Comment on “Pre-Collapse Space Geodetic Observations of Critical Infrastructure: The Morandi Bridge, Genoa, Italy” by Milillo et al. (2019)

: We present in this comment a Multi-Temporal SAR Interferometry (MT-InSAR) analysis showing that the results published by Milillo et al. (2019) in the Remote Sensing Journal, presenting the evidence of space geodetic observations relevant to displacements occurring before the collapse of the Morandi Bridge, happened in Genova (Italy) on the 14 August 2018, are questionable. In particular, we focus on the InSAR results obtained by Milillo et al. (2019) by processing the 3 m × 3 m resolution COSMO-SkyMed (CSK) data collected from ascending and descending orbits on the area of interest. These results, thanks to the high spatial resolution and the short revisit time characterizing this multi-orbit SAR dataset, represent the cornerstone of their analysis. The main ﬁndings of their study allow Milillo et al. to conclude that the InSAR processing of this COSMO-SkyMed dataset reveals the increased deformation magnitude over time of points located near the strands of the deck next to the collapsed pier, between 12 March 2017 and August 2018. In this comment, we show the results obtained by the IREA-CNR SAR team after processing the same ascending and descending CSK dataset, but by using two alternative and independent processing techniques: the Small BAseline Subset (SBAS) and the Advanced Tomographic SAR (TomoSAR) approaches, respectively. Our analysis shows that, although both the SBAS and the TomoSAR analyses allow achieving denser coherent pixel maps relevant to the Morandi bridge, nothing of the pre-collapse large displacements reported in Milillo et al. (2019) appears in our results, leading us to deeply disagree with the ﬁndings of their InSAR analysis.


Introduction
On the 14 August 2018, the Polcevera viaduct-known as the Morandi bridge-in the city of Genova (Italy), which connects the eastern and western sides of the Genova metropolitan area and represents the main highway to link Southern France and Northern Italy, suffered the collapse of a section of the deck approximately 240 m long, adjacent to pier 9, close to the Polcevera river, causing the death of 43 people, and leaving 11 injured [1].
for the representation of the ground coverage of the two CSK frames), over the time interval extending from January 2011 to August 2018 (immediately before the Morandi bridge collapse) on an area close to the Morandi bridge, in the city of Genova (North Italy). More specifically, the two datasets are composed of 132 and 134 images for the ascending and descending passes, respectively. For both datasets, the acquisition mode is the H-IMAGE, corresponding to the standard Stripmap acquisition mode [21], characterized by a ground spatial resolution of about 3 m × 3 m in azimuth (along-track) and slant-range (cross-track) directions [22] The reference acquisition beam is the H4-05 for the ascending dataset and H4-01 for the descending one, corresponding to an incidence angle at scene center of about 34 • and 27 • , respectively. The key parameters for the two SAR datasets exploited in this work are briefly summarized in Table 1.
Remote Sens. 2020, 12, x FOR PEER REVIEW 3 of 17 area close to the Morandi bridge, in the city of Genova (North Italy). More specifically, the two datasets are composed of 132 and 134 images for the ascending and descending passes, respectively. For both datasets, the acquisition mode is the H-IMAGE, corresponding to the standard Stripmap acquisition mode [21], characterized by a ground spatial resolution of about 3 m × 3 m in azimuth (along-track) and slant-range (cross-track) directions [22] The reference acquisition beam is the H4-05 for the ascending dataset and H4-01 for the descending one, corresponding to an incidence angle at scene center of about 34° and 27°, respectively. The key parameters for the two SAR datasets exploited in this work are briefly summarized in Table 1.  The distributions of the selected interferometric SAR acquisitions in the temporal/perpendicular baseline plane for the two CSK datasets are pictorially shown in Figure 2. It is worth remarking the temporal correlation of the perpendicular baseline distribution (and thus of the corresponding orbits), which has to be properly taken into account throughout the interferometric processing.
In this paper, we focus on the analysis of the pre-collapse displacements associated with the Morandi bridge (see the white rectangle in Figure 1 for its location), which are extensively Figure 1. Representation of the two exploited COSMO-SkyMed (CSK) frames acquired from ascending (black rectangle) and descending (red rectangle) orbits, superimposed on a Google Earth optical image. Note that the white rectangle falling within the two frames represents the area of interest surrounding the Polcevera viaduct close to the Morandi bridge. The distributions of the selected interferometric SAR acquisitions in the temporal/perpendicular baseline plane for the two CSK datasets are pictorially shown in Figure 2. It is worth remarking the temporal correlation of the perpendicular baseline distribution (and thus of the corresponding orbits), which has to be properly taken into account throughout the interferometric processing.
Remote Sens. 2020, 12, x FOR PEER REVIEW 4 of 17 investigated through two independent MT-InSAR techniques referred to as the Small BAseline Subset (SBAS) and the Advanced Tomographic SAR (TomoSAR) approaches. A brief summary of these two techniques is reported in the following two paragraphs. Figure 2. SAR data representation in the temporal/perpendicular baseline plane for the two CSK datasets relevant to the area of interest. The black triangles and the red diamonds represent the ascending and descending acquisitions, respectively. The reference (master) images selected for the processing of the ascending and descending CSK datasets, relevant to the 29 October 2014 and to the 3 February 2015 acquisitions, respectively, are identified by the arrows.

The SBAS Technique
The available ascending and descending CSK datasets were processed through the SBAS technique [5][6][7], which is an advanced SAR Interferometry method that allows retrieving useful information on the spatial and temporal patterns of the detected radar Line of Sight (LOS)-projected displacements, through the generation of deformation time series and mean velocity maps. The SBAS approach relies on a proper selection of the SAR data pairs used to generate a multi-temporal sequence of differential interferograms (i.e., the phase difference between two SAR images collected over the same area at different epochs), which are characterized by a small separation between the acquisition orbits (short temporal and perpendicular baselines).
The use of small spatial and temporal baselines of the generated interferograms permits mitigating the noise effects (decorrelation phenomena) that affect the interferometric pairs [23], thus increasing the spatial density of the retrieved InSAR measurements (number of coherent pixels), especially in semi-urbanized and rural areas. The deformation time series (and the corresponding mean velocity maps) are then computed by solving a linear system of equations in a least-squares sense, with a minimum norm energy constraint (the Singular Value Decomposition-SVD technique is applied in presence of independent acquisition subsets separated by large baselines), and with an accuracy of about 1-2 mm/year for what concerns the mean deformation velocity information and 5-10 mm for the single deformation measurement [3,24]. Moreover, within the SBAS processing chain, a filtering operation is performed to detect and remove atmospheric artifacts from the displacement time series [5].
One key point of the SBAS algorithm is the capability of performing both multi-sensor [7,25] and multi-scale [6,10] deformation analysis; in particular, the latter allows generating InSAR products at two spatial resolution scales, referred to as a regional and a local one. This is accomplished by dealing with multi-look interferograms (regional scale analysis with a typical spatial resolution of about 30-90 m), and full-resolution interferograms (local scale analysis, generated from the single-look data, with the full spatial resolution, 3 m for this CSK case), respectively. The regional scale is suitable to detect low-pass signal components, such as deformation related to large areas or atmospheric artifacts, whereas the local scale one is particularly appropriate to investigate high-pass signals, such . SAR data representation in the temporal/perpendicular baseline plane for the two CSK datasets relevant to the area of interest. The black triangles and the red diamonds represent the ascending and descending acquisitions, respectively. The reference (master) images selected for the processing of the ascending and descending CSK datasets, relevant to the 29 October 2014 and to the 3 February 2015 acquisitions, respectively, are identified by the arrows.
In this paper, we focus on the analysis of the pre-collapse displacements associated with the Morandi bridge (see the white rectangle in Figure 1 for its location), which are extensively investigated through two independent MT-InSAR techniques referred to as the Small BAseline Subset (SBAS) and the Advanced Tomographic SAR (TomoSAR) approaches. A brief summary of these two techniques is reported in the following two paragraphs.

The SBAS Technique
The available ascending and descending CSK datasets were processed through the SBAS technique [5][6][7], which is an advanced SAR Interferometry method that allows retrieving useful information on the spatial and temporal patterns of the detected radar Line of Sight (LOS)-projected displacements, through the generation of deformation time series and mean velocity maps. The SBAS approach relies on a proper selection of the SAR data pairs used to generate a multi-temporal sequence of differential interferograms (i.e., the phase difference between two SAR images collected over the same area at different epochs), which are characterized by a small separation between the acquisition orbits (short temporal and perpendicular baselines).
The use of small spatial and temporal baselines of the generated interferograms permits mitigating the noise effects (decorrelation phenomena) that affect the interferometric pairs [23], thus increasing the spatial density of the retrieved InSAR measurements (number of coherent pixels), especially in semi-urbanized and rural areas. The deformation time series (and the corresponding mean velocity maps) are then computed by solving a linear system of equations in a least-squares sense, with a minimum norm energy constraint (the Singular Value Decomposition-SVD technique is applied in presence of independent acquisition subsets separated by large baselines), and with an accuracy of about 1-2 mm/year for what concerns the mean deformation velocity information and 5-10 mm for the single deformation measurement [3,24]. Moreover, within the SBAS processing chain, a filtering operation is performed to detect and remove atmospheric artifacts from the displacement time series [5].
One key point of the SBAS algorithm is the capability of performing both multi-sensor [7,25] and multi-scale [6,10] deformation analysis; in particular, the latter allows generating InSAR products at two spatial resolution scales, referred to as a regional and a local one. This is accomplished by dealing with multi-look interferograms (regional scale analysis with a typical spatial resolution of about 30-90 m), and full-resolution interferograms (local scale analysis, generated from the single-look data, with the full spatial resolution, 3 m for this CSK case), respectively. The regional scale is suitable to detect low-pass signal components, such as deformation related to large areas or atmospheric artifacts, whereas the local scale one is particularly appropriate to investigate high-pass signals, such as those related to very small displacements associated with infrastructures or parts of them. Such a pioneer peculiarity of the two-scale SBAS approach has influenced the development of various MT-InSAR techniques and has deeply contributed to its wide dissemination within the Solid Earth science community, since it reveals to be particularly suitable for a wide range of applications, from civil protection scenarios (natural hazard risk prevention and mitigation, volcano, seismic events, landslides) to anthropogenic contexts (extended urban areas, archaeological and historical sites, oil-gas extraction, structures and transport infrastructures, mine activity, etc.) [13,14,[26][27][28][29][30][31][32][33][34].
In this work, the two-scale SBAS approach [6] was exploited to process both the ascending and descending CSK datasets of our case study, allowing us to deeply investigate the spatial and temporal evolution of the possible pre-event displacements affecting the Morandi bridge. Moreover, additional information on the residual topography (with respect to the used DEM) of the coherent pixels, as well as on the thermal dilation components associated with the daily temperatures, is also retrieved.

The TomoSAR Technique
SAR Tomography (TomoSAR) [9] is a processing technology that extends the Persistent Scatterers Interferometry (PSI) [35,36] method for the analysis of pixels at the full available resolution of the data. It brings the imaging concept of the synthetic aperture in the along-track direction, which allows achieving azimuth high-resolution SAR images, also in a third dimension, referred to as elevation, associated with the spatial (perpendicular) separation of the orbits and orthogonal to the azimuth/slant-range plane. SAR Tomography switches the processing from the matching of the measured phases with the expected model related to the parameters of interest, i.e., height and velocity, firstly introduced by PSI phase matching into the imaging of the backscattered profile [8,9]. SAR Tomography extends the synthetic aperture processing, which allows achieving high spatial resolution images, to the height direction based on the availability of multiple observations from slightly different orbits. Therefore, the signal backscattered from ground objects can be focused at high resolution also in the third dimension (height), hence, the name 3D imaging [37][38][39]. By including in the processing also the temporal diversity of the acquisitions, space/velocity model (4D imaging) is derived by extending the 3D imaging concept into the time domain to measure the deformation parameters of any temporal coherent Persistent Scatterer (PS) in the focused 3D space [8,40]. The imaging domain can be further extended to even monitor the slight movements induced by the thermal dilation of the imaged structures [19,41,42].
In this paper, TomoSAR processing is performed according to the procedure described in [9]. It is based on a cascade of two main blocks operating at different spatial scales: the first block, operating on a set of spatially averaged interferograms belonging to a grid set according to the Small-Baseline criteria [5,43], allows carrying out the estimation and compensation of the atmospheric phase patterns, as well as an estimation of the non-linear deformation at a lower resolution. These two-phase component estimates provide the necessary input for the phase calibration of the data which is mandatory for the implementation of the TomoSAR imaging at the full resolution. This latter, which represents the second and core step of the processing, is implemented through the procedure in [8] based on the exploitation of the Beamforming filter, i.e., the matched filter, which measures the correlation of the measured phase vector with the expected phase model accounting for the geophysical parameters of interest.
For this specific analysis, the TomoSAR processing was carried out through the implementation of 5D imaging, accounting for residual topography, linear temporal deformation and possible deformation components correlated with the daily average temperature of the area induced by thermal dilation of materials [41]. The analysis was limited to the single scatterers: the detection of output measurements points was implemented by applying the Generalized Likelihood Ratio Test (GLRT) decision scheme [44] which is based on the comparison of the normalized correlation among the measurement vector and the expected phase model, with a threshold set according to the designed probability of false alarm. The test is applied to every image pixel: neither pre-processing candidates selection nor post-processing outlier filtering is performed.

Results
The results of the CSK analysis at the full available spatial resolution obtained through the independent implementation of the SBAS and of the TomoSAR processing techniques in the area of interest, including the Morandi bridge, are reported in Figure 3.

Results
The results of the CSK analysis at the full available spatial resolution obtained through the independent implementation of the SBAS and of the TomoSAR processing techniques in the area of interest, including the Morandi bridge, are reported in Figure 3. This figure shows the estimated LOS deformation mean velocity maps for the ascending (upper row) and descending (bottom row) datasets relevant to both the SBAS (left column) and TomoSAR (right column) processing techniques. The maps are represented in the North-oriented geometry; for all the results, the colormap is set according to the estimated mean velocity with the typical convention that negative velocities are associated with scatterers that are moving away from the sensor along the LOS. The images show a very high density of measurement points, with a good agreement between the results obtained through the two MT-InSAR processing techniques, both in terms of coverage and density, and measured deformation. A comparable density of measurement points is finally achieved on both ascending and descending datasets, thus proving a satisfactory sampling of the data over the two orbits. Note also that the reference pixels for the two MT-InSAR processing are independently selected in different stable areas far away from the bridge: the agreement of the achieved results demonstrates the robustness of the applied processing approaches with respect to the spatial reference selection. The position of the Morandi bridge is easily recognizable: being a raised structure, it causes the presence of a sort of straight strip, induced by the shadow effect, crossing the image from left to right, characterized by the absence of measurement points. Bridge points are located on the southern and northern boundaries of that strip for ascending and descending data, respectively. By analyzing the maps, no significant deformation is, at that scale, identified. This figure shows the estimated LOS deformation mean velocity maps for the ascending (upper row) and descending (bottom row) datasets relevant to both the SBAS (left column) and TomoSAR (right column) processing techniques. The maps are represented in the North-oriented geometry; for all the results, the colormap is set according to the estimated mean velocity with the typical convention that negative velocities are associated with scatterers that are moving away from the sensor along the LOS. The images show a very high density of measurement points, with a good agreement between the results obtained through the two MT-InSAR processing techniques, both in terms of coverage and density, and measured deformation. A comparable density of measurement points is finally achieved on both ascending and descending datasets, thus proving a satisfactory sampling of the data over the two orbits. Note also that the reference pixels for the two MT-InSAR processing are independently selected in different stable areas far away from the bridge: the agreement of the achieved results demonstrates the robustness of the applied processing approaches with respect to the spatial reference selection. The position of the Morandi bridge is easily recognizable: being a raised structure, it causes Remote Sens. 2020, 12, 4011 7 of 16 the presence of a sort of straight strip, induced by the shadow effect, crossing the image from left to right, characterized by the absence of measurement points. Bridge points are located on the southern and northern boundaries of that strip for ascending and descending data, respectively. By analyzing the maps, no significant deformation is, at that scale, identified.
The detailed analysis of the results over the Morandi bridge is now in order. To assess the bridge deformation behavior in the observed period, we firstly isolated, from the overall results in Figure 3, the measurement points located at the bridge deck. Specifically, we firstly isolated measurement points located on the northern roadway, approximately framed within the yellow rectangle in Figure 4. A threshold of 40 m on the estimated height, above the sea level, of the coherent pixels was also applied to isolate the raised points, only. The distribution of the detected pixels shows a concentration near the support towers. The lack of coherent points in the central part of the suspended deck may be induced by large non-linear displacements possibly induced by vibration effects and wind.
Remote Sens. 2020, 12, x FOR PEER REVIEW 7 of 17 The detailed analysis of the results over the Morandi bridge is now in order. To assess the bridge deformation behavior in the observed period, we firstly isolated, from the overall results in Figure 3, the measurement points located at the bridge deck. Specifically, we firstly isolated measurement points located on the northern roadway, approximately framed within the yellow rectangle in Figure  4. A threshold of 40 m on the estimated height, above the sea level, of the coherent pixels was also applied to isolate the raised points, only. The distribution of the detected pixels shows a concentration near the support towers. The lack of coherent points in the central part of the suspended deck may be induced by large non-linear displacements possibly induced by vibration effects and wind. The estimated height of the selected pixels, represented by red diamonds, is provided in the plots depicted in Figure 5. The horizontal axis represents the longitudinal coordinate of the yellow rectangle, i.e., the length from its origin, conventionally set at its eastern end and approximately corresponding to the initial part of the bridge, as also pointed out by the graduated scale in Figure 4. The straight distribution of pixels at almost constant height on the bridge deck is easily recognizable; moreover, the triangular shape, particularly evident in the descending dataset results, also allows recognizing the first tower from East. Remarkably, the very high-resolution capability of the CSK sensors allows monitoring even pixels located along the tie rods cables. In the same figure, the black diamonds represent the coherent pixels that do not exceed the height threshold which likely belong to ground structures: the rectangular shape determined by the pillars of the second tower is recognizable at about 250-300 m from the origin of the line. By comparing the results of the first and second row, the different coverage induced by the sensor LOS visibility over the two opposite orbits is evident. On the contrary, no significant differences are met by the comparison of the coherent pixel distributions, for the same dataset, achieved by the two processing techniques. Finally, the plot of the estimated deformation mean velocity over the selected points, along the longitudinal coordinate of The estimated height of the selected pixels, represented by red diamonds, is provided in the plots depicted in Figure 5. The horizontal axis represents the longitudinal coordinate of the yellow rectangle, i.e., the length from its origin, conventionally set at its eastern end and approximately corresponding to the initial part of the bridge, as also pointed out by the graduated scale in Figure 4. The straight distribution of pixels at almost constant height on the bridge deck is easily recognizable; moreover, the triangular shape, particularly evident in the descending dataset results, also allows recognizing the first tower from East. Remarkably, the very high-resolution capability of the CSK sensors allows monitoring even pixels located along the tie rods cables. In the same figure, the black diamonds represent the coherent pixels that do not exceed the height threshold which likely belong to ground structures: the rectangular shape determined by the pillars of the second tower is recognizable at about Remote Sens. 2020, 12, 4011 8 of 16 250-300 m from the origin of the line. By comparing the results of the first and second row, the different coverage induced by the sensor LOS visibility over the two opposite orbits is evident. On the contrary, no significant differences are met by the comparison of the coherent pixel distributions, for the same dataset, achieved by the two processing techniques. Finally, the plot of the estimated deformation mean velocity over the selected points, along the longitudinal coordinate of the yellow rectangle, is provided in Figure 6: it confirms that no significant deformation was detected along the whole northern roadway of the bridge deck. Results are very consistent among the different processing outputs and datasets. Note that the coherent pixel detected in the results of the TomoSAR processing of the descending dataset, i.e., the plot in Figure 6d, at approximately 450 m from the line origin, which exhibits a linear deformation of more than 1.5 cm/year, is very likely related to a false alarm which, however low was the desired probability of false alarm, passed the GLRT detection stage. It is worth remembering that, as already mentioned, no further post-processing outlier filtering is performed.
The point is surrounded by a large density of coherent pixels concentrated in a few meters which exhibit no significant deformation. It is very unlikely that, given this spatial distribution of close PSs, a ground pixel could exhibit such a significantly different deformation behavior from its neighboring PS. Moreover, this anomalous behavior is detected neither in the SBAS results on the same dataset, nor in both SBAS and TomoSAR results related to the ascending dataset.
Remote Sens. 2020, 12, x FOR PEER REVIEW 8 of 17 along the whole northern roadway of the bridge deck. Results are very consistent among the different processing outputs and datasets. Note that the coherent pixel detected in the results of the TomoSAR processing of the descending dataset, i.e., the plot in Figure 6d, at approximately 450 m from the line origin, which exhibits a linear deformation of more than 1.5 cm/year, is very likely related to a false alarm which, however low was the desired probability of false alarm, passed the GLRT detection stage. It is worth remembering that, as already mentioned, no further post-processing outlier filtering is performed.
(a) (b) (c) (d) Figure 5. Plots of the estimated height of the coherent pixels located within the yellow rectangle in Figure 4 for (a) SBAS and (b) TomoSAR results achieved for the ascending dataset, and (c) SBAS and (d) TomoSAR results relevant to the descending dataset. Red and black diamonds correspond to measurement points located on the roadway, pillars and cables and at the ground level, respectively. Horizontal axis corresponds to the longitudinal distance from the origin of the yellow rectangle located at its southern/eastern end, as reported by the white graduated scale in Figure 4.
The point is surrounded by a large density of coherent pixels concentrated in a few meters which exhibit no significant deformation. It is very unlikely that, given this spatial distribution of close PSs, a ground pixel could exhibit such a significantly different deformation behavior from its neighboring PS. Moreover, this anomalous behavior is detected neither in the SBAS results on the same dataset, nor in both SBAS and TomoSAR results related to the ascending dataset.  Figure 4 for (a) SBAS and (b) TomoSAR results achieved for the ascending dataset, and (c) SBAS and (d) TomoSAR results relevant to the descending dataset. Red and black diamonds correspond to measurement points located on the roadway, pillars and cables and at the ground level, respectively. Horizontal axis corresponds to the longitudinal distance from the origin of the yellow rectangle located at its southern/eastern end, as reported by the white graduated scale in Figure 4. located at its southern/eastern end, as reported by the white graduated scale in Figure 4. The point is surrounded by a large density of coherent pixels concentrated in a few meters which exhibit no significant deformation. It is very unlikely that, given this spatial distribution of close PSs, a ground pixel could exhibit such a significantly different deformation behavior from its neighboring PS. Moreover, this anomalous behavior is detected neither in the SBAS results on the same dataset, nor in both SBAS and TomoSAR results related to the ascending dataset. A close-up analysis of the collapsed pillar is now in order: Figure 7 shows a zoom in the area of interest of the collapsed pillar of the measurement points of Figure 4, i.e., located on the northern roadway of the bridge for (a) SBAS and (b) TomoSAR processing techniques over the ascending dataset, (c) SBAS and (d) TomoSAR processing ones over the descending dataset. For these selected points, the time series of the estimated deformation are shown in Figure 8. Apart from a typical seasonal trend limited to a 3 cm range, all measurement points do not undergo any significant deformation. Again, a very high correlation of estimated time series is found. In Figure 8d the single anomalous deformation trend is related to the false alarm point whose anomaly was previously discussed. The time series associated with other neighboring points do not show any significant behavior: this further confirms the hypothesis of the false alarm detection.  A close-up analysis of the collapsed pillar is now in order: Figure 7 shows a zoom in the area of interest of the collapsed pillar of the measurement points of Figure 4, i.e., located on the northern roadway of the bridge for (a) SBAS and (b) TomoSAR processing techniques over the ascending dataset, (c) SBAS and (d) TomoSAR processing ones over the descending dataset. For these selected points, the time series of the estimated deformation are shown in Figure 8. Apart from a typical seasonal trend limited to a 3 cm range, all measurement points do not undergo any significant deformation. Again, a very high correlation of estimated time series is found. In Figure 8d the single anomalous deformation trend is related to the false alarm point whose anomaly was previously discussed. The time series associated with other neighboring points do not show any significant behavior: this further confirms the hypothesis of the false alarm detection.
Remote Sens. 2020, 12, x FOR PEER REVIEW 9 of 17 (c) (d) Figure 6. Plots of the estimated deformation mean velocity of the coherent pixels located at the northern roadway level within the yellow rectangle in Figure 4 for the (a) SBAS and (b) TomoSAR processing results achieved for the ascending dataset, and (c) SBAS and (d) TomoSAR results relevant to the descending dataset. Horizontal axis corresponds to the longitudinal distance from the origin of the yellow rectangle located at its southern/eastern end, as reported by the white graduated scale in Figure 4.
A close-up analysis of the collapsed pillar is now in order: Figure 7 shows a zoom in the area of interest of the collapsed pillar of the measurement points of Figure 4, i.e., located on the northern roadway of the bridge for (a) SBAS and (b) TomoSAR processing techniques over the ascending dataset, (c) SBAS and (d) TomoSAR processing ones over the descending dataset. For these selected points, the time series of the estimated deformation are shown in Figure 8. Apart from a typical seasonal trend limited to a 3 cm range, all measurement points do not undergo any significant deformation. Again, a very high correlation of estimated time series is found. In Figure 8d the single anomalous deformation trend is related to the false alarm point whose anomaly was previously discussed. The time series associated with other neighboring points do not show any significant behavior: this further confirms the hypothesis of the false alarm detection.  A similar analysis was performed on the southern roadway of the bridge deck. Similar to the northern roadway case, the coherent pixels located at the bridge deck on the southern roadway were isolated. The results are provided in Figure 9, in which the analysis was limited to the points located within the red rectangle and characterized by an estimated height greater than 40 m above the sea level. The plots of the estimated height for the selected pixels, represented as red diamond, and for the ground ones (black diamonds) are reported in Figure 10. Again, the very high CSK resolution allows easily recognizing the first tower and, particularly on the descending dataset, the stay from the top of the tower to the bridge deck, as well as pillars of second and third towers. These plots also highlight the significant different coverage between the ascending and descending datasets on the southern flank of the bridge: because of the significant shadow effect on the descending dataset, the density is drastically reduced on the bridge deck whereas even almost no coherent pixel is detected on the ground segment underling the southern roadway of the bridge. As for the northern roadway, the plots of the estimated deformation mean velocity within the red rectangle, provided in Figure 11, do not show any significant deformation of interest neither on the pillars, nor at the deck level and specifically in correspondence with the connections of the stay-cables. It is worth noting that, despite the choice of the reference pixels far away from the area of interest, the coverage over the bridge is fairly dense.
The stable behavior is also confirmed by the close-up analysis in Figure 12, where the pixels located in correspondence to the tower of interest were isolated. For each of these points, the estimated time series depicted in Figure 13 confirm the prevailing seasonal behavior.
In order to further provide information about a possible precursor, our displacement study also includes the analysis of the detected PS located at the ground level in correspondence with the collapsed tower. These results, provided in Figures S1 and S2 in the Supplementary Materials, further confirm that no significant displacements were detected.
As a final point, it is once again worth remarking that the results achieved by exploiting the two independent SBAS and TomoSAR approaches are in very good agreement, regardless of the reference pixels selection, thus confirming the robustness of the applied MT-InSAR processing approaches. A similar analysis was performed on the southern roadway of the bridge deck. Similar to the northern roadway case, the coherent pixels located at the bridge deck on the southern roadway were isolated. The results are provided in Figure 9, in which the analysis was limited to the points located within the red rectangle and characterized by an estimated height greater than 40 m above the sea level. The plots of the estimated height for the selected pixels, represented as red diamond, and for the ground ones (black diamonds) are reported in Figure 10. Again, the very high CSK resolution allows easily recognizing the first tower and, particularly on the descending dataset, the stay from the top of the tower to the bridge deck, as well as pillars of second and third towers. These plots also highlight the significant different coverage between the ascending and descending datasets on the southern flank of the bridge: because of the significant shadow effect on the descending dataset, the density is drastically reduced on the bridge deck whereas even almost no coherent pixel is detected on the ground segment underling the southern roadway of the bridge. As for the northern roadway, the plots of the estimated deformation mean velocity within the red rectangle, provided in Figure 11, do not show any significant deformation of interest neither on the pillars, nor at the deck level and specifically in correspondence with the connections of the stay-cables. It is worth noting that, despite the choice of the reference pixels far away from the area of interest, the coverage over the bridge is fairly dense. (c) (d) Figure 9. Highlight on the measurement points located within the red frame, corresponding to the southern roadway for (a) SBAS and (b) TomoSAR processing results related to the ascending dataset, (c) SBAS and (d) TomoSAR results relevant to the descending dataset. Colormap is set according to the estimated velocity with the convention that negative values correspond to departure from the sensor along the LOS. The white graduated axis represents the longitudinal coordinate of the red rectangle whose origin is set at the southern/eastern end of the bridge.
(a) (b) (c) (d) Figure 10. Plots of the estimated height of the coherent pixels located within the red rectangle in Figure 9 for the (a) SBAS and (b) TomoSAR processing results related to the ascending dataset, and (c) SBAS and (d) TomoSAR results relevant to the descending dataset. Red and black squares correspond to measurement points located at the roadway, pillars and cables and at the ground level, respectively. Horizontal axis corresponds to the longitudinal distance from the origin of the red rectangle located at its southern/eastern end, as reported by the white graduated scale in Figure 9. Figure 10. Plots of the estimated height of the coherent pixels located within the red rectangle in Figure 9 for the (a) SBAS and (b) TomoSAR processing results related to the ascending dataset, and (c) SBAS and (d) TomoSAR results relevant to the descending dataset. Red and black squares correspond to measurement points located at the roadway, pillars and cables and at the ground level, respectively. Horizontal axis corresponds to the longitudinal distance from the origin of the red rectangle located at its southern/eastern end, as reported by the white graduated scale in Figure 9. (c) (d) Figure 11. Plots of the estimated mean deformation velocity of the coherent pixels located at the southern roadway level within the red rectangle in Figure 9 for (a) SBAS and (b) TomoSAR processing results of the ascending dataset, and (c) SBAS and (d) TomoSAR results relevant to the descending dataset. The horizontal axis corresponds to the longitudinal distance from the origin of the red rectangle located at its southern/eastern end, as reported by the white graduated scale in Figure 9.
(a) (b) Figure 11. Plots of the estimated mean deformation velocity of the coherent pixels located at the southern roadway level within the red rectangle in Figure 9 for (a) SBAS and (b) TomoSAR processing results of the ascending dataset, and (c) SBAS and (d) TomoSAR results relevant to the descending dataset. The horizontal axis corresponds to the longitudinal distance from the origin of the red rectangle located at its southern/eastern end, as reported by the white graduated scale in Figure 9.
The stable behavior is also confirmed by the close-up analysis in Figure 12, where the pixels located in correspondence to the tower of interest were isolated. For each of these points, the estimated time series depicted in Figure 13 confirm the prevailing seasonal behavior.
(c) (d) Figure 11. Plots of the estimated mean deformation velocity of the coherent pixels located at the southern roadway level within the red rectangle in Figure 9 for (a) SBAS and (b) TomoSAR processing results of the ascending dataset, and (c) SBAS and (d) TomoSAR results relevant to the descending dataset. The horizontal axis corresponds to the longitudinal distance from the origin of the red rectangle located at its southern/eastern end, as reported by the white graduated scale in Figure 9.  In summary, our analysis of the PS detected on the bridge deck shows a great similarity of the results in terms of coverage and measurements and specifically, all the results agree on the fact that, apart from a seasonal trend, no significant deformation trend was observed. This is quite evident by the analysis of the deformation time series and related mean velocities in correspondence to the Morandi bridge obtained by exploiting both ascending and descending data and both the SBAS and the TomoSAR MT-InSAR approaches.

Discussion and Conclusions
The monitoring of civil infrastructures as road and railroad bridges is envisaged in order to follow their health conditions and to sort out possible actions for their safety and maintenance over time, thus reducing material and economic losses and, most of all, preserving human lives. Recent In order to further provide information about a possible precursor, our displacement study also includes the analysis of the detected PS located at the ground level in correspondence with the collapsed tower. These results, provided in Figures S1 and S2 in the Supplementary Materials, further confirm that no significant displacements were detected.
As a final point, it is once again worth remarking that the results achieved by exploiting the two independent SBAS and TomoSAR approaches are in very good agreement, regardless of the reference pixels selection, thus confirming the robustness of the applied MT-InSAR processing approaches.
In summary, our analysis of the PS detected on the bridge deck shows a great similarity of the results in terms of coverage and measurements and specifically, all the results agree on the fact that, apart from a seasonal trend, no significant deformation trend was observed. This is quite evident by the analysis of the deformation time series and related mean velocities in correspondence to the Morandi bridge obtained by exploiting both ascending and descending data and both the SBAS and the TomoSAR MT-InSAR approaches.

Discussion and Conclusions
The monitoring of civil infrastructures as road and railroad bridges is envisaged in order to follow their health conditions and to sort out possible actions for their safety and maintenance over time, thus reducing material and economic losses and, most of all, preserving human lives. Recent estimates demonstrate that the number of bridges in Italy to be assessed after a 50-year service life runs around 50,000 [45]. Currently, the structural health monitoring is mainly based on a visual inspection and accurate in-situ and embedded measurement systems directly installed on the investigated structure, resulting in complex and expensive surveys, which inevitably limit the surveillance to very restricted areas or to single infrastructures. The widespread development of Earth Observation (EO) technologies over the last decades has allowed the detection and monitoring from space of surface displacements related to very large areas and with high accuracy. In this framework, the innovation brought by the MT-InSAR techniques represents very valuable solutions to provide systematic, large scale displacement measurements related to infrastructures, able to simultaneously detect and analyze the deformation of hundreds of road and railroad bridges at relatively low costs. Such measurements, properly integrated with in-situ investigations and damage assessment models derived from structural engineering, may support actions for pre-emptive bridge rehabilitation and decision-making actors [11].
However, to properly move from a scientific context to a fully operative scenario, it is mandatory to deal with consolidated integrated procedures that are able to reduce the false alarm probability within the retrieved InSAR products, by minimizing the possible false positive and negative detections. Indeed, the false alarm issues strongly impact the suitability of InSAR techniques in operative contexts that can support the decision-makers. In particular, the false positives can make negligible the benefits of EO-based techniques in supporting well-focused in-situ surveys and, furthermore, can generate a strong impact on the population when not carefully verified information are disseminated, whereas the false negatives strongly jeopardize the monitoring capability of this technology by leading, in the worst cases, to the lack of identification of possible pre-event signals of future structural failures, with consequent human casualties.
These issues have to be addressed with a twofold approach, involving both the maturity of the exploited InSAR techniques and the related expertise of the end-users. The former concerns the development of accurate and highly verified InSAR techniques, as well as a deep scientific assessment of their maturity in a branched out number of use-cases to demonstrate their usefulness and effectiveness in real operative scenarios. The latter requires knowledge and consolidated experience of the users when dealing with InSAR techniques and products, at least in terms of technology limits, noise sources and accuracy analyses, thus leading to properly identity possible anomalies and correctly interpret the InSAR-derived results.
Unfortunately, the broad development of advanced InSAR techniques in the last decades has fostered their massive exploitation among a wide community of end-users coming from different scientific contexts, even not specifically experts in the InSAR field. Such a large variety of possible users dealing with InSAR deformation measurements associated with built-on structures has to be preliminarily qualified and, in some cases, guided by expert "eyes" throughout the interpretation of the InSAR-derived products. This is particularly demanded in the case of man-made structures and infrastructures interested in failures or collapses that provoked casualties with serious legal implications, as for the case of the Morandi bridge, where the InSAR analyses should be carefully presented only after a deep and accurate assessment of the results and their interpretation.
In this study, we exploited the well established and extensively tested SBAS and TomoSAR MT-InSAR techniques to reprocess the same ascending and descending CSK datasets, relevant to the Morandi bridge, analyzed by Milillo et al. [2]. Our results show that, although we achieve denser coherent pixel maps relevant to the Morandi bridge, there is no evidence at all, even not for the results relevant to one of the processed orbits dataset or to the exploited SBAS and TomoSAR processing techniques, of the pre-collapse displacements reported in [2] in pixels located near the strands of the deck next to the collapsed pier. Accordingly, in our opinion, the pre-event InSAR study relevant to the Morandi bridge carried out by Milillo et al. [2] is an example of an InSAR deformation analysis affected by a false alarm.
Supplementary Materials: The following are available online at http://www.mdpi.com/2072-4292/12/24/4011/s1, Figure S1. Deformation mean velocity for the measurement points located at the ground level in correspondence to the area of interest of the collapsed pillar, (a) SBAS and (b) TomoSAR processing over the ascending dataset, (c) SBAS and (d) TomoSAR processing over the descending dataset; Figure S2. Plots of the estimated deformation time series for the measurement points at the ground level in correspondence to the area of interest of the collapsed pillar provided in Figure S1 for (a) SBAS and (b) TomoSAR processing over the ascending dataset, (c) SBAS and (d) TomoSAR processing over the descending dataset.