The development of social, economic, cultural and also technological aspects over the last 80 years has changed the organization of the territory, making it more fragile. From the perspective of sustainable development, and to prevent possibly calamitous events from having a disastrous impact, it is necessary to spatialize the processes not only to obtain qualitative but also quantitative information concerning the shape of the landscape. The most modern combined aerial and terrestrial survey techniques, such as satellite remote sensing, UAV photogrammetry, and focused range-based scanning, allow us to interpret the territory in a complex way as they allow us to combine and cross-reference data from heterogeneous sources, also involving geoinformation data management.
2.1 Prevention and Preparedness
The first step in studying the landscape is not only to date the signs and traces, but to establish meaningful historical thresholds [57]. This delimitation is not arbitrary but derives from the research questions and themes. From the literature review, we can identify three significant historical phases for studying the relationship between settlements and natural hazards in Limone Valley. It is also noteworthy to highlight, in relation to par. 1.1, that the time frame identified for Limone Piemonte is valid for many other valleys in the Alpine arc and in the Italian Apennines. The first phase corresponds to the years up to the post-World War II period, and as a reference map we identified the digitized Reference Cartography produced by the Istituto Geografico Militare (Military Geographical Institute) in 1960 in raster format at a map scale of 1:25000 (Geoportale Nazionale). The contemporary period, from the 1960s to the present, was divided into two phases. For what concerns the period from 1960 to 1990 the Carta Tecnica Regionale Numerica (CTRN) updated to 1991 in vector format at a scale of 1:10000 (Geoportale Piemonte, 2020) has been used; while the Base Dati Territoriale di Riferimento degli Enti 2019 (BDTRE) was the topographical multiscale datasets archive used to study the development of the last 30 years [58]. Spatial analyses and related analytical comparisons of data from the three cartographic data sources made it possible to determine when the buildings present today were constructed. As an intermediate product, a multi-temporal map related to the evolution of the urban fabric was designed. The BDTRE was used as the reference data, to which a semantic attribute was added to indicate the year of the source and, thus, the presence of the buildings in the area on that date. Managing the reprojection of the reference systems used in the different epochs for the different cartographic products, it was possible to highlight, on the most recent dataset, buildings already present in 1960 and 1995 (fig. 4 a) [59].
A first analysis is qualitative and allows us to establish constants in settlement patterns through the description of urban morphologies. Prior to 1960, buildings are clustered in well-defined cores, and routes are halfway, away from the river. From Vernante to Limone Piemonte, the borgate, an Italian name for hamlets, are located at the intersection of this route and small streams: Tetti Salet, Tetti Mariné, Tetti Mezzavia. From Limone Piemonte, going up the valley to the last borgata, all the cores are located at the intersection of the halfway route and the ridge line, beginning with Limone Piemonte itself, Tetti Mecci, and Limonetto. All these cores are not only distant from the river, but also from the base of the small valleys that are loaded with snow in winter and with rain in spring and fall. As noted in some historical photos (fig. 4b, c [60]) incoherent types of deposit materials are accumulated in these valleys but were consolidated through planting and management of forests. After World War II, new settlements were built in the paleoalveum of the Vermegnana River (fig. 5), which now is partially lithified with control works, such as levees and barriers, when not silting of watercourses in tunnels.
This development also has changed the type of soil: no longer meadows but artificial impermeable materials.
In fig. 4a, the built-up areas have been divided according to the period of occurrence in the cartographic sources. Once georeferenced data are obtained, it is also possible to have quantitative information, which is necessary to identify any buildings possibly involved or vulnerable to damages because of the flood event (fig.6). As shown in the graphs of fig. 7, a significant increase in land area lasted from the 1960s to 2019 highlighting how the land area occupied by built-up areas more than tripled in this timeframe. This figure is quite incredible even considering the ground surface area alone, but if we also consider, and above all, that between the 80s and 90s the new buildings, including hotels and residences, settled on a multi-store building typology of up to 8 -10 floors above ground, the figure becomes quite impressive.
In order to map and identify buildings exposed to a possible flood event, some of the data from the Flood Risk Management Plan, established by the European Directive 2007/60/EC [61], was used. In particular, datasets of flood event hazard scenarios were used to accomplish the map shown in fig. 6, based on these data, and generated in a GIS environment using the slope raster dataset as a base map. These portions of land could be affected by flooding according to scenarios of high, medium or low probability of water disaster events. According to which built-up surfaces in Limone Piemonte insist on areas of known flood hazard, a series of graphs (fig. 7a – 7d) presents and enables reasoning about surfaces and percentages of buildings classified according to the construction period compared to the cross-reference probability of being exposed to flooding. (NULL probability grey, medium probability orange and high probability red).
The graph of fig. 7 explicates which percentages of built-up area on the ground exposed to flood hazard compared to the total. It shows how 2.5% of the built-up area is exposed to a medium flood probability while 10% is exposed to a high probability.
In the graph of fig. 7, the built-up areas have been divided according to the period of occurrence in the cartographic sources and presented their percentage of probability of being exposed to flooding (low, medium, high). it is evident how, since the 1960s, the city has developed in areas of high flood hazard. Note how the medium probability zones, representing 2.5 percent of the entire built ground area, were more than 58 percent built after '61 (49.5% + 8.4%). The built-up areas exposed to high probability turn out to be 10% of the total, even of these more than 85% were occupied after '61.
2.2 Emergency response
To support the emergency recovery phase, multi-sensor and multi-scale geomatics surveys were carried out, since the choice of the sensors and nominal map scale were tailored depending on the specific needs of each area of interest. The planned workflow complied with consolidated approaches from experiences and literature: firstly, a small scale analysis based on satellite images aimed at identifying hot spots; secondly, more accurate UAV photogrammetric flights enabling very high spatial resolution (GSD up to a few centimeters) on the main affected areas, aimed at assessing damages to buildings and infrastructures; then a series of high-resolution terrestrial range based surveys to provide a more detailed 3D dataset of the network infrastructure, to increase the thematic accuracy of the damage assessment.
2.2.1 Analysis from spaceborne images classification
The multi-spectral satellite image used was acquired by the GeoEye-1 satellite at an off-nadir angle of 23.3°, consisting of 3 spectral bands in the visible range of the e.m. spectrum (Red, Green, Blue) and one in the near-infrared part, with a nominal GSD of 0.5 meters. This acquisition was made on October 5, 2020, at 10:27 UTC, 3 days after the flood event, as there were better atmospheric conditions (minimal cloud cover of the affected area). The satellite data were a second-level product, thus already radiometrically corrected. However, further geometric correction, such as the orthoprojection, was needed to be properly georeferenced.
In order to carry out this operation, the 2011 Digital Elevation Model made available by the Piedmont Region geoportal was used as the terrain model, which is characterised by a geometric resolution of 5 m and an altimetric accuracy that varies from 0.3 m to 0.6 m depending on the characteristics of the territory (flat and regular areas have a better accuracy). As for the GCPs (Ground Control Points) and CPs (Check points), these were acquired with about 5 cm accuracy during a survey campaign through an NRTK (Network Real Time kinematic) approach using GNSS receivers connected to the national network of permanent GNSS stations. Image orthoprojection was carried out using ArcGIS Pro software via a non-parametric method—the Rational Polynomial Coefficients.
Next step was value-added information extraction process resulting from three main operations: image segmentation, analytical clustering, and supervised classification. The output of each operation is the input for subsequent operations, in which the goal is to produce a map representing thematic information classes. Training samples related to six main classes of land use have been selected for the automatic recognition: water bodies, forest, buildings, river beds, meadows, being buildings and breakdowns the most sensitive for distress assessment. By checking the preliminary result, multiple classes representing a single entity (i.e. the same category of shaded and unshaded surface) were merged. The classification result was further refined. First, the edges of the class boundaries were smoothed (since the classification is based on raster data). Then the image was filtered, by tools that assign isolated regions consisting of a few (or single) pixels) to the spatially adjacent thematic class.
Fig. 8 represents the result of the semi-automatic classification, i.e. a considerable advantage for estimating damage (potentially by comparison with pre-event data) with respect to photointerpretation, which is considerably more time-consuming and it is carried out manually by image analysts. As far as thematic accuracy is concerned, a manual classification (considered as ground truth) of the buildings and breakdowns classes, was compared with the automatic one in two test areas. The result was positive for the breakdown class and for the non-densely built-up test area (correspondence of correctly classified damaged surface around 80%, false positives around 20%). The automatic classification for the building class proved to be less satisfactory, especially in the more densely built test area, which is affected by shaded areas (although the correspondence of correctly classified surfaces stood at 89%, the false positives were 45 % in one case and 60% in the other). A quick visual inspection of the satellite classification in fig. 8 confirms the extensive areas of flooding of the Vermenagna stream and its tributaries, confirming that the damage to buildings and infrastructures was concentrated on extremely vulnerable elements, exactly as predicted by the flood hazard map.
2.2.2 Integrated 3D survey from UAV photogrammetry
Previous analysis and processing of satellite data were aimed at identifying the areas most affected in the flood event based on 2D products (e.g. image classification and visual interpretation). The following analyses are mainly based on 3D data (or 2.5D derivative products), since UAV photogrammetry enables the extraction of 3D information in the form of point clouds.
Starting from November 2020, the UAV photogrammetric surveys were carried out to support the activities of damage documentation and related water supply restoration activities in Limone Piemonte and Vernante municipalities along the Vermenagna river. The flights were therefore designed to cover the entire area shown in fig. 3a and sectors in fig. 9 and fig. 10, with a buffer strip of on average 80 m along the river and State Road No. 20, to generate orthoimagery and DSM products with accuracy suitable for 1:500 nominal map scale (expected precision of final products below 10 cm and accuracy lower than 20 cm (at 95% probability). A DJI Phantom 4 RTK UAV platform was used: it is equipped with a 20-Mpixel resolution CMOS sensor and a multi-frequency and multi-constellation RTK GNSS receiver. An in-depth examination of the 3D positional accuracies depending on different image orientation strategies was conducted. The findings affirm that if an RTK-enabled platform and an appropriate workflow are followed, the use of GCPs does not significantly improve the overall positional accuracy [23]. This result is also due to the integrated use of nadiral and oblique poses which make the bundle adjustment of the photogrammetric process more robust [63]. The overall process allowed to reduce the time required by acquisition compliant with the rapid mapping approach (fig. 9, fig. 10). To concisely conclude the topic of mapping via UAV photogrammetry, the table 1 is also reported that compares the quality parameters of flights processed in standard mode, i.e. with the aid of GCPs, and a flight oriented via direct georeferencing.
The UAV images acquisitions were integrated with higher resolution terrestrial surveys of hot spots to monitor infrastructures that are as important as they are fragile (an example of comparison of different resolution aerial products is shown in fig. 11). . In order to deliver more accurate documentation of damages, an articulated series of laser scans using Cam2's Focus3D Terrestrial Laser Scanner was acquired in the area of the collapsed parking lot and bridge in Limone Piemonte (Via Cuneo). A photogrammetric flight at the altitude of about 20 m above the ground from DJI Mini Mavic micro-drone was also planned in this area to obtain orthophotos and DSMs at very high accuracy and resolution, from the perspective of a multiscale survey strategy (fig. 12). Limone Piemonte Municipality has used these data afterwards to restore destroyed bridges and roads.
2.3 Recovery
2.3.1 3D models and change detection analysis
After this process, we have clear information about the disaster extent. Our concern was also to go into insights about what and where the damaged areas were for an impact assessment on the landscape.
First, to assess the damage that occurred during the event in more depth, it is possible to use a change detection strategy, which implements, as said, a process of identifying changes that occur on the land or on built heritage using automatic approaches, in this case, based using 2.5 D surfaces. Therefore, comparing the pre-event DEM by the Piedmont Region with the DSM generated by the UAV photogrammetric process is helpful. As mentioned above, they have different features (i.e., resolution and accuracy) due to different final scales and main purposes. For this reason, the first emergency mapping was more qualitative and based on operator skills and experience to recognize soil classes, while the latter, with a better ground resolution, is useful for a recovery plan.
Table 1 Comparison among accuracy values represented by RMSE in flights processed using a standard approach and direct georeferencing, enabling to save time acquisition.
|
Flying altitude (m)
|
GSD (cm/pixel)
|
GCPs RMSE (cm)
|
CPs RMSE (cm)
|
Area 1
|
87
|
4.40
|
1.53
|
4.00
|
Area 2
|
95
|
2.37
|
1.60
|
2.49
|
Area 3
|
117
|
2.91
|
2.98
|
6.60
|
Area 4
|
120
|
3.04
|
3.51
|
7.36
|
Area limone centre
(direct georeferencing)
|
108
|
2.97
|
2.53
|
4.69
|
Despite the different types of elevation data, differences in elevation can be analyzed to identify areas that suffered damage during the flood event. The elevation comparison was performed in the GIS environment by a subtraction operation between the two elevation models: that is the DEM was subtracted from the DSM (fig. 13 a, b). This operation results in positive values for buildings and vegetation and negative values for eroded surfaces. The result obtained is a map where areas having a lower elevation than the conditions recorded before the event are represented in gradations from yellow to red. These parts of the territory are where obviously damage occurred. Evidently, the roads abutting the river are in a highly threatened area and repeated damages over time make its recovery unsustainable. In fact, the entire road network is highly threatened with disruption while the most damaged or completely destroyed buildings are those close to the riverbanks.
Analysing the textured 3D geometric mode derived from the photogrammetric process, and particularly deriving architectural scale section profiles and projected representation on vertical planes, this change in landscape is even more evident. . A textured 3D geometric model was derived from the photogrammetric process by drone-acquired images and the survey of GCP. Sections at a scale of 1:100 (fig. 14 a ,b, c and fig. 15 a, b, c) were made on the post-event DSM (accuracy 0.05m) and the DTM (accuracy 0.60m) representing the pre-event condition: they were then combined to make a comparison between the two investigated situations. These products, characterized by geometric resolution and centimeter accuracy, quantify the major changes suffered by the land and the extensive damage to the built environment: the result is not only a different soil but also a widening of the riverbed
2.3.2 Rapid tool supporting local authorities recovery activities
One of the most significant issues related to the products derived from geomatic acquisition techniques and connected geoinformation processing is the usability of the data, both due to the weight of the files and above all to the need for users to have at their disposal the software necessary for visualization and hopefully management.
This reason also pushes research toward adequate and readily available tools to increase the diffusion and correct use of data in the direction of user-oriented webGIS solutions or even a digital AMS (asset management system) platform [64].
In the case of the post-emergency documentation experience of Limone Piemonte, whose recipients were the province water management society (ACDA - Azienda Cuneese dell'Acqua) and the municipality, the archive of documents delivered included the orthophotos and DSM of all areas, calculated using different resolutions to guarantee the use of the corresponding products in a more versatile way: the orthophotos were calculated with resolutions (GSD) equal to 3, 6, 12 cm, while the DSMs with resolutions equal to 6, 12, 24 cm
In 2020, and in the first months of 2021, when local authorities were busy restoring the usability of roads and infrastructures, as is known, the covid sars pandemic made free mobility difficult (in fact, it must be taken into consideration that this mountain area had remained free from the contagion until before the flood which brought many rescue and civil protection teams to those places, including the Covid sars).
The availability of orthophotos and DSM archives, when displayed in a GIS environment, can allow the route of the new sewer pipe to be designed without further inspections or, at most, reducing them to a minimum. This opportunity was created by online sharing with ACDA technicians.
In fact, GIS tools enable to evaluate the exact position of the excavation for the pipeline, enabling to check both the planimetric position on orthophotos and the trend of the section profile calculated on the DSM, both obviously derived from processing from UAV data. The images in the following fig. 16 a, b, c, d illustrate the phases of extracting a section profile operated on the DSM, once the exact planimetric position has been identified with the control of the high resolution orthophoto (3 cm).
The road section profiles enriched by the representation of retaining walls, small walls, bridges, buildings, road crossings, manholes, are certainly together with many elevation points, the usual technical preparations that the construction company carrying out the restoration must have at its disposal. The availability of DSM and orthophotos of such accuracy and resolution from which to derive the drawings is certainly considerable an innovation because it eliminates or reduces the need to carry out additional surveys in the field and therefore, certainly proves to be a sustainable aspect in the context of recovery