Keywords

Introduction

Systematic investigation and characterization of landslide phenomena at regional scale represent an essential requirement for risk assessment and for consequent planning of proper hazard-reduction resources and actions. Inventory maps of landslide processes deal with the detection of ground mass movements since they provide spatial distribution and parameters of each landslide (e.g. extension, typology, and state of activity).

Landslide inventories are usually based on conventional geomorphologic studies, which rely on intensive field surveys and on time-consuming manual interpretation of aerial and satellite optical images. Technical problems of spatial accessibility and related economical efforts of field campaigns and surveys can be considerably reduced thanks to the contribution of high precision, wide area coverage, non-invasiveness, and cost-efficiency of remote sensing technologies.

Persistent Scatterer Interferometry (PSI) is a multi-temporal interferometric technique which analyzes long temporal series of satellite SAR (Synthetic Aperture Radar) data, and provides annual velocities and time series of ground deformation on dense grids of point-wise targets, the so-called PS or Persistent Scatterers (e.g. Ferretti et al. 2001; Colesanti and Wasowski 2006). The PS analysis has recently demonstrated to be a valuable and useful tool to detect surface dis placements with millimetre precision, and has turned out tobe successfully used for identification and mapping of slow-moving landslide phenomena and for evaluation of their state of activity and intensity, at regional and local scale (e.g. Farina et al. 2006; Meisina et al. 2006; Casagli et al. 2008; Pancioli et al. 2008; Herrera et al. 2009; Cigna et al. 2010; Notti et al. 2010; Greif and Vlcko 2011; Righini et al. 2011).

In this paper we present an innovative methodological approach for the updating and improvement of landslide inventories by means of PSI analysis, using radar images acquired by satellites ERS1/2 in 1992–2001 and RADARSAT-1 in 2003–2010.

Test Area and Input Data

The test area is the northern part of the Calabria Region, southern Italy. This area has an extension of 4,410 km2 and includes part of the provinces of Cosenza (CS) and Crotone (KR). The geological background of the Calabrian peninsula shows a sheet pile of basement thrust units piled up and overthrust toward the Apulian foreland during the Neogene (Cucci and Cinti 1998). The outcropping terrains consist of Hercinian crystalline and metamorphic basement units and overlying Mesozoic and Tertiary sedimentary rocks (Amodio Morelli et al. 1976), covered by recent (mostly Upper Neogene) sediments. Northern Calabria is characterized by mountainous and hilly morphology and is extremely susceptible to landsliding because of high slope gradient, intense rainfalls, presence of weathered and faulted crystalline-metamorphic rocks and weakly-consolidated sedimentary sequences; these factors, when associated with extensive human activity (including unsuitable urban expansion and systematic deforestation), have played an important role in producing steep and highly unstable slopes (Carrara and Merenda 1976).

The available input data for the Northern Calabria case study include ancillary data, two pre-existing landslide inventory databases and PSI data. The ancillary data consist of topographic maps (1:10,000 and 1:25,000), geological map (1:25,000), DTM (Digital Terrain Model) with 20 m resolution and its derived maps, and aerial orthophotos with 1 m resolution (Volo Italia IT2000). The accessible pre-existing inventory maps are PAI (Piano Stralcio di Assetto Idrogeologico) and IFFI (Inventario Fenomeni Franosi in Italia). These inventories include spatial distribution of mass movements with related records (e.g. type, extension and state of activity of each phenomenon). They are based on aerial photo-interpretation, field campaigns and local databases and their quality depends strongly on the scale of resolution. PAI inventory reference date is 2001, while IFFI inventory was published afterward, updating PAI database, in 2007. IFFI inventory includes a total of 1,167 landslides, 371 (32 %) of which are classified as ‘active’ (including active, suspended and reactivated phenomena, according to the multilingual landslide glossary WP/WLI 1993), and 796 (68 %) as dormant. Stabilized phenomena (artificially stabilized, abandoned and relict; WP/WLI 1993) are not mapped. Landslides cover a total area of 168.4 km2, corresponding to 4.1 % of the whole investigated territory and the most represented landslide typologies are roto-translational slides, complex movements, deep-seated and shallow landslides; these landslide phenomena mainly involve crystalline metamorphic rocks, terrigenous tectonized flysches and recent marine and continental deposits.

PSI ground deformation data consist of present and historical measures derived from two different satellites: present RADARSAT-1 data in the interval 2003–2010 and processed by e-GEOS with the PSP (Persistent Scatterers Pairs) technique (Costantini et al. 2000), and historical ERS1/2 data, acquired in 1992–2001 and processed with the PSInSAR (Permanent Scatterers Interferometric SAR) technique by TRE S.r.l. (Ferretti et al. 2001), in the framework of the PST-A (Piano Straordinario di Telerilevamento Ambientale) project. Both ERS and RADARSAT products are characterized by the map distribution of PS benchmarks, and the information provided for each target are: geographic coordinates (latitude, longitude, elevation), LOS (Line Of Sight) yearly velocity and displacement time series.

Within Northern Calabria, we selected two sample sites that are geologically and morphologically representative of most of the region and that are significant because of their landslide assortment in terms of typology, activity and intensity: Verbicaro site and San Lorenzo Bellizzi site, both located in CS province, respectively in the north-west and in the north-central portion of the test area (Figs. 1a and 2a).

Fig. 1
figure 00151

Sample site of Verbicaro, Northern Calabria (Italy). (a) Geological map, 1:25,000; (b) Pre-existing and updated inventory map showing landslide state of activity; (c) Updated inventory map showing landslide intensity

Full size image
Fig. 2
figure 00152

Sample site of S. Lorenzo Bellizzi, Northern Calabria (Italy). (a) Geological map, 1:25,000; (b) Pre-existing and updated inventory map showing landslide state of activity; (c) Updated inventory map showing landslide intensity

Full size image

PSI-based Methodological Procedure

The approach is based on photo- and radar-interpretation and combines conventional thematic data (topographic, geological and land use maps, and optical images) with deformation measures obtained with the PSI analysis. This integrated analysis leads to: (1) landslide detection and mapping, which consist in the boundaries confirmation or updating of pre-mapped phenomena and in the recognition of new landslides; (2) landslide activity and intensity assessment, which consists in the evaluation of the state of activity and intensity of each mapped phenomenon.

The entire process of acquisition, analysis and interpretation of all the available data is carried out in a Geographic Information System (GIS) environment, which properly combines radar measurements with the other maps and thematic layers.

Landslide Detection and Mapping

Landslide detection and mapping is carried out through the evaluation of boundaries and spatial distribution of each landslide within the test area. This spatial investigation relies on the ancillary data integrated with the pre-existing landslide inventory and with PS ground displacement measurements.

On the basis of the pre-existing inventory map and PS distribution, the methodology consists of the radar-interpretation and aerial photo-interpretation procedures (Farina et al. 2006; Casagli et al. 2008; Pancioli et al. 2008; Herrera et al. 2009) that allow us to evaluate and spatially extend the point-wise ground motion rates provided by the PS deformation data.

Landslide mapping is mainly obtained by the individuation of geomorphologic evidences related to landslide phenomena, through the use of aerial imagery and through land deformation measurements over the space grid of PS points. Starting from the already mapped landslides, the presence of PS data within (or very close to) their boundaries can result in two conditions: confirmation of boundaries, when the inventory information is in agreement with PS data, or change (usually enlargement) of boundaries, when PS data suggest the modification of the pre-mapped area. No additional information can be given when PS data are not sufficient for radar interpretation or even when they are not present at all inside the already mapped area (Not Classified, N.C.). The presence of PS data outside the previous mapped areas can result in new landslide detections, when significant PS velocities and also geological and geomorphologic evidences suggest the presence of a landslide.

Landslide Activity and Intensity Assessment

Once mapped the spatial distribution of landslides in the area of interest, representative ground displacement values are determined, for each landslide, through the analysis of available PS data.

Landslide activity and intensity assessment is then performed by a PSI-based matrix approach, an advanced methodology for the evaluation of the state of activity and intensity of slow moving landslides, exploiting PS measures of present RADARSAT (2003–2010) and historical ERS (1992–2001) data. The evaluation of activity and intensity is computed by analyzing separately RADARSAT and ERS data for each landslide, and by calculating the average velocity of the PS identified within its boundary, V P and V H which are referred, respectively, to the two temporal coverages, Present 2003–2010 and Historical 1992–2001. These representative velocities (V P and V H ) are then compared with the activity and intensity thresholds, which are chosen for this area as ±2 mm/year and ±10 mm/year, away or towards the satellite.

The state of activity of landslides is assigned through two activity matrices: the first one (Fig. 3a) is used for pre-mapped phenomena and it is based on the information coming from the pre-existing inventory map (reference data: 2007) and present RADARSAT PS measures (2003–2010). The second one (Fig. 3b) is used for new phenomena and it is based on PS measures derived from two distinct time intervals, present RADARSAT 2003–2010 and historical ERS 1992–2001. In the resulting inventory, using a simplified version of the official classification of the landslide states of activity as defined in WP/WLI 1993, two states of activity are distinguished: ACTIVE (currently moving) – including ‘active continuous’ (constantly moving or suspended) and ‘reactivated’ (moving after being inactive) phenomena – and INACTIVE (not moving in the last 12 months) – including ‘stabilized’ (not affected by their original causes anymore) and ‘dormant’ (potentially being reactivated) phenomena. The reference year of the classified activity is also pointed out in every cell-grid of the matrices.

Fig. 3
figure 00153

Activity matrices and intensity scale based on the exploitation of PSI measures. (a) Matrix defined for landslides mapped in the pre-existing inventories; (b) Matrix employed for new detections, comparing historical ERS and recent RADARSAT measures; (c) Intensity scale

Full size image

The evaluation of the intensity of landslides is performed by an intensity scale (Fig. 3c), which is based on PS mean velocity (V P or V H ) and consists of three classes: negligible (V < 2 mm/year), extremely slow (2 mm/year ≤ V < 10mm/year) and very slow (V ≥ 10 mm/year). When PS number or density is insufficient within the landslide boundaries, the phenomenon is not classified (N.C.).

Results and Discussion

Sample Site 1 – Verbicaro Site (CS)

The first sample site lies in the municipality of Verbicaro (CS province) and is characterized by a mountainous morphology with a dense network of narrow rivers and waterways (Fig. 1a). The site is affected by several large landslides, classified in the pre-existing landslide inventory as roto-translational slides and complex movements. These phenomena mainly involve low-grade metamorphic rocks and occasionally well-stratified limestones.

Pre-existing landslide delineation turned out to be incomplete, and photo- and radar-interpretation complementarity allowed the recognition of four new phenomena (Table 1). Monoscopic aerial photos, together with topographic maps, DTM and derived layers, allowed the detection of soil erosion evidences, vegetation coverage anomalies, and counterslopes along the longitudinal profile as landslide indicators; conventional photo-interpretation permitted to map land movements in natural and rural environments where geomorphologic features were easy to be observed, but where a low density of radar targets was identified. On the other hand, radar-interpretation turned out to be very useful and suitable for mapping ground displacements in urbanized and artificial areas where much more PS benchmarkers (buildings or other infrastructures) were retrieved. In particular, aerial photos contribute to better identify landslides in the vegetated areas outside Verbicaro village, such as to the NW, where one new phenomenon was detected, even if it was classified as N.C. because of the presence of only one radar target (Figs. 1b and 3b). In fact, in order to reach a reliability of representative velocity evaluations, at least three radar targets were considered for both state of activity and intensity assessment (Meisina et al. 2006; Notti et al. 2010), and those phenomena not meeting the requirement were not classified (N.C.). The other three new landslides were identified through the orthophotos and the PS data, and show average velocities of 8–16 mm/year, which permit to classify two of them as very slow, and one as extremely slow (Figs. 1c and 3c). It is worth highlighting that the acquisition parameters of the currently available SAR sensors (i.e. temporal sampling and signal wavelength) do not allow the measurements and consequent classification of deformations that are faster than few tens of cm/year (LOS rates in C-band satellites up to 15 cm/year, falling in the class of very slow phenomena, i.e. 16 mm/year to 1.6 m/year, according to Cruden and Varnes 1996).

Table 1 Updating the pre-existing landslide inventory in Verbicaro (Site 1) and San Lorenzo Bellizzi (Site 2): number and percentage of landslide state of activity, intensity and main modifications implemented to the pre-existing inventory
Full size table

Moreover, some pre-mapped landslide boundaries were modified and enlarged by means of PS ground deformation data, especially in the Verbicaro urban fabric, which turned out to be unstable (Fig. 1b). For those PS located outside pre-mapped phenomena, when geomorphologic and geological setting and/or the PS velocity distribution did not show the presence of a landslide, no additional phenomenon was mapped. Other geological processes (e.g. subsidence or erosion) may, in fact, induce spatially distributed deformation; moreover, single unstable targets may not be indicative of a real landslide process, but more likely can be due to settlements of single structures.

In the western portion of the urbanized area, the state of activity of many pre-mapped phenomena was changed from dormant to reactivated; while in the eastern portion three landslides were classified as stabilized/dormant since their average ground deformation velocities ranged from 0.9 to 1.5 mm/year. This ambiguous classification is due to the “conservative” approach proposed in the activity matrix, which shows a double name in the two lowermost-left cells (Fig. 3a). For pre-mapped landslides, if PSI data register movements lower than ±2 mm/year, the state of activity recorded in the pre-existing inventory should be theoretically lowered from dormant to stabilized, or from active to dormant, respectively; however, real lowering of the state of activity can be eventually performed and confirmed afterwards, by means of field evidences and focused in situ monitoring (Cigna et al. 2010).

Sample Site 2 – San Lorenzo Bellizzi Site (CS)

The second sample site, located in the municipality of San Lorenzo Bellizzi (CS province), lies at the mountainside of Mt. Pollino and is characterized by steep slopes which reach heights of 1,300 m a.s.l., to both W (at Mt. Timpa di Cassano) and E (at Mt. Serra di Paola) (Fig. 2a). The western sector of Mt. Serra di Paola is affected by several landslides, which are predominantly deep-seated slope gravitational deformations, complex movements and roto-translational slides, as recorded in the pre-existing landslide inventory. These landslides involve terrigenous flysches and were confirmed or modified by means of photo- and radar-interpretation (Table 1): the state of activity of three phenomena was changed from dormant to reactivated and the deep-seated landslide located W of San Lorenzo Bellizzi village was enlarged to include the urbanized area, which showed a ground average velocity of 6–7 mm/year (Fig. 2b). PSI analysis also allowed the identification of two new phenomena E of San Lorenzo Bellizzi village: one of these was classified as Active/Reactivated with reference year 2001 because of the presence of historical ERS data, but absence of any recent RADARSAT targets. Regarding new detections, the lack of one of the two temporal evaluations of activity – past or present – of ground displacements, due to absence of benchmarks – RADARSAT or ERS respectively – prevents to discern between stabilized and dormant, or between active continuous and reactivated phenomena; for this reason, four cells in the activity matrix in Fig. 3b show double classification, reflecting the uncertainty in the assessment of the state of activity.

On outcropping rocks of Mt. Timpa di Cassano, which is oriented E-NE, some RADARSAT descending PS show positive velocities (blue colour; motion toward the satellite; Fig. 2b), while available ascending ERS data show opposite rates. This is because PSI measures are the one dimension (1D) projection along the satellite LOS, of 3D ground deformation. Slope orientation combined with LOS orientation may cause significant underestimation of land motions, or even may prevent at all the identification of PS within the observed scene, due to foreshortening, layover and shadowing effects (Colesanti and Wasowski 2006). The geometrical visibility of an area depends on its slope and on its aspect with respect to the satellite LOS orientation. The use of appropriate satellite geometry (i.e. along ascending or descending orbits) for the investigated area is therefore fundamental for each PSI-based analysis: generally, ascending acquisition modes are suitable for E-facing slopes, while descending modes are suitable for W-facing slopes.

Conclusions

The improvement of landslide inventory map in Northern Calabria was successfully carried out by means of the PSI technique. The availability of recent RADARSAT PS data (2003–2010) allowed the updating of the landslide inventory up to 2010. The comparison of recent PS data with the information coming from the pre-existing landslide database referred to 2007 and with historical PS measures (PS ERS data spanning 1992–2001) permitted the spatial detection and, when it was possible, the evaluation of present-day and past landslide activity and intensity of the phenomena in the area of interest. In particular, the PSI-based methodological procedure, based on the exploitation of activity matrices and intensity scales, lead to the assessment of the spatial distribution, state of activity and intensity of a total of 1,196 landslides, that cover 177.5 km2 (4.3 % of the whole investigated territory).

This PSI-based methodology demonstrated to be highly valuable and helpful for improving landslide inventory maps, especially in urban areas, in order to support the management and planning of risk-reduction measures and strategies.

A future enhancement of the proposed PSI-based methodology would consist in the exploitation of new SAR data (e.g. TerraSAR-X) with a higher spatial resolution (up to 1 m) and with a better acquisition frequency (up to 3–4 measures per month), allowing the identification of more recent and faster landslide phenomena with improved resolution and precision.