In recent decades, both high-resolution digital photographs and structure from motion (SfM) software have enabled the generation of high-quality topographic information. In geosciences many studies have been dedicated to morphological processes (Castillo et al., 2012; James and Robson, 2012; James and Varley, 2012; Amici et al., 2013b; Casella et al., 2014; Gomez-Gutierrez et al., 2014; James and Robson, 2014; Lucieer et al., 2014; Ryan et al., 2015; Westoby et al., 2015; Woodget et al., 2015; Eltner et al., 2015; Dietrich, 2016b; Smith et al., 2016; Javernick et al., 2016; Walter et al., 2018; Deng et al., 2019; Johnson et al., 2014). Applications include runoff laboratory trials (Morgan et al., 2017), applied geology (Niethammer et al., 2012; Russell, 2016; Saito et al., 2018), geomorphology (Bemis et al., 2014; Javernick et al., 2014; Snapir et al., 2014; Dietrich, 2014; Smith and Vericat, 2015; Bakker and Lane, 2015; Dietrich, 2016a, b; Mercer and Westbrook, 2016; Pearson et al., 2017; Prosdocimi et al., 2017; Marteau et al., 2016; Balaguer-Puig et al., 2017; Vinci et al., 2017; Heindel et al., 2018; Seitz et al., 2018), glaciology (Immerzeel et al., 2017; Piermattei et al., 2016), coastal morphology (James and Robson, 2012; Casella et al., 2016; Brunier et al., 2016), volcanology (James and Robson, 2012; Bretar et al., 2013; Müller et al., 2017; Giordan et al., 2017, 2018; Carr et al., 2018; Favalli et al., 2018; Witt et al., 2018; Andaru and Rau, 2019; Bonali et al., 2019; De Beni et al., 2019) and geophysics (Amici et al., 2013a; Greco et al., 2016; Di Felice et al., 2018; Zahorec et al., 2018; Federico et al., 2019). SfM is commonly used in the cultural heritage field for 3D reconstruction (Sapirstein, 2016, 2018; Sapirstein and Murray, 2017; Jalandoni et al., 2018). The monitoring of active geological processes is a preventive action in risk mitigation (Stöcker et al., 2017; Turner et al., 2017b; Diefenbach et al., 2018; Rosa et al., 2018; Deng et al., 2019). Disasters occur when two factors – hazard and vulnerability – coincide. The risk is proportional to the magnitude of the hazards and the vulnerability of the affected population. Among the deformation monitoring systems, the photogrammetry technique from unmanned aerial vehicles (UAVs) is becoming more widely used thanks to the high efficiency in data acquisition, the low cost compared to traditional techniques and the acquisition of high-resolution images (Harwin and Lucieer, 2012; James and Robson, 2012; Westoby et al., 2012; Fonstad et al., 2013; Javernick et al., 2014; Johnson et al., 2014; James et al., 2017a, b, 2020). This technique is important for studying catastrophic natural events such as floods, earthquakes, landslides, etc. Different acquisition methods and the ability to obtain high spatial (centimeter) and temporal resolution (hours or days) (Boccardo et al., 2015) enable the acquisition of detailed information on the evolution of the landscape; therefore UAVs are an effective and complementary tool for field investigations. Furthermore, UAVs have other advantages including (i) the ability to fly at low altitudes, (ii) the ability to reach remote locations, (iii) the ability to host multiple sensors (cameras, lidar, thermal imaging cameras, navigation/inertial sensors, etc.), (iv) the ability to capture images at different angles and (v) the flexibility to carry out monitoring operations on a small, medium and large scale (Jordan et al., 2018). Ground control points (GCPs) are used to improve the accuracy of the resulting data. Therefore, recognizable points in the UAV imagery are measured with a high-precision surveying device to georeference the data. In this process a correct number of GCPs is required which leads to a greater accuracy of the resulting data (point clouds, 3D grid, orthomosaic or digital surface model, DSM). The precision of the resulting data is also controlled by other variables, such as the focal distance of the camera, flight path and flight altitude, the orientation of the camera, the picture quality, the processing chain, and the category of UAV system (fixed or rotary wings).

In this paper, we present the results and analysis of the surface deformation monitoring of the mud volcano of Santa Barbara (Caltanissetta, central Sicily) (Fig. 1). We have applied the statistical analysis of significant changes with a level of detection at 95 % confidence (LoD95%). In detail, we used precision maps and the M3C2-PM (Lague et al., 2013; James et al., 2017b) algorithm to determine the surface variations. The statistical analysis allows us to verify (i) the uncertainty between the different surveys, (ii) the spatial variability in the accuracy in the surveys (James et al., 2017b) and (iii) the quality of the georeferencing of the surveys based on the number of GCPs.

The mud volcano of Santa Barbara is located within the Caltanissetta foredeep basin of the Apennines–Maghrebian collisional chain which developed from the late Miocene to the Quaternary along the border of the converging Eurasia–Nubia plate (Catalano et al., 2008; Dewey et al., 1989; Serpelloni et al., 2007). This structural domain is formed by a foreland fold and thrust belt involving the deposition of clastic sediments which were gradually deformed from the late Miocene to the Pleistocene (Monaco and Tortorici, 1996; Lickorish et al., 1999, and references therein).

According to Madonia et al. (2011) mud volcanoes are most of the time in stasis, but they represent a preferential way for rising fluids rich in methane and sludge; therefore they can be considered a risk to urbanized areas or sites with an economy dedicated to natural attractions.

At several mud volcanoes (e.g., Ayaz-Akhtarma and Khara Zira island mud volcanoes in Azerbaijan), certain geomorphic and/or structural features have been observed within the year preceding a paroxysmal event (Antonielli et al., 2014; Madonia et al., 2011). The area of the Santa Barbara volcano was also affected in 2008 by a paroxysmal mud eruption which was preceded by deformation features (Fig. 2). Moreover, the surface of the mud volcanic cone is incised by a drainage system (Fig. 1) characterized by hydrographic basins with elongated dendritic geometry arranged for a centrifugal development from the areas of the summit craters towards the lower slopes of the volcano complex. The higher order of the thalwegs presents deep recessed meanders (landscape rejuvenation process). This morphometric structure is typical of uplifting areas and therefore relative decrease in the base level. This suggests an inflection process of the volcano ground surface induced by an increase in fluid pressure inside of a shallower stagnation chamber which is located at a depth of about 30 m (Imposa et al., 2018). The stagnation chamber has a “sill-like” geometry, a radius of about 50 m and a thickness of about 30 m. This morphostructural configuration supported by geophysical data configures the active geological structure as a high potential geological hazard.

On the surface, fractures and shear lineaments extend outside the erupted mud area (Madonia et al., 2011; Bonini et al., 2012; INGV, 2008; Regione Siciliana, 2008), and they highlight the high stress and strain environment induced by the mud volcano (Fig. 2). Such structures have been detected in 2002 and 2008, and we speculate that they are still active (Fig. 2). This development has often been a precursor of paroxysmal events such as the 11 August 2008 event (INGV, 2008; Regione Siciliana, 2008).

As illustrated in Sect. 2.2, the results show that using between 40 % and 60 % of the GCPs (as control points) the RMSE value has minimal variation. Thus, the optimal minimum number of GPCs is between 12 and 18. When the threshold of 60 % is exceeded, there are no significant improvements (Fig. 12). This result has been confirmed in all campaigns.

The results obtained from the photogrammetric comparisons supported by geodetic topographic survey have an average uncertainty of about 6.4 cm, with a minimum of about 2 cm and a maximum of about 12 cm, relative to an area of 42 700 m2 (Fig. 13a). The uncertainty of the central area has an average value of about 3.9 cm, with a minimum of about 2 cm and maximum of 10 cm, on an extension of 360 m2 (Fig. 13b).

During the monitoring period, in addition to natural changes, we recorded anthropogenic action due to the relocation of objects (garbage) in the area after unauthorized access. These changes have a decimetric order of magnitude and are easily detected by the technique used (Fig. 14).

Two types of analysis were carried out: semi-quantitative and quantitative. The surveys on 29 July 2019 and 15 June 2020 (time interval of about 1 year) were chosen to perform the semi-quantitative analysis. This analysis was carried out on the whole mud volcano area, and the objective is to detect significant deformations (on the order of decimeters). In order to visualize this deformation (Fig. 15), the values of M3C2 distance ranging between -2 and 2 cm were excluded according to the minimum value of distance uncertainty. This range was verified instrumentally with the use of callipers.

Comparing the 29 July 2019 and 15 June 2020 surveys (Fig. 15), we observe that the surface of the volcanic cone is not affected by deformations. Only morphological changes in the volcanic structures can be underlined, such as small eruptive cones, gryphons, sauces and mud pools.

The quantitative analysis was performed on small central portions of the mud volcano. The aim of the quantitative analysis is to estimate the trend and evolution of the deformation. To assess the deformation and the local morphostructural evolution, two temporal series have been developed in two areas: Z1 (collapse zone) and Z2 (uplifting zone) (Figs. 16 and 17). The zones were chosen in relation to the significant change. For the selected zone, the average distances between points with significant change were computed by M3C2-PM. Z1 has an extension of 1 m2 (Fig. 16), and Z2 has an extension of 0.84 m2 (Fig. 17).

The campaign of 29 July 2019 (T0) was chosen as the master (Table 2).

In Fig. 16, the subsidence (about 10 cm per 60 d) of gryphon is represented by the southwest migration of the mud pool edge. In Fig. 17, the uplift (about 4 cm per 60 d) of a blind gryphon is represented by the development of radial fractures on the cone surface. Close to the growing gryphon, another indication of deformation is the deviation of a mudslide flowing in the north–south direction towards the southern sector of the analyzed area. Finally, the sudden appearance of the gryphon is well highlighted in the orthophoto of the last survey (Fig. 17).

UAV technology combined with SfM is a valuable tool in geological risk assessment and monitoring; however, some issues must be considered.

The first is the quantity and quality of the acquired GCPs. According to the results, the minimum optimal number for georeferencing is 12 GCPs. A slight improvement is observable up to 18 GCPs, while no appraisable improvement is detected for a higher number of GCPs (Fig. 12). In addition, the method of acquiring GCPs reduces their error by using the total station theodolite (Table 1). The combined use of high-precision topographic instruments with an optimal number of GCPs improves the reliability of the datasets.

The second aspect to be considered is the evaluation of distance uncertainty when two surveys are compared. The distance uncertainty between two datasets (surveys) can be considered as an estimate of the sensitivity of the methods to detect measurable topographic changes. The results of the M3C2-PM show that the average distance uncertainty between the first and the last surveys is about 6.5 cm over the whole area at the 95 % confidence level (Fig. 13a). Furthermore, considering a smaller portion of the area, the uncertainty decreases to about 3.9 cm at the 95 % confidence level (Fig. 13b). This allows us to analyze certain morphological changes and anthropogenic activity on the mud surface (Fig. 14). The anthropogenic activity determined a height decrease of about 16 cm where a wood platform was previously located. Moreover, a height increase of about 14 cm is recorded at the place where the wood platform has been relocated. The values recorded are consistent with the real thickness of the object detected, differing by 2 cm (Fig. 14). The callipers show that it is possible to measure changes of at least 2 cm (Fig. 11).

After the uncertainty and sensitivity of the surveys were computed, the time series were made. Considering the significant changes and their errors, we computed the trend of deformation of two areas in order to reconstruct the evolution of the phenomena which generated these changes (Figs. 16 and 17).

Considering the results obtained by SfM, we propose a monitoring system of the Santa Barbara mud volcano based on a semi-quantitative approach. We defined three states by setting a space-time range:

normal state if the deformation value does not exceed 5 cm in the range between 1 and 2 months

To define the state of the activity of the mud volcano, the area affected by deformation must be a significant area of the total surface of the volcano. The definition of the activity is an important aspect of the monitoring of hazard because there are many natural phenomena that can produce deformation in a small area (like a gryphon) or in a larger portion (like the sedimentation occurring at the border of the mud volcano). The significant area affected by deformation must be correctly defined by an operator (public agency, university, institute of research or Dipartimento della Protezione Civile) which really knows the natural phenomena that can occur on the Santa Barbara mud volcano.

The methodological approach is valuable and efficient from the point of view of quantity and quality of data collected in relation to the work and time spent. This monitoring technique is a useful tool to detect the early unrest phase of the mud volcano usually induced by changes in pressure and volume of fluid rising from the stagnation chamber.

The pre-eruptive deformation consists of a marked uplift and occasional small subsidence which are probably related to the redistribution of the subsoil of the pressurized fluids (Antonelli et al., 2014). According to Antonelli et al. (2014), soil uplift can occur up to a year before the eruption.

In hazard management, the SfM technique (Gomez and Purdie, 2016; Kaab, 2000; Fugazza et al., 2018; Giordan et al., 2017, 2018) is starting to be widely used by the scientific community. In the monitoring of potentially dangerous active sites, the UAVs are very advantageous because they are not used only as a support in the post-disaster events (Rokhmana and Andaru, 2017; Hisbaron et al., 2018) but also for the pre-event monitoring.

According to Kopf (2002), Antonielli et al. (2014), Madonia et al. (2011), INGV (2008), and Regione Siciliana (2008), the deformations of the surface shell of mud volcanoes can occur up to 1 year before the paroxysmal event with doming and the development of structural lineaments with an order of magnitude from centimeters to decimeters.

The results allow us to define the criteria for monitoring and analyzing the study area. For the mud volcano of Santa Barbara, the monitoring criteria are as follows:

Monitoring interval is between 1 and 2 months.

The frequency of the campaigns depends on the status of activity, while the other criteria depend on the object/structure of the monitoring.

These criteria allow us to detect events with deformation of at least 2 cm. In the case of anomalous values detected, the monitoring campaigns must be improved. This involves extensive monitoring, such as (i) developing time series localized in key areas and (ii) combining different methodologies, e.g., micro-seismicity monitoring and 3D geophysical prospecting (Imposa et al., 2016), to improve the monitoring system of the active geological process.