Measuring Urban Subsidence in the Rome Metropolitan Area (Italy) with Sentinel-1 SNAP-StaMPS Persistent Scatterer Interferometry

Abstract

Land subsidence in urban environments is an increasingly prominent aspect in the monitoring and maintenance of urban infrastructures. In this study we update the subsidence information over Rome and its surroundings (already the subject of past research with other sensors) for the first time using Copernicus Sentinel-1 data and open source tools. With this aim, we have developed a fully automatic processing chain for land deformation monitoring using the European Space Agency (ESA) SentiNel Application Platform (SNAP) and Stanford Method for Persistent Scatterers (StaMPS). We have applied this automatic processing chain to more than 160 Sentinel-1A images over ascending and descending orbits to depict primarily the Line-Of-Sight ground deformation rates. Results of both geometries were then combined to compute the actual vertical motion component, which resulted in more than 2 million point targets, over their common area. Deformation measurements are in agreement with past studies over the city of Rome, identifying main subsidence areas in: (i) Fiumicino; (ii) along the Tiber River; (iii) Ostia and coastal area; (iv) Ostiense quarter; and (v) Tivoli area. Finally, post-processing of Persistent Scatterer Inteferometry (PSI) results, in a Geographical Information System (GIS) environment, for the extraction of ground displacements on urban infrastructures (including road networks, buildings and bridges) is considered.

1. Introduction

Since the launch of Copernicus Sentinel-1A on 8 April 2014, a new era of continuous monitoring using spaceborne Synthetic Aperture Radar (SAR) sensors has started. Sentinel-1 constitutes a significant improvement from previous European C-band SAR missions, European Remote Sensing (ERS) satellites and Environmental Satellite (ENVISAT), since it reduced the temporal revisit time from 35 to six days, at best using the two satellite segments A and B, with a large swath coverage of 250 km. The scientific communities as well as Earth Observation (EO) practitioners were thus given the means to extend the use of spaceborne SAR data to land applications.

In support to the EO community, the European Space Agency (ESA) continued developing appropriate tools for the utilization of the Copernicus Sentinel data. By evolving existing tools, such as the Next ESA SAR Toolbox (NEST) as well as integrating others, the SeNtinel Application Platform (SNAP) [1] becomes a multi-mission toolbox supporting both SAR and optical data processing. Newly implemented on Sentinel-1, the Terrain Observation by Progressive Scans (TOPS) acquisition mode [2] required further development in terms of interferometric handling to ensure robust results. The SNAP TOPSAR) capabilities were made available to users at an early stage, just before the start of Sentinel-1 data dissemination, while SNAP TOPS Interferometric SAR (InSAR) development were first communicated at the ESA Fringe 2015 consultation meeting. Currently, TOPS InSAR processing is sufficiently documented [3,4,5,6,7] and SNAP remains a widely used end-to-end open source tool for processing of Sentinel-1 data.

Further development of SNAP was carried out to include exports to software packages supporting more advanced interferometric analysis, such as Stanford Method for Persistent Scatterers (StaMPS) [8]. StaMPS is a freely distributed package for research purposes with a large user community that incorporates Persistent Scatterer Interferometry (PSI) and Small Baseline methods to measure ground displacements from time series of SAR acquisitions. The idea was for the open source InSAR processor to be used together with StaMPS PSI, boosting the utilization of the Copernicus Sentinel-1 data for geohazard-related applications. The potential of SNAP-StaMPS integration has been already demonstrated at [9], for which the authors had also published a set of scripts to support utilization by the scientific community [10].

In this study we employ such open tools to analyze the urban deformation of the Rome metropolitan area for the first time using Sentinel-1 data, and combined ascending and descending orbits to depict the vertical urban deformation with special attention to different areas. Ground deformation analyses of Rome have already been undertaken in the past [11,12], but these have not been updated recently, and have not included Sentinel-1 SAR observations. Here we intend to update existing knowledge with contemporary information regarding ground deformation in the Rome metropolitan area using open data and tools. Previous relevant studies are indirectly used to verify our findings and to allow us to understand which subsidence patterns correspond to already identified phenomena and which new sources of deformation that would require further attention. Finally, a dedicated analysis was performed highlighting vertical displacements along urban infrastructures, including road networks, buildings and bridges.

1.1. Study Area

The study area includes the city of Rome and its surroundings, in the Lazio region of central Italy (Figure 1). The geology of the region is characterized by volcanic deposits (mainly pyroclastic tuff) from the Albano volcano district to the southeast and the Sabatino volcano district to the northwest, with alluvial sediments along the Tiber valley in between the two [13]. The topography gradually decreases from these two volcanic districts towards the Tiber, with valleys carved by fluvial erosion. The variability of heights in the Rome metropolitan plain does not exceed 100 m.

The southwest of the study area is dominated by the Tiber River delta and coastal plain. Both are comprised of alluvial sediments with a flat morphology [12]. The northeast of the study area includes the beginning of the Apennines mountain chain, comprising mainly sedimentary limestone and dolomite rocks [13].

Many cases of land subsidence have already been identified over the study area, and quantified through various InSAR techniques, a detailed review of which is presented in [14]. Subsidence of buildings on the alluvial sediments along the Tiber River in Rome has been measured using PSI, Small Baseline Subsets (SBAS), Interferometric Point Target Analysis (IPTA) and 4D SAR imaging techniques in many studies, e.g., by [11,15,16,17] and others, using ERS-1, 2 and ENVISAT ASAR data. The main cause of subsidence in this case is the weight of relatively recent construction on the unconsolidated alluvial material, especially in areas such as Grotta Perfetta, in the southwestern outskirts of the city [11,12,13,14,15,16].

Other studies, such as [14,18,19], focused on quantifying displacement which may affect the structural integrity of archaeological monuments in the historical center of the city, using SBAS, PSI and SqueeSAR with ERS-1 and -2, ENVISAT ASAR [18], Radarsat-1, 2 [19] and COnstellation of small Satellites for the Mediterranean basin Observation (COSMO-SkyMed) data [14].

Another type of ground displacement in the region of interest, which has been measured using InSAR techniques, has been identified in the Acque Albule Plain [20], in the northeastern part of the study area. Here a combination of groundwater extraction for mining and the presence of compressible soils has led to ground subsidence in the area. This has been quantified and studied with ERS and ENVISAT data using PSI and Quasi-PS InSAR (QPS) techniques by [21]. QPS is based on a different set of filtered interferograms (multi-master configuration) and is weighted by interferometric coherence.

More recently, subsidence affecting the area surrounding the Rome Fiumicino (FCO) airport, in particular over the third runway, has been studied with the PSI technique applied to ERS, ENVISAT and COSMO-SkyMed data by [12]. The authors showed how the varying rate of subsidence in the area correlates with the age of overlying man-made constructions and the nature of the underlying geology.

2. Materials and Methods

2.1. Open Source Toolboxes

This work has been carried out using the open source ESA SNAP and StaMPS software packages. The SNAP Graph Builder operator can be used to create processing chains which can be called using the batch mode Graph Processing Tool (GPT). We have exploited this utility to create the several templates necessary for creating single master TOPSAR coregistration and interferogram generation.

Finally, in order to fully automate the single master interferogram generation, we have developed and made available, based on well-designed SNAP graphs, a set of scripts called “snap2stamps”. These scripts enable automatic processing after setting some parameters in a configuration file. In fact, they are python wrappers which use the aforementioned templates based mainly on SNAP TOPSAR interferometric operators and whose outputs are compatible with StaMPS PSI chain. The snap2stamps scripts are available via the Zenodo repository [10]. Latest versions of the scripts (not verified by the developing team) can also be found on the GitHub repository (https://github.com/mdelgadoblasco/snap2stamps). The authors had released a first version of the snap2stamps package in July 2018, which automates the TOPSAR single master Differential InSAR (DInSAR) processing, fully compatible with StaMPS PSI [9], allowing the creation of stacks of single master interferograms in batch mode, just by defining some simple settings such as project folder, subswath to process and defining the bounding box coordinates of the area of interest.

Additionally, for the removal of the Atmospheric Phase Screen (APS), we have employed the Toolbox for Reducing Atmospheric InSAR Noise (TRAIN) [22] and applied the linear approach (topography versus phase) integrated in the aforementioned open source package.

2.2. Data and Processing

For the data processing we have employed an ESA RSS CloudToolbox which is a Virtual Machine provided by the ESA Research and Service Support [23] with access to collocated Sentinel-1 data via Copernicus Data and Information Access Services (DIAS). This has the advantage of eliminating the data download time, as the data is locally accessible and ready to use. The resources employed were 8 vCPUs, 32 GB RAM and 1TB disk space, resulting in a total processing time of approximately 15 days, including the post-analysis of PSI results.

For the interferometric processing, the Advanced Land Observation Satellite (ALOS) World 3D (AW3D30) Digital Surface Model (DSM) [24], of 30 m spatial resolution, was utilized, while for examining the geolocation accuracy as well as the interpretation of PSI results we employed a very high resolution DSM (5 m/pixel), as extracted from CartoSat-1 satellite data [25].

2.2.1. Copernicus Sentinel-1 Data

We limited our analysis to Sentinel-1A data only (12-days repeat cycle) (

Table 1. Sentinel-1 data employed for processing, with first and last image of each dataset, orbit pass, track and number of acquisitions.

Satellite	First Image	Last Image	Orbit Pass	Track	N Acquisitions
S1A	2015/03/24	2018/04/13	Descending	22	82
S1A	2015/03/30	2018/04/19	Ascending	117	87

), which is sufficient given the expected magnitude of ground displacements and the availability of a large number of acquisitions over the area of interest. Some details on the Sentinel-1 data employed for the processing are shown in Table 1. It should be noted that since Sentinel-1 products are not spatially synchronized, meaning that their starting and ending times may vary within each orbit, often more than one scene is required to fully cover our area of interest. This introduces additional storage and computational requirements, as consecutive scenes, for the same acquisition date, need to be downloaded and assembled into single products before proceeding with the interferometric processing. Our area of interest (AOI) and the extent of Sentinel-1 ascending (A117) and descending (D022) orbits is illustrated in Figure 1.

Among the different options, we have selected ascending and descending tracks, 117 and 022, respectively, for which the area of interest is mapped with comparable incidence angles. By ensuring combination of similar viewing geometry, i.e., sensitivity to vertical motion, we facilitate a more robust extraction of the vertical motion component, of interest for our investigation.

2.2.2. SNAP-StaMPS PSI Processing

The PSI processing is split into two independent workflows: (i) single master DInSAR processing using ESA SNAP; and (ii) the PSI processing using StaMPS.

Firstly, the master scene is selected from the beginning of the data series, as Sentinel-1 has orbit control which guarantees any interferometric combination among the data. Additionally, as we want to obtain PS points over urban infrastructure, we expect that temporal baseline will not greatly affect the number of PS obtained. Master image splitting and update of orbit state vectors follow, also using the SNAP Graphical User Interface (GUI), to ensure proper selection of bursts covering our AOI. These steps are critical since they optimize time and resources for the rest of the processing.

Table 2. Main characteristics of the selected Sentinel-1A master scene.

Track	Acquisition Date	Mean Inc. Angle (rad/degrees)	Sub-Swath	Polarization	Initial Burst	Last Burst
D022	2015/05/23	0.75/42.97	IW3	VV	5	8
A117	2015/08/09	0.67/38.39	IW2	VV	5	7

details the parameters involved in master image splitting for burst selection over the AOI.

The next step involves generating all single master interferograms using the snap2stamps scripts, by following an automatic processing scheme implemented in four steps:

• Slave preparation. In this step, the Sentinel-1 Single Look Complex (SLC) data are sorted by acquisition date while checking if SLC assembly (concatenation procedure) is necessary, depending on whether the defined AOI is covered by more than one scene per acquisition date.
• Slave splitting. To enable processing in batch mode, the SNAP Graph Processing Tool (GPT) is used, which runs already-defined processing chains (graphs in xml format). For this step, the TOPSAR-Splitting and Apply Orbit operators are called, to update the annotated orbit information with more precise ones according to their availability (restituted or precise). These orbits are automatically downloaded by SNAP. The corresponding graph is illustrated in Figure 2, part A.
• Coregistration and interferogram computation. This is the most computationally demanding step, as it performs the coregistration of the TOPSAR data (Back-geocoding with Enhanced Spectral Diversity [26] refinement) and produces the interferograms with the Flat-Earth and topographic phase contributions removed. Optionally, a finer subset can be applied over an AOI, as defined in the project configuration file. If no information is provided by the user, the full burst interferograms are generated. The outputs of this step are two debursted stacks of master-slave Single Look Complex (SLC) files and the master-slave interferogram. Supplementary data files required by StaMPS are also generated, including elevation band and orthorectified latitude and longitude coordinates as independent products. The graph employed for this step is shown in Figure 2, part B.
• Stamps export. This is the final step of the single master DInSAR processing, which converts previous processing results into binary raster files compatible with StaMPS readers. Graph shown in Figure 2, part C.

The following step involves the ingestion of SNAP exports into StaMPS using a specific script, called mt_prep_snap, available in the distribution. Subsequently, the StaMPS PSI processing chain is run from step 1 to 7 as described in the StaMPS User Manual [27].

In this case, we additionally applied the integrated TRAIN, using the linear tropospheric correction approach, to mitigate the topography-correlated atmospheric phase.

In order to properly merge the results from both ascending and descending tracks in subsequent post-processing steps, we selected the same reference point in both cases, corresponding to a permanent European Reference Frame (EUREF) Global Navigation Satellite System (GNSS) station (M0SE00ITA), located at the Aerospace Engineering Faculty of the University of Rome “La Sapienza”. Based on the EUREF solution [28], the station seems stable, with no evident vertical motion (Vx = −0.7 ± 0.1 mm/yr, Vy = −1.3 ± 0.1 mm/yr and Vz =0.5 ± 0.1 mm/yr in ETRF2014) during the entire observation period.

We ran StaMPS PSI three times from the merging of the different patches (step 5) onwards, each time with different grid options [27]: (i) no merging; (ii) merging by 20 m grid; and (iii) merging by 40 m grid. For each run, the merging of PS candidates was performed with the same threshold selected for the phase noise filtering in StaMPS step 4.

2.2.3. Post-Processing

Having both ascending and descending PSI measurements, we combined them to calculate the vertical component for each individual Persistent Scatter (PS) point (see Figure 3) using Equation (1) and (2), as described in [29]:

$$\left[\begin{array}{c}{d}_{LOS}{}^{asc}\\ d_{LOS}{}^{desc}\end{array}\right]=A \left[\begin{array}{c}{d}_{up}\\ d_{h_{ald}}\end{array}\right]$$

(1)

with

$$A=\left[\begin{array}{cc}\cos {\theta }^{asc}& \frac{\sin {\theta }^{asc}}{\cos \Delta \alpha }\\ \cos {\theta }^{desc}& \sin {\theta }^{desc}\end{array}\right]$$

(2)

where d_LOS is displacement along Line-Of-Sight (LOS). d_up is the vertical displacement. d_hald is the projection of horizontal displacement in descending azimuth look direction (ALD). θ is the incident angle. Δα is the satellite heading, difference between ascending and descending orbit.

The combination of ascending and descending PS points was performed in the vector domain avoiding any rasterization option, as described in [30]. The method combines PS targets based on their geographic proximity, while attributes transfer and decomposition of motion is done within the features geodatabases. This leads to a higher number of final PS points and reduction of error budget introduced by spatial interpolation and rasterization procedures. PS points of one geometry having no neighbouring targets by the opposite geometry, within a defined search radius, are being excluded. The selection of the maximum search radius for the combination, in our case 40 m, was based on the statistical analysis of the distances between PS targets from the independent LOS solutions.

After obtaining the vertical component of the deformation rates, by using GIS capabilities, we overlaid the deformation information over buildings, roads, highways and railways vector layers to calculate the maximum observed deformation for each of these elements. In such a way, we were able to provide subsidence information over critical infrastructure (roads, bridges etc.) as well as on individual building blocks. For the special case of the road networks, and in order to depict the spatial variability of motion along them, a segmentation procedure was applied considering a distance of 20 m for each segment. The displacement value is then calculated based on PS points located at a specified distance across each road segment. To reduce overlaps between successive segments, we considered a square buffering option, i.e., buffer zones do not exceed the segments’ start and end points.

Since StaMPS does not update PS heights, inaccuracies in the geolocation of PS targets over urban environments, especially for high buildings, might occur. To compensate for the above mentioned issue, during the calculation of the deformation statistics for each vector element (road, buildings etc.) a 20 m buffer was considered. The choice of the buffer distance was based on the visual interpretation of the results.

3. Results

We have obtained the average PSI LOS deformation rates for both ascending and descending tracks (Figure 4). An indicator of the compatibility of solutions between acquisition geometries is the standard deviation of the mean LOS deformation velocities, which correspond to 1.04 mm/yr and 1.17 mm/yr for A117 and D022 tracks, respectively.

Given the difference in area covered by each track, the relatively larger number of PS points in the descending solution could be explained (

Table 3. Total number of PS points obtained by StaMPS processing (LOS) and after the decomposition to actual motion component (Vertical).

Orbit	No Merging LOS/Vertical	20 m LOS/Vertical	40 m LOS/Vertical
Ascending (A117)	1065328/947386	486188/418481	264024/211999
Descending (D022)	1342924/1061976	580578/439738	311615/217237

). Apart from the effect of area coverage, it seems that the overall numbers of PS points is comparable. We attribute this to the common observation period and incidence angles considered for both ascending and descending datasets.

After decomposing the ascending and descending LOS measurements, we obtained the vertical deformation rate map presented in Figure 5. It can be easily seen that various deformation patterns exist, attributed mainly to the different subsidence mechanisms acting in the metropolitan area of Rome. There are several areas undergoing significant subsidence, such as (i) FCO airport; (ii) along the Tiber River; (iii) the coastal zone of Ostia; (iv) Ostiense quarter within Rome and; (v) Tivoli area, while the rest of the region exhibits relatively low ground deformation rates. In the following sections we provide a more detailed analysis of the PSI results.

As we ran StaMPS using different merging options, we obtained different numbers of PS points for each solution. In Table 3, we summarize the total points obtained for each merging configuration and those remaining after the vertical decomposition. It should be noted that, for the decomposition, only the overlapping area between ascending and descending results is exploited.

For the full resolution datasets, we obtained over 1 million point targets for both orbits (Table 3), a fact which poses some difficulties in handling the dataset for post analysis purposes. The decision to reduce the initial number of PS points or not depends actually on the application at hand. For example, while working on infrastructure monitoring, it may be relevant to maintain full resolution results, as the number of points decreases rapidly after merging. For the needs of our work, the 20 m merged solutions offered a reasonable trade-off between density of PS targets and computational requirements for post-processing.

The obtained PS density after the decomposition, calculated at 200 m grid, is shown in Figure 5. The increase of density over the built-up area of Rome, reaching 1700 points/km² in the center of the city, is evident, while several rural urban centers also exhibit relatively high PS densities varying from 300 to 600 points/km². The locations of ascending and descending PS targets over an urban fabric of moderate density are presented (Figure 6), showing the advantages of the applied decomposition approach in retaining larger numbers of PS targets that allows proper characterization of on-going deformation phenomena. The final average velocity map of vertical deformation over the study area is shown in Figure 5, where the different areas with higher subsidence can be clearly identified.

3.1. Critical Urban Infrastructures: Global Road Network

As mentioned in Section 2, we separated the vertical deformation of the different elements such as roads, highways, railways and buildings using the extracted OpenStreetMap shapefile layers available in [31], and here in Figure 7 we show the deformation for the different roads and highways over our AOI. Significant deformation is revealed near the Fiumicino area, along the Tiber River and in the eastern part outside the Grande Raccordo Anulare (GRA). However, other roads show stable behavior or irrelevant deformation.

Systematic monitoring of human infrastructure, particularly critical for communication and transportation (highways, railways, roads, bridges, viaducts) or providing resources (electricity plants, dikes, dams) can be used for maintenance planning activities as well as for infrastructure risk assessment.

3.2. Subsidence along the Tiber River

This type of subsidence phenomena is quite common on areas constructed over alluvial deposits along river floodplains.

As mentioned in Section 1.1, the subsidence of buildings on the alluvial sediments along the Tiber River in Rome has been studied using similar techniques. Some notable studies include [11,12,13,14,15,16,17]. These all used ERS-1, 2 and ENVISAT ASAR data. The main cause of subsidence in this case is the weight of relatively recent construction on the unconsolidated alluvial material [11,12,13,14,15,16]. The older constructions in the city center of Rome on the other hand display less movement, as their position has consolidated over time. Findings from the PanGeo project [17], and the studies mentioned above, reported subsidence along the Tiber River and its tributaries determined using PSI and other techniques applied to ERS SAR and ENVISAT ASAR data acquired from 2002 to 2005.

Similar patterns over these alluvial deposits are present in our results (Figure 8). It is difficult to compare the deformation velocities between the studies, due to the different datasets employed and periods analyzed. We measure strong subsidence in several areas such as Ostiense and Santa Victoria quarter, with maximum vertical deformation rate of −7.2 mm/yr. Also remarkable is the subsidence along the Tiber River and its tributaries (with more than 52k PS points), with a maximum deformation rate of −8.7 mm/yr and an average of −1.4 mm/yr.

3.3. Fiumicino Airport and Ostia Coastal Region

The deformation over the Fiumicino area is also known, and consistent with previous studies (e.g., [12]), in which similar subsidence patterns are found. There are some points to highlight on the deformation over this area: (i) the airport runway oriented north-south; (ii) the highway; and (iii) the harbor area.

Regarding the runway, part of the track is located over an ancient lake. This has a different degree of consolidation compared to the other part, located over alluvial deposits. A very different behavior is thus shown between the two parts [12]. Also, the highway from Rome towards Fiumicino airport suffers from a high rate of vertical subsidence. This poses a higher risk than comprehensive subsidence over the entire area as large variations in deformation rate can lead to cracks in infrastructures.

Figure 9 illustrates the spatial distribution of the vertical deformation on the Fiumicino area, where the Rome-Fiumicino highway deformation is highlighted in (A), the full time series deformation of a smaller area for both ascending and descending orbits is shown in (B) and in (C) the total accumulated vertical deformation of the portion of highway inside the black ellipse found in (A). Figure 9C has been obtained by considering the linear velocity of the vertical motion for each PS point inside the dashed lines in Figure 9A from West to East. The horizontal axis is the longitude and vertical axis refers to the total accumulated vertical motion. There are visible transitions of blue to red and vice versa between adjacent points, where blue corresponds to no motion or the PS point in that position, as there is not always a PS point per each 20 m of highway.

Moving from West to East, the first area with strong vertical deformation is near the Mediterranean coast of Ostia-Fiumicino, where more than 20k PS are obtained, measuring a maximum deformation rate of −9 mm/yr. Further East, the nearest area with strong deformation is the Fiumicino airport, where only on the airport North–South oriented runway are located 270 PS points, with a maximum vertical deformation of −14.8 mm/yr, and an average deformation along the track of −6.6 mm/yr. Next to the Fiumicino airport, in the Ponte Galleria industrial area, with more than 3700 PS points, the maximum subsidence value is less than −19 mm/yr while the average vertical deformation is less than −5 mm/yr. Continuing towards Rome city, the highway Rome-Fiumicino also suffers strong vertical deformation, having a maximum deformation of −19 mm/yr and an average vertical deformation of almost −7 mm/yr.

3.4. Other Cases of Strong Displacement Patterns

Other important patterns are found over highways and roads, for which it is worth highlighting that the areas where strong transitions between subsidence rates occur are generally more dangerous as these are responsible for cracks in infrastructures. Hence, these areas with strong variations in subsidence behavior, and with high spatial frequency are the ones we would suggest to closely monitor, as well as the areas with significant subsidence, as continuous subsidence may also pose a risk.

In Figure 10, we show an example of the road subsiding while the bridges seem to remain more stable. Specifically, near Settebagni, on the “A1 Diramazione Roma Nord” there are more than 700 PS, measuring a maximum vertical deformation of −7.8 mm/yr with an average of around −4 mm/yr.

Finally, on the Eastern part of Rome (see Figure 6), we find Tivoli and Tivoli Terme, with more than 17k PS, measuring a maximum vertical deformation of −12 mm/yr, with an average of −3 mm/yr, similar to values obtained in [20].

4. Discussion

We present ground deformation in the Rome metropolitan area for the first time, using Copernicus Sentinel-1 mission data and open source toolboxes, paving the way for the broader utilization of the proposed chain by EO practitioners in geohazards applications.

Despite the availability of additional satellite data, limiting the analysis to Sentinel-1A acquisitions (repeat of 12 days) was considered sufficient, given the relatively low deformation rates in the area and the expected linear behavior of motion in time.

The area exhibits diverse deformation patterns, the spatial expression of which indirectly suggests, at least for some cases, the underlying deformation mechanism. The demonstrated quality of interferometric products obtained by Sentinel-1, both in terms of spatial density and uncertainties of displacement estimates is of key importance for the interpretation of motion and the phenomena involved. In our case, the density of the results reaches 1700 point/km² (~70 PS targets on a 200 m × 200 m grid) for dense urban centers, 300–600 point/km² for the suburban environments, while for the entire area of interest we calculated (including areas with no PS) on average 250 point/km². It is worth mentioning that we exploit only the co-pol. (VV) channel of Sentinel-1, yet by considering both polarizations (VV+VH) an increase in the number of PS targets is expected. As a result of the high PS density in built-up areas, and thus, availability of multiple PS targets within each building block, it is possible to differentiate between deformation of the buildings themselves from those related to soil foundations. Yet, given the uncertainties in the location of the PS points and the resolution of Sentinel-1, such separation of motion should be handled with care.

Among the most pronounced deformation patterns is the subsidence at Fiumicino airport reaching −20 mm/yr and along the Tiber River and its tributaries with motion in the order of −5mm/yr. Several other cases with a magnitude of motion worth mentioning, in the range of −12 to −3 mm/yr, are found along the coastal zone of Ostia as well as in the Tivoli region. The presence of relatively high deformation gradients for the above mentioned sites is well documented in several studies [11,12,13,14,15,16,17,18,20]. A more detailed analysis is provided in the PanGeo project report about the Geohazard description of Rome [17]. In [20], an analysis integrating geological and hydrogeological modelling provided insights on the relation between ground displacement, variations in the groundwater table and geotechnical properties of the subsoil for the specific case of Fiumicino area. For most of the cases, ground deformation can be attributed to the local geological conditions, such as the compaction of soft sediments, whereas loading by urban constructions should be one of the major reinforcing factors. However, further investigation is required to characterize the on-going subsidence induced phenomena and their temporal evolution compared to past ground displacement measurements.

Computed average vertical displacement rates for the various lithological types, as described in the 1/25,000 scale geological map of Lazio [24] are shown in (Figure 11). As expected, unconsolidated deposits (e.g., sands, clays and other alluvial material) show higher subsidence rates compared to basement formation such as marls, limestones and dolomites.

Deformation over urban infrastructures was also detected, with the case of the GRA, the ring highway surrounding the city of Rome, showing subsidence rates of up to −7.8 mm/yr. In fact, the proposed segmentation approach enabled the localization of the deformation along the entire road network of the Rome metropolitan area, a valuable option to support planners.

The above-mentioned findings were verified by inter-comparison with previous studies. Further validation activities were not considered necessary, since InSAR techniques have already undergone a long period of validation [32], confirming their capacity to measure surface motion. Actually, the majority of the observed subsidence has been already reported in the past [11,12,13,14,15,16,17,18]. However, the updated status as provided by Sentinel-1 is of significant importance, since the continuation of ground deformations without addressing any mitigation actions may lead to undesirable consequences in the future.

Finally, our ability to detect and measure ground movements with substantial accuracy within a short time should be underlined. Comparable studies in the past would require long observation periods to collect sufficient satellite data to obtain robust solutions. At the same time, our capacity to actually monitor phenomena has been increased by the systematic availability of the Copernicus Sentinel-1 data.

5. Conclusions

We have demonstrated the utilization of open and free data from the Copernicus Sentinel-1 mission using open source toolboxes for advanced interferometric processing. We provide details on a dedicated package for the automation of SNAP-StaMPS PSI processing, which we make available to EO practitioners in response to the growing need for EO-based solutions for monitoring geohazards. The integration of such a chain on a cloud processing environment will enable EO practitioners to respond to the ever-increasing volume of satellite data and high processing capacity requirements.

We verified the results by inter-comparison with previously published studies. To facilitate openness, we have made the PSI measurements over Rome available online, encouraging further analysis and interpretation as well as promoting collaboration between research communities.