Digital Rock Mass Analysis for the Evaluation of Rockfall Magnitude at Poorly Accessible Cliffs

Abstract

The analysis of a digital rock cliff model, built by airborne photogrammetric data and infrared thermal images, is herein presented as an alternative tool for rock mass study in restricted and poorly accessible areas. Photogrammetric and infrared thermography techniques were combined for the geostructural and morphological characterization of an unstable cliff located in a nature reserve, where the rock mass extension and the environmental preservation rules required the use of minimally invasive surveying solutions. This methodological approach provided quantitative and qualitative data on both the spatial orientation of discontinuities and the location of major structural features, jutting blocks and past rockfall source areas. The digitally derived spatial data were used to carry out a rock mass kinematic analysis, highlighting the most recurring unstable failure patterns. Thermal images were overlapped to the photogrammetric cliff model to exploit the data combination and to analyze the presence of protruding rock mass volumes to be referred to as potential unstable volumes. Based on this activity, rock volumes were quantified on the digital model and the results were used to provide a zonation map of the potential magnitude of future rockfalls threatening the reserve. Digital data were validated by a field surveying campaign, which returned a satisfactory match, proving the usefulness and suitability of the approach, as well as allowing the quick and reliable rock mass characterization in the frame of practical use and risk management purposes.

1. Introduction

Landslides are acknowledged among the major threats to nature reserves, geosites and cultural heritage, e.g., [1,2,3,4,5,6]. Among these, rockfalls strongly affect geotourism accessibility [7,8,9], and several literature studies have underlined the importance of addressing the geological aspect when dealing with the fruition of natural tourist environments, e.g., [10,11,12,13,14,15]. Nevertheless, performing conventional rock mass surveys in restricted areas can be limited by both the preservation measures in force and local poorly accessible settings. In this frame, conventional surveys carried out at the lowest face of high and wide rock masses might suffer from low data representativeness. In this perspective, the use of remote surveying approaches has become a reliable and valuable solution to overcome these limitations. Aerial photogrammetry by unmanned aerial vehicles (UAVs) is one of the growing approaches for landslide mapping and monitoring, as well as for hazard and risk monitoring [16,17,18,19,20]. UAV photogrammetry applied to rock masses allows the building of digital models of the surveyed outcrop to be used for geostructural and geomechanical purposes. According to the literature, the remote photogrammetric characterization of rock mass, through discontinuity data extraction from dense point clouds, is a growing procedure that is gaining the trust of the international scientific community [21,22,23,24,25,26,27,28,29]. In particular, airborne photogrammetry has largely been employed for such purposes, e.g., [30,31,32,33], providing interesting results and proving a useful technology to achieve a detailed knowledge of the main features of rock masses. The available literature studies have proven the reliability of such approach applied to wide rock slopes: Mineo et al. [34], for example, compared three different procedures to extract discontinuity spatial data from RGB dense point clouds, highlighting the existence of a good match between remote and field data, if a precisely georeferenced point cloud is available. Salvini et al. [35] used dense point clouds and orthophotos to reconstruct the geometry of a quarry and to map the numerous discontinuities, while Menegoni et al. [36] compared the rock volume displaced by a landslide through 3D models. Similarly, infrared thermography (IRT) is a remote surveying procedure, which has been gaining positive feedbacks when applied to the survey of rock masses, even in combination with other non-contact methodologies [3,37,38,39,40,41,42]. Based on such background, this paper aims to address the rockfall threat at a nature reserve hosting a poorly accessible sub-vertical cliff, affected by numerous rockfalls over the past years, by analyzing an UAV-derived photogrammetric digital model and IRT images for the evaluation of the rockfall potential magnitude. The study area selected for this research is the Marinello Lakes reserve, located in northeastern Sicily (Italy) and characterized by a nature-created lagoon of white sand and gravel at the foot of the Tindari rock cliff. Past rockfalls have led to the permanent interdiction of some spots and paths, among which an uncoated tunnel anciently excavated in the rock mass to access the reserve when the sea used to break on the cliff. Both the cliff height and the logistical limitation for its complete survey make this case suitable for the scientific purpose aimed by this study. In fact, the achievement of a digital rock mass model could solve such a limitation and would allow reliable cliff characterization for stability evaluation. Through an integrated UAV–IRT photogrammetric campaign, the extensive geostructural characterization of the cliff was performed, aiming at locating the sources of rockfall threats, modeling the rock mass setting, and evaluating the associated potential in terms of likely unstable volumes. To achieve such goals, airborne RGB photogrammetric surveys were carried out to build detailed 3D models of the cliff; automatically extracted spatial data were statistically analyzed on stereograms and a kinematic analysis was carried out by taking into account the main rock failure patterns [43]. Photogrammetric data were supported by the analysis of IRT images, which were matched with the 3D model of the cliff, bringing novel information to the analyzed digital model, such as the recognition of the major discontinuity sets crossing the slope, the easier location of past rockfall detachment zones and the detection of loose blocks. This latter aspect was further analyzed on the photogrammetric dense point cloud by a specific application tool for the geometrical size estimation, to provide an evaluation of the potential magnitude of future rockfalls by measuring the range of volumes belonging to detected rock blocks and providing a cliff zonation based on their spatial distribution and statistics.

2. The Study Area

The Marinello Lakes nature reserve is located along the easternmost Tyrrhenian coastline of Sicily, overlooking the Patti gulf, at the foot of the Cape Tindari promontory (Figure 1). From the geological point of view, the area is set in the central–northern sector of the Peloritani Mountains, which represent the southernmost edge of the Calabria-Peloritani Orogen, an arcuate belt consisting of a set of metamorphic nappes and remnants of Mesozoic–Cenozoic covers [44]. Cape Tindari promontory is made up of intensely fractured two-mica marbles and paragneiss crossed by intersecting systems of strike-slip NW–SE trending faults and NE–SW trending normal faults [45] (Figure 1). Recent tectonics have raised up the metamorphic basement, giving rise to steep cliffs and escarpments [46] and, in turn, to unique landscapes and peculiar natural spots. The reserve hosts five salty lakes, aligned along a NW–SE direction and sub-parallel to the Cape Tindari coastal cliff, and a split characterized by continuous shape modifications (Figure 1). The rock cliff maximum height is about 250 m a.s.l. and its morphology is characterized by sub-vertical walls locally interrupted, at the highest sectors, by gentle inclined spots where vegetation occurs, likely due to the presence of talus. The sub-vertical rock face impends over the Marinello Lakes and some debris accumulations, arising from previous landslides, locally occur at its foot. Numerous rockfalls have occurred in past, as testified also by the widespread presence of blocks accumulated at both the foot of the slope and inside the lakes, with cumulative volumes of hundreds of cubical meters (Figure 1). In July 2017, for example, a rockfall injured a person, prompting the local authorities to forbid access to the reserve, including an underground path, from now on referred to as “tunnel”, which was drilled to access the reserve when the rock cliff was overhanging the sea. Moreover, the walking paths running parallel to the foot of the cliff are still forbidden due to the rockfall threat.

3. Materials and Methods

Due to the extension of the unstable rock slope and to the poor accessibility field conditions for conventional surveys, a remote airborne photogrammetric survey was carried out in the southern sector of the Tindari cliff, with the aim of building a detailed three-dimensional digital rock mass model for geostructural purposes (Figure 2). In this paper, airborne photogrammetric surveys were carried out by a Parrot Anafi Thermal quadricopter equipped with a CMOS (complementary metal–oxide semiconductor) 21 Mpixel with a 26 mm focal length digital sensor. The Sony IMX230 sensor was 1/2.4 inch in size, with a CROP factor of 7.02. Two flight plans were set to achieve frames overlapping for about 70% (Figure 3a). In particular, the first was a nadiral flight aimed at obtaining a perpendicular overview of the survey area. The second flight allowed acquiring frontal and oblique images (45°) of both the sub-vertical and inclined (35–50°) rock facets, also according to other literature studies mapping very steep areas [47]. The flight altitude was kept constant at 40 m for oblique and frontal flights and a slightly higher altitude was reached for nadiral flights, so that each photo covered an area of about 45 m². This resulted in an average ground sample equivalent distance (GSD) of about 1.98 cm/pixel. Nineteen ground control points (GCPs) were homogeneously arranged at the foot of the rock mass for the correct model georeferencing (Figure 3b), achieving an average RMSE on the x-, y- and z-axes of 2.3, 1.8 and 4.2 cm, respectively. All the frames acquired during the flights were post-processed to build a dense point cloud by the SfM technique [48,49,50], in order to achieve a three-dimensional model of the surveyed rock mass (Figure 2). The image processing was carried out by using a computer program allowing the following 4 steps: (1) image import and matching for the definition of the binding points between the input images; (2) camera alignment; (3) construction of the sparse point cloud; (4) construction of the dense point cloud. The resulting point cloud was further processed by the open-source point cloud and mesh processing software CloudCompare to extract the dip-directions of discontinuity planes by the “scalar field” objective and semiautomatic approach. Data were then plotted on stereograms and statistically treated to group the main sets, and a kinematic analysis was carried out to define the potentially unstable patterns threatening the reserve fruition. A field validation campaign was carried out to verify the reliability of digitally surveyed data according to the International Society for Rock Mechanics (ISRM) recommendations [51].

The digital photogrammetric analysis was aided by IRT surveys, carried out at the most exposed cliff spots to fruition, to acquire focused information on the geostructural and morphological features occurring along the cliff, in addition to data retrieved by the dense point cloud (Figure 2). In particular, IRT is a non-contact surveying methodology allowing the measurement of the surface temperature of an object characterized by temperatures above absolute zero based on its emissivity. In fact, according to Stefan–Boltzman’s law, the radiation emitted by an object is directly proportional to its temperature. This allows estimating the surface temperature of a material, with a known emissivity, through thermal cameras, which provide color-scale images showing the surface temperature variation of the framed subjects. Such a technological approach, herein applied to the qualitative analysis of thermal images, still has a limited background for rock mass surveying and monitoring. Pappalardo et al. [52] proved that IRT outcomes are strongly linked to the rock mass degree of fracturing and that it is a reliable tool for the detection of fractures, rockfall source areas and cavities. In this study, IRT surveys were carried out by using a 320–240 pixel thermal camera operating within a −20–650 °C temperature range (with ±2 °C accuracy). Thermograms were acquired in dark environmental conditions (after sunset), when the influence of parasite radiation is lower and the best thermal outcomes of the rock mass can be achieved [52]. The emissivity was set to 0.93 according to previous accounts in the literature [52], which can be regarded as a representative value of the whole rock face, where elements affected by different emissivity values usually occur (e.g., bare rock, vegetation, debris, weathering film, wet/dry sectors). Thermograms were post-processed by the software FlirTools, which allows the analysis of specific ranges of temperatures to find the best outcomes in terms of displayed information. Results arising from the two remote surveying approaches were then combined by overlapping selected thermal images on the digital RGB model through the open-source software Meshlab. It allows superimposing and aligning raster images, thus achieving an integrated tool for a combined analysis, with a cumulative reprojection error lower than 10 pixels in terms of distance. The two-dimensional thermal images were, indeed, textured on the corresponding portions of the digital model and aligned by using the raster alignment tool, based on homologous points. The expected results of this activity are the identification, on the RGB point cloud, of the most open/persistent discontinuities, which likely represent the structural elements driving the instability (Figure 2). Moreover, based on the thermal variations arising from the morphological rock face irregularity, IRT was also useful to locate, on the RGB point cloud, the protruding rock mass volumes to be considered as potential detachment volumes, as these appear as cold regions due to the greater rock exposition to the environmental cooling [52]. Due to the extent of the studied rock mass and to its complex degree of fracturing, the potentially detaching rock volumes were numerous and greatly variable in size. Therefore, their average volume was estimated on the digital rock mass model through the “Compute 2, 5D volume” tool implemented in the CloudCompare software. This allows rasterizing the dense point cloud by dividing it, with a fixed step, into square cells in the XY plane and associating a Z elevation value, thus creating a digital elevation model (DEM) [53]. Such an analysis was extended to the whole digital cliff model, and for each selected block an approximate detaching plane was rasterized to calculate the difference in terms of elevation between the plane and the protruding rock block, and to turn it into a volume value. The reliability of such an approach has already been proved in the literature, even for the estimation of topographic volume changes applied to slope morphological analysis, e.g., [54,55]. Once the most evident unstable blocks on the point cloud were located and their volumes estimated, a database was created, and a rock volume distribution map was provided for practical purposes.

4. UAV Photogrammetry Outcomes

The three-dimensional cliff model was built on the basis of 276 aerial images, ac-quired according to the methodology reported above, covering a 57,120 m² outcrop area (Figure 4). Frames were overlapped to generate a 104,000-point scattered point cloud, which was turned into a 114-million-point dense point cloud with a pixel accuracy of 1.33 cm/pixel by the SFM procedure, e.g., [56,57,58,59]. The good resolution combined with the presence of GCPs returned a georeferenced cliff model to be exploited for geostructural purposes. In particular, the discontinuity extraction was carried out at eight representative surveying windows (

Table 1. Surveying windows’ size and number of discontinuity poles extracted by the dense point cloud.

Surveying Window	Surveying Window Size (m)		Number of Extracted Poles
	Height	Width
W1	17	12	117
W2	18	20	171
W3	30	35	192
W4	19	40	156
W5	37	49	275
W6	16	38	166
W7	28	36	198
W8	12	24	154

), which were located at key rock mass sectors where the main instability features were suspected (Figure 4). The results showed that some discontinuity sets systematically occurred at almost all the surveying windows (Figure 4). These were S1, S2, S3 and S5, which, according to their orientation, could be correlated to the main regional fault systems, such as both the N–S direct fault lines cropping out in the westernmost sector of the study area, and the NW–SE trending faults crossing the studied promontory (Figure 1). On the other hand, further minor discontinuity sets were identified at some of the surveying windows, where their occurrence was more frequent, suggesting a structural complexity of the cliff. Among the surveyed discontinuities, bedding and shear planes were recognized as well. From the kinematic point of view, the geometrical relationship between the slope face and the structural sets suggests that all the surveyed rock mass sectors suffer from instability features due to the unfavorable kinematic orientation of the discontinuities. Both planar sliding and toppling failure patterns were, indeed, recognized. The first condition occurs when a discontinuity plane dips at a flatter angle than the rock face, with a dip direction differing from the slope dip direction no more than about 20°, and it was mainly verified for the S1 and S2 sets. By taking into account the abrupt variation of the slope immersion, other sets can locally daylight and become involved in such a failure pattern. Moreover, there were some spots where the sub-vertical cliff face dipped towards southwest (W1 and W6), and the planar sliding kinematic instability involved the S3 and S5 sets. Similarly, potential toppling was detected where the discontinuity surfaces dipped into the slope face, such as in the S3 and S5 sets, followed by S7 at W4 and by S1 and S2 at W1 and W6. Moreover, the intersection between two or three discontinuity planes gave rise to unstable wedges, mainly characterized by a symmetrical configuration, with some asymmetrical patterns at W2, W3, W5 and W7. In this latter case, wedge sliding would occur along the most unfavorably daylighting plane. It must be highlighted that all the previously reported failure mechanisms also involved some random planes not included in the statistical set contouring, suggesting that the instability may locally occur also along occasional discontinuities.

5. Field Rock Mass Data Validation

Data automatically extracted by the digital cliff model were validated in the field by random rock mass surveys (Figure 5a) aimed at verifying the goodness of the automatic data sampling, thus supporting the reliability of the achieved outcomes. This operation was affected by the poor accessibility of the cliff and by local restrictions arising from the active state of the slope in terms of rockfall occurrence. In particular, the discontinuity spatial orientation was measured according to the conventional ISRM subjective criterion at accessible portions of the cliff, and the collected data were then compared to those digitally extracted by the photogrammetric model. Moreover, some geostructural measurements were carried out within the ancient tunnel, where the three-dimensional prosecution of the selected discontinuity planes could be observed at both the gallery vault and flanks (Figure 5b,c). The geostructural setting of surveyed spots was characterized by the presence of variously oriented discontinuity traces and planes, with a moderate spacing and a persistence ranging from low to very high. The discontinuities showed different aperture values, with extremely local wide to cavernous openings and the local presence of both soft and hard filling material [51].

The stereogram resulting from the field measurements satisfactorily matched with the digital measurements performed on the cliff model, thus allowing a successful validation of the presented data (Figure 5d). Some scattered random poles occurred in accordance with the random values highlighted by the digital model analysis. A comparable geostructural setting was outlined within the tunnel, where six main sets were grouped, including a series of NW-dipping structures, likely related to the regional fault system affecting the study area and the whole cliff. It is self-evident that validation plots provide a synthesis of geostructural data collected at different spots; therefore, the relationship between field and UAV orientation data has to be determined for the whole cliff face rather than at each digitally analyzed rock mass sector reported in Figure 4, where some discontinuity might be more or less evident in one sector than in another.

6. IRT Outcomes

Photogrammetric data were integrated by IRT surveys carried out at selected rock mass sectors (Figure 6a), located at the most exposed spots to fruition (i.e., tunnel mouth and path running at the foot of the cliff), to evaluate the thermal anomalies and to experimentally relate them to specific morphological features of the rock mass. Post-processed thermograms were overlapped to the digital 3D cliff model so as to highlight the thermal anomalies’ three-dimensional variation occurring along the framed slope. The results showed that the cliff portion overhanging the tunnel mouth (Figure 6b) is affected by variable thermal features related to both the presence of vegetation and the slope morphology. Vegetation holds the lowest surface temperatures (up to 9.4 °C), thus providing a lower boundary of temperature range to consider for the qualitative thermogram interpretation. From the stability point of view, vegetation usually occurs at the loose-weathered bedrock, which has not been involved in recent rockfalls. Besides this feature, low-to-intermediate surface temperatures (9.4–10.1 °C) label the most protruding rock mass sectors, which are represented by rock volumes more exposed to cooling processes. This interpretation is in accordance with what verified in previous studies, e.g., [52] and, in this specific case, it allows the quick location of the jutting rocks to be regarded as potential rockfall-prone volumes. Based on the same principle, the highest surface temperatures (12.5–14 °C) label the hollowest rock mass sectors, benefiting from a morphological surface temperature preservation. In this case, hollow areas left by past detachments can be quickly located on the digital model, thus allowing a sort of zonation of the landslide-affected spots along the cliff. Moreover, past source areas are characterized by regular geometries (Figure 6b), suggesting the preponderant role of the rock mass fracturing in the isolation of unstable rock mass volumes. For example, at the W2 rock mass sector, the discontinuity sets mainly driving these specific instability features are S3, S7 and S10, as resulting from the match between IRT image and automatically extracted orientation data (Figure 6b). A further geostructural datum suggested by IRT is the presence of linear positive anomalies at some discontinuity traces related to sub-vertical planes (dip-immersion 070/80–250/80), likely responsible for rock mass “slicing”, thus representing a further key instability feature (Figure 6b). In fact, the literature regarding the application of IRT to rock masses suggests that open and persistent discontinuities are labeled by positive thermal anomalies [60,61], with specific reference to images acquired in dark environmental conditions [34,62], thus allowing differentiating the main systems in heavily jointed rock masses. This leads to the recognition of further discontinuity intersections along the surveyed rock slope, such as those produced by S4 and S6 and S7 (Figure 6c). Moreover, the IRT outcomes showed that some great emptied areas occur where persistent fractures cross the rock mass. In Figure 6c, for example, the persistent fracture highlighted by the highest surface temperature has a NE–SW trend, matching well with one of the regional fault systems, thus suggesting the influence of tectonics on slope instability. Similarly, in Figure 6d, the highest surface temperatures retrace two shear planes, characterized by a great persistence. Wedge kinematic patterns are recognized as well, although, in some cases, the presence of vegetation tends to hide the thermal contrast between rock and intersecting discontinuities.

Moreover, the intersection between sub-horizontal and sub-vertical planes gives rise to a sequence of empty sectors and potential detaching volumes, to be regarded as a high hazard zone, also due to the presence of undercutting (Figure 6d).

7. Digital Estimation of Block Volumes

Having ascertained the unstable kinematic rock mass setting and exploited the combined IRT–photogrammetry potential for detecting jutting rock volumes and past rockfall source areas, the digital model was further analyzed for the quantitative rock volume estimation. This represents a crucial activity aimed at providing an estimation of the potential magnitude of future events, in the perspective of a practical utility of the methodological approach developed herein. Starting from the protruding rocks and cavities highlighted by IRT at the rock mass sector overhanging the tunnel mouth (Figure 7a), these were identified on the IRT image and then transposed to the RGB photogrammetric dense point cloud, where they were selected and measured. In particular, for each selected item (block or cavity), a reference plane was rasterized according to the methodology previously explained. In the case of jutting rocks, this plane is assumed as a potential detaching surface and the elevation difference between the protruding block surface and the plane itself is a measure of the boulder width, allowing calculating its volume. Similarly, for cavities, representing signs of past rockfalls in the form of deep scars in the rock wall (Figure 7a), the rasterized plane coincides with the external rock face. The elevation difference between the plane and the inner cavity walls is a measure of the cavity depth. So, the estimated jutting volumes ranged from 2.2 to 16 m³, where the greatest value was related to a rock volume made up of different adjacent jutting blocks, which could likely have been mobilized in a single potential rockfall event. On the other hand, the cavities showed a volume of up to 7.1 m³, proving the past occurrence of large rockfall events (Figure 7a). The calculated volumetric values were characterized by a > 90% match between the created plane and the block, with errors lower than 10% of the unmatched cells. It is therefore believed that the error of the volume calculation for each block is negligible since the volumetric estimation accuracy mainly relies on the scaling factor of the point cloud.

This volume estimation procedure was extended to the whole digital cliff model, suggesting that the studied rock mass hosts countless widespread isolated rock blocks, which could be regarded as potential future rockfalls, with volumes up to >10m³ (Figure 7b). It is underlined that, in some cases, greater volumes could be mobilized both as single blocks and as a set of heterometric rock material simultaneously driven downstream by the major discontinuity planes. Moreover, measured cavities and hollow rock mass sectors are, in some cases, characterized by peculiar shapes retracing the main kinematic failure patterns.

According to these outcomes, the block distribution along the surveyed cliff can be categorized to analyze the potential magnitude of future rockfalls (Figure 8). Six main con-centration sectors were defined on the digital rock mass model, grouping block volumes into four size classes, from small (<0.5m³) to extra-large (>2.5m³) (Figure 8). This allows providing a practical tool, useful in the perspective of protection measure design, by focusing on specific rock mass sections, which might be affected by different unstable rock volumes. In particular, at sectors 1 and 2, about half of the detected blocks had volumes <0.5 m³, and only 11–18% of the detected blocks exceeded 2.5 m³; at sector 3, volumes were balanced around one quarter for each category, similarly to sector 4 where a slightly lower percentage of volumes between 0.5 and 1 m³ was found (Figure 7). Contrarywise, sector 5 was mainly affected by rock volumes ranging from 1 to 2.5 m³, with a lower percentage of small volumes, while at sector 6, no volume greater than 2.5 m³ was recognized (Figure 8).

8. Discussion

The combined analysis of the digital rock mass photogrammetric model and thermal images paves the way to a discussion of both the specifically studied unstable cliff, which affects the safe fruition of a nature reserve, and of the scientific methodological approach applied in the study. These two aspects are mutually dependent, since the resulting data, surely related to the local contest of the study area, confirmed the reliability of the methodological approach, as well as its suitability for the specific setting of a restricted area. Starting from the achieved outcomes, the airborne photogrammetry allowed surveying a wide and high cliff, overcoming the logistical limitations linked to the direct accessibility of the site for conventional surveys. UAV photogrammetrically derived data were used to build a georeferenced detailed 3D model of the cliff, which was used for quantitative data processing. In particular, the automatic extraction of discontinuity spatial data further proved to be a practical and useful tool for the quick reconstruction of the geostructural setting. This is in accordance with literature studies highlighting the goodness of quantitative remote sensed data if supported by field georeferencing and validation, e.g., [25,33,34,63,64]. Nevertheless, this approach suffers from a potential underestimation of the low-recurring discontinuity planes, whose statistical relevance could be concealed by the abundance of automatically extracted data, resulting in incomplete stereograms, especially due to the extension of the survey. In this view, collected data have to be verified and validated in the field by surveys along accessible rock mass portions and carried out according to international standards [51]. Besides this aspect, which is easily solvable with a meticulous approach, the analysis of the UAV digital model provided a useful basis for the kinematic and stability analyses, which, in the case presented herein, returned a rock slope showing multiple unstable kinematic failure patterns as a direct consequence of the intense degree of rock mass fracturing. It was revealed that these kinematic patterns had driven the previous slope instability, as testified by the occurrence of empty rock mass zones geometrically defined by the major discontinuity planes. This aspect was further highlighted by IRT surveys carried out at key rock mass sectors, showing that the highest-temperature linear anomalies retraced the major geostructural systems. Thanks to the IRT images, indeed, it was possible to better define major discontinuity intersections, also supporting the literature outcomes on the use of IRT for the study of wedge kinematic patterns in jointed rock masses [65]. The use of IRT also allowed considerations of the cliff face morphology, thanks to the possibility of locating the most protruding rock volumes, in the form of negative thermal anomalies (low surface temperatures), and cavities left by previous detachments (high surface temperatures). This result represents the achievement of a further goal among those already accomplished by the use of IRT for rock mass survey, thus shedding light on new potential applications of this methodology. Moreover, the combination of the outcomes arising from the two surveying methodologies allowed matching specific thermal aspects to geomorphological features of the cliff face, providing a useful tool for rock mass analysis. In fact, the possibility of a quick remote location of jutting blocks and cavities along wide rock slopes turns the scientific methodology application into a practical opportunity to be used in several settings. This procedure can be, indeed, repeated in other areas, as there is no specific dependency of the used approaches on the local setting. In fact, thermal images were acquired in dark environmental condition, confirming what previous literature accounts proved, i.e., that the absence of solar light (late evening/nighttime) is the ideal environmental setting for such-aimed IRT surveys, because no interferences arising from direct or indirect lighting condition are present [52]. It must be underlined that this approach is not fully automatic since thermal images have to be processed before being overlapped on the point cloud. In this frame, the experience of the operator plays a key role in achieving the best highlight of the searched features both during the processing of the raw IR images and during the combined analysis of the outcomes, when the IRT-recognized features have to be matched with specific rock mass elements. Advances in this application should involve the construction of a thermal dense point cloud allowing thermogram post-processing directly on the 3D model, so as to enhance the global contextualization of specific thermal anomalies at the whole rock mass scale. Some recent pioneering studies attempted to challenge this issue by building thermal dense point clouds from IRT images [34,40], although further research is needed.

Moreover, the volume of located blocks and cavities can be quantitatively estimated on the digital model, thus providing information on the potential magnitude of future rockfalls. This is a key aspect, as the magnitude of a landslide is linked to its hazard. Therefore, the methodological approach presented in study can be considered also a reference starting point for digital-model-based hazard assessment procedures to develop at natural sites. The final output of the study is, indeed, an overall map of potential future rockfall sources accompanied by statistical considerations on the volume distribution along the cliff. This also represents a practical and useful tool for the reserve management in the perspective of planning mitigation measures. Its main utility relies on the possibility of sub-dividing the rock cliff into areas of intervention where protection works can be chosen according to the specific potential rockfall magnitude.

9. Conclusions

In this paper, the analysis of a digital cliff model, built by both airborne photogrammetric data and infrared thermography images, was presented as an alternative tool for the evaluation of the potential rockfall magnitude in poorly accessible cliffs that also require low invasiveness in terms of the possibility of performing field surveys. The combination of photogrammetry and infrared thermography allowed the remote quantitative and qualitative characterization of a wide rock mass portion, which can be easily repeated in other settings worldwide, leading to the achievement of the following conclusions:

(1)

A georeferenced photogrammetry-derived digital rock mass model can be exploited for the quantitative extraction of discontinuity spatial orientations, thus allowing the related stereographic statistical processing and kinematic analysis.

(2)

Infrared thermography confirmed its utility in mapping the major discontinuity traces and analyzing the rock face morphology, with good results achieved by acquiring thermal images in dark environmental conditions.

(3)

The combined analysis of photogrammetric and thermal data can be exploited to locate unstable projecting rock blocks and detachment areas of past rockfalls. Their volume can be quantitatively assessed, shedding light on an innovative methodological approach for the evaluation of the potential magnitude of future rockfalls.

(4)

The spatial and statistical distribution of quantified rock volumes can be reported on a distribution map, suggesting practical potential for hazard assessment studies and the planning of remedial measures.