Landslides and Subsidence Assessment in the Crati Valley (Southern Italy) Using InSAR Data

: In this work, we map surﬁcial ground deformations that occurred during the years 2004–2010 in the Crati Valley (Southern Italy). The valley is in one of the most seismically active regions of the Italian peninsula, and presents slope instability and widespread landslide phenomena. We measured ground deformations by applying the small baseline subset (SBAS) technique, a multi-temporal synthetic aperture radar interferometry (InSAR) methodology that is used to process datasets of synthetic aperture radar (SAR) images. Ground displacements are only partially visible with the InSAR technique. Visibility depends on the geometry of the acquisition layout, such as the radar acquisition angle view, and the land use. These two factors determine the backscattering of the reﬂected signal. Most of the ground deformation detected by InSAR can be attributed to the gravitational mass movements of the hillslopes (i.e., landslides), and the subsidence of the quaternary deposits ﬁlling the valley. The movements observed along the valley slopes were compared with the available landslide catalog. We also identiﬁed another cause of movement in this area, i.e., ground subsidence due to the compaction of the quaternary deposits ﬁlling the valley. This compaction can be ascribed to various sources, such as urban population growth and sprawl, industrial water withdrawal, and tectonic activity.


Introduction
Ground deformation can be measured by SAR interferometry (InSAR; [1]), a technique applied on data acquired by satellite synthetic aperture radar (SAR) sensors. InSAR involves the processing of two images of a target area acquired by a satellite at different times. More recently, differential InSAR (DInSAR) algorithms, such as permanent scatterers (PS; [2]), the small baseline subset (SBAS; [3]), or a combination of the two [4,5], have been developed. These algorithms, which are based on the interferometric analysis of large image datasets, allow us to follow ground displacement in time.
Our study site, the Crati Valley (CV, Figure 1), which is located in Southern Italy (Calabria Region), is one of the most risk-prone areas of the Italian peninsula, as defined by the official Italian territory seismic classification (Official Gazette No. 105 of 8 May 2003; Figure 2). The CV is a north-south (N-S) oriented graben where landslides along valley slopes and vertical displacement (subsidence) in the

Geological Setting
The CV is a N-S oriented graben bounded by two morphostructural highs: the Sila Massif to the east, and the Coastal Range to the west. The graben is controlled by an array of normal faults along both sides (Figure 1), the so-called Crati fault system [18]. This area, which is part of the Siculo-Calabrian rift zone [19], is the result of the back arc extension due to the opening of the Tyrrhenian Sea [20]. The relationship between the evolution of the Calabrian Arc and Southern Apennines is still debated [21][22][23][24]. Tansi, C. et al. interpret the CV as an active transextensional area that developed since the late Pliocene, and as the southeast termination of the Falconara-Carpanzano strike-slip fault [23]. Spina, V. et al. explained the development of the CV, since the middle Pleistocene, as due to the activity of the N-S oriented, east and west-dipping normal faults bounding the valley [24]. These faults probably developed as a response to the uplift of the orogenic edifice, and the extension of the Tyrrhenian back arc.
The faults bounding the CV show persistent activity with medium-grade seismicity (5.5 < M < 6; [18,21,23,25]). Earthquakes occurred during the time covered by our SAR data, as reported in the INGV (National Institute of Geophysics and Volcanology) catalogue (ISIDe working group 2016; Figure 2). Focal mechanisms of recent earthquakes ( Figure 2; [25][26][27]) agree with the extensional tectonics attributed by several authors to this sector of the northern Calabrian Arc (e.g., [24,28]). An extensional trend along the east-west (E-W) axis is also shown by permanent GNSS (Global Navigation Satellite Systems) stations [29].
The stratigraphy of the CV, starting from the Upper Miocene to the Holocene, is characterized by deposits covering a nonconformity that overlies the Palaeozoic crystalline bedrock. The deposits have increasing thickness from the Coastal Range towards the Sila Massif [24,30] (Figure 1). The Upper Miocene succession, which also includes Messinian evaporites [31], outcrops discontinuously along the western side of the CV. The Pliocene and Plio-Pleistocene successions, which are visible along the western side of the valley, are separated by angular unconformities and constituted by deltaic (Gilbert-type and shelf-type deltas) and coastal deposits [32,33]. The Plio-Pleistocene sediments outcrop along both sides of the valley where they lie, along a nonconformity, directly on Paleozoic bedrock [18]. The alluvial plane of the CV presents the subsidence of recent deposits, which is similar to the other Quaternary basins of Calabria (Sibari and Gioia Tauro plains) that are adjacent to the CV ( Figure  1). In particular, the Gioia Tauro Plain is similar to the CV in its geological evolution. Both basins belong to the so-called Siculo-Calabrian rift zone (SCRZ) [19], are characterized by crustal extension, and are bounded by normal faults that have produced strong earthquakes in historical times. In the Gioia Tauro Plain, subsidence rates of 10-15 mm/year between 1992-2006 are considered to be caused exclusively by groundwater withdrawal [37]. In the Sibari Plain, where the CV estuary is located, subsidence rates up to 20 mm/year are ascribed to a coaction of natural and anthropogenic factors [38]. High relief energy, steep slopes, and severe tectonic fracturing drive shallow and deep-seated gravitational deformations along both the east and west sides of the CV [34][35][36].
The alluvial plane of the CV presents the subsidence of recent deposits, which is similar to the other Quaternary basins of Calabria (Sibari and Gioia Tauro plains) that are adjacent to the CV ( Figure 1). In particular, the Gioia Tauro Plain is similar to the CV in its geological evolution. Both basins belong to the so-called Siculo-Calabrian rift zone (SCRZ) [19], are characterized by crustal extension, and are bounded by normal faults that have produced strong earthquakes in historical times. In the Gioia Tauro Plain, subsidence rates of 10-15 mm/year between 1992-2006 are considered to be caused exclusively by groundwater withdrawal [37]. In the Sibari Plain, where the CV estuary is located, subsidence rates up to 20 mm/year are ascribed to a coaction of natural and anthropogenic factors [38].  (Italian Seismological Instrumental and parametric Data-base (ISIDe) database). Black lines represent "capable faults" (Ithaca database [39]

Remote Sensing Processing Technique
SAR interferometry (InSAR) is a technique that, starting from airborne or spaceborne synthetic aperture radar (SAR) image data, measures the ground surface movements that take place within two acquisitions (e.g., passages of the satellite) over the same area. It is based on the radar concept, that is, the phase of the radar signal returned at the satellite conveys quantitative information on the change in the sensor-to-ground distance caused by the deformation of the detected surface [1,41,42]. By subtracting the phase of the two images, a differential interferogram containing displacement signals is formed after the removal of the phase contributions due to: the topographic relief (using a digital elevation model, or DEM); the earth curvature; the atmospheric contribution due to the atmospheric humidity, temperature, and pressure change between the two acquisitions; the noise introduced by temporal change of the scatterers; a different look angle; and volume scattering [1,41].  (Italian Seismological Instrumental and parametric Data-base (ISIDe) database). Black lines represent "capable faults" (Ithaca database [39]

Remote Sensing Processing Technique
SAR interferometry (InSAR) is a technique that, starting from airborne or spaceborne synthetic aperture radar (SAR) image data, measures the ground surface movements that take place within two acquisitions (e.g., passages of the satellite) over the same area. It is based on the radar concept, that is, the phase of the radar signal returned at the satellite conveys quantitative information on the change in the sensor-to-ground distance caused by the deformation of the detected surface [1,41,42]. By subtracting the phase of the two images, a differential interferogram containing displacement signals is formed after the removal of the phase contributions due to: the topographic relief (using a digital elevation model, or DEM); the earth curvature; the atmospheric contribution due to the atmospheric humidity, temperature, and pressure change between the two acquisitions; the noise introduced by temporal change of the scatterers; a different look angle; and volume scattering [1,41].
With InSAR, a non-intrusive and non-destructive technology, the accuracy in the measurement of the surface movements depends on various factors (atmospheric effects, orbital effects, stability of ground scatterers, unwrapping errors, the wavelength of the signal, etc.). In favorable cases, relative displacement can be detected over time within sub-centimeter accuracy [43]. An order of magnitude improvement in accuracy has been achieved with the development of time series differential SAR interferometry (DInSAR) down to~1 mm/year [44].
In this work, we adopted the SBAS multi-temporal InSAR technique [3] to obtain the displacement time series and the mean ground velocity map. The SBAS algorithm is based on a combination of many SAR differential interferograms that are generated by applying constraints on the temporal and perpendicular baselines. The inversion of the interferometric phase, performed with the singular value decomposition (SVD) technique, produces a ground displacement time series for each coherent pixel, by minimizing possible topographic, atmospheric, and orbital artifacts [2,3,44,45].
We used the SBAS algorithm on the Geohazard Exploitation Platform (GEP), which was developed by ESA (https://geohazards-tep.eo.esa.int). This SBAS version performs three-dimensional (3D) unwrapping through the algorithm proposed by Pepe, A. and Lanari, R. [46]. Atmospheric contribution is removed by "standard" filtering in time and space. In the current online version, no external atmospheric model is implemented. Detailed information about the SBAS version working on the GEP is available at https://geohazards-tep.eo.esa.int, and in De Luca, C. et al. [47].
The processing of Envisat data produces velocity maps in the line of sight (LoS) of satellite SAR geometry (data acquisitions in both ascending and descending orbits). Vertical (up-down) and horizontal (E-W) ground velocity components can be derived by combining ascending and descending LoS velocities. We considered two Envisat datasets (Table 1)  For each dataset, we selected pairs of images by applying constraints on the maximum orbital separation, and the temporal distance between the two passages. The effect of this constraint is to minimize spatial and temporal decorrelation effects [3]. In particular, we chose 350 m and 700 days as the maximum values for the perpendicular and temporal baselines, respectively. We used the 90-m SRTM (Shuttle Radar Topographic Mission) digital elevation model (DEM) for the topography subtraction step ( [48]; http://www2.jpl.nasa.gov/srtm). Processing was performed, for both the ascending and descending data, by applying a multi-looking factor equal to 20 for the azimuth, and 4 for the range direction, resulting in a final ground resolution of 90 m. DInSAR observations derived from ascending and descending orbits allow us to decompose the E-W and the vertical components of the detected deformation. The sum and the difference applied to the mean deformation velocities, which was computed for the ascending and the descending orbits, are calculated for all of the pixels that are coherent in both geometries. Given the namely near-polar sensor orbit direction, the north-south (N-S) component of the deformation cannot be reliably derived.
To retrieve the E-W and vertical components of deformation, the following assumptions are made: (i) ascending and descending radar LoS directions (LoS Asc and LoS Desc, respectively) can be considered as lying on the plane containing the east and z directions; and (ii) the sensor incidence angle ϑ is the same for both ascending and descending geometries. Then, assuming the availability of the displacement measurements along both the ascending (d LoS_Asc ) and the descending (d LoS_Desc ) radar LoS directions, on the basis of simple geometric considerations, we can calculate d East , which is the E-W component of the measured surface deformation for a target that is "observed" from both the ascending and the descending satellite passages: while the vertical component, d Up , is: where θ is the incidence angle at center swath (23 • for ASAR-Envisat data), and d LoS_Desc and d LoS_Asc are the descending and ascending velocities for all of the overlapping pixels, respectively. We validated InSAR LoS velocities through comparison with measurements at CGPS (Continuous Global Positioning System) sites [44] projected on the same ground elements. GPS velocity and its error were projected on the ascending and descending SAR LoS. To compare the two measurements, we averaged the InSAR velocities in a circular buffer (200-m radius) centered on each GPS benchmark. The velocity differences did not exceed ±1.5 mm/year for both the ascending ( Figure 3c the E-W component of the measured surface deformation for a target that is "observed" from both the ascending and the descending satellite passages: while the vertical component, dUp, is: where θ is the incidence angle at center swath (23° for ASAR-Envisat data), and dLoS_Desc and dLoS_Asc are the descending and ascending velocities for all of the overlapping pixels, respectively. We validated InSAR LoS velocities through comparison with measurements at CGPS (Continuous Global Positioning System) sites [44] projected on the same ground elements. GPS velocity and its error were projected on the ascending and descending SAR LoS. To compare the two measurements, we averaged the InSAR velocities in a circular buffer (200-m radius) centered on each GPS benchmark. The velocity differences did not exceed ±1.5 mm/year for both the ascending (  Ground visibility depends on the angle formed by the satellite's LoS and the surface slope. Areas located along the eastern CV sides are more visible on Envisat ascending tracks, while areas on the western side are more visible on the descending tracks. The quality of our data for the CV area, then, depends on acquisition geometry and visibility, which in turn depend on the alluvial plain's disposition in space. (Figure 4). Ground visibility depends on the angle formed by the satellite's LoS and the surface slope. Areas located along the eastern CV sides are more visible on Envisat ascending tracks, while areas on the western side are more visible on the descending tracks. The quality of our data for the CV area, then, depends on acquisition geometry and visibility, which in turn depend on the alluvial plain's disposition in space. (Figure 4). The spatial distribution of pixels appears as a regular grid after rounding the coordinates to significant digits. The "capable faults" of the Ithaca database [38] are plotted on the maps.

Alluvial Plain
The vertical component ( Figure 5a) shows several subsiding areas, of approximately −7 mm/year, in the flat part of the valley near populated areas such as the suburbs of Rende town (Quattromiglia; inset in Figure 5b). The Quattromiglia urban area underwent a fast growth during the last 50 years, as confirmed by images acquired at different times ( Figure 6a). The ascending time series shows an increase of movement away from the sensor (mainly related to subsidence) in two of the five sites within zones 1 and 2 in the Quattromiglia area (Figure 6b), where urban sprawl has occurred. Only a small displacement is present in zone 3 (industrial area), and stability (~0 mm/year) is detected in zones 4 and 5, which have not been affected by urban expansion (Figure 6b). The descending time series (Figure 6b) confirms that the largest displacement is in zones 1 and 2, and that displacement is small or absent in the other zones (3, 4, and 5). The spatial distribution of pixels appears as a regular grid after rounding the coordinates to significant digits. The "capable faults" of the Ithaca database [38] are plotted on the maps.

Alluvial Plain
The vertical component (Figure 5a) shows several subsiding areas, of approximately −7 mm/year, in the flat part of the valley near populated areas such as the suburbs of Rende town (Quattromiglia; inset in Figure 5b). The Quattromiglia urban area underwent a fast growth during the last 50 years, as confirmed by images acquired at different times ( Figure 6a). The ascending time series shows an increase of movement away from the sensor (mainly related to subsidence) in two of the five sites within zones 1 and 2 in the Quattromiglia area (Figure 6b), where urban sprawl has occurred. Only a small displacement is present in zone 3 (industrial area), and stability (~0 mm/year) is detected in zones 4 and 5, which have not been affected by urban expansion (Figure 6b). The descending time series (Figure 6b) confirms that the largest displacement is in zones 1 and 2, and that displacement is small or absent in the other zones (3, 4, and 5). To check for hydrogeological effects in the same area, we compared the water table level of 1979 (Celico, F. et al. [49]) with the piezometric levels of two more recent periods, 1990-1992 and 2000-2005, using the "Indagini Nel Sottosuolo" dataset (http://sgi.isprambiente.it/GMV2/index.html). Our goal is the reconstruction of the long term variations of the water table without taking into account seasonal oscillations, since they cannot be investigated with the available piezometric data. From these data, we derived a map of water   To check for hydrogeological effects in the same area, we compared the water table level of 1979 (Celico, F. et al. [49]) with the piezometric levels of two more recent periods, 1990-1992 and 2000-2005, using the "Indagini Nel Sottosuolo" dataset (http://sgi.isprambiente.it/GMV2/index.html). Our goal is the reconstruction of the long term variations of the water table without taking into account seasonal oscillations, since they cannot be investigated with the available piezometric data. From these data, we derived a map of water      No significant surface deformation was observed in the correspondence of Cosenza old town with COSMO-SkyMed and ground-based radar sensors data by [50]. In particular, we observed vertical component values of 0.23, 0.02, and −0.37 mm/year (Figure 8a) close to the Saint Augustine Monumental compound. This area was investigated in detail by Montuori, A. et al. [50]. In the western sector of Cosenza, the vertical component showed a weak uplift (max value 2.3 mm/year; mean value 0.76 mm/year) for parts of the town that were built on Pliocene deposits (clay, sandstone, and conglomerate) and the Quaternary conglomerates representing fluvial terraces (Figure 8b). On the other hand, for the eastern sector, a weak subsidence (up to 1.6 mm/year) was instead observed for the areas that were built on Quaternary alluvial deposits.  No significant surface deformation was observed in the correspondence of Cosenza old town with COSMO-SkyMed and ground-based radar sensors data by [50]. In particular, we observed vertical component values of 0.23, 0.02, and −0.37 mm/year (Figure 8a) close to the Saint Augustine Monumental compound. This area was investigated in detail by Montuori, A. et al. [50]. In the western sector of Cosenza, the vertical component showed a weak uplift (max value 2.3 mm/year; mean value 0.76 mm/year) for parts of the town that were built on Pliocene deposits (clay, sandstone, and conglomerate) and the Quaternary conglomerates representing fluvial terraces (Figure 8b). On the other hand, for the eastern sector, a weak subsidence (up to 1.6 mm/year) was instead observed for the areas that were built on Quaternary alluvial deposits.  No significant surface deformation was observed in the correspondence of Cosenza old town with COSMO-SkyMed and ground-based radar sensors data by [50]. In particular, we observed vertical component values of 0.23, 0.02, and −0.37 mm/year (Figure 8a) close to the Saint Augustine Monumental compound. This area was investigated in detail by Montuori, A. et al. [50]. In the western sector of Cosenza, the vertical component showed a weak uplift (max value 2.3 mm/year; mean value 0.76 mm/year) for parts of the town that were built on Pliocene deposits (clay, sandstone, and conglomerate) and the Quaternary conglomerates representing fluvial terraces (Figure 8b). On the other hand, for the eastern sector, a weak subsidence (up to 1.6 mm/year) was instead observed for the areas that were built on Quaternary alluvial deposits.

Eastern and Western Sides (CV Slopes)
Subsidence is also locally present along the eastern and western boundaries of the valley ( Figure  5a CV slopes are affected by widespread gravitative (landslides) phenomena. The PAI map ( Figure  9; Table 2), which is based on geomorphological criteria, defines three states of activity ("active", "dormant", and "inactive"), and identifies several typologies of landslide phenomena. In particular, for CV, the landslide inventory distinguishes 586 rotational slides, nine earth flows, six rock falls, two deep-seated gravitational movements, 139 complex landslides, 115 surface landslide areas, and 223 deep landslide areas. Over a total number of 1080 landslides, the CV slopes are characterized by a prevalence of dormant phenomena (81.3%) with respect to active (18.2%) and inactive (0.5%).

Eastern and Western Sides (CV Slopes)
Subsidence is also locally present along the eastern and western boundaries of the valley (Figure 5a CV slopes are affected by widespread gravitative (landslides) phenomena. The PAI map ( Figure 9; Table 2), which is based on geomorphological criteria, defines three states of activity ("active", "dormant", and "inactive"), and identifies several typologies of landslide phenomena. In particular, for CV, the landslide inventory distinguishes 586 rotational slides, nine earth flows, six rock falls, two deep-seated gravitational movements, 139 complex landslides, 115 surface landslide areas, and 223 deep landslide areas. Over a total number of 1080 landslides, the CV slopes are characterized by a prevalence of dormant phenomena (81.3%) with respect to active (18.2%) and inactive (0.5%).      The CLC categories are reduced to three classes: "urbanized areas and bare rocks" (class I); "cultivated areas and bare soils" (class II); and "vegetated areas and inland waters" (class III). Class II covers 63.5% of the total, while classes III and I occupy 33.3% and 3.2%, respectively. For each class, we computed the number and density of coherent pixels derived from InSAR ( Figure 10). For both satellite tracks, the highest coherent pixel density is recorded in class I (102.46 pixels/km 2 and 20.66 pixels/km 2 , respectively). Classes II and III correspond to a density of 6.11 pixels/km 2 and 1.25 pixels/km 2 for the ascending orbit, and 1.22 pixels/km 2 and 0.19 pixels/km 2 for the descending orbit, respectively. This analysis confirms that the urbanized areas (class I) are the best in terms of InSAR pixel coherence, while the vegetated areas (class III) are the worst.  The CLC categories are reduced to three classes: "urbanized areas and bare rocks" (class I); "cultivated areas and bare soils" (class II); and "vegetated areas and inland waters" (class III). Class II covers 63.5% of the total, while classes III and I occupy 33.3% and 3.2%, respectively. For each class, we computed the number and density of coherent pixels derived from InSAR ( Figure 10). For both satellite tracks, the highest coherent pixel density is recorded in class I (102.46 pixels/km 2 and 20.66 pixels/km 2 , respectively). Classes II and III correspond to a density of 6.11 pixels/km 2 and 1.25 pixels/km 2 for the ascending orbit, and 1.22 pixels/km 2 and 0.19 pixels/km 2 for the descending orbit, respectively. This analysis confirms that the urbanized areas (class I) are the best in terms of InSAR pixel coherence, while the vegetated areas (class III) are the worst. We also focused our attention on Lungro village, which is located in the northwest (Figure 11a), where both subsidence and landslides occur [51][52][53][54]. Landslides affecting the urban areas have been known to occur in this area since historical times [51,52]. By integrating geological/geomorphological studies, inclinometer measurements, and InSAR data, Antronico, L. et al. [53] explain the damage of buildings as due to slow landslides in the historical center and landslide remobilization in the new urban area (San Leonardo). Guerricchio, A. et al. [54] recorded subsidence of several tens of centimeters affecting San Leonardo hill (Figure 11b).
The Lungro salt mine at the San Leonardo foothill (Figure 11a) has been exploited from the Magna Grecia epoch (starting in the 8 th century BC) up until 1976 [55]. The Messinian halite deposits extracted by means of dry mining operations along five levels at different depths that vary from −90 m to −240 m, have a total length of ~400 m, and a width of ~250 m. For this site, only ascending data We also focused our attention on Lungro village, which is located in the northwest (Figure 11a), where both subsidence and landslides occur [51][52][53][54]. Landslides affecting the urban areas have been known to occur in this area since historical times [51,52]. By integrating geological/geomorphological studies, inclinometer measurements, and InSAR data, Antronico, L. et al. [53] explain the damage of buildings as due to slow landslides in the historical center and landslide remobilization in the new urban area (San Leonardo). Guerricchio, A. et al. [54] recorded subsidence of several tens of centimeters affecting San Leonardo hill (Figure 11b).
The Lungro salt mine at the San Leonardo foothill (Figure 11a) has been exploited from the Magna Grecia epoch (starting in the 8th century BC) up until 1976 [55]. The Messinian halite deposits extracted by means of dry mining operations along five levels at different depths that vary from −90 m to −240 m, have a total length of~400 m, and a width of~250 m. For this site, only ascending data are available. Three different time series (Figure 11c) were acquired simultaneously in the San Leonardo hill and along the topographic survey profile of Guerricchio, A. et al. [54] (Figure 11b). Ground displacements up to −20 mm were observed in the central sector, near the salt mine (time series 2), and smaller ones were observed at the hill borders (time series 1 and 2; Figure 11c). Our observations are in agreement with Guerricchio, A. et al. [54] (Figure 11b).  (Figure 11c) were acquired simultaneously in the San Leonardo hill and along the topographic survey profile of Guerricchio, A. et al. [54] (Figure 11b). Ground displacements up to −20 mm were observed in the central sector, near the salt mine (time series 2), and smaller ones were observed at the hill borders (time series 1 and 2; Figure 11c). Our observations are in agreement with Guerricchio, A. et al. [54] (Figure 11b). Figure 11. (a) Ascending ground velocity map for Lungro village. Violet triangles represent the measurements of electrical conductivity (mS/cm) along the streams, testifying the still active halite dissolution. See the inset in Figure 3a for the area location; (b) monitoring of subsidence phenomena in the San Leonardo locality by means of topographic survey (from Guerricchio, A. et al. [54]); (c) Envisat ascending time series for the investigated area; their location is shown by black circles in Figure 11a. The gray background represents the temporal interval covered also by topographic data.

Discussion
We now try to explain the deformation patterns in the CV by comparing our InSAR observations with available geological, tectonic, and geomorphological information.

Subsidence, CV Plain
Subsidence affects the CV alluvial plain where important infrastructures are present, such as the A3 highway (Salerno-Reggio Calabria, Figure 4a). Although the CV graben is bounded by active N- Figure 11. (a) Ascending ground velocity map for Lungro village. Violet triangles represent the measurements of electrical conductivity (mS/cm) along the streams, testifying the still active halite dissolution. See the inset in Figure 3a for the area location; (b) monitoring of subsidence phenomena in the San Leonardo locality by means of topographic survey (from Guerricchio, A. et al. [54]); (c) Envisat ascending time series for the investigated area; their location is shown by black circles in Figure 11a. The gray background represents the temporal interval covered also by topographic data.

Discussion
We now try to explain the deformation patterns in the CV by comparing our InSAR observations with available geological, tectonic, and geomorphological information.

Subsidence, CV Plain
Subsidence affects the CV alluvial plain where important infrastructures are present, such as the A3 highway (Salerno-Reggio Calabria, Figure 4a). Although the CV graben is bounded by active N-S striking normal faults that cause widespread seismicity (Figure 2), we are not able to detect a tectonic signal associated with subsidence. This is probably due to the resolution and short time span of our satellite observations compared to the time scale of the subsidence and tectonic processes in the region.
Then, we need to focus our attention on other detectable causes of ground motion. Water depletion in compressible soils could increase the long-term subsidence processes, as observed in other Quaternary basins [56,57]. We do find a correspondence between areas characterized by groundwater drawdowns and the main subsidence rates detected by InSAR (Figure 7). Compaction of the sedimentary infill in the alluvial plain, which thickens towards the Sila foothills (Spina, V. et al. [24]), is a predominant factor affecting ground movement. Groundwater exploitation in this area is testified by the presence of numerous active wells (Figure 2).
Another factor behind sediment consolidation and subsidence in the CV is urban expansion (e.g., the Quattromiglia area, Figure 6). This link has also been observed in other quaternary basins (see for example Polcari, M. et al. [58]).
At Cosenza, the transition between weak uplift in the eastern sector and weak subsidence in the west could be explained by the lateral passage from the Pliocene deposits and Quaternary conglomerates to the more compressible alluvial Quaternary sediments. More detailed information, such as hydrogeological data, is necessary to better explain the weak uplift observed in eastern Cosenza sector, but it is not currently available.
Finally, we detected another cause of ground motion. From our detailed analysis, we detected ground instability close to San Leonardo hill in the Lungro village area, which is most likely related to salt mining. The high water electrical conductivity (around 6 mS/cm) recorded downstream from the salt mine in the Burrone della Salina stream (5.7 mS/cm) and the Fiumicello River (6.3 mS/cm; Figure 11a) are indicative of the active halite dissolution in the mine. From the ascending data, we deduced the greatest displacement in the middle sector of the San Leonardo hill, which we explain as due to subsidence after taking into account the topographic survey of Guerricchio, A. et al. [54]. The observed deformation and the still active halite dissolution suggest the key role of subsidence as a triggering and controlling factor for landslides. This type of subsidence has been observed in other old mines, such as for example the Wieliczka salt mine (Poland) [59] and the Pasquasia potassic salt mine (Italy) [60].

Landslides, CV Slopes
The visibility of landslide-affected areas depends on the SAR acquisition geometry. The eastern slope is more visible on the ascending Envisat track, while the western slope is more visible on the descending track ( Figure 12). Along the slopes, ground movement is observed for an area extending 86 km 2 : Of this activity, 46.2% is visible in the ascending track, and 17.7% is visible in the descending track. Only a small part of the landslides mapped by the PAI inventory is covered by coherent pixels (recording the slope movement). In particular, ascending coherent pixels cover 19.7% of the PAI landslides, while descending coherent pixels cover 6.5% of the landslides in PAI. In this calculation, the landslide area should include at least one coherent pixel.
We take 1.5 mm/year as the displacement rate threshold to discriminate movement from absence of movement, in agreement with Cascini, L. et al. [10]. Under this assumption, some of our observations point to slope movement in areas that the PAI identifies as dormant landslides, or an absence of movement in areas where landslides are classified as active ( Figure 13). This disagreement could be due to the evolution of the landslides over time (i.e., reactivation and stabilization) following the state of activity that was defined in the PAI (compiled in 2001, last update in 2011).   (b) evidence of movement or no movement according to ascending data for a portion of the investigated area. We compared the state of activity inferred by the PAI catalogue and the velocities of the SAR pixels included in the landslides bodies (e.g., Figure 12a).
To further explore this finding, we selected and analyzed the time series of four cases in the group of landslides that are classified as dormant in PAI, but show evidence of movement from SAR. We examined the ascending time series (Figure 14b) of MM1, MM12 (both are deep landslide areas located close to Marano Marchesato village), and RG7 (a complex landslide area close to Rota Greca municipality), and the descending time series (Figure 14c  (b) evidence of movement or no movement according to ascending data for a portion of the investigated area. We compared the state of activity inferred by the PAI catalogue and the velocities of the SAR pixels included in the landslides bodies (e.g., Figure 12a).
To further explore this finding, we selected and analyzed the time series of four cases in the group of landslides that are classified as dormant in PAI, but show evidence of movement from SAR. We examined the ascending time series (Figure 14b  The increase in 2009 can be correlated to the rainfall peak recorded by the Cosenza and Montalto rain gauges (Arpacal-http://www.cfd.calabria.it/index.php/dati-stazioni). Also, the descending time series of the ROS13 landslide (Figure 14b) shows increasing displacement, with the maximum rate occurring between the end of 2008 and the beginning of 2009. This maximum value is associated to a rainfall peak occurring during the same period (recorded near San Pietro in the Guarano rain gauge of ArpaCal). The activity of these landslides (MM1 and ROS13) is testified by the deformation of ground and man-made structures (Figure 14d,e). Furthermore, the increase of landslide displacement, recorded during the winter of 2008-2009, is likely due to hydrogeological instability caused by intense rainfall in the Calabria region during the same period [61].
Finally, the mean displacement rates inferred by SAR suggest that some landslides mapped by PAI as dormant might in fact be active, similarly to the Rende and San Fili areas, as proposed by Bianchini, S. et al. [62].

Conclusions
The Crati Valley (CV) in the Calabria Region (Southern Italy) is located in one of the most riskprone areas of the Italian peninsula in terms of seismic and hydrogeological hazard. Observation of ground deformation from space has become a fundamental tool in hazard evaluation. Our analysis of remote sensed SAR data shows that the spatial and temporal evolution of ground deformation in The increase in 2009 can be correlated to the rainfall peak recorded by the Cosenza and Montalto rain gauges (Arpacal-http://www.cfd.calabria.it/index.php/dati-stazioni). Also, the descending time series of the ROS13 landslide (Figure 14b) shows increasing displacement, with the maximum rate occurring between the end of 2008 and the beginning of 2009. This maximum value is associated to a rainfall peak occurring during the same period (recorded near San Pietro in the Guarano rain gauge of ArpaCal). The activity of these landslides (MM1 and ROS13) is testified by the deformation of ground and man-made structures (Figure 14d,e). Furthermore, the increase of landslide displacement, recorded during the winter of 2008-2009, is likely due to hydrogeological instability caused by intense rainfall in the Calabria region during the same period [61].
Finally, the mean displacement rates inferred by SAR suggest that some landslides mapped by PAI as dormant might in fact be active, similarly to the Rende and San Fili areas, as proposed by Bianchini, S. et al. [62].

Conclusions
The Crati Valley (CV) in the Calabria Region (Southern Italy) is located in one of the most risk-prone areas of the Italian peninsula in terms of seismic and hydrogeological hazard. Observation of ground deformation from space has become a fundamental tool in hazard evaluation. Our analysis of remote sensed SAR data shows that the spatial and temporal evolution of ground deformation in the CV mainly occurs as subsidence in the alluvial plane and landslides on the slopes. We consider InSAR data processed from more than 70 images collected along both ascending and descending Envisat tracks during April 2003-September 2010. A main objective of this work was to determine the possible causes behind ground motion by comparing SAR data with other independent observations. Subsidence is the result of several mechanisms. In our case, the compaction of the recent alluvial deposits seems to be the main cause. There is an anthropogenic contribution to the observed subsidence, namely compaction by groundwater exploitation and fast urban sprawl. We are not able to detect the tectonic contribution to the signal associated with subsidence, although we cannot exclude the presence of vertical movements due to the activity of normal faults bounding the valley.
A comparison of our observations with the PAI landslide catalog shows discrepancies between the activity classified in the catalog, and the displacement measured by InSAR. In particular, some landslide activity might be underestimated in the PAI. We believe that satellite-based techniques represent a useful tool to update large landslides catalogs, as also suggested by other authors (e.g., Cascini, L. et al. [63]. Our detailed analysis of the S. Leonardo locality (Lungro village) shows that a landslide near the old salt mine caused westward and high subsidence velocities. This particular case is interesting, since it shows the interaction between subsidence and landslide activity. Here, halite dissolution and the collapse due to mine excavations represent both a triggering and control mechanism for the S. Leonardo landslide.
The next steps to increase our knowledge of ground deformation space-time evolution in this area are: (1) to extend the satellite SAR displacement time series; and (2) to add data from other missions, especially from the higher resolution and/or recent acquisition systems (e.g., COSMO-SkyMed and Sentinel-1 constellations).
This effort might allow for further discrimination of the different causes of ground displacement, in particular the tectonic contribution.