# GPR microwave tomography for diagnostic analysis of archaeological sitesthe case of a highway construction in Pontecagnano Southern Italy.

код для вставкиСкачатьArchaeological Prospection Archaeol. Prospect. 16, 203–217 (2009) Published online 29 July 2009 in Wiley InterScience (www.interscience.wiley.com) DOI: 10.1002/arp.362 GPR MicrowaveTomography for Diagnostic Analysis of Archaeological Sites: the Case of a Highway Construction in Pontecagnano (Southern Italy) R. CASTALDO1, L. CROCCO2, M. FEDI1, B. GAROFALO1, R. PERSICO3, A. ROSSI4 AND F. SOLDOVIERI2* 1 Dipartimento di Scienze DellaTerra, Universita' ‘Federico II’di Napoli, Largo San Marcellino 10, 80138 Napoli, Italy 2 Istituto per il Rilevamento Elettromagnetico dell’Ambiente, Consiglio Nazionale delle Ricerche,Via Diocleziano 328, 80124, Napoli, Italy 3 Istituto per i Beni Archeologici e Monumentali, Consiglio Nazionale delle Ricerche, Prov.le Lecce-Monteroni, - 73100 Lecce, Italy 4 Dipartimento di Beni Culturali, Universita' degli Studi di Salerno,Via Ponte Don Melillo, 84080, Fisciano (SA), Italy ABSTRACT Interpretation of ground-penetrating radar (GPR) data usually involves data processing similar to that used for seismic data analysis, including also migration techniques. Alternatively, in the past few years, microwave tomographic approaches exploiting more accurate models of the electromagnetic scattering have gained interest, owing to their capability of providing accurate results and stable images. Within this framework, this paper deals with the application of a microwave tomography approach, based on the Born Approximation and working in the frequency domain. The case study is a survey performed during the realization ofthe thirdlane ofthe most important highway in southern Italy (the Salerno-Reggio Calabria, near Pontecagnano, Italy). It is shown that such an inversion approach produces well-focused images, from which buried structures can be more easily identified by comparison to traditional radar images. Moreover, the visualization of the reconstruction results is further enhanced through a three-dimensional volumetric rendering of the surveyed region, simply achieved by staggering the reconstructed GPR two-dimensional profiles. By means of this rendering it is possible to follow the spatial continuity of the buried structures in the subsurface thus obtaining a very effective geometrical characterization. The results are very useful in our case where, due to important civil engineering works, a fast diagnosis of the archaeological situation was needed. The quality of our GPR data modelling was confirmed by a test excavation, where a corner of a building and the eastern part of another house, with its courtyard, were found at the depth and horizontal position suggested by our interpretation. Copyright # 2009 John Wiley & Sons, Ltd. Key words: Archaeologicalprospecting; microwavetomography; inverse scattering; three-dimensional target rendering; ground-penetrating radar * Correspondence to: F. Soldovieri, Istituto per il Rilevamento Elettromagnetico dell’Ambiente, Consiglio Nazionale delle Ricerche, Via Diocleziano 328, 80124, Napoli, Italy. E-mail: soldovieri.f@irea.cnr.it Copyright # 2009 John Wiley & Sons, Ltd. Received 8 April 2009 Accepted 8 June 2009 204 Introduction Ground-penetrating radar (GPR) is one of the most widely adopted tools for monitoring shallow subsurface structures, as it allows a non-invasive diagnosis of the investigated domain in a fast and simple way. As such, it represents a convenient technology for archaeological prospecting applications. Ground-penetrating radar works by emitting a modulated electromagnetic pulse into the ground and by recording the strength of the echo, produced by the interaction between the impinging waves and the buried objects, received at the air–ground interface, usually in a monostatic or bistatic configuration. In the former case the locations of the transmitting and the receiving antennae are coincident, whereas they are different in the latter case. By moving the antenna along a selected profile above the ground surface, a two-dimensional reflection profile (radargram) is obtained in which the delay time of the recorded echoes (that can be related to the depth of the underground reflectors) is drawn versus the antenna position (Daniels, 1996). In order to extract information from the radargram, the interpretation of GPR data usually takes advantage of migration techniques and thus exploits much of the standard processing used for reflection seismic data analysis (Stolt, 1978; Yilmaz, 2001). Although this kind of processing is widespread, a possible alternative processing strategy has gained an increasing interest in the past few years: the so-called microwave tomographic (MT) approach (Crocco and Soldovieri, 2003; Leone and Soldovieri, 2003; Catapano et al., 2006). The MT approaches rely on suitable models of the electromagnetic scattering that properly describe the interactions between the wave and the target. By so doing, these methods are in principle capable not only of detecting, locating and retrieving the shape of the buried objects, but also in providing information on their electromagnetic properties (dielectric permittivity and conductivity). To pursue this goal MT approaches tackle the GPR data processing as an inverse scattering problem, in which the measurement of the electromagnetic field backscattered by the buried objects are the data and the electromagnetic properties of the objects are the unknowns. In particular, the Copyright # 2009 John Wiley & Sons, Ltd. R. Castaldo et al. electromagnetic properties of the buried targets are searched for as ‘anomalies’ with respect to the background (ground). However, the solution of an inverse scattering problem is a very difficult issue that poses several mathematical challenges (Colton and Kress, 1992). This kind of problem is ill-conditioned and non-linear. The ill-conditioning arises from the limited information contained in the data, and this is further worsened by the finiteness of the number of GPR measurements and from the fact that the antennae do not encircle the scene to be investigated, due to the configuration of the measurement in reflection mode. Accordingly, the information that can be gained by the GPR measurements and then exploited in the inverse problem has a finite content, so that one should determine an optimal way to exploit it in order to achieve a reliable, accurate and stable image of the investigated scene. This is an open question that can be rigorously answered only in particular situations. The second difficulty is related to the non-linearity of the inverse scattering problem, which entails the existence of several solutions that correspond (within the considered accuracy) to the same dataset. As a consequence, it may happen that a solution completely different from the ‘ground truth’ (a false solution) is achieved from the inverse problem. Since one cannot a priori appraise the quality of the obtained solution, this circumstance means that the results of the overall diagnostics procedure may be completely wrong. Of course, it is easy to work out that such an issue becomes even more relevant when the available information is reduced (owing to the aforementioned constraints). Due to these mathematical difficulties, the validity of MT approaches for GPR imaging based on exact models of the electromagnetic scattering has been proved only in simulated cases or under controlled situations (Catapano et al., 2006) and they do not seem at this stage viable in practical conditions. Nevertheless, it is possible to take advantage of some of the features of MT methods by considering reconstruction algorithms based on simplified models. This is for instance the case of the Born Approximation (BA) or the Kirchhoff Approximation (KA) (Chew, 1995), which overcome the nonlinearity Archaeol. Prospect. 16, 203–217 (2009) DOI: 10.1002/arp GPR Microwave Tomography for Diagnostic Analysis of the inverse problem thus allowing remarkable simplification. Approaches of this kind have been widely adopted in GPR surveys for different application frameworks such as the civil engineering and sub-services detection and mapping (e.g. Soldovieri et al., 2007; Pettinelli et al., 2009) and the obtained results confirm the capabilities of these methods as compared to traditional ones. This paper aims at indicating how simplified MT methods are suitable for GPR in archaeological prospecting. In particular, the inverse scattering approach we consider here is based on the BA in the frequency domain. The BA leads to a very effective solution algorithm able to deal with large investigation domains (in terms of the radiated wavelength) in a quasi-real time and thus is well suited for archaeological purposes. The adoption of mathematical tools that are specifically designed to handle linear inverse problems (Bertero and Boccacci, 1998), makes it possible to mitigate the effect of noise on data and uncertainties on the investigated scenario. We applied the BA based inversion method on data collected during a GPR survey performed at an archaeological site in Pontecagnano, in southern Italy. The aim of this archaeological survey was to infer the presence of buried remains near to an excavated zone within which ancient artefacts and walls dating to before the Pre-Roman period were discovered. The GPR investigation aimed at establishing whether remains of archaeological interest exist within the area as it lies on the proposed third lane of the most important highway in southern Italy (the Salerno-Reggio Calabria). By referring to the real data collected in such a GPR survey, we first present in detail the results achieved by the microwave tomographic approach (in two-dimensional geometry). The achieved results show that the tomographic inversion algorithm provides well-focused reconstructions. The importance of adopting suitable parameters in the regularization of the linear inverse problem has been investigated allowing us to improve the reconstruction, both in terms of spatial resolution and depth of investigation. Results of the overall survey are visualized as parallel fence diagrams obtained by staggering the two-dimensional tomographic reconstructions (GPR profiles). This simple post-processing Copyright # 2009 John Wiley & Sons, Ltd. 205 makes the archaeological interpretation of the tomographic reconstruction results easier to be interpreted, and proved to be consistent with the outcomes of subsequent excavation. Archaeological site and measurement survey The investigated settlement in Pontecagnano was at the height of its commercial and cultural development in the –ninth to third centuries BC, when it was part of the territory of the Etruscan and then of the Sannita people. Between the seventh and fourth centuries BC, the settlement was divided into two areas: the western one where the civil, administrative and religious structures were present and the eastern one comprising the commercial and public sites (Strøm, 1993). The seventh century BC was a phase of great expansion, but after the foundation of the Greek city of Paestum-Posidonia, Pontecagnano started to lose its commercial relevance. In 268 BC the Romans founded Picentia on the ruins of Pontecagnano (Giglio, 2001). At the Pontecagnano site, excavations were carried out as preliminary operations during the construction of the third lane of the SalernoReggio Calabria motorway. These excavations revealed ancient artefacts and wall structures dating before the Pre-Roman period (sixth to fourth centuries BC) (Cinquantaquattro, 2000). A geophysical survey was subsequently performed to provide indirect evidence of how the buried remains were linked to the excavated ones (Fedi et al., 2008). The occurrence of so many ancient remains in the area created the necessity to investigate the non-excavated areas thereby avoiding possible destruction of buried remains of archaeological interest. Both excavations and the GPR survey were performed, in the framework of a collaboration between the archaeologists of the University of Salerno and the group of the ‘Mobile Laboratory of Archaeo-geophysics’ of the Regional Centre of Competence INNOVA. The GPR survey was carried out with the SIR3000 system and antennae of 400 MHz and 600 MHz centre frequency (in air), with the two antennae operated in a continuous mode. The Archaeol. Prospect. 16, 203–217 (2009) DOI: 10.1002/arp R. Castaldo et al. 206 Table 1. The parameters adopted in the GPR survey Mode Samples Rate (scan s1) Gain (dB) Range (ns) Time 512 64 13, 13, 39, 39, 40 50 acquisition parameters are reported in Table 1. The shallow layers of the ground were investigated to a maximum depth of investigation of a few metres. The material is composed of pyroclastic material interspersed with limestone, where an electromagnetic wave velocity range of 9– 11 cm ns1 was estimated by hyperbola fitting, corresponding to a relative dielectric permittivity around 10. The velocity estimate is in agreement with those found in the literature for similar materials (Gomez-Ortiz et al., 2005). The survey was performed close to the excavated zone and consisted of 11 profiles of varying lengths spaced 50 cm apart (Figure 1). Figure 2 depicts the structures discovered in Trench 29. The microwave tomographic approach The MT approach adopted in the data processing simplifies the problem as linear inverse scattering by means of the Born Approximation (Chew, 1995; Leone and Soldovieri, 2003; Crocco and Soldovieri, 2003). The approach is described in several papers (e.g. Leone and Soldovieri, 2003; Persico et al., 2005). Here, we briefly outline the main features of the approach. The approach is formulated with respect to the two-dimensional geometry that is depicted in Figure 3. The inhomogeneous background scenario is modelled as two homogeneous half-spaces separated by a planar interface at z ¼ 0. The upper half-space is the air (dielectric permittivity e0), whereas the lower one represents the ground, having a relative dielectric permittivity eb and an electrical conductivity sb. The incident field source is simply modelled as a time-harmonic (time dependence exp(j2pft)) filamentary y-directed electric current, invariant along the y axis. The data are collected under a multifrequency Figure 1. The location ofthe GPR profiles and the excavated trenches.This figure is available in colouronline at www.interscience. wiley.com/journal/arp Copyright # 2009 John Wiley & Sons, Ltd. Archaeol. Prospect. 16, 203–217 (2009) DOI: 10.1002/arp GPR Microwave Tomography for Diagnostic Analysis 207 Figure 2. The remains discovered at theTrench 29.This figure is available in colouronline at www.interscience.wiley.com/journal/arp multimonostatic configuration in the band [fmin, fmax], where the transmitting and receiving antennae are placed at the same point. The targets to be imaged are assumed to be invariant along the y axis and their cross-section is enclosed in the rectangular investigation domain. In particular, the unknown relative dielectric permittivity profile and the conductivity profile are assumed to be located inside the investigation domain D and represent a variation with respect to the characteristics of the background scenario (in this case the hosting ground of parameters eb and sb). Accordingly, the inverse problem is recast in terms of the contrast function, which embeds such an ‘alteration’ from the background and is defined as: xðx0 ; z0 Þ ¼ "eq ðx0 ; z0 Þ "eqb ; "eqb (1) with sb and "eq ðx0 ; z0 Þ ¼ "0 "r ðx0 ; z0 Þ 2pf sðx0 ; z0 Þ j 2pf "eqb ¼ "0 "b j being the equivalent complex dielectric permittivity of the targets and of the ground, respectively. Notice that x ¼ 0 when (x0 ,z0 ) does not belong to the target. Under the Born Approximation, the relationship in the frequency domain between the unknown contrast function and the scattered field data is provided by a linear integral equation of Fredholm type (first kind): Z 2 Es ðxs ; vÞ ¼ ks Ge ðxs ; v; ~ r0 ÞEinc ðxs ; v; ~ r0 Þxð~ r0 Þd~ r0 ; Figure 3. Geometry of the Inverse problem. This figure is available in colour online at www.interscience.wiley.com/ journal/arp Copyright # 2009 John Wiley & Sons, Ltd. D (2) Archaeol. Prospect. 16, 203–217 (2009) DOI: 10.1002/arp R. Castaldo et al. 208 where v ¼ 2pf is the angular frequency, Es(xs,v) is the scattered electric field along the air–ground interface collected at the abscissa xs and frequency f and ks is the wave-number in the soil. By definition, the scattered field Es is given by the ‘difference’ between the total field and the unperturbed field. The total field is the sum of two contributions: the field reflected by the soil and the field backscattered by the buried objects. The unperturbed field coincides with the field reflected by the soil when the objects are absent, and therefore accounts for the reflection/transmission at the air–ground interface. Under the measurement configuration at hand, the scattered field is collected over a rectilinear observation domain at the air–soil interface with xs ranging from –xM to xM (see Figure 3). Ge ðxs ; v; ~ r0 Þ is the Green’s function of the problem, while Einc is the unperturbed or incident field in the investigation domain D. Both these quantities are known after defining the sources, the reference scenario and the measurement configuration. The numerical implementation of the solution algorithm requires the discretization of Equation (2). This task is pursued by resorting to the method of moments (MoM) (Harrington, 1961). In particular, the linear integral relationship in Equation (2) is discretized into a linear algebraic system, where the unknowns are the expansion coefficients of the contrast function along the chosen functional basis and point-matching is adopted in the data space (Crocco and Soldovieri, 2003; Persico et al., 2005). The inversion of the resulting matrix L is performed by a scheme based on the truncated singular value decomposition (TSVD) (Bertero and Boccacci, 1998), which provides robust solutions with respect to the uncertainties and the noise on data. This ‘regularized solution’ is given by: ~ ð~ x rÞ ¼ N X 1 hEs ; vn iun s n¼0 n (3) where fs n ; un ; vn gKn¼0 is the singular system of the matrix L, h; i denotes the scalar product in the data space; K denotes the minimum between the number of measurements and the pre-chosen number of expansion coefficients for the discretization of the problem, and N K. Copyright # 2009 John Wiley & Sons, Ltd. In particular, sn are singular values, while vn and un are the singular vectors in the data and unknown spaces respectively (Bertero and Boccacci, 1998). The choice of the N index is performed to ensure the trade-off between the contrasting needs of accuracy and resolution (that tends to increase such an index) and the stability of the solution (that limits the N index). Setting the tomographic approach to tackle the experimental data The description of the BA tomographic approach given above is referred to the general case of subsurface sensing tackled under a reflection measurement configuration. This section will detail the steps that have been necessary to apply the method in practice to the GPR data gathered during the archaeological survey carried out at Pontecagnano. Preprocessing of the measured data The aim of this preprocessing stage is to ‘translate’ the time-domain GPR measurements to the frequency-domain scattered field data suitable for tomographic inversion. This has been done for each of the GPR profiles collected and has been accomplished by means of the following procedure. The first step consists of time zero correction and background removal. A zero time correction is performed to correct for the effects related to the propagation path of the signal within the cables and the antennae of the GPR. As a result, the corrected signal takes into account only the propagation of the electromagnetic signal from the transmitting antenna to the structure and then to the receiving antenna. Background removal performs the reduction of the spatially constant signal in the GPR profile related to the direct coupling and the reflection from the air– ground interface. Within a fixed-length window (for instance 100 traces) the average signal is then subtracted from the current trace. Figure 4 depicts four of eleven profiles (25 m in length) of GPR survey before and after the zero time correction and background removal. Archaeol. Prospect. 16, 203–217 (2009) DOI: 10.1002/arp GPR Microwave Tomography for Diagnostic Analysis 209 Figure 4. Raw and time-domain processed realdata: zero time and backgroundremoval.This figure is available in colouronline at www.interscience.wiley.com/journal/arp The second step of the preprocessing consists in the Fourier transform of the frequency domain within which the inversion algorithm works. Figure 5 depicts the frequency spectrum of the 400 and 600 MHz antennae for the same spatial point. From this figure we can first note that, as usually occurs when the antenna radiates in the soil, a lowering of the peak frequency (with respect to the nominal frequency in air) arises. In addition, we observe that the working frequency band is larger for the 400 MHz antenna compared with the 600 MHz antenna; this spectral beha- Copyright # 2009 John Wiley & Sons, Ltd. viour has dictated the choice of the different frequency bands exploited in the tomographic inversion presented in the section below. Single profile reconstruction The MT approach has been applied separately to each of the preprocessed GPR profiles. Here, we present some reconstruction results in terms of tomographic images of the GPR profiles for both of the antennae deployed in the survey. Archaeol. Prospect. 16, 203–217 (2009) DOI: 10.1002/arp 210 R. Castaldo et al. Figure 4. (Continued). 400 MHz antenna The parameters adopted in the inversion are reported in Table 2. The main computational effort required in performing the tomographic approach consists in the evaluation of the SVD of the matrix L (given by the discretization of the integral equation in Equation (2)). In particular, such an effort increases with the size of the investigated domain, so that the imaging of the whole region underlying a GPR profile can be exceedingly time consuming. Therefore, in order to keep the required computational burden as low as possible, we have tackled the inversion of the data as a sequence of inversions by sub- Copyright # 2009 John Wiley & Sons, Ltd. dividing the GPR profile into 24 (or 16) 1-m subprofiles, each one providing the tomographic image of the domain beneath it. When moving along the GPR profile, the domain is shifted accordingly and the corresponding segment of measurements provides the input for the inversion algorithm. By assuming that the background characteristics do not change along the profile, we can tackle each subprofile by means of the same matrix L (and its SVD), which for the case at hand is computed with respect to a rectangular investigation domain D having a horizontal extent of 1 m and a depth ranging from 0 and 1.5 m. For the inversion, the contrast function in Archaeol. Prospect. 16, 203–217 (2009) DOI: 10.1002/arp GPR Microwave Tomography for Diagnostic Analysis 211 Figure 5. Frequency spectrum of the antenna: 400 MHz (A) and 600 MHz (B). This figure is available in colour online at www. interscience.wiley.com/journal/arp Equation (1) has been expanded within the investigation domain D and built up of 33 Fourier harmonics along the horizontal direction and 33 vertical steps (Leone and Soldovieri, 2003). Finally, the overall tomographic reconstruction corresponding to the GPR profiles is obtained by juxtaposing the ‘partial’ reconstructions obtained by moving along the entire length of the profile. In TSVD inversion (see Equation (3)), the choice of the truncation index N has been performed by observing the singular values behaviour of the appropriate matrix (see Figure 6). In particular, we chose the index N ¼ 300 that represents the ‘boundary’ between the two different behaviours of the singular value curve: in fact, for lower values of the index, the singular values have a smooth decay. This procedure allows us to exploit all the main singular value information in TSVD, which improves the accuracy of the solution. For index values higher than this ‘boundary index’, the decay of the singular values becomes faster and this entails that any improvement of the accuracy (increase of the index N) will result in a reduction of the stability. It is worth noting that the computational cost needed to the offline filling of the matrix requires about four hours on a standard PC and the evaluation of its SVD requires about five minutes. Table 2. Parameters of the inversion for the 400 MHz antenna Model relative dielectric permittivity "b Model conductivity s b Effective frequency band Frequency step Df Measurement domain Spatial step Dx Investigation domain 10 0.005 S m1 100^604 MHz 28 MHz 1m 0.02 m 1m [0^1.5] m Copyright # 2009 John Wiley & Sons, Ltd. Figure 6. Singular values behaviour for the case of 400 MHz antenna. This figure is available in colour online at www. interscience.wiley.com/journal/arp Archaeol. Prospect. 16, 203–217 (2009) DOI: 10.1002/arp 212 Afterwards, the inversion itself and is performed as multiplication of vectors and requires only a few seconds. The images are given in terms of the modulus of the contrast function, normalized with respect to the maximum achieved throughout the profile. For the sake of interpretation, the regions where the modulus of the contrast function is significantly different from zero correspond to an estimate of the position and geometry of the buried anomalies. Figure 7 describes the tomographic reconstructions of the four GPR profiles of the survey for which time domain data have been presented in the Figure 4. As can be seen, the tomographic processing allows for focalizing various structures, which are persistent across the different profiles. The main anomalies are localized between 5 and 10 m. In particular, we see a sharp contrast in all four profiles around x ¼ 8 m and another one R. Castaldo et al. around x ¼ 13 m. Their depth is approximately 0.9 m, which is about 0.2 m deeper than expected from nearby excavations. This difference is possibly due to the fact that the (constant) electromagnetic velocity assumed for the background (the half-space model depicted in Figure 3) is slightly different from the actual one. 600 MHz antenna The same analysis has been performed with the data collected by the 600 MHz antenna. The parameters adopted in the inversion are reported in the Table 3. In this case, we have chosen a smaller working frequency band compared with that for the 400 MHz data. This choice arises from the investigation of the spectral behaviour of the data collected by the two antennae, as presented earlier in Figure 5. Figure 7. The tomographic reconstruction of the first four GPR profiles with the 400 MHz antenna.This figure is available in colour online at www.interscience.wiley.com/journal/arp Copyright # 2009 John Wiley & Sons, Ltd. Archaeol. Prospect. 16, 203–217 (2009) DOI: 10.1002/arp GPR Microwave Tomography for Diagnostic Analysis 213 Table 3. Parameters of the inversion for the 600 MHz antenna Model relative dielectric permittivity "b Model conductivity s b Effective frequency band Frequency step Df Measurement domain Spatial step Dx Investigation domain 10 0.005 S m1 100^520 MHz 30 MHz 1m 0.02 m 1m [0^1.5] m We present for comparison the first four GPR profiles reconstructed by the tomographic approach. For the TSVD inversion, we have retained the first 205 singular values (see Figure 8 for the singular values behavior). By examining these reconstruction results (Figure 9), we can gain information about the behaviour of the anomalies. In these reconstructed images we observe two main anomalies at depths of 0.65 and 0.9 m, especially in Figure 8. Singular values behaviour for the case of 600 MHz antenna. This figure is available in colour online at www. interscience.wiley.com/journal/arp Figure 9. The tomographic reconstruction of the first four GPR profiles with the 600 MHz antenna.This figure is available in colour online at www.interscience.wiley.com/journal/arp Copyright # 2009 John Wiley & Sons, Ltd. Archaeol. Prospect. 16, 203–217 (2009) DOI: 10.1002/arp 214 R. Castaldo et al. Figure 10. Pseudo-three-dimensionalvisualizationofthereconstructionby 400 MHzdata.Thisfigureisavailableincolouronlineat www.interscience.wiley.com/journal/arp Figure 11. Pseudo-three-dimensionalvisualizationofthereconstructionby 600 MHzdata.Thisfigureisavailablein colouronlineat www.interscience.wiley.com/journal/arp Copyright # 2009 John Wiley & Sons, Ltd. Archaeol. Prospect. 16, 203–217 (2009) DOI: 10.1002/arp GPR Microwave Tomography for Diagnostic Analysis 215 Figure 12. Test excavation: verification of the results. This figure is available in colour online at www.interscience.wiley.com/ journal/arp the regions with x ranging from 5 m to 10 m and with x ranging from 20 m to the end of the profiles. As a possible interpretation, these two anomalies can be associated to the upper and lower sides of the same structure. If we compare these reconstructions with those of the 400 MHz antenna, we can see that the 400 MHz antenna is able to resolve only a significant anomaly at the rough depth of 0.9 m. Interpretation In this section we present the tomographic reconstruction results in terms of a pseudo-three-dimensional visualization of the subsurface by simply staggering the obtained two-dimensional after a suitable normalization (Catapano et al., 2006; Solimene et al., 2007). The first result concerns measurements collected using the 400 MHz antenna. Figure 10 shows the three-dimensional representation. We are confident that the strong anomalies at x ¼ 6– 7 m and x ¼ 15 m, spanning spatially both x and y directions, are representative of walls. The strongest anomaly around x ¼ 8–10 m is in our interpretation associated with a road, which is Copyright # 2009 John Wiley & Sons, Ltd. the continuation of a road visible in the excavated trenches. In addition, the other anomalies that are detected give indications of the foundations of some structures. The anomalies detected by GPR correspond to two structures, built by small limestone blocks, directed NE–SW, crossed by a road, 4.50 m in width, that take advantage of a slight morphological dip of the travertine (Rossi, 2005). The second interpretation concerns the measurements collected by the 600 MHz antenna. Figure 11 shows a pseudo-three-dimensional representation: the main anomalies already pointed out for the 400 MHz antenna are evident also for the 600 MHz antenna reconstruction. We have again evidence of a significant anomaly between x ¼ 8 and x ¼ 10 m, associated with a buried road. In addition, more anomalies are detected and they can be associated to the foundations of some structure. Finally, we point out also an anomaly in the zone at x ¼ 21–24 m, which was not detected in the 400 MHz antenna results. Conclusions In this paper we have described the application of a microwave tomography approach to the Archaeol. Prospect. 16, 203–217 (2009) DOI: 10.1002/arp R. Castaldo et al. 216 processing of the GPR data gathered at the archaeological site of Pontecagnano. The inverse scattering approach exploits a consistent, although simplified, model of the electromagnetic scattering and allows well-focused images from which features of the buried structures can be properly inferred. The interpretation is further supported by means of a simple three-dimensional rendering made by staggering the reconstructed GPR profiles, so providing an overall visualization of the investigated site. The results of GPR survey, elaborated by the inverse scattering approach, have been tested by excavation 5 m northeast of Trench 28 (Figure 12), where a corner of the building identified by our GPR data modelling was found. This information allows the reconstructing the plan of the ancient house found inside this trench (Santoriello and Rossi, 2005). In the same way, on the east side of Trench 29 archaeological excavation confirmed the GPR survey by locating the eastern part of another house with its courtyard. With regard to the computational cost of the inverse scattering algorithm, the offline construction of each matrix required about 5 hours on a PC and the calculation of its SVD required about 5 minutes, although the inversion in itself is a real-time procedure. As a final comment, it can be said that the inverse scattering technique, with respect to the standard GPR procedure of processing, implicitly contains a focalization procedure and allows us to adopt regularization schemes that make the inversion potentially more robust against noise and parametric uncertainties. Therefore, this technique appears to be extremely significant in supporting the experience of the human operator, since it has a definite advantage when creating reliable and more interpretable images. As a future development, we will conduct a comparison between the microwave tomography approach and the state of art traditional processing. On the other hand, it has to be said that, due the complexity of the scenario usually dealt with, one cannot completely rely on a single processing approach. Accordingly, multisensor and multiprocessing approaches can be exploited in order to further help the operator through the comparison of the results arising from different techniques. Copyright # 2009 John Wiley & Sons, Ltd. References Bertero M, Boccacci P. 1998. Introduction to Inverse Problems in Imaging. Institute of Physics: Bristol. Catapano I, Crocco L, Persico R, Pieraccini M, Soldovieri F. 2006. Linear and nonlinear microwave tomography approaches for subsurface prospecting: validation on real data. Antennas and Wireless Propagation Letters 5: 49–53. Chew WC. 1995. Waves and Fields in Inhomogeneous Media. Institute of Electrical and Electronica1 Engineers: Piscataway, NJ. Cinquantaquattro T. 2000. Pontecagnano (SA): saggi stratigrafici nell’abitato antico. Bollettino di Archeologia 28–30: 121–171. Colton D, Kress R. 1992. Inverse Acoustic and Electromagnetic Scattering Theory. Springer-Verlag: Berlin. Crocco L, Soldovieri F. 2003. GPR prospecting in a layered medium via microwave tomography. Annals of Geophysics 46: 559–572. Daniels DJ. 1996. Surface Penetrating Radar. The Institution of Electrical Engineers: London. Fedi M, Florio G, Garofalo B, La Manna M, Pellegrino C, Rossi A, Soldovieri MG. 2008. Integrated geophysical survey to recognize ancient Picentia’s buried walls, in the archaeological park of Pontecagnano-Faiano (southern Italy). Annals of Geophysics 51(5–6): 867–875. Giglio M. 2001. Picentia, fondazione romana? Quaderni degli Annali dell’Instituto Orientale di Napoli (AION), 8: 119–131. Gómez-Ortiz D, Martı́n-Velázquez S, Martı́n-Crespo T, Márquez A, Lillo J, López I, Carreño F. 2005. Characterization of volcanic materials using ground penetrating radar: a case study at Teide volcano (Canary Islands, Spain). Journal of Applied Geophysics 59: 63–78. Harrington RF. 1961. Time-Harmonic Electromagnetic Fields. McGraw Hill. Leone G, Soldovieri F. 2003. Analysis of the distorted Born approximation for subsurface reconstruction: truncation and uncertainties effects. IEEE Transactions Geosciences Remote Sensing 41: 66–74. Persico R, Bernini R, Soldovieri F. 2005. On the configuration of the measurements in inverse scattering from buried objects under the distorted Born approximation. IEEE Transactions on Antennas and Propagation 53: 1875–1886. Pettinelli E, Di Matteo A, Mattei E, Crocco L, Soldovieri F, Redman JD, Annan AP. 2009. GPR response from buried pipes: measurement on field site and tomographic reconstructions. IEEE Transactions on Geoscience and Remote Sensing DOI: 10.1109/TGRS-2009-2018301 Rossi A. 2005. Contesto ambientale e dinamiche insediative tra l’ età del Ferro e l’età Arcaica. Annali Istituto Orientale di Napoli – sezione Archeologia e Storia Antica (N.S.) 11–12: 225–234. Archaeol. Prospect. 16, 203–217 (2009) DOI: 10.1002/arp GPR Microwave Tomography for Diagnostic Analysis Santoriello A, Rossi A. 2005. Aspetti e problemi sulle divisioni agrarie nella piana di Pontecagnano (Salerno): una prima riflessione. Annali Istituto Orientale di Napoli - sezione Archeologia e Storia Antica (N.S.) 11–12: 245–257. Soldovieri F, Hugenschmidt J, Persico R, Leone G. 2007. A linear inverse scattering algorithm for realistic GPR applications. Near Surface Geophysics 5: 29–42. Solimene R, Soldovieri F, Prisco G, Pierri R. 2007. Three-dimensional microwave tomography by a Copyright # 2009 John Wiley & Sons, Ltd. 217 2-D slice-based reconstruction algorithm. IEEE Geoscience and Remote Sensing Letters 4: 556– 560. Stolt RH. 1978. Migration by Fourier transform. Geophysiscs 43: 23–48. Strøm I. 1993. Pontecagnano-Picentia. A Hellenistic town in the former Etruscan Campania, The Danish Excavations. Acta Hyperborea 5: 107– 126. Yilmaz Ö. 2001. Seismic Data Analysis. Society of Exploration Geophysicists (SEG): Tulsa, OK. Archaeol. Prospect. 16, 203–217 (2009) DOI: 10.1002/arp

1/--страниц