includes terraces and roads (CI_T as output), and the modified DTM (DTM_NoT), where terraces were removed (CI_NoT as output).

4.1. Connectivity Index Results for Different Target Scenarios

Figure 5 shows the CI results for the Rupinaro basin using original terraced DTM with the following options: (A) catchment outlet (without a specific target), (B) hydrographic network as a target, (C) road network as a target. And the DTM with terraced removed with the following options: (D) catchment outlet without a target, DTM_NoT; (E) hydrographic network as a target, DTM_NoT; (F) road network as a target, DTM_NoT. It also presents the Connectivity Index (CI) results for the Rupinaro basin under three scenarios: catchment outlet (without a specific target), hydrographic network as the target, and road network as the target. In these cases, we used the same DTMs as the weights that were used as input.

Catchment outlet (without a specific target): This Scenario (Figure 5A,D) represents natural sediment transport without a predefined deposition point. The comparison between CI_T and DTM_NoT highlights the role of anthropogenic features in shaping sediment flow paths. In DTM_T, terraces and roads create several flat surfaces that act as sediment barriers, resulting in lower CI values. These artificial structures alter natural connectivity by disrupting sediment transport pathways, particularly in the northern sector of the basin, where terraces are densely distributed. On the other hand, sediment flow is diverted and may become concentrated along specific paths, mainly through the road network. However, as discussed later, the effect of an artificial drainage network is not always represented at the local scale, even when using a high-resolution DTM. In contrast, DTM_NoT shows an increase in sediment connectivity, as the removal of terraces and roads eliminates artificial obstacles, allowing for the sediment to follow shorter and more direct downslope flow paths. This effect is particularly noticeable in steeper areas, where natural slopes regain their function in facilitating sediment transport.

The hydrographic network target scenario (Figure 5B,E) aims to subset the analysis of CI along the main streams, enhancing the processes at the sub-basin scale. When the hydrographic network is set as the target, the CI values generally increase towards the path of the stream, indicating that more sediment is routed towards the drainage system. This effect is more pronounced in DTM_NoT, as removing terraces reduces artificial interruptions in sediment pathways, leading to higher connectivity values along the main channels. In DTM_T, terraces and roads cause local sediment accumulation, restricting sediment transfer to the hydrographic network.

The road network target scenario (Figure 5C,F) enables us to understand the interaction between roads and sediment transport. In DTM_T, roads act as both barriers and conduits, depending on their position and the interaction with the slope. Such results enable us to focus on the segments of the roads where the potential for sediment transport is higher. The CI values are constrained by the road imposed as a target; they are lower downslope, as the roads reduce direct sediment transfer by intercepting flow paths, which are the starting points for the sediment connectivity calculation. At the same time, while they are higher up, the roads are obviously the target. A slope with a road featuring multiple hairpin turns can introduce bias in the results, as the upper road segment may artificially filter the data for the lower segment.

In DTM_NoT, connectivity is slightly enhanced; this scenario allows us to simulate a case in which agriculture terraces are removed and the roads intercept the sediment flow along its natural path.

4.2. Land Use-Weighted Connectivity Index

The results of CI calculations using land use as a weighting factor instead of terrain-based roughness are shown in Figure 6. This approach incorporates Manning’s flow resistance values, reflecting how different land covers may behave in terms of flow impedance. In this case, we used the catchment outlet (without a specific target) scenario. In Figure 6A (the original DTM), connectivity is controlled by land use types, with urbanized areas and forests exhibiting lower CI values, while grasslands and abandoned agricultural areas show higher connectivity. In Figure 6B (DTM_NoT), removing terraces slightly increases connectivity values, particularly in areas where terraces previously acted as stabilizing structures.

Figure 7 compares the statistical analysis of the CI area using the Kolmogorov–Smirnov test for the whole study area. The value of CI_NoT indicates a general shift toward higher values, as confirmed by the KS test, with a p-value of 0.33. Moreover, the distribution is less concentrated due to interpolation and smoothing.

In this scenario, we also overlap the shallow landslide database; at first, at the basin scale, there are no significant links between the landslide distributions and CI. The density of landslides is mainly defined by lithological, slope, and land use conditions, as described in [31]. In the next section, a more in-depth statistical analysis will be employed to elucidate the relationship between shallow landslides and CI.

4.3. Influence of Lithology on CI

The box plots in Figure 8A,B show the distribution of CI for land use weight solutions for the main lithology of the study area (except for the alluvial deposits of central valleys). Slightly higher CI values are observed in regions dominated by slate and limestone formations (1A) compared to those dominated by shale and flysch. This is related to the consequences of lithology on morphology and land use: on slate and limestone, the slopes are steeper and have a low degree of anthropogenic disturbance, which is reflected in a higher correlation with sediment transport processes; meanwhile, on flysch and shale formations, especially on the southern slope, where the bedding orientation matches the slope, the low gradient promoted low sediment connectivity, while simultaneously increasing the degree of anthropogenic disturbance. Conversely, areas with more urbanization and infrastructure exhibit lower CI values, as built environments disrupt natural sediment flow.

This is also evident from the difference in CI (Figure 8C), where, in the case of shale flysch formation, the removal of terraces and roads caused a higher increment of CI compared to other formations.

The highest CI values were observed in areas underlain by shale–flysch formations, characterized by steep slopes and limited urbanization. These regions exhibit high sediment transport potential, as steeper slopes facilitate rapid sediment movement and increase connectivity values. However, the statistical analysis suggests that lithology and terrain steepness are more strongly influenced by landslide density than by CI alone.

5. Discussion

The results of the sediment connectivity analysis and the statistical tests indicate a complex relationship between Connectivity Index (CI) values and shallow landslide occurrences. While no clear correlation between CI and landslide density was observed at a broad scale, local-scale analyses suggest that certain anthropogenic elements, such as roads and terraces, influence sediment flow and landslide susceptibility in specific areas.

Local effects of roads on sediment transport and landslide occurrences.

Roads, particularly downslope sections, play a significant role in sediment redistribution. As shown in Figure 9A,B, shallow landslides are often concentrated in areas with high CI values, where roads may act as sediment conduits, directing flow downslope and increasing the likelihood of slope failures. However, the comparison with ground-truth data suggests that, in some cases, the actual source of sediment may not align perfectly with the modeled connectivity values.

When terraces were removed in the CI_NoT scenario, sediment connectivity remained high in many areas affected by landslides, particularly in road-adjacent slopes (Figure 9C,D). This suggests that, while terraces contribute to sediment redistribution, the removal of terraces does not drastically alter the connectivity patterns in landslide-prone areas. Instead, other topographic and hydrological factors, such as road drainage systems and artificial fill, appear to play a more prominent role in sediment transport and slope instability.

The presence of artificial drainage systems along roads can influence connectivity in ways that are not fully captured by high-resolution DTMs. Slight elevation differences, particularly those created by human-modified drainage infrastructure, may not be well represented in the DTM, leading to potential discrepancies between predicted sediment pathways and actual sediment flow observed in the field. An example is shown in Figure 9E,F, where post-event orthophotos and a Google Street View image captured on a rainy day highlight the role of artificial fill in landslide triggering. These observations emphasize the importance of integrating field data and geomorphological surveys with connectivity models to improve accuracy.

The different effects of roads and terraces on the CI

Figure 10 shows a comparison along a profile of normalized CI_T and CI_NoT. Terraces contribute to sediment redistribution, but their removal does not drastically alter overall connectivity patterns. It is possible to appreciate the variation in the CI in terraced slopes, which have trends related to steps and flat terraces.

Instead, roads, particularly in downslope sections, appear to act as key sediment transport pathways in some cases, as shown in the profile in Figure 10C. The roads break sediment connectivity along the maximum slope, causing a decrease in the CI downslope. However, in many cases (e.g., Figure 9), the roads deviate and concentrate flow, increasing the likelihood of slope instability [44,45]. The comparison between modeled CI values and real-world observations highlights the need for integrating field-based validation into sediment connectivity analyses.

The Connectivity Index influences landslide density compared to other factors

We analyzed the relationships the CI had with the main effect of terraces and roads and their correlation with shallow landslides. Firstly, we examined the impact of different landscape types on the Connectivity Index. Using a 10 × 10 m grid and the input data (terrace skeleton, roads, buildings, and the hydrographic network) shown in Figure S5, we classified the AOI into three main classes: (i) terraced landscape, where the skeleton mask of terraces (Figure 4B) is prevalent; (ii) non-terraced landscape, where the skeleton is not significant; (iii) roads network, where the map of the landscape type is displayed in Figure S6A of the Supplementary Materials. We also extrapolate the (iv) hydrograph network and (v) buildings to remove them from the statistical analysis. The box plots in Figure S6B of the Supplementary Materials show that the CI_T (land use weight scenario) did not change significantly between terraced and non-terraced landscapes, while detecting a high value of CI_T in the road network. While Figure S6C shows that the CINoT does not reveal significant differences between landscape classes, these results confirm the higher influence of roads compared to terraces on the deviation of sediment flow.

Then, we analyzed the density of landslide trigger points in relation to four factors: (i) landscape type, (ii) lithology, (iii) slope, and (iv) CI_T class. The results of the statistical analysis are shown in Figure 11. The lithology (Figure 11B) and slope gradient (Figure 11C) have a strong control on landslide distribution, as previously found by other studies [31]. At the same time, the landslide trigger point density shows a slightly higher density in terraced areas compared to non-terraced areas. In contrast, roads exhibit high density (Figure 11A), confirming that roads both influence sediment flow and facilitate the triggering of landslides (both upslope and downslope). Finally, plot Figure 11D shows a slightly high density of landslide trigger points (TP) in a low class of CI_T (in the land use weight scenario) that corresponds to the CI_T class of the road (Figure S6B of the Supplementary Materials). Such a result suggests that CI is a low predictor of the location of shallow landslides compared to other factors; conversely, it is not its main factor. However, the CI help to identify the critical road section that can be a sediment source for shallow landslides.

Further analysis also shows that landslides are more common on mid-to-upper slopes than near rivers or streams. This suggests that, in this study area, water runoff along roads and terraces significantly triggers landslides compared to direct river erosion. Most shallow landslides occurred at a moderate distance from the main channels, indicating that hillslope water flow, rather than river erosion, is the main driver of sediment movement in the Rupinaro catchment.

Kolmogorov–Smirnovand Mann–Whitney Test for shallow landslides Trigger Points

The Supplementary Materials (Figures S1–S4) report the results of the Kolmogorov–Smirnov test of CI for the main type of trigger points (terraces, roads upslope, and roads downslope) in all scenarios and both the terraced and non-terraced cases.

The Kolmogorov–Smirnov test revealed that the distribution of CI within trigger point values between the terraced and non-terraced scenarios did not exhibit significant shifts from the results calculated for the entire AOI.

This suggests that removing terraces did not fundamentally alter sediment transport dynamics across the basin and is not fundamental a predisposing factor for shallow landslides. Figure 12, considering the KS statistic (0.429) of the trigger point, shows a somewhat higher value compared to the entire basin (0.33).

However, at a more localized scale, specific areas, especially downslope road sections, showed noticeable differences in connectivity patterns; in such cases, anthropogenic modification plays a high role in contributing to landslide triggers.

Similarly, the Mann–Whitney U test (

Table 3. Mann–Whitney U test results for trigger points classified as terraces.

Scenario	U Statistic	p-Value
Scenario Catchment outlet (without a specific target)	468	7.24 × 10−15
Scenario Hydrographic network	1677	2.27 × 10−2
Scenario Road	852	4.44 × 10−9

) revealed that, while statistically significant differences in CI values existed between terraced and non-terraced DTMs, these differences were not substantial enough to suggest a strong, direct correlation between terraces and sediment connectivity in all cases. Instead, terraces appear to play a modifying rather than a dominant role in sediment transfer, altering connectivity in specific locations, but not fundamentally changing the overall sediment transport structure.

6. Conclusions

This study investigates the impact of human-made terraces and roads on sediment connectivity in the Rupinaro catchment, a landslide-prone basin in Liguria, Italy. Using high-resolution LiDAR-derived DTMs, we implemented a methodology to remove terraces, simulating a pre-anthropogenic landscape and allowing for a direct comparison between natural and human-modified connectivity patterns. We assessed how different landscape features influence sediment transport processes by analyzing sediment connectivity across multiple scenarios—without a target, with road and hydrographic networks as targets, and using land use as a weighting factor.

The results indicate that a combination of topography, lithology, land use, and human interventions has a strong influence on sediment connectivity. Terraces and roads alter natural sediment dynamics by introducing flat areas that act as local sinks or flow pathways, reducing and modifying overall sediment transfer efficiency. The application of high-resolution DTM processing techniques for terrace removal proved to be effective in isolating the influence of anthropogenic features on sediment connectivity. A statistical test showed an increase in the CI in the simulated natural relief on the whole basin compared to the original DTM. In particular, the comparison indicates that roads are the element that most affects the CI, while terraces have a reduced influence on CI. The visual analysis of the results helped to assess natural sediment pathways more effectively and distinguish the impact of human modifications.

Despite these changes, statistical analyses suggest that, while sediment connectivity influences sediment transport, landslide occurrence and density appear to be more closely linked to lithology, structural settings, and slope rather than connectivity alone. However, on a local scale, our analysis highlighted that roads, particularly in downslope positions, can contribute to sediment flow concentration and influence landslide initiation. While large-scale patterns suggest a limited correlation between connectivity and landslide density, specific cases demonstrate how localized anthropogenic modifications can amplify sediment transport processes.

These findings underscore the importance of integrating sediment connectivity analysis with high-resolution terrain processing to enhance our understanding of landscape dynamics, particularly in regions where human interventions significantly alter natural geomorphic processes. While this study focuses on a terraced catchment in Liguria, its methodological framework can be applied to other geomorphic and land use contexts to further explore the link between terrain modification and sediment landslide interactions.