The lower Pliocene fossiliferous layer where the fossil sirenian was found is exposed in a small area near Arcille (Grosseto, Italy) (Fig. 1a). The succession outcropping in this area consists of shallow marine siliciclastic deposits. They contain a planktonic foraminiferal assemblage consistent with their attribution to the lower part of the Zanclean, in particular to the MPl2 zone of Cita (1975) dated between 5.08 and 4.52 Ma (age after Lourens et al., 2004). The succession, affected by systems of normal faults, is dominated by yellowish, locally pebbly, fossiliferous sandstone overlain by greyish mudstone (Tinelli et al., 2011). In the geological map, the sediments where sirenians were discovered are reported as “Argille, argille sabbiose e sabbie marine con livelli conglomeratici presenti alla base. Pliocene Inferiore”, without any formal definition of the formation (Carobbi et al., 1996).

Three partially articulated sirenian fossil skeletons were found in 2007 in a sand quarry located in this area and they are now kept in the museum of the Gruppo Avis Mineralogia e Paleontologia di Scandicci (GAMPS) near Florence. Their catalogue numbers are GAMPS 62M, GAMPS 63M and GAMPS 64M. All specimens belong to Metaxytherium subapenninum, a halitheriine dugongid (Mammalia: Sirenia) that lived in the Mediterranean Basin and became extinct in the upper part of the Pliocene because of the progressive climatic cooling that occurred after 3.1 Ma (Sorbi et al., 2012 and Sorbi et al., 2008).

The discovery of these three sirenian skeletons is a result of an excavation in the quarry, during which fossil bones were exposed. Unfortunately this activity also caused the partial destruction of the fossils. Two of them (GAMPS 62M and GAMPS 63M) were found in the upper part of the succession, in the proximity of two shell beds (sb1 and sb2, Fig. 1b), including fragmented and decalcified shells of bivalves, rare gastropods and scaphopods. GAMPS 62M consists of a partial skeleton composed of an incomplete skull and mandible, some teeth, several vertebrae, ribs and sternum; GAMPS 63M is more fragmentary and is composed of teeth, ribs and vertebrae. The third specimen (GAMPS 64M), represented by few ribs and vertebrae, came from sandstone beds whose correlation with the succession containing the other sirenians is unclear. The fourth specimen, which is the main object of this study, also belonging to M. subapenninum, was discovered in 2010 in a sunflower field a few ten meters north-west of the quarry (Fig. 1c) in the upper part of the succession near sb2 (Fig. 1b). During palaeontological prospecting, we observed several fossil bones (mostly fragmented ribs) on the field surface particularly concentrated in a small area near a draw. Considering the morphology of the field, we presumed that these bones had not suffered any main displacement but were brought to the surface during the plowing of the field. Our hypothesis was supported by the previous discovery of the three sirenian specimens in the quarry: all of them were represented by more or less complete single skeletons, while no accumulation of bones from different individuals or isolated sirenian bones were collected.

Therefore we thought that this small area of the sunflower field was particularly suitable to test the GPR technique. This fourth specimen is now kept in the Museo di Storia Naturale e del Territorio, Università di Pisa (MSNTUP) and its catalogue number is MSNTUP I15892.

Because Ground Penetrating Radar is based on the propagation and reflection of electromagnetic (EM) waves, it is sensitive to variations of the EM parameters in the subsoil, especially the dielectric constant and electric conductivity (Davis and Annan, 1989). The lower the frequency of EM waves propagating into the subsurface, the greater their penetration. The latter varies from a few meters in conductive materials to tens of metres for low conductivity media (Annan, 2009, Davis and Annan, 1989 and Smith and Jol, 1995). The capability to resolve targets vertically (vertical resolution) increases with the antenna frequency up to centimetre values (> 200 MHz), while it is strongly reduced (several decimetres) when a < 100 MHz antenna is adopted. Lateral resolution depends also on the geometry of acquisition (step size, e.g. the distance between each point where a measurement is made along a GPR profile) and can reach a sub-centimetric resolution. The Nyquist sampling interval, e.g. one-quarter of the wavelength in the ground, is the base value to which the step size of the acquisition refers in order to avoid spatial aliasing effects (Annan, 2009 and Davis and Annan, 1989). Despite its relatively low penetration depth (especially in conductive materials), GPR high resolution (lateral and vertical) makes this technique successful in studies of shallow stratigraphy, structural geology and archaeology (Basile et al., 2000, Bini et al., 2010, Grandejean and Gourry, 1996, Grasmueck, 1996, Grasmueck et al., 2004, Leckebusch, 2003, Leucci, 2006, Orlando, 2007 and Soldovieri and Orlando, 2009). Further significant advantages are reached by adopting a grid of radar profiles, allowing a pseudo-3D or full-3D visualisation of the subsurface, and facilitating the interpretation of geometric structures, such as joints patterns, geological contacts and archaeological remains (Malagodi et al., 1996 and Nuzzo et al., 2002). In this respect, “time slice” (or depth slice) maps are used to display the pattern of radar data at variable depths (Goodman et al., 1995).

Contrary to the bistatic collection mode, where transmitter and receiver antennas are placed side by side on the ground and one progressively increases the distance between them from a fixed point (called Common Mid Point), in the common offset collection mode both antennas are placed at fixed distance and moved at fixed increments along the survey line. The bistatic collection mode is performed with a bistatic antenna and it is useful for the determination of EM wave velocity. The common offset collection mode needs a monostatic antenna whereby transmitter and receiver maintain a constant reciprocal distance, suitable for mapping the underground reflection along the survey line. The transmitter and receiver of a monostatic antenna are commonly contained in a box no larger than a few decimetres, that is easily moved onto the investigated surface.

In this study, the GPR survey was performed using the Radar System device of IDS Company^© (www.ids-spa.it), equipped with a monostatic antenna of 200 MHz of nominal peak frequency and HH-polarised. The choice of this antenna frequency was determined by the need to investigate, at a resolution of some centimetres, a fossil likely located in the first 1.5 m of subsurface, or perhaps in a lower depth due to a steep inclination of some fossil bones.

An odometer wheel was used to control the distance of measurement stations of ground response to EM signal along the survey line (step size). A grid of orthogonal survey lines was made with a spacing of 0.2 m. The raw GPR data were processed following a standard procedure (described below), first tested on a single line and then applied to the rest of the data. Finally, the interpretation was based both on the most relevant reflective areas visible in the time slices at various depths and the intersecting vertical radargrams.

To compare the GPR map and the location of fossil remains, we constructed a grid system of squares (0.40 m × 0.40 m) where the relative position of each bones was recorded using an alphanumeric code for each square, as show in Fig. 7a.

The GPR prospecting undertaken near the out cropping of several fossil bones on the field surface indicated the presence of reflective zones in the subsurface at various depths. The excavation confirmed that these zones mostly corresponded to buried bones. In order to compare reflectors and fossil remains in the subsurface, during the excavation each fossil bone was mapped using a grid system. By comparing geophysical and palaeontological data, we observed the effectiveness of GPR for detecting fossil bones, with clear reflections corresponding to the skull, the mandibles, some cervical and dorsal vertebrae and ribs.

However, we noted that some reflections were not generated by fossil remains. These types of reflections are mostly related to a concretionary level in the silty sands or soil clumps locally associated with voids caused by agricultural activity. In other cases, fossil remains either did not reflect back electromagnetic energy, or the shape of the backscattered energy is not fully consistent with the shape of the fossil assemblage. Heterogeneous bone density and GPR vertical resolution may be the explanation of these missed or partial targets. In general, sirenians have heavy skeletons that help them to stay submerged. Sirenian bones are both swollen (pachyostotic) and dense (osteosclerotic), especially the ribs, which are often found as fossils (Domning, 2002). We observed that the central reflective zone (A) corresponds to a high concentration of ribs: therefore we inferred that due to their massive and dense structure, a GPR antenna could very easily detect the sirenian ribs. However, laboratory experiments to determine EM characteristics of different types of bones (mainly electric permittivity) are needed to support this hypothesis. Following the first experiments of Keller (1987) and the indications of Main and Hammon (2003), new tests should be performed directly on bones and sediments in order to constrain the contrasting EM proprieties and simulate different situations through synthetic diagrams.

Vertical GPR resolution (e.g. the minimum vertical distance between two objects in the subsurface to be distinguished with two separate reflections) is assumed to be 1/4 of the signal wavelength (λ). In our case, this means that bones vertically spaced of less than 0.12–0.13 m may be represented by a single reflection instead two distinct events in the radargram. Hence, the missed reflections of some undetected bones may have been convoluted in the reflective event of shallower objects. In order to overcome this problem, a higher frequency antenna should be used, provided that the fossil remains are assumed to be found at a very shallow depth (< 1.0–1.5 m), thus avoiding the loss of signal caused by the EM wave attenuation with depth. In this respect, we must consider the attenuation of GPR signal when a conductive material, e.g. clay and saline sediments, is crossed by EM waves. Indeed, the ease of electron movement typical of conductive material led to a decrease of the of initial energy because it is partly converted into heat. A multifrequency radar system might supply the better compromise between resolution and penetration depth.

Based on all the above considerations, we conclude that GPR surveys, applied to palaeontological field research, turned out to be a time-cost effective solution and a tool for the local planning. Despite the encouraging results, in order to refine this application in vertebrate palaeontology we deem that further experimentations need to be conducted in different geological and palaeontological contexts and with higher frequency antennae (> 200 MHz), especially when we presume that fossil remains lie at shallow depths. However, considering that geophysical technologies and techniques evolve continuously, it is clear that much more testing will have to be conducted to further improve the use of this methodology in vertebrate palaeontology.

Finally, the most strictly palaeontological result of this study is the discovery of one of the most complete Metaxytherium specimens ever found in the Mediterranean area. In fact, all the sirenian remains discovered in Area 1 and Area 2 surely belong to a single individual considering that all the identified unpaired bones are single (skull, mandible, atlas and axis) and paired bones were recognised (a scapula and a humerus). Moreover several bones (e.g. the ribs) are in relative anatomical position and the others are disarticulated but still closely associated (Fig. 7). The detailed systematic and taphonomic study of this specimen will sensibly increase knowledge about extinct sirenian anatomy, biology and behaviour. This new discovery also further improves the relevance of the Arcille area as an extraordinary deposit of marine vertebrates, also because, besides the four sirenian skeletons, fishes (sharks and teleosteans) and odontocete cetaceans have been recently collected in this area.