Advanced GPR imaging of sedimentary features: integrated attribute analysis applied to sand dunes

Advanced GPR imaging of sedimentary features: integrated attribute analysis applied to sand dunes Summary We evaluate the applicability and the effectiveness of integrated GPR attribute analysis to image the internal sedimentary features of the Piscinas Dunes, SW Sardinia, Italy. The main objective is to explore the limits of GPR techniques to study sediment-bodies geometry and to provide a non-invasive high-resolution characterization of the different subsurface domains of dune architecture. On such purpose, we exploit the high-quality Piscinas data-set to extract and test different attributes of the GPR trace. Composite displays of multi-attributes related to amplitude, frequency, similarity and textural features are displayed with overlays and RGB mixed models. A multi-attribute comparative analysis is used to characterize different radar facies to better understand the characteristics of internal reflection patterns. The results demonstrate that the proposed integrated GPR attribute analysis can provide enhanced information about the spatial distribution of sediment bodies, allowing an enhanced and more constrained data interpretation. Electromagnetic theory, Ground penetrating radar, Image processing INTRODUCTION Aeolian processes produce landscape that can be spectacular in specific regions, where the sand has apparently organized itself into structures, such as, for example, coastal dunes and other wind-related features (see e.g. Bagnold 1941; Lorenz & Zimbelman 2014). Accurate field observations are required to understand actual dune geomorphology, and to analyse sand layers that record the history of their deposition and migration. GPR is a fast, efficient and non-invasive imaging system to characterize shallow subsurface structures based on changes in the electromagnetic (EM) properties of the materials (Baker & Jol 2007; Cassidy 2009). Dunes are a suitable environment for the application of GPR techniques because they are usually made of sands that exhibit low electrical conductivity and low magnetic permeability allowing high depths of penetration (Bristow 1995; Bristow & Jol 2003; Bristow et al.2010). The successful application of GPR to dunes is due to the contrast in the relative dielectric properties, caused by moisture, porosity, grain size and mineral content (Van Dam & Schlager 2000; Bristow 2009). As GPR is non-destructive and can overcome the restrictions of limited outcrops, in the last 30 yr GPR has spread as a suitable technique to characterize shallow stratigraphy and deformation (see e.g. Jol et al.1996; Busby & Merritt 1999; Buynevich & Fitzgerald 2001; Jol & Bristow 2003) and characteristics of coastal barriers and of sand dunes (see e.g. Schenk et al.1993; Van Heteren et al.1996; Tercier et al.2000; Moore et al.2004), formed and shaped in modern and ancient fluvial, aeolian and shallow-marine sediment environments. The combination of GPR and traditional geomorphic and geological investigations allows detailed studies of sand dunes development and evolution (see e.g. Harari 1996; Bristow et al.2000a, 2005; Adetunji et al.2008; Gómez-Ortiz et al.2009; Santalla et al.2009; Clemmensen & Nielsen 2010; Guillemoteau et al.2012; Tillmann 2014). Coastal dunes build up where a number of conditions are met such as width of the beach sufficient to allowing the accumulation of wind-blown sand, prevailing winds from the sea and obstacles (e.g. vegetation and pebbles), which tend to slow down the wind and lead to the deposition of sand grains (Bigarella et al.1969; Goldsmith 1973; Hesp 1988). Storms can reduce the height of coastal dunes due to erosion. The first applications of GPR to the study of coastal dunes date back to the late 1980s (e.g. Leatherman 1987). Nielsen et al. (1995) and Clemmensen et al. (1996) studied the stratigraphy of the late Holocene eolian deposits of the west coast of Jutland (Denmark) by using GPR. The radar survey further helped calculating the sand budget of the dunefield and analysing dunefield dynamics in relation to wind, sediment supply, sea level and human influence. Several examples in literature indicate that GPR can accurately image the stratigraphy and internal sedimentary structure of coastal barriers, spits and strandplains, both above and below a fresh groundwater table (Neal & Roberts 2000; Nobes et al.2016; Rockett et al.2016). Good examples are, for example, radar surveys along the coast of Venice that did not only provide images of internal dune bedding, but also clues about the evolutional history of the coast line and the depth of transition from fresh water to brackish-salt water (Galgaro et al.2000). A detailed study of the Atlantic Galician Coast (NW Spain) through 2-D/3-D GPR surveys reveals the internal structure of depositional lobes, the pattern of which was related mainly to changes in the vegetation cover (González-Villanueva et al.2011). In southern Portugal, GPR reconstructs the conditions of a coastal segment and allows the interpretation of depositional and erosional evidences of the AD 1755 tsunami in this region (Costa et al.2016). Reconstructions of coastal process by means of GPR can further be found in Oliver & Woodroffe (2016), which provide a history of storm events for the Callala Barrier (SE Australia) from ∼4000 yr ago to present. The resolution provided by GPR images is nonetheless constrained by the depth of interest and is highest for the shallower targets (Jol & Bristow 2003). Moreover, aeolian deposits are often characterized by moderate variations in physical properties that result in small contrasts of EM impedance. Such characteristics affect the dynamic part of the radar response, that is, they are linked to low reflectivity and small amplitude GPR responses. The coastal environment is further characterized by conditions related to thin sedimentary lenses, prograding units and cross-cutting erosion surfaces. GPR can image such sedimentary features and allow the reconstruction of relative chronology, with particular reference to the identification of periods of accretion, progradation, erosion, reworking, stabilization and other major events in the dune history (e.g. Bailey & Bristow 2000; Bristow et al.2000b; McGourty & Wilson 2000; Havholm et al.2004; Flor-Blanco et al.2016; Forde et al.2016; Rockett et al.2016). Nonetheless, the information provided by conventional radar scans [i.e. amplitude versus two-way-time (TWT)] is not always sufficient to resolve low-amplitude contrasts in geometrically complex subsurface conditions. Attribute analysis has been incorporated into GPR data interpretation to extract more subsurface information from the processed GPR data-sets and to at least partly overcome the above discussed limits (e.g. Orlando 2002; Tronicke et al.2006; McClymont et al.2008; Forte et al.2012; Zhao et al.2013, 2016a,b; Ercoli et al.2015; Nobes & Jol 2015; Nobes et al.2016). In this paper, we focus on GPR attribute analysis to studyparabolic climbing sand dunes and to identify specific radar facies on the basis of the signature observed in different domains (i.e. the original Amplitude-TWT, and the calculated Attributes-TWT). The test site is located at Piscinas (SW Sardinia, Italy), an area characterized by a well-developed dune system. Radar facies are defined in terms of reflection continuity, shape, dip, internal reflection configuration and external form (Neal 2004), following the terms used in seismic stratigraphy (Mitchum et al.1977), and have been widely used to describe and classify different sedimentary structures (e.g. Van Heteren et al.1998; Van Overmeeren 1998; Beres et al.1999; Ekes & Hickin 2001; Moysey et al.2006; Franke et al.2015). We use a multi-attribute comparative analysis to study the Piscinas dunes and, in particular, to better analyse the reflection patterns and to characterize the different radar facies. Every single attribute can actually capture details related to different physical or geometrical properties of the subsurface (Chopra & Marfurt 2007) and their integration (multi-attribute analysis) can enhance the understanding of dunes stratigraphy and evolution. Moreover, we introduce composite displays (i.e. displays of multiple attributes in a single image) to highlight local continuity or variability of the GPR response and to identify and correlate specific features of the Piscinas dunefield. Study area The dunes of Piscinas, located along the Arburese coast, SW Sardinia, Italy (Fig. 1), are one of the largest and most spectacular transgressive dunefields of the Mediterranean basin. The system is formed by inactive and active parabolic bodies, developing from the present beach toward the inland. The longest moved more than 2 km inland, reaching a maximum elevation of more than 100 m. The sand dunes, composed of Quaternary lithified aeolian sandstones superimposed by coastal-aeolian loose sands (Palmerini & Ulzega 1969), are still active, modeled by the prevailing Mistral wind from the northwest and influenced by the vegetation pattern. In fact, the shape of the dunes is a non-linear function of coast exposure to the predominant winds and waves: the northern part of the dunefield is in the regressive phase, while the southern portion is transgressive, under the influence of the wind that pushes the sand mostly toward the southeast (Atzeni & Ginesu 1993). In particular, the coastal landscape of the area surrounded by green hills overlooking the sea, presents morphologies that reflect various geological contexts, where the general features of the sedimentation environment are recognized as microtidal coastal facies dominated by wave action and comprising the shoreface, the foreshore and the backshore, fluvial facies and aeolian facies (Caredda et al.1999; Costamagna & Barca 2008). Figure 1. View largeDownload slide Location map, photos and aerial picture of the study area, in the Piscinas Dunes, SW Sardinia, Italy. The orange line shows the GPR survey line having a total length of 386.5 m. Figure 1. View largeDownload slide Location map, photos and aerial picture of the study area, in the Piscinas Dunes, SW Sardinia, Italy. The orange line shows the GPR survey line having a total length of 386.5 m. Data acquisition and processing We performed a 2-D GPR survey with a ProExMalå Geoscience GPR system equipped with 250 MHz central frequency shielded antennas, which provided the optimum trade-off between depth of penetration and resolution. In this paper, we focus on an NW-SE oriented profile, because the optimum imaging conditions are met when the profile is perpendicular to the dunes (i.e. parallel to the dominant wind direction). The sector crossed by the profile includes two dune lobes and is paradigmatic of the entire dunefield (Fig. 1). We obtained 7730 traces on a profile length of 386.5 m at a constant 0.05 m sampling interval. The trace length is set to 300 ns and the sampling rate was 0.398 ns, considerably smaller than the Nyquist limit to optimize the results of attribute analysis. We applied the following GPR processing sequence: Data editing: to analyse data characteristics, identify noisy traces and anomalies (such as, e.g. zero-time drift, polarity inversions, isolated spikes and band-limited noise components). Dewow filter: to remove the low-frequency ‘wow’ component that may cover reflections. Design and application of a bandpass filter: to select the optimal band based on a signal-to-noise ratio (S/N) criterion, and the corner frequencies of the zero-phase Ormsby filter were defined after narrow-band (10 MHz) filter scans as 50–70 and 500–600 MHz. Windowed background removal: to remove horizontally coherent components primarily due to antenna-ground impedance mismatch. Amplitude analysis: to analyse amplitude decay (1-D) and evaluate lateral variations of the attenuation (2-D). True amplitude recovery: to correct for geometrical divergence and attenuation, and it requires knowledge of the velocity of radar waves in the volume to evaluate the optimal parameters to compensate for the effects of amplitude decay. Topographic correction (static correction) and depth conversion: to correct for near-surface effects. F–k migration: to focus diffracted energy and reconstruct the real position of subsurface features. A dedicated GPS survey in NRTK (Network Real Time Kinematik) mode provided digital elevation data with centimetric precision in latitude/longitude and an average elevation error of less than 4 cm. An average velocity of 0.13 m ns−1 was determined from diffraction hyperbolas analysis and utilized for topographic correction and time–depth conversion. The subsurface velocity field is rather homogeneous with maximum and minimum values respectively equal to 0.140 and 0.115 m ns−1. This is probably due to the small and rather constant amount of moisture within the dunes, which do not host any continuous aquifer, and to the relatively homogeneous characteristics of the sediments. Attribute analysis The S/N of the GPR data-set is rather high in the study area, notwithstanding the relatively low reflectivity due to the homogeneous characteristics of sediments and the small and constant moisture content (see previous comments about the velocity field). We applied GPR attribute analysis to integrate the basic interpretation of amplitude data with quantitative information about the physical property variations and to classify domains characterized by homogeneous GPR response. The choice of attribute analysis is mainly due to the locally low reflectivity and to the geometrical characteristics of the aeolian deposits. The former imply imaging difficulties through conventional amplitude analysis (or analysis of individual attributes sensitive to amplitude). The latter, related to conflicting and steep dips in hummocky cross-stratification or prograding deposits and thin or tapering out layers (i.e. at or below the resolution limits of the method) require different strategies to help the identification of volumes characterized by homogeneous response. Different attribute categories, including amplitude-/phase-/frequency-related, coherence and textural ones, have been therefore comparatively tested and evaluated. We particularly concentrate on textural attributes that enhance the identification of repeating patterns of local variations in image intensity. Such attributes were originally applied to 2-D image analysis, and have been successively introduced to GPR interpretation to improve identification of complex stratigraphic features (e.g. Zhao et al.2016b). We adopt the grey-level co-occurrence matrix (GLCM) algorithm to generate second-order statistical textural characteristics (Haralick et al.1973). A brief mathematical background about textural attributes is provided in the  Appendix, while hereafter just the definition and a short comment on each used attribute is provided. Amplitude first derivative: it is the time derivative of the instantaneous amplitude (i.e. the time rate of change of the envelope), which is in turn calculated by the Hilbert transform. Interfaces and discontinuities can be highlighted by this attribute. Instantaneous frequency: it is also calculated through the Hilbert transform and outputs, sample by sample, the local highest frequency component of the signal. It is sensitive to both wave propagation effects and depositional characteristics; hence it is a physical attribute that can be used to discriminate between materials having different intrinsic attenuation due to variable EM properties. Similarity: it is one of the coherence attributes that expresses how much two or more trace segments look alike. A similarity of 1 means the trace segments are completely identical in waveform and amplitude. A similarity of 0 means they are completely different. Low S/N segments exhibit low similarity. Assuming two trace segments u(x, y, t) and v(x, y, t) with sample length N, such attribute is defined for a specific analysis window as   \begin{equation*} 1 - \frac{{\sqrt {\sum\limits_{i = 1}^N {{{({u_i} - {v_i})}^2}} } }}{{\sqrt {\sum\limits_{i = 1}^N {u_i^2} } + \sqrt {\sum\limits_{i = 1}^N {v_i^2} } }} \end{equation*} Every single attribute can partly capture details related to different physical or geometrical properties of the subsurface. In order to highlight local continuity or variability of GPR data-sets and to identify and correlate specific sedimentary layers and interfaces, we performed integrated multi-attributes analysis through the combination of different attributes into a single image. We adopt two different schemes to display composite images: (i) one attribute of interest is plotted in colour (or in grey scale preferably) to form a background, and another attribute is plotted in variable area format and overlaid on the background attribute; (ii) each pixel in composite image is expressed by the weighted red/green/blue, representing three different attributes, and in this way three attributes can be displayed simultaneously. RESULTS Fig. 2 shows the processed GPR amplitude profile over two overlying asymmetric sand dunes. A clear reflection (indicated by the cyan line) can be easily tracked at a maximum depth of about 8 m below the ground surface, calculated from the estimated velocity of 0.13 m ns−1, corresponding to a marked unconformity possibly related to post-Tyrrhenian palaeosoil (Palmerini & Ulzega 1969), herein referred as PTP. Horizontal and dipping reflections with different angles above the PTP indicate a complex history of aeolian erosion and deposition, while the signal is totally attenuated below the PTP. Figure 2. View largeDownload slide Processed GPR amplitude profile. The cyan line indicates a marked unconformity interpreted as a post-Tyrrhenian palaeosoil (PTP), while the four zones marked by white rectangles will be analysed in detail hereafter. Figure 2. View largeDownload slide Processed GPR amplitude profile. The cyan line indicates a marked unconformity interpreted as a post-Tyrrhenian palaeosoil (PTP), while the four zones marked by white rectangles will be analysed in detail hereafter. Fig. 3 shows the attribute results calculated on the processed GPR data: amplitude first derivative, instantaneous frequency and similarity. We can note that: amplitude first derivative can identify and emphasize the main lateral changes, at the expense of weak reflections (Fig. 3a); the resolution of the instantaneous frequency profile seems a little lower compared with amplitude-related attributes, but it is more sensitive to the actual target response (i.e. low-frequency component in Fig. 3b, the value of which is close to 230 MHz); high similarity values are likely correlated with coherent reflections and low similarity values are related with more chaotic patterns and low S/N sectors of the radar section (see e.g. the sector below PTP in Fig. 3c). Fig. 4 shows two textural attributes, that is, dissimilarity and homogeneity. We can see that the textural attributes locally exhibit opposite characteristics, for example, the high-amplitude continuous reflections are characterized by relatively high dissimilarityand low homogeneity. Figure 3. View largeDownload slide The attribute results calculated on the processed GPR data. (a) Amplitude first derivative; (b) instantaneous frequency and (c) similarity. Figure 3. View largeDownload slide The attribute results calculated on the processed GPR data. (a) Amplitude first derivative; (b) instantaneous frequency and (c) similarity. Figure 4. View largeDownload slide Textural attributes calculated on the processed GPR data. (a) Dissimilarity and (b) homogeneity. Figure 4. View largeDownload slide Textural attributes calculated on the processed GPR data. (a) Dissimilarity and (b) homogeneity. Every attribute can reflect different physical or geometrical properties of the subsurface target. For example, the PTP (marked in Fig. 2) can be tracked down as a continuous reflector between 90 and 170 m in the similarity profile but exhibits different characteristics in the amplitude first derivative. In order to understand the relationships between amplitude radargrams and calculated attributes, four typical radar facies configurations are selected from the processed GPR data, considering reflection continuity, shape, dip and external form, which is also important for GPR sedimentary interpretation and definition of sedimentary facies (Reading 1996). The dimensions of all the selected image samples characterized by radar facies are 2.75 × 0.5 m (horizontal × vertical), the specific locations of which can be found in Fig. 2. They are identified as: Zone A: continuous dipping reflections, generated by sets of cross-stratification. Zone B: continuous horizontal reflections, generated by lateral sand layers. Zone C: discontinuous horizontal reflections, generated by high-angle discontinuities. Zone D: reflection-free, generated by massive homogeneous units with low S/N. Fig. 5 shows the results of four attributes (plus amplitude) and their corresponding histograms calculated on the selected zones, and Table 1 summarizes the qualitative analysis of histogram distributions of attributes calculated from the four typical radar facies from Fig. 5. We can see that different radar facies are characterized by different attribute patterns. All the attributes maintain continuous characteristics for continuous reflections, while they have lateral variations when the reflection is discontinuous. Figure 5. View largeDownload slide Four typical radar facies, the corresponding attributes and their corresponding histograms. All the analysed portions have dimensions equal to 2.75 × 0.5 m, and their specific locations can be found in Fig. 2. Figure 5. View largeDownload slide Four typical radar facies, the corresponding attributes and their corresponding histograms. All the analysed portions have dimensions equal to 2.75 × 0.5 m, and their specific locations can be found in Fig. 2. Table 1. Results of the histogram analysis of attributes calculated from the four typical radar facies from Fig. 5. Zone  Processed GPR data  Amplitude first derivative  Instantaneous frequency  Similarity  Homogeneity  Zone A  Low  Low  240 MHz concentrated  High  Medium  Zone B  High  High  240 MHz concentrated  High  Medium  Zone C  Medium  Medium  220 MHz concentrated  Mixed  Low  Zone D  Very low  Very low  Dispersed  Dispersed  High  Zone  Processed GPR data  Amplitude first derivative  Instantaneous frequency  Similarity  Homogeneity  Zone A  Low  Low  240 MHz concentrated  High  Medium  Zone B  High  High  240 MHz concentrated  High  Medium  Zone C  Medium  Medium  220 MHz concentrated  Mixed  Low  Zone D  Very low  Very low  Dispersed  Dispersed  High  View Large Amplitude-related attributes can measure the overall reflectivity and recognize major changes in sediments. Frequency is very sensitive to attenuation, but there are no significant frequency variations in Zones A, B and C. Zone D exhibits different and more dispersed frequency characteristics and the histogram accordingly shows a wider distribution. The patterns of instantaneous frequency of continuous reflections (i.e. Zones A and B) are similar to the amplitude first derivative. A clear correlation is lacking for discontinuous reflections (Zone C) and a sort of blurring shows up in both displays (frequency/Amplitude first derivative). Blurring is the main characteristics for all attributes in diffuse scattering and low S/N zones (i.e. Zone D). Similarity is very sensitive to continuous reflections, particularly to horizontal ones—as in the case of Zone B, which shows a concentrated distribution in the histogram. The textural attribute homogeneity can reflect the overall smoothness in image intensity, and it seems a more stable parameter, in comparison to other attributes, when reflectors exhibit a laterally constant response. This is not only related to the presence of fractures/faults, but also to lateral variations of attenuation in the overburden. Fig. 6 provides various composite displays of combined attributes. Different reflection patterns within specific zones, associated with different sediment characteristics can be highlighted: major reflections can be recognized in any composite image, including amplitude-related attributes, such as the PTP marked with cyan lines in Figs 6(a) and (c), interpreted as the base of the present depositional aeolian system; two continuous dipping reflection zones are emphasized in Fig. 6(b), which is an attribute combination, without considering any amplitude-related attribute; two irregular sand lenses can be displayed in Figs 6(b) and (c), within the white polygons, and can be associated with more homogeneous sand units having a grain size slightly different from the surrounding materials. Figure 6. View largeDownload slide Example of composite multi-attributes displays. (a) Homogeneity overlaid on amplitude first derivative; (b) dissimilarity overlaid on similarity; (c) homogeneity, amplitude first derivative and instantaneous frequency mixed with RGB model. Figure 6. View largeDownload slide Example of composite multi-attributes displays. (a) Homogeneity overlaid on amplitude first derivative; (b) dissimilarity overlaid on similarity; (c) homogeneity, amplitude first derivative and instantaneous frequency mixed with RGB model. DISCUSSIONS AND CONCLUSIONS GPR can effectively image complex shallow sedimentary features in parabolic climbing sand dunes. The integrated analysis of different attributes of the radar trace may offer enhanced opportunities of understanding the spatial distribution of sedimentary features and their characteristics. Our study shows that attribute analysis can successfully integrate the basic amplitude interpretation and allow constrained studies of lateral continuity and variability of sediments. Every attribute can partly reflect different physical or geometrical properties of the subsurface target at different resolution levels and, in some cases, at resolution lower than the original amplitude data (see e.g. Fig. 5). Some attributes, such as, for example, similarity, are more sensitive to lateral continuity regardless of the amplitude of the radar response (see also the relevant Zones A and B histograms in Fig. 5). This is particularly important in the study of aeolian deposits and coastal dune systems that may be characterized by moderate variations of physical properties and by consequently small EM impedance contrasts. The interpretation of the resulting low-reflectivity records can greatly benefit from the integration provided by attributes that are less sensitive to the amplitude factor. Moreover, the combination of several attributes into a single image, namely a multi-attribute composite display, can visually highlight groups of distinct reflection patterns, and enhance the identification of volumes characterized by homogeneous response. A GPR homogeneity, lato sensu, that is, inferred from the integrated analysis of the different attributes, can be linked, in space, to peculiar depositional environments and, in time, to different phases in the development of coastal geomorphology (Fig. 6). In fact, by using integrated attributes and in particular combined textural ones, we can attain not only a clearer subsurface image but even more important an enhanced discrimination between different sedimentological facies/domains. Some attributes are particularly sensitive on such purpose, such as, for example, similarity, which is related to reflector's continuity and stability of response. Localized variations of physical properties and sedimentological characteristics (e.g. grain size and arrangement) can be responsible for large similarity oscillations that are easily identified and can be effectively correlated to sedimentological facies changes. The relationships among different attributes and their corresponding sedimentological and radar facies deserve further study. In particular, the study of the distributions of the attributes (see Fig. 5 and Table 1) should be tested on different case studies to evaluate its consistency as an estimator and the possibility to use it as a classification tool of sediment bodies and aeolian landforms. This study is limited to a single test site and leaves such issue open. Nonetheless, the GPR study of the Piscinas dunes indicates that the combination among GPR attributes, ground truthing and results of laboratory analysis, can establish a new route towards a more complete non-invasive sedimentological interpretation. ACKNOWLEDGEMENTS This study has been carried out within the RITMARE Flagship Project, coordinated by CNR and funded by the Italian Ministry of University and Research. We gratefully acknowledge the support of the International Centre for Theoretical Physics (ICTP, Trieste, Italy) Training on Research in Italian Laboratories (TRIL) programme, which sponsored the scholarship of the first author. We thank Halliburton Landmark for the license of the SeisSpace ProMAX Seismic Processing Software provided through an academic grant and dGB Earth Sciences for the OpendTect open source seismic data analysis software. A. Bezzi, G. Casagrande, D. Martinucci and S. Pillon are kindly acknowledged for their help in the topographic survey and the GPR data acquisition. REFERENCES Adetunji A.Q., Al-Shuhail A., Korvin G., 2008. Mapping the internal structure of sand dunes with GPR: a case history from the Jafurah sand sea of eastern Saudi Arabia, Leading Edge , 27( 11), 1446– 1452. https://doi.org/10.1190/1.3011016 Google Scholar CrossRef Search ADS   Atzeni A., Ginesu S., 1993. Evoluzione dei litorali della Sardegna e interventi di riequilibrio, La difesa dei litorali Ital. , 34, 215– 231. Bagnold R.A., 1941. The Physics of Blown Sand and Desert Dunes . Methuen, London, 265 pp. Bailey S., Bristow C.S., 2000. Structure of coastal dunes: observations from ground penetrating radar (GPR) surveys, in 8th International Conference on Ground Penetrating Radar , pp. 660– 665. SPIE, University of Queensland, Australia. Baker G.S., Jol H.M., Eds., 2007. Stratigraphic Analyses using GPR, Special Paper . Geological Society of America, Boulder, CO. Beres M., Huggenberger P., Green A.G., Horstmeyer H., 1999. Using two- and three-dimensional georadar methods to characterize glaciofluvial architecture, Sediment. Geol. , 129( 1–2), 1– 24. https://doi.org/10.1016/S0037-0738(99)00053-6 Google Scholar CrossRef Search ADS   Bigarella J.J., Becker R.D., Duarte G.M., 1969. Coastal dune structures from Paraná (Brazil), Mar. Geol. , 7( 1), 5– 55. https://doi.org/10.1016/0025-3227(69)90002-4 Google Scholar CrossRef Search ADS   Bristow C., 1995. Facies analysis in the Lower Greensand using ground-penetrating radar, J. geol. Soc. , 152( 4), 591– 598. https://doi.org/10.1144/gsjgs.152.4.0591 Google Scholar CrossRef Search ADS   Bristow C.S., 2009. Ground penetrating radar in aeolian dune sands, in Ground Penetrating Radar: Theory and Applications , pp. 273– 297. ed. Jol H.M., Elsevier, Amsterdam, The Netherlands. Google Scholar CrossRef Search ADS   Bristow C.S., Augustinus P.C., Wallis I.C., Jol H.M., Rhodes E.J., 2010. Investigation of the age and migration of reversing dunes in Antarctica using GPR and OSL, with implications for GPR on Mars, Earth planet. Sci. Lett. , 289( 1–2), 30– 42. https://doi.org/10.1016/j.epsl.2009.10.026 Google Scholar CrossRef Search ADS   Bristow C.S., Bailey S.D., Lancaster N., 2000a. The sedimentary structure of linear sand dunes, Nature , 406( 6791), 56– 59. https://doi.org/10.1038/35017536 Google Scholar CrossRef Search ADS   Bristow C.S., Chroston P.N., Bailey S.D., 2000b. The structure and development of foredunes on a locally prograding coast: insights from ground-penetrating radar surveys, Norfolk, UK, Sedimentology , 47( 5), 923– 944. https://doi.org/10.1046/j.1365-3091.2000.00330.x Google Scholar CrossRef Search ADS   Bristow C.S., Jol H.M., 2003. An introduction to ground penetrating radar (GPR) in sediments, Geol. Soc. Lond. Spec. Publ. , 211( 1), pp. 1– 7. https://doi.org/10.1144/GSL.SP.2001.211.01.01 Google Scholar CrossRef Search ADS   Bristow C.S., Lancaster N., Duller G.A.T., 2005. Combining ground penetrating radar surveys and optical dating to determine dune migration in Namibia, J. geol. Soc. , 162( 2), 315– 321. https://doi.org/10.1144/0016-764903-120 Google Scholar CrossRef Search ADS   Busby J., Merritt J., 1999. Quaternary deformation mapping with ground penetrating radar, J. appl. geophys. , 41( 1), 75– 91. https://doi.org/10.1016/S0926-9851(98)00050-0 Google Scholar CrossRef Search ADS   Buynevich I.V., Fitzgerald D.M., 2001. Styles of coastal progradation revealed in subsurface records of paraglacial barriers: Duxbury, Massachusetts, USA. J. Coast. Res. , 34, 194– 208. Caredda A.M., Cristini A., Ferrara C., Lobina M.F., Baroli M., 1999. Distribution of heavy metals in the Piscinas beach sediments (SW Sardinia, Italy), Environ. Geol. , 38( 2), 91– 100. https://doi.org/10.1007/s002540050405 Google Scholar CrossRef Search ADS   Cassidy N.J., 2009. Ground penetrating radar data processing, modelling and analysis, in Ground Penetrating Radar: Theory and Applications , pp. 141– 176, ed. Jol H.M., Elsevier, Amsterdam, The Netherlands. Chopra S., Marfurt K.J., 2007. Seismic Attributes for Prospect Identification and Reservoir Characterization . SEG/EAGE, 464 pp. Google Scholar CrossRef Search ADS   Clemmensen L.B., Andreasen F., Nielsen S.T., Sten E., 1996. The late Holocene coastal dunefield at Vejers, Denmark: characteristics, sand budget and depositional dynamics. Geomorphology , 17( 1–3), 79– 98. https://doi.org/10.1016/0169-555X(95)00096-N Google Scholar CrossRef Search ADS   Clemmensen L.B., Nielsen L., 2010. Internal architecture of a raised beach ridge system (Anholt, Denmark) resolved by ground-penetrating radar investigations, Sediment. Geol. , 223( 3–4), 281– 290. https://doi.org/10.1016/j.sedgeo.2009.11.014 Google Scholar CrossRef Search ADS   Costa P.J. et al.  , 2016. How did the AD 1755 tsunami impact on sand barriers across the southern coast of Portugal?, Geomorphology , 268, 296– 311. https://doi.org/10.1016/j.geomorph.2016.06.019 Google Scholar CrossRef Search ADS   Costamagna L.G., Barca S., 2008. Depositional architecture and sedimentology of the Tuppa Niedda Conglomerates (Late Carboniferous, Arburese, SW Sardinia, Italy), Boll. Soc. Geol. Ital. , 127, 625– 636. Eichkitz C.G., Davies J., Amtmann J., Schreilechner M.G., de Groot P., 2015. Grey level co-occurrence matrix and its application to seismic data, First Break , 33, 71– 77. Ekes C., Hickin E.J., 2001. Ground penetrating radar facies of the paraglacial Cheekye Fan, southwestern British Columbia, Canada, Sediment. Geol. , 143( 3–4), 199– 217. https://doi.org/10.1016/S0037-0738(01)00059-8 Google Scholar CrossRef Search ADS   Ercoli M., Pauselli C., Cinti F.R., Forte E., Volpe R. 2015. Imaging of an active fault: comparison between 3D GPR data and outcrops at the Castrovillari fault, Calabria, Italy. Interpretation , 3( 3), SY57– SY66. https://doi.org/10.1190/INT-2014-0234.1 Google Scholar CrossRef Search ADS   Flor-Blanco G., Rubio-Melendi D., Flor G., Fernández-Álvarez J.P., Jackson D.W.T., 2016. Holocene evolution of the Xagó dune field (Asturias, NW Spain) reconstructed by means of morphological mapping and ground penetrating radar surveys. Geo-Mar. Lett. , 36( 1), 35– 50. https://doi.org/10.1007/s00367-015-0427-1 Google Scholar CrossRef Search ADS   Forde T.C., Nedimović M.R., Gibling M.R., Forbes D.L., 2016. Coastal evolution over the past 3000 years at Conrads Beach, Nova Scotia: the influence of local sediment supply on a paraglacial transgressive system. Estuar. Coast , 39( 2), 363– 384. https://doi.org/10.1007/s12237-015-0016-6 Google Scholar CrossRef Search ADS   Forte E., Pipan M., Casabianca D., Di Cuia R., Riva A., 2012. Imaging and characterization of a carbonate hydrocarbon reservoir analogue using GPR attributes. J. appl. geophys. , 81, 76– 87. https://doi.org/10.1016/j.jappgeo.2011.09.009 Google Scholar CrossRef Search ADS   Franke D., Hornung J., Hinderer M., 2015. A combined study of radar facies, lithofacies and three-dimensional architecture of an alpine alluvial fan (Illgraben fan, Switzerland), Sedimentology , 62( 1), 57– 86. https://doi.org/10.1111/sed.12139 Google Scholar CrossRef Search ADS   Galgaro A., Finzi E., Tosi L., 2000. An experiment on a sand-dune environment in Southern Venetian coast based on GPR, VES and documentary evidence, Ann. Geophys. , 43( 2), 289– 295. Goldsmith V., 1973. Internal geometry and origin of vegetated coastal sand dunes, J. Sediment. Res. , 43( 4), 1128– 1143. Gómez-Ortiz D., Martín-Crespo T., Rodríguez I., Sánchez M.J., Montoya I., 2009. The internal structure of modern barchan dunes of the Ebro River Delta (Spain) from ground penetrating radar, J. appl. geophys. , 68( 2), 159– 170. https://doi.org/10.1016/j.jappgeo.2008.11.007 Google Scholar CrossRef Search ADS   González-Villanueva R., Costas S., Duarte H., Pérez-Arlucea M., Alejo I., 2011. Blowout evolution in a coastal dune: using GPR, aerial imagery and core records, J. Coast. Res. , 64, 278– 282. Guillemoteau J., Bano M., Dujardin J.R., 2012. Influence of grain size, shape and compaction on georadar waves: examples of aeolian dunes, Geophys. J. Int. , 190( 3), 1455– 1463. https://doi.org/10.1111/j.1365-246X.2012.05577.x Google Scholar CrossRef Search ADS   Haralick R.M., Shanmugam K., Dinstein I.H., 1973. Textural features for image classification. IEEE Trans. Syst. Man Cybern. , 6( 6), 610– 621. https://doi.org/10.1109/TSMC.1973.4309314 Google Scholar CrossRef Search ADS   Harari Z., 1996. Ground-penetrating radar (GPR) for imaging stratigraphic features and groundwater in sand dunes, J. appl. geophys. , 36( 1), 43– 52. https://doi.org/10.1016/S0926-9851(96)00031-6 Google Scholar CrossRef Search ADS   Havholm K.G. et al.  , 2004. Stratigraphy of back-barrier coastal dunes, northern North Carolina and southern Virginia, J. Coast. Res. , 20( 4), 980– 999. https://doi.org/10.2112/03503A2.1 Google Scholar CrossRef Search ADS   Hesp P., 1988. Morphology, dynamics and internal stratification of some established foredunes in southeast Australia, Sediment. Geol. , 55( 1–2), 17– 41. https://doi.org/10.1016/0037-0738(88)90088-7 Google Scholar CrossRef Search ADS   Jol H.M., Smith D.G., Meyers R.A., 1996. Digital ground penetrating radar (GPR): a new geophysical tool for coastal barrier research (Examples from the Atlantic, Gulf and Pacific Coasts, U.S.A.), J. Coast. Res. , 12, 960– 968. Jol H.M., Bristow C.S., 2003. GPR in sediments: advice on data collection, basic processing and interpretation, a good practice guide. Geol. Soc., Lond. Spec. Publ. , 211, 9– 27. Google Scholar CrossRef Search ADS   Leatherman S.P., 1987. Coastal geomorphological applications of ground-penetrating radar, J. Coast. Res. , 3( 3), 397– 399. Lorenz R.D., Zimbelman J.R., 2014. Dune Worlds. Springer Praxis Books, Berlin/Heidelberg, 308 pp. Love P.L., Simaan M., 1985. Segmentation of a seismic section using image processing and artificial intelligence techniques, Pattern Recog. , 18( 6), 409– 419. https://doi.org/10.1016/0031-3203(85)90011-1 Google Scholar CrossRef Search ADS   McClymont A.F. et al.  , 2008. Visualization of active faults using geometric attributes of 3D GPR data: an example from the Alpine Fault Zone, New Zealand, Geophysics , 73( 2), B11– B23. https://doi.org/10.1190/1.2825408 Google Scholar CrossRef Search ADS   McGourty J., Wilson P., 2000. Investigating the internal structure of Holocene coastal sand dunes using ground-penetrating radar: example from the north coast of Northern Ireland, Proceedings of SPIE–The International Society for Optical Engineering, 4084 , pp. 14– 19. Mitchum R.M. Jr, Vail P.R., Sangree J.B., 1977. Seismic stratigraphy and global changes of sea level, part 6: stratigraphic Interpretation of seismic reflection patterns in depositional sequences, Geophys. Res. Lett. , 165( 22), 117– 133. Moore L.J., Jol H.M., Kruse S., Vanderburgh S., Kaminsky G.M., 2004. Annual Layers Revealed by GPR in the Subsurface of a Prograding Coastal Barrier, Southwest Washington, U.S.A., J. Sediment. Res. , 74( 5), 690– 696. https://doi.org/10.1306/021604740690 Google Scholar CrossRef Search ADS   Moysey S., Knight R.J., Jol H.M., 2006. Texture-based classification of ground-penetrating radar images, Geophysics , 71( 6), K111– K118. https://doi.org/10.1190/1.2356114 Google Scholar CrossRef Search ADS   Neal A., 2004. Ground-penetrating radar and its use in sedimentology: principles, problems and progress. Earth-Sci. Rev. , 66( 3–4), 261– 330. https://doi.org/10.1016/j.earscirev.2004.01.004 Google Scholar CrossRef Search ADS   Neal A., Roberts C.L., 2000. Applications of ground-penetrating radar (GPR) to sedimentological, geomorphological and geoarchaeological studies in coastal environments, in Pye K, Allen J.R.L. eds, Coastal and Estuarine Environments . Sedimentology, Geomorphology and Geoarchaeology , Vol. 175, pp. 139– 171. Geological Society Special Publication. Google Scholar CrossRef Search ADS   Nielsen S.T., Clemmensen L.B., Andreasen F., 1995. The middle and late Holocene barrier spit system at Vejers, Denmark: structure and development, Bull. geol. Soc. Denmark , 42( 1), 105– 119. Nobes D.C., Jol H.M., 2015. Enhancing form and structure: complex attributes as aids for ground penetrating radar interpretation, in Near-Surface Asia Pacific Conference ,, Waikoloa, Hawaii, pp. 312– 315, Society of Exploration Geophysicists, Australian Society of Exploration Geophysicists, Chinese Geophysical Society, Korean Society of Earth and Exploration Geophysicists, and Society of Exploration Geophysicists of Japan. Nobes D.C., Jol H.M., Duffy B., 2016. Geophysical imaging of disrupted coastal dune stratigraphy and possible mechanisms, Haast, South Westland, New Zealand, N. Z. J. Geol. Geophys. , 59( 3), 426– 435. https://doi.org/10.1080/00288306.2016.1168455 Google Scholar CrossRef Search ADS   Oliver T.S., Woodroffe C.D., 2016. Chronology, morphology and GPR-imaged internal structure of the Callala Beach prograded barrier in Southeastern Australia, J. Coast. Res. , 75( suppl.1), 318– 322. https://doi.org/10.2112/SI75-064.1 Google Scholar CrossRef Search ADS   Orlando L., 2002. Detection and analysis of LNAPL using the instantaneous amplitude and frequency of ground-penetrating radar data, Geophys Prospect. , 50( 1), 27– 41. https://doi.org/10.1046/j.1365-2478.2002.00288.x Google Scholar CrossRef Search ADS   Palmerini V., Ulzega A., 1969. Sedimentologia e Geomorfologia del settore costiero tra la foce del Rio Piscinas e Capo Pecora (Sardegna sud-occidentale). Rendiconti del Seminario della Facoltà di Scienze dell’Università di Cagliari , Vol. 39, pp. 1– 38. Reading H.G., 1996. Sedimentary Environments: Processes, Facies and Stratigraphy . Blackwell Publishing, Oxford, 687 pp. Rockett G.C., Barboza E.G., Rosa M.L.C., 2016. Ground penetrating radar applied to the characterization of the Itapeva Dunefield, Torres, Brazil. J. Coast. Res. , 75 ( suppl. 1), 323– 327. https://doi.org/10.2112/SI75-065.1 Google Scholar CrossRef Search ADS   Santalla I.R., García M.J.S., Montes I.M., Ortiz D.G., Crespo T.M., Raventos J.S., 2009. Internal structure of the aeolian sand dunes of El Fangar spit, Ebro Delta (Tarragona, Spain). Geomorphology , 104( 3–4), 238– 252. https://doi.org/10.1016/j.geomorph.2008.08.017 Google Scholar CrossRef Search ADS   Schenk C.J., Gautier D.L., Olhoeft G.R., Lucius J.E., 1993. Internal structure of an aeolian dune using ground-penetrating radar, in Pye K, Lancaster N., eds, Aeolian Sediments , pp. 61– 69. Blackwell Publishing Ltd., Oxford, UK. Google Scholar CrossRef Search ADS   Soh L.-K., Tsatsoulis C., 1999. Texture analysis of SAR sea ice imagery using gray level co-occurrence matrices, IEEE Trans. Geosci. Remote Sens. , 37( 2), 780– 795. https://doi.org/10.1109/36.752194 Google Scholar CrossRef Search ADS   Tercier P., Knight R., Jol H., 2000. A comparison of the correlation structure in GPR images of deltaic and barrier?spit depositional environments, Geophysics , 65( 4), 1142– 1153. https://doi.org/10.1190/1.1444807 Google Scholar CrossRef Search ADS   Tillmann T., 2014. Landscape development of Amrum's west coast (Southern North Sea): GPR and sedimentology, in 15th International Conference on Ground Penetrating Radar (GPR) , pp. 250– 254, IEEE. Tronicke J., Villamor P., Green A.G., 2006. Detailed shallow geometry and vertical displacement estimates of the Maleme Fault Zone, New Zealand, using 2D and 3D georadar. Near Surf. Geophys. , 4( 3), 155– 161. Van Dam R.L., Schlager W., 2000. Identifying causes of ground-penetrating radar reflections using time-domain reflectometry and sedimentological analyses, Sedimentology , 47( 2), 435– 449. https://doi.org/10.1046/j.1365-3091.2000.00304.x Google Scholar CrossRef Search ADS   Van Heteren S., Fitzgerald D.M., Barber D.C., Kelley J.T., Belknap D.F., 1996. Volumetric Analysis of a New England Barrier System Using Ground-Penetrating-Radar and Coring Techniques, J. Geol. , 104( 4), 471– 483. https://doi.org/10.1086/629840 Google Scholar CrossRef Search ADS   Van Heteren S.V., Fitzgerald D.M., Mckinlay P.A., Buynevich I.V., 1998. Radar facies of paraglacial barrier systems: coastal New England, USA, Sedimentology , 45( 1), 181– 200. https://doi.org/10.1046/j.1365-3091.1998.00150.x Google Scholar CrossRef Search ADS   Van Overmeeren R.A., 1998. Radar facies of unconsolidated sediments in The Netherlands: a radar stratigraphy interpretation method for hydrogeology, J. appl. geophys. , 40( 1–3), 1– 18. https://doi.org/10.1016/S0926-9851(97)00033-5 Google Scholar CrossRef Search ADS   West B.P., May S.R., Eastwood J.E., Rossen C., 2002. Interactive seismic facies classification using textural attributes and neural networks, Leading Edge , 21( 10), 1042– 1049. https://doi.org/10.1190/1.1518444 Google Scholar CrossRef Search ADS   Zhao W., Forte E., Colucci R.R., Pipan M., 2016a. High-resolution glacier imaging and characterization by means of GPR attribute analysis, Geophys. J. Int. , 206( 2), 1366– 1374. https://doi.org/10.1093/gji/ggw208 Google Scholar CrossRef Search ADS   Zhao W., Forte E., Pipan M., 2016b. Texture attribute analysis of GPR data for archaeological prospection, Pure appl. Geophys. , 173( 8), 2737– 2751. https://doi.org/10.1007/s00024-016-1355-3 Google Scholar CrossRef Search ADS   Zhao W., Forte E., Pipan M., Tian G., 2013. Ground penetrating radar (GPR) attribute analysis for archaeological prospection, J. appl. geophys. , 97, 107– 117. https://doi.org/10.1016/j.jappgeo.2013.04.010 Google Scholar CrossRef Search ADS   APPENDIX: TEXTURAL ATTRIBUTES Textural analysis considers an ensemble of traces as an image with the goal to mathematically describe and quantify the distribution of values (pixels) within a portion of data. This means computing the spatial organization of reflections (or any other event) in terms of their continuity, smoothness, coherence, extracting and highlighting the global ‘signature’ of the analysed region. Following Haralick et al. (1973), a 2-D digitized image (I), including in this category any seismic or GPR profile, is simply a 2-D array or ordered values referred to as time (usually a two-way traveltime) and spatial samples (or pixels/voxels), when considering the two image axes. If R(x) = {1, 2, …, Nx} and C(y) = {1, 2, …, Ny} are the x and y spatial domains (often reported as rows and columns, respectively), then R(x) × C(y) is the set of resolution cells and (I) is a function assigning grey-tone values G ∈ {1, 2, …, Ng} to each singular resolution cell. (G) is called GLCM and it is a measure of how often different combinations of pixel amplitude values occur in (I) being Ng the quantized grey tones. Because typically two samples are compared, GLCM is referred to as a second-order texture classification method, but the previous consideration is still valid for any data dimension. We can assume that the texture information in (I) is represented by the ‘global’ spatial relationships among grey levels. So, this information is specified by a matrix of the relative frequencies (i.e. probabilities) Pi, j of two neighbouring resolution cells, separated by a distance d and having one grey tone i and the other grey tone j. Pi, j matrices are therefore function of the distance between neighbouring resolution cells and angular relationships between them. If angles ϑ are quantized by 45° steps, there are horizontal and vertical GLCM directions analysis, respectively, for angles equal to 0° and 90° and two diagonal directions for angles equal to 45° and 135°. So it is possible to calculate four relative frequency matrices P(i, j, d, ϑ) for the previous described angle values. If N ∈ {R(x) × C(y)} is the number of elements in the considered set defined by N[(k, l); (m, n)], then P(i, j, d, ϑ) for ϑ = 0°; 90°; 45°, 135° are respectively:   \begin{eqnarray*} P \left( {0^\circ } \right) &=& \ N\{ ( {k,l;m,n} )\ \times ( {R( x ) \times C( y)} )| {k - m}\nonumber\\ & =& { 0;\ | {l - n} | = \ d;I ( {k,l} ) = \ i,I ( {m,n} ) = \ j} \} \end{eqnarray*}   \begin{eqnarray*} P ( {90^\circ } ) &=& \ N\{ ( {k,l;m,n} ) \times ( {R( x ) \times C( y )} )| | {k - m}|\nonumber\\ &=& d;l - n\ = \ 0;I ( {k,l} ) = \ i,I ( {m,n} ) = \ j \} \end{eqnarray*}   \begin{eqnarray*} &&{P ( {45^\circ } ) = \ N\Bigg\{ ( {k,l;m,n} ) \times ( {R( x ) \times C( y )} )\Bigg|}\\ &&{\begin{array}{@{}*{1}{c}@{}} {( {k - m\ = \ d;l - n\ = \ - d} )}\\ \quad{{\rm or}\ ( {k - m\ = \ - d;l - n\ = \ d} );I ( {k,l} ) = \ i,I ( {m,n} ) = \ j} \end{array}} \Bigg\} \end{eqnarray*}   \begin{eqnarray*} &&{P ( {135^\circ } ) = N\Bigg\{ ( {k,l;m,n} ) \times ( {R( x ) \times C( y )} )\Bigg|}\\ &&{\begin{array}{@{}*{1}{c}@{}} {( {k - m\ = \ d;l - n\ = \ d} )}\\ \quad{{\rm {or}}\ ( {k - m\ = \ - d;l - n\ = \ - d} );I ( {k,l} ) = \ i,I ( {m,n} ) = \ j} \end{array} \Bigg\}} \end{eqnarray*} From the initial assumption that all the texture information of (I) is contained in the GLCM, it is possible to statistically derive several textural attributes. Haralick et al. (1973) proposed 14 attributes, while Soh & Tsatsoulis (1999) developed 10 additional quantities. All these attributes are usually assigned in the central point of each analysis running window, and repeating the calculation for all the subregion selected it is possible to obtain an attribute image (I'), R(x) × C(y) wide. The mostly used attributes for seismic data analysis are defined by the following equations, in which Ng is the number of distinct grey levels considered in the quantized image and the GLCM are normalized. For each of them, we also give a brief qualitative description and the maximum possible range (Rg).   \begin{equation*} Contrast\!:{C_{i,j}} = \mathop \sum \limits_{i,j}^{{N_g}} {\left| {i,j} \right|^2}\ p\left( {i,j} \right) \end{equation*} It returns a measurement if the intensity contrast between a pixel and its neighbour; (Rg) = [0, ((R(x) × C(y)) − 1)2]. Ci, j is 0 for a constant amplitude image.   \begin{equation*} Energy\!:{\rm{\ }}{E_{i,j}} = \mathop \sum \limits_{i,j}^{{N_g}} p{\left( {i,j} \right)^2} \end{equation*} (Rg) = [0, 1]. Ei, j is 1 for a constant amplitude image.   \begin{equation*} Homogeneity\!:{H_{i,j}} = \mathop \sum \limits_{i,j}^{{N_g}} \frac{{p\left( {i,j} \right)}}{{1 + \left| {i - j} \right|}} \end{equation*} It returns a measurement of the closeness of the distribution of the elements in GLCM as compared with the GLCM diagonal; (Rg) = [0, 1]. Hi, j is 1 for a diagonal GLCM.   \begin{equation*} Dissimilarity\!:{D_{i,j}} = \mathop \sum \limits_{i,j}^{{N_g}} \left| {i - j} \right|\ p\left( {i,j} \right) \end{equation*} It is a measurement that defines the variation of grey-level pairs in an image. It is the closest to Ci, j with adifference in the weight because Ci, j unlike Di, j grows quadratically. (Rg) = [0, 1].   \begin{equation*} Entropy\!:{T_{i,j}} = \mathop \sum \limits_{i,j}^{{N_g}} p\left( {i,j} \right){\rm{log}}\ p\left( {i,j} \right) \end{equation*} It is a measurement of spatial disorder. A completely random distribution would have very high entropy, while a constant amplitude image would have an entropy value of 0. Seismic textural attributes were first introduced into geophysical exploration by Love &Simaan (1985), which fist highlighted some of the possible advantages for data analysis and interpretation. For seismic and GPR data, the number of grey levels typically considered is between 4 Bits (i.e. 16 levels) and 8 Bits (i.e. 256 levels, West et al.2002; Eichkitz et al.2015), but more accurate (and time-consuming) analysis can be performed with higher sampled dynamic ranges. The previous calculations refer to amplitude seismic (or GPR) section, however it is possible to apply all the algorithms to virtually any input quantity including, for instance, smoothed versions of the original data, frequency- or phase-related attributes, data obtained after the application of edge detection techniques, and so on. © The Author(s) 2017. Published by Oxford University Press on behalf of The Royal Astronomical Society. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Geophysical Journal International Oxford University Press

Advanced GPR imaging of sedimentary features: integrated attribute analysis applied to sand dunes

Loading next page...
 
/lp/ou_press/advanced-gpr-imaging-of-sedimentary-features-integrated-attribute-LlHyOBDmXC
Publisher
Oxford University Press
Copyright
© The Author(s) 2017. Published by Oxford University Press on behalf of The Royal Astronomical Society.
ISSN
0956-540X
eISSN
1365-246X
D.O.I.
10.1093/gji/ggx541
Publisher site
See Article on Publisher Site

Abstract

Summary We evaluate the applicability and the effectiveness of integrated GPR attribute analysis to image the internal sedimentary features of the Piscinas Dunes, SW Sardinia, Italy. The main objective is to explore the limits of GPR techniques to study sediment-bodies geometry and to provide a non-invasive high-resolution characterization of the different subsurface domains of dune architecture. On such purpose, we exploit the high-quality Piscinas data-set to extract and test different attributes of the GPR trace. Composite displays of multi-attributes related to amplitude, frequency, similarity and textural features are displayed with overlays and RGB mixed models. A multi-attribute comparative analysis is used to characterize different radar facies to better understand the characteristics of internal reflection patterns. The results demonstrate that the proposed integrated GPR attribute analysis can provide enhanced information about the spatial distribution of sediment bodies, allowing an enhanced and more constrained data interpretation. Electromagnetic theory, Ground penetrating radar, Image processing INTRODUCTION Aeolian processes produce landscape that can be spectacular in specific regions, where the sand has apparently organized itself into structures, such as, for example, coastal dunes and other wind-related features (see e.g. Bagnold 1941; Lorenz & Zimbelman 2014). Accurate field observations are required to understand actual dune geomorphology, and to analyse sand layers that record the history of their deposition and migration. GPR is a fast, efficient and non-invasive imaging system to characterize shallow subsurface structures based on changes in the electromagnetic (EM) properties of the materials (Baker & Jol 2007; Cassidy 2009). Dunes are a suitable environment for the application of GPR techniques because they are usually made of sands that exhibit low electrical conductivity and low magnetic permeability allowing high depths of penetration (Bristow 1995; Bristow & Jol 2003; Bristow et al.2010). The successful application of GPR to dunes is due to the contrast in the relative dielectric properties, caused by moisture, porosity, grain size and mineral content (Van Dam & Schlager 2000; Bristow 2009). As GPR is non-destructive and can overcome the restrictions of limited outcrops, in the last 30 yr GPR has spread as a suitable technique to characterize shallow stratigraphy and deformation (see e.g. Jol et al.1996; Busby & Merritt 1999; Buynevich & Fitzgerald 2001; Jol & Bristow 2003) and characteristics of coastal barriers and of sand dunes (see e.g. Schenk et al.1993; Van Heteren et al.1996; Tercier et al.2000; Moore et al.2004), formed and shaped in modern and ancient fluvial, aeolian and shallow-marine sediment environments. The combination of GPR and traditional geomorphic and geological investigations allows detailed studies of sand dunes development and evolution (see e.g. Harari 1996; Bristow et al.2000a, 2005; Adetunji et al.2008; Gómez-Ortiz et al.2009; Santalla et al.2009; Clemmensen & Nielsen 2010; Guillemoteau et al.2012; Tillmann 2014). Coastal dunes build up where a number of conditions are met such as width of the beach sufficient to allowing the accumulation of wind-blown sand, prevailing winds from the sea and obstacles (e.g. vegetation and pebbles), which tend to slow down the wind and lead to the deposition of sand grains (Bigarella et al.1969; Goldsmith 1973; Hesp 1988). Storms can reduce the height of coastal dunes due to erosion. The first applications of GPR to the study of coastal dunes date back to the late 1980s (e.g. Leatherman 1987). Nielsen et al. (1995) and Clemmensen et al. (1996) studied the stratigraphy of the late Holocene eolian deposits of the west coast of Jutland (Denmark) by using GPR. The radar survey further helped calculating the sand budget of the dunefield and analysing dunefield dynamics in relation to wind, sediment supply, sea level and human influence. Several examples in literature indicate that GPR can accurately image the stratigraphy and internal sedimentary structure of coastal barriers, spits and strandplains, both above and below a fresh groundwater table (Neal & Roberts 2000; Nobes et al.2016; Rockett et al.2016). Good examples are, for example, radar surveys along the coast of Venice that did not only provide images of internal dune bedding, but also clues about the evolutional history of the coast line and the depth of transition from fresh water to brackish-salt water (Galgaro et al.2000). A detailed study of the Atlantic Galician Coast (NW Spain) through 2-D/3-D GPR surveys reveals the internal structure of depositional lobes, the pattern of which was related mainly to changes in the vegetation cover (González-Villanueva et al.2011). In southern Portugal, GPR reconstructs the conditions of a coastal segment and allows the interpretation of depositional and erosional evidences of the AD 1755 tsunami in this region (Costa et al.2016). Reconstructions of coastal process by means of GPR can further be found in Oliver & Woodroffe (2016), which provide a history of storm events for the Callala Barrier (SE Australia) from ∼4000 yr ago to present. The resolution provided by GPR images is nonetheless constrained by the depth of interest and is highest for the shallower targets (Jol & Bristow 2003). Moreover, aeolian deposits are often characterized by moderate variations in physical properties that result in small contrasts of EM impedance. Such characteristics affect the dynamic part of the radar response, that is, they are linked to low reflectivity and small amplitude GPR responses. The coastal environment is further characterized by conditions related to thin sedimentary lenses, prograding units and cross-cutting erosion surfaces. GPR can image such sedimentary features and allow the reconstruction of relative chronology, with particular reference to the identification of periods of accretion, progradation, erosion, reworking, stabilization and other major events in the dune history (e.g. Bailey & Bristow 2000; Bristow et al.2000b; McGourty & Wilson 2000; Havholm et al.2004; Flor-Blanco et al.2016; Forde et al.2016; Rockett et al.2016). Nonetheless, the information provided by conventional radar scans [i.e. amplitude versus two-way-time (TWT)] is not always sufficient to resolve low-amplitude contrasts in geometrically complex subsurface conditions. Attribute analysis has been incorporated into GPR data interpretation to extract more subsurface information from the processed GPR data-sets and to at least partly overcome the above discussed limits (e.g. Orlando 2002; Tronicke et al.2006; McClymont et al.2008; Forte et al.2012; Zhao et al.2013, 2016a,b; Ercoli et al.2015; Nobes & Jol 2015; Nobes et al.2016). In this paper, we focus on GPR attribute analysis to studyparabolic climbing sand dunes and to identify specific radar facies on the basis of the signature observed in different domains (i.e. the original Amplitude-TWT, and the calculated Attributes-TWT). The test site is located at Piscinas (SW Sardinia, Italy), an area characterized by a well-developed dune system. Radar facies are defined in terms of reflection continuity, shape, dip, internal reflection configuration and external form (Neal 2004), following the terms used in seismic stratigraphy (Mitchum et al.1977), and have been widely used to describe and classify different sedimentary structures (e.g. Van Heteren et al.1998; Van Overmeeren 1998; Beres et al.1999; Ekes & Hickin 2001; Moysey et al.2006; Franke et al.2015). We use a multi-attribute comparative analysis to study the Piscinas dunes and, in particular, to better analyse the reflection patterns and to characterize the different radar facies. Every single attribute can actually capture details related to different physical or geometrical properties of the subsurface (Chopra & Marfurt 2007) and their integration (multi-attribute analysis) can enhance the understanding of dunes stratigraphy and evolution. Moreover, we introduce composite displays (i.e. displays of multiple attributes in a single image) to highlight local continuity or variability of the GPR response and to identify and correlate specific features of the Piscinas dunefield. Study area The dunes of Piscinas, located along the Arburese coast, SW Sardinia, Italy (Fig. 1), are one of the largest and most spectacular transgressive dunefields of the Mediterranean basin. The system is formed by inactive and active parabolic bodies, developing from the present beach toward the inland. The longest moved more than 2 km inland, reaching a maximum elevation of more than 100 m. The sand dunes, composed of Quaternary lithified aeolian sandstones superimposed by coastal-aeolian loose sands (Palmerini & Ulzega 1969), are still active, modeled by the prevailing Mistral wind from the northwest and influenced by the vegetation pattern. In fact, the shape of the dunes is a non-linear function of coast exposure to the predominant winds and waves: the northern part of the dunefield is in the regressive phase, while the southern portion is transgressive, under the influence of the wind that pushes the sand mostly toward the southeast (Atzeni & Ginesu 1993). In particular, the coastal landscape of the area surrounded by green hills overlooking the sea, presents morphologies that reflect various geological contexts, where the general features of the sedimentation environment are recognized as microtidal coastal facies dominated by wave action and comprising the shoreface, the foreshore and the backshore, fluvial facies and aeolian facies (Caredda et al.1999; Costamagna & Barca 2008). Figure 1. View largeDownload slide Location map, photos and aerial picture of the study area, in the Piscinas Dunes, SW Sardinia, Italy. The orange line shows the GPR survey line having a total length of 386.5 m. Figure 1. View largeDownload slide Location map, photos and aerial picture of the study area, in the Piscinas Dunes, SW Sardinia, Italy. The orange line shows the GPR survey line having a total length of 386.5 m. Data acquisition and processing We performed a 2-D GPR survey with a ProExMalå Geoscience GPR system equipped with 250 MHz central frequency shielded antennas, which provided the optimum trade-off between depth of penetration and resolution. In this paper, we focus on an NW-SE oriented profile, because the optimum imaging conditions are met when the profile is perpendicular to the dunes (i.e. parallel to the dominant wind direction). The sector crossed by the profile includes two dune lobes and is paradigmatic of the entire dunefield (Fig. 1). We obtained 7730 traces on a profile length of 386.5 m at a constant 0.05 m sampling interval. The trace length is set to 300 ns and the sampling rate was 0.398 ns, considerably smaller than the Nyquist limit to optimize the results of attribute analysis. We applied the following GPR processing sequence: Data editing: to analyse data characteristics, identify noisy traces and anomalies (such as, e.g. zero-time drift, polarity inversions, isolated spikes and band-limited noise components). Dewow filter: to remove the low-frequency ‘wow’ component that may cover reflections. Design and application of a bandpass filter: to select the optimal band based on a signal-to-noise ratio (S/N) criterion, and the corner frequencies of the zero-phase Ormsby filter were defined after narrow-band (10 MHz) filter scans as 50–70 and 500–600 MHz. Windowed background removal: to remove horizontally coherent components primarily due to antenna-ground impedance mismatch. Amplitude analysis: to analyse amplitude decay (1-D) and evaluate lateral variations of the attenuation (2-D). True amplitude recovery: to correct for geometrical divergence and attenuation, and it requires knowledge of the velocity of radar waves in the volume to evaluate the optimal parameters to compensate for the effects of amplitude decay. Topographic correction (static correction) and depth conversion: to correct for near-surface effects. F–k migration: to focus diffracted energy and reconstruct the real position of subsurface features. A dedicated GPS survey in NRTK (Network Real Time Kinematik) mode provided digital elevation data with centimetric precision in latitude/longitude and an average elevation error of less than 4 cm. An average velocity of 0.13 m ns−1 was determined from diffraction hyperbolas analysis and utilized for topographic correction and time–depth conversion. The subsurface velocity field is rather homogeneous with maximum and minimum values respectively equal to 0.140 and 0.115 m ns−1. This is probably due to the small and rather constant amount of moisture within the dunes, which do not host any continuous aquifer, and to the relatively homogeneous characteristics of the sediments. Attribute analysis The S/N of the GPR data-set is rather high in the study area, notwithstanding the relatively low reflectivity due to the homogeneous characteristics of sediments and the small and constant moisture content (see previous comments about the velocity field). We applied GPR attribute analysis to integrate the basic interpretation of amplitude data with quantitative information about the physical property variations and to classify domains characterized by homogeneous GPR response. The choice of attribute analysis is mainly due to the locally low reflectivity and to the geometrical characteristics of the aeolian deposits. The former imply imaging difficulties through conventional amplitude analysis (or analysis of individual attributes sensitive to amplitude). The latter, related to conflicting and steep dips in hummocky cross-stratification or prograding deposits and thin or tapering out layers (i.e. at or below the resolution limits of the method) require different strategies to help the identification of volumes characterized by homogeneous response. Different attribute categories, including amplitude-/phase-/frequency-related, coherence and textural ones, have been therefore comparatively tested and evaluated. We particularly concentrate on textural attributes that enhance the identification of repeating patterns of local variations in image intensity. Such attributes were originally applied to 2-D image analysis, and have been successively introduced to GPR interpretation to improve identification of complex stratigraphic features (e.g. Zhao et al.2016b). We adopt the grey-level co-occurrence matrix (GLCM) algorithm to generate second-order statistical textural characteristics (Haralick et al.1973). A brief mathematical background about textural attributes is provided in the  Appendix, while hereafter just the definition and a short comment on each used attribute is provided. Amplitude first derivative: it is the time derivative of the instantaneous amplitude (i.e. the time rate of change of the envelope), which is in turn calculated by the Hilbert transform. Interfaces and discontinuities can be highlighted by this attribute. Instantaneous frequency: it is also calculated through the Hilbert transform and outputs, sample by sample, the local highest frequency component of the signal. It is sensitive to both wave propagation effects and depositional characteristics; hence it is a physical attribute that can be used to discriminate between materials having different intrinsic attenuation due to variable EM properties. Similarity: it is one of the coherence attributes that expresses how much two or more trace segments look alike. A similarity of 1 means the trace segments are completely identical in waveform and amplitude. A similarity of 0 means they are completely different. Low S/N segments exhibit low similarity. Assuming two trace segments u(x, y, t) and v(x, y, t) with sample length N, such attribute is defined for a specific analysis window as   \begin{equation*} 1 - \frac{{\sqrt {\sum\limits_{i = 1}^N {{{({u_i} - {v_i})}^2}} } }}{{\sqrt {\sum\limits_{i = 1}^N {u_i^2} } + \sqrt {\sum\limits_{i = 1}^N {v_i^2} } }} \end{equation*} Every single attribute can partly capture details related to different physical or geometrical properties of the subsurface. In order to highlight local continuity or variability of GPR data-sets and to identify and correlate specific sedimentary layers and interfaces, we performed integrated multi-attributes analysis through the combination of different attributes into a single image. We adopt two different schemes to display composite images: (i) one attribute of interest is plotted in colour (or in grey scale preferably) to form a background, and another attribute is plotted in variable area format and overlaid on the background attribute; (ii) each pixel in composite image is expressed by the weighted red/green/blue, representing three different attributes, and in this way three attributes can be displayed simultaneously. RESULTS Fig. 2 shows the processed GPR amplitude profile over two overlying asymmetric sand dunes. A clear reflection (indicated by the cyan line) can be easily tracked at a maximum depth of about 8 m below the ground surface, calculated from the estimated velocity of 0.13 m ns−1, corresponding to a marked unconformity possibly related to post-Tyrrhenian palaeosoil (Palmerini & Ulzega 1969), herein referred as PTP. Horizontal and dipping reflections with different angles above the PTP indicate a complex history of aeolian erosion and deposition, while the signal is totally attenuated below the PTP. Figure 2. View largeDownload slide Processed GPR amplitude profile. The cyan line indicates a marked unconformity interpreted as a post-Tyrrhenian palaeosoil (PTP), while the four zones marked by white rectangles will be analysed in detail hereafter. Figure 2. View largeDownload slide Processed GPR amplitude profile. The cyan line indicates a marked unconformity interpreted as a post-Tyrrhenian palaeosoil (PTP), while the four zones marked by white rectangles will be analysed in detail hereafter. Fig. 3 shows the attribute results calculated on the processed GPR data: amplitude first derivative, instantaneous frequency and similarity. We can note that: amplitude first derivative can identify and emphasize the main lateral changes, at the expense of weak reflections (Fig. 3a); the resolution of the instantaneous frequency profile seems a little lower compared with amplitude-related attributes, but it is more sensitive to the actual target response (i.e. low-frequency component in Fig. 3b, the value of which is close to 230 MHz); high similarity values are likely correlated with coherent reflections and low similarity values are related with more chaotic patterns and low S/N sectors of the radar section (see e.g. the sector below PTP in Fig. 3c). Fig. 4 shows two textural attributes, that is, dissimilarity and homogeneity. We can see that the textural attributes locally exhibit opposite characteristics, for example, the high-amplitude continuous reflections are characterized by relatively high dissimilarityand low homogeneity. Figure 3. View largeDownload slide The attribute results calculated on the processed GPR data. (a) Amplitude first derivative; (b) instantaneous frequency and (c) similarity. Figure 3. View largeDownload slide The attribute results calculated on the processed GPR data. (a) Amplitude first derivative; (b) instantaneous frequency and (c) similarity. Figure 4. View largeDownload slide Textural attributes calculated on the processed GPR data. (a) Dissimilarity and (b) homogeneity. Figure 4. View largeDownload slide Textural attributes calculated on the processed GPR data. (a) Dissimilarity and (b) homogeneity. Every attribute can reflect different physical or geometrical properties of the subsurface target. For example, the PTP (marked in Fig. 2) can be tracked down as a continuous reflector between 90 and 170 m in the similarity profile but exhibits different characteristics in the amplitude first derivative. In order to understand the relationships between amplitude radargrams and calculated attributes, four typical radar facies configurations are selected from the processed GPR data, considering reflection continuity, shape, dip and external form, which is also important for GPR sedimentary interpretation and definition of sedimentary facies (Reading 1996). The dimensions of all the selected image samples characterized by radar facies are 2.75 × 0.5 m (horizontal × vertical), the specific locations of which can be found in Fig. 2. They are identified as: Zone A: continuous dipping reflections, generated by sets of cross-stratification. Zone B: continuous horizontal reflections, generated by lateral sand layers. Zone C: discontinuous horizontal reflections, generated by high-angle discontinuities. Zone D: reflection-free, generated by massive homogeneous units with low S/N. Fig. 5 shows the results of four attributes (plus amplitude) and their corresponding histograms calculated on the selected zones, and Table 1 summarizes the qualitative analysis of histogram distributions of attributes calculated from the four typical radar facies from Fig. 5. We can see that different radar facies are characterized by different attribute patterns. All the attributes maintain continuous characteristics for continuous reflections, while they have lateral variations when the reflection is discontinuous. Figure 5. View largeDownload slide Four typical radar facies, the corresponding attributes and their corresponding histograms. All the analysed portions have dimensions equal to 2.75 × 0.5 m, and their specific locations can be found in Fig. 2. Figure 5. View largeDownload slide Four typical radar facies, the corresponding attributes and their corresponding histograms. All the analysed portions have dimensions equal to 2.75 × 0.5 m, and their specific locations can be found in Fig. 2. Table 1. Results of the histogram analysis of attributes calculated from the four typical radar facies from Fig. 5. Zone  Processed GPR data  Amplitude first derivative  Instantaneous frequency  Similarity  Homogeneity  Zone A  Low  Low  240 MHz concentrated  High  Medium  Zone B  High  High  240 MHz concentrated  High  Medium  Zone C  Medium  Medium  220 MHz concentrated  Mixed  Low  Zone D  Very low  Very low  Dispersed  Dispersed  High  Zone  Processed GPR data  Amplitude first derivative  Instantaneous frequency  Similarity  Homogeneity  Zone A  Low  Low  240 MHz concentrated  High  Medium  Zone B  High  High  240 MHz concentrated  High  Medium  Zone C  Medium  Medium  220 MHz concentrated  Mixed  Low  Zone D  Very low  Very low  Dispersed  Dispersed  High  View Large Amplitude-related attributes can measure the overall reflectivity and recognize major changes in sediments. Frequency is very sensitive to attenuation, but there are no significant frequency variations in Zones A, B and C. Zone D exhibits different and more dispersed frequency characteristics and the histogram accordingly shows a wider distribution. The patterns of instantaneous frequency of continuous reflections (i.e. Zones A and B) are similar to the amplitude first derivative. A clear correlation is lacking for discontinuous reflections (Zone C) and a sort of blurring shows up in both displays (frequency/Amplitude first derivative). Blurring is the main characteristics for all attributes in diffuse scattering and low S/N zones (i.e. Zone D). Similarity is very sensitive to continuous reflections, particularly to horizontal ones—as in the case of Zone B, which shows a concentrated distribution in the histogram. The textural attribute homogeneity can reflect the overall smoothness in image intensity, and it seems a more stable parameter, in comparison to other attributes, when reflectors exhibit a laterally constant response. This is not only related to the presence of fractures/faults, but also to lateral variations of attenuation in the overburden. Fig. 6 provides various composite displays of combined attributes. Different reflection patterns within specific zones, associated with different sediment characteristics can be highlighted: major reflections can be recognized in any composite image, including amplitude-related attributes, such as the PTP marked with cyan lines in Figs 6(a) and (c), interpreted as the base of the present depositional aeolian system; two continuous dipping reflection zones are emphasized in Fig. 6(b), which is an attribute combination, without considering any amplitude-related attribute; two irregular sand lenses can be displayed in Figs 6(b) and (c), within the white polygons, and can be associated with more homogeneous sand units having a grain size slightly different from the surrounding materials. Figure 6. View largeDownload slide Example of composite multi-attributes displays. (a) Homogeneity overlaid on amplitude first derivative; (b) dissimilarity overlaid on similarity; (c) homogeneity, amplitude first derivative and instantaneous frequency mixed with RGB model. Figure 6. View largeDownload slide Example of composite multi-attributes displays. (a) Homogeneity overlaid on amplitude first derivative; (b) dissimilarity overlaid on similarity; (c) homogeneity, amplitude first derivative and instantaneous frequency mixed with RGB model. DISCUSSIONS AND CONCLUSIONS GPR can effectively image complex shallow sedimentary features in parabolic climbing sand dunes. The integrated analysis of different attributes of the radar trace may offer enhanced opportunities of understanding the spatial distribution of sedimentary features and their characteristics. Our study shows that attribute analysis can successfully integrate the basic amplitude interpretation and allow constrained studies of lateral continuity and variability of sediments. Every attribute can partly reflect different physical or geometrical properties of the subsurface target at different resolution levels and, in some cases, at resolution lower than the original amplitude data (see e.g. Fig. 5). Some attributes, such as, for example, similarity, are more sensitive to lateral continuity regardless of the amplitude of the radar response (see also the relevant Zones A and B histograms in Fig. 5). This is particularly important in the study of aeolian deposits and coastal dune systems that may be characterized by moderate variations of physical properties and by consequently small EM impedance contrasts. The interpretation of the resulting low-reflectivity records can greatly benefit from the integration provided by attributes that are less sensitive to the amplitude factor. Moreover, the combination of several attributes into a single image, namely a multi-attribute composite display, can visually highlight groups of distinct reflection patterns, and enhance the identification of volumes characterized by homogeneous response. A GPR homogeneity, lato sensu, that is, inferred from the integrated analysis of the different attributes, can be linked, in space, to peculiar depositional environments and, in time, to different phases in the development of coastal geomorphology (Fig. 6). In fact, by using integrated attributes and in particular combined textural ones, we can attain not only a clearer subsurface image but even more important an enhanced discrimination between different sedimentological facies/domains. Some attributes are particularly sensitive on such purpose, such as, for example, similarity, which is related to reflector's continuity and stability of response. Localized variations of physical properties and sedimentological characteristics (e.g. grain size and arrangement) can be responsible for large similarity oscillations that are easily identified and can be effectively correlated to sedimentological facies changes. The relationships among different attributes and their corresponding sedimentological and radar facies deserve further study. In particular, the study of the distributions of the attributes (see Fig. 5 and Table 1) should be tested on different case studies to evaluate its consistency as an estimator and the possibility to use it as a classification tool of sediment bodies and aeolian landforms. This study is limited to a single test site and leaves such issue open. Nonetheless, the GPR study of the Piscinas dunes indicates that the combination among GPR attributes, ground truthing and results of laboratory analysis, can establish a new route towards a more complete non-invasive sedimentological interpretation. ACKNOWLEDGEMENTS This study has been carried out within the RITMARE Flagship Project, coordinated by CNR and funded by the Italian Ministry of University and Research. We gratefully acknowledge the support of the International Centre for Theoretical Physics (ICTP, Trieste, Italy) Training on Research in Italian Laboratories (TRIL) programme, which sponsored the scholarship of the first author. We thank Halliburton Landmark for the license of the SeisSpace ProMAX Seismic Processing Software provided through an academic grant and dGB Earth Sciences for the OpendTect open source seismic data analysis software. A. Bezzi, G. Casagrande, D. Martinucci and S. Pillon are kindly acknowledged for their help in the topographic survey and the GPR data acquisition. REFERENCES Adetunji A.Q., Al-Shuhail A., Korvin G., 2008. Mapping the internal structure of sand dunes with GPR: a case history from the Jafurah sand sea of eastern Saudi Arabia, Leading Edge , 27( 11), 1446– 1452. https://doi.org/10.1190/1.3011016 Google Scholar CrossRef Search ADS   Atzeni A., Ginesu S., 1993. Evoluzione dei litorali della Sardegna e interventi di riequilibrio, La difesa dei litorali Ital. , 34, 215– 231. Bagnold R.A., 1941. The Physics of Blown Sand and Desert Dunes . Methuen, London, 265 pp. Bailey S., Bristow C.S., 2000. Structure of coastal dunes: observations from ground penetrating radar (GPR) surveys, in 8th International Conference on Ground Penetrating Radar , pp. 660– 665. SPIE, University of Queensland, Australia. Baker G.S., Jol H.M., Eds., 2007. Stratigraphic Analyses using GPR, Special Paper . Geological Society of America, Boulder, CO. Beres M., Huggenberger P., Green A.G., Horstmeyer H., 1999. Using two- and three-dimensional georadar methods to characterize glaciofluvial architecture, Sediment. Geol. , 129( 1–2), 1– 24. https://doi.org/10.1016/S0037-0738(99)00053-6 Google Scholar CrossRef Search ADS   Bigarella J.J., Becker R.D., Duarte G.M., 1969. Coastal dune structures from Paraná (Brazil), Mar. Geol. , 7( 1), 5– 55. https://doi.org/10.1016/0025-3227(69)90002-4 Google Scholar CrossRef Search ADS   Bristow C., 1995. Facies analysis in the Lower Greensand using ground-penetrating radar, J. geol. Soc. , 152( 4), 591– 598. https://doi.org/10.1144/gsjgs.152.4.0591 Google Scholar CrossRef Search ADS   Bristow C.S., 2009. Ground penetrating radar in aeolian dune sands, in Ground Penetrating Radar: Theory and Applications , pp. 273– 297. ed. Jol H.M., Elsevier, Amsterdam, The Netherlands. Google Scholar CrossRef Search ADS   Bristow C.S., Augustinus P.C., Wallis I.C., Jol H.M., Rhodes E.J., 2010. Investigation of the age and migration of reversing dunes in Antarctica using GPR and OSL, with implications for GPR on Mars, Earth planet. Sci. Lett. , 289( 1–2), 30– 42. https://doi.org/10.1016/j.epsl.2009.10.026 Google Scholar CrossRef Search ADS   Bristow C.S., Bailey S.D., Lancaster N., 2000a. The sedimentary structure of linear sand dunes, Nature , 406( 6791), 56– 59. https://doi.org/10.1038/35017536 Google Scholar CrossRef Search ADS   Bristow C.S., Chroston P.N., Bailey S.D., 2000b. The structure and development of foredunes on a locally prograding coast: insights from ground-penetrating radar surveys, Norfolk, UK, Sedimentology , 47( 5), 923– 944. https://doi.org/10.1046/j.1365-3091.2000.00330.x Google Scholar CrossRef Search ADS   Bristow C.S., Jol H.M., 2003. An introduction to ground penetrating radar (GPR) in sediments, Geol. Soc. Lond. Spec. Publ. , 211( 1), pp. 1– 7. https://doi.org/10.1144/GSL.SP.2001.211.01.01 Google Scholar CrossRef Search ADS   Bristow C.S., Lancaster N., Duller G.A.T., 2005. Combining ground penetrating radar surveys and optical dating to determine dune migration in Namibia, J. geol. Soc. , 162( 2), 315– 321. https://doi.org/10.1144/0016-764903-120 Google Scholar CrossRef Search ADS   Busby J., Merritt J., 1999. Quaternary deformation mapping with ground penetrating radar, J. appl. geophys. , 41( 1), 75– 91. https://doi.org/10.1016/S0926-9851(98)00050-0 Google Scholar CrossRef Search ADS   Buynevich I.V., Fitzgerald D.M., 2001. Styles of coastal progradation revealed in subsurface records of paraglacial barriers: Duxbury, Massachusetts, USA. J. Coast. Res. , 34, 194– 208. Caredda A.M., Cristini A., Ferrara C., Lobina M.F., Baroli M., 1999. Distribution of heavy metals in the Piscinas beach sediments (SW Sardinia, Italy), Environ. Geol. , 38( 2), 91– 100. https://doi.org/10.1007/s002540050405 Google Scholar CrossRef Search ADS   Cassidy N.J., 2009. Ground penetrating radar data processing, modelling and analysis, in Ground Penetrating Radar: Theory and Applications , pp. 141– 176, ed. Jol H.M., Elsevier, Amsterdam, The Netherlands. Chopra S., Marfurt K.J., 2007. Seismic Attributes for Prospect Identification and Reservoir Characterization . SEG/EAGE, 464 pp. Google Scholar CrossRef Search ADS   Clemmensen L.B., Andreasen F., Nielsen S.T., Sten E., 1996. The late Holocene coastal dunefield at Vejers, Denmark: characteristics, sand budget and depositional dynamics. Geomorphology , 17( 1–3), 79– 98. https://doi.org/10.1016/0169-555X(95)00096-N Google Scholar CrossRef Search ADS   Clemmensen L.B., Nielsen L., 2010. Internal architecture of a raised beach ridge system (Anholt, Denmark) resolved by ground-penetrating radar investigations, Sediment. Geol. , 223( 3–4), 281– 290. https://doi.org/10.1016/j.sedgeo.2009.11.014 Google Scholar CrossRef Search ADS   Costa P.J. et al.  , 2016. How did the AD 1755 tsunami impact on sand barriers across the southern coast of Portugal?, Geomorphology , 268, 296– 311. https://doi.org/10.1016/j.geomorph.2016.06.019 Google Scholar CrossRef Search ADS   Costamagna L.G., Barca S., 2008. Depositional architecture and sedimentology of the Tuppa Niedda Conglomerates (Late Carboniferous, Arburese, SW Sardinia, Italy), Boll. Soc. Geol. Ital. , 127, 625– 636. Eichkitz C.G., Davies J., Amtmann J., Schreilechner M.G., de Groot P., 2015. Grey level co-occurrence matrix and its application to seismic data, First Break , 33, 71– 77. Ekes C., Hickin E.J., 2001. Ground penetrating radar facies of the paraglacial Cheekye Fan, southwestern British Columbia, Canada, Sediment. Geol. , 143( 3–4), 199– 217. https://doi.org/10.1016/S0037-0738(01)00059-8 Google Scholar CrossRef Search ADS   Ercoli M., Pauselli C., Cinti F.R., Forte E., Volpe R. 2015. Imaging of an active fault: comparison between 3D GPR data and outcrops at the Castrovillari fault, Calabria, Italy. Interpretation , 3( 3), SY57– SY66. https://doi.org/10.1190/INT-2014-0234.1 Google Scholar CrossRef Search ADS   Flor-Blanco G., Rubio-Melendi D., Flor G., Fernández-Álvarez J.P., Jackson D.W.T., 2016. Holocene evolution of the Xagó dune field (Asturias, NW Spain) reconstructed by means of morphological mapping and ground penetrating radar surveys. Geo-Mar. Lett. , 36( 1), 35– 50. https://doi.org/10.1007/s00367-015-0427-1 Google Scholar CrossRef Search ADS   Forde T.C., Nedimović M.R., Gibling M.R., Forbes D.L., 2016. Coastal evolution over the past 3000 years at Conrads Beach, Nova Scotia: the influence of local sediment supply on a paraglacial transgressive system. Estuar. Coast , 39( 2), 363– 384. https://doi.org/10.1007/s12237-015-0016-6 Google Scholar CrossRef Search ADS   Forte E., Pipan M., Casabianca D., Di Cuia R., Riva A., 2012. Imaging and characterization of a carbonate hydrocarbon reservoir analogue using GPR attributes. J. appl. geophys. , 81, 76– 87. https://doi.org/10.1016/j.jappgeo.2011.09.009 Google Scholar CrossRef Search ADS   Franke D., Hornung J., Hinderer M., 2015. A combined study of radar facies, lithofacies and three-dimensional architecture of an alpine alluvial fan (Illgraben fan, Switzerland), Sedimentology , 62( 1), 57– 86. https://doi.org/10.1111/sed.12139 Google Scholar CrossRef Search ADS   Galgaro A., Finzi E., Tosi L., 2000. An experiment on a sand-dune environment in Southern Venetian coast based on GPR, VES and documentary evidence, Ann. Geophys. , 43( 2), 289– 295. Goldsmith V., 1973. Internal geometry and origin of vegetated coastal sand dunes, J. Sediment. Res. , 43( 4), 1128– 1143. Gómez-Ortiz D., Martín-Crespo T., Rodríguez I., Sánchez M.J., Montoya I., 2009. The internal structure of modern barchan dunes of the Ebro River Delta (Spain) from ground penetrating radar, J. appl. geophys. , 68( 2), 159– 170. https://doi.org/10.1016/j.jappgeo.2008.11.007 Google Scholar CrossRef Search ADS   González-Villanueva R., Costas S., Duarte H., Pérez-Arlucea M., Alejo I., 2011. Blowout evolution in a coastal dune: using GPR, aerial imagery and core records, J. Coast. Res. , 64, 278– 282. Guillemoteau J., Bano M., Dujardin J.R., 2012. Influence of grain size, shape and compaction on georadar waves: examples of aeolian dunes, Geophys. J. Int. , 190( 3), 1455– 1463. https://doi.org/10.1111/j.1365-246X.2012.05577.x Google Scholar CrossRef Search ADS   Haralick R.M., Shanmugam K., Dinstein I.H., 1973. Textural features for image classification. IEEE Trans. Syst. Man Cybern. , 6( 6), 610– 621. https://doi.org/10.1109/TSMC.1973.4309314 Google Scholar CrossRef Search ADS   Harari Z., 1996. Ground-penetrating radar (GPR) for imaging stratigraphic features and groundwater in sand dunes, J. appl. geophys. , 36( 1), 43– 52. https://doi.org/10.1016/S0926-9851(96)00031-6 Google Scholar CrossRef Search ADS   Havholm K.G. et al.  , 2004. Stratigraphy of back-barrier coastal dunes, northern North Carolina and southern Virginia, J. Coast. Res. , 20( 4), 980– 999. https://doi.org/10.2112/03503A2.1 Google Scholar CrossRef Search ADS   Hesp P., 1988. Morphology, dynamics and internal stratification of some established foredunes in southeast Australia, Sediment. Geol. , 55( 1–2), 17– 41. https://doi.org/10.1016/0037-0738(88)90088-7 Google Scholar CrossRef Search ADS   Jol H.M., Smith D.G., Meyers R.A., 1996. Digital ground penetrating radar (GPR): a new geophysical tool for coastal barrier research (Examples from the Atlantic, Gulf and Pacific Coasts, U.S.A.), J. Coast. Res. , 12, 960– 968. Jol H.M., Bristow C.S., 2003. GPR in sediments: advice on data collection, basic processing and interpretation, a good practice guide. Geol. Soc., Lond. Spec. Publ. , 211, 9– 27. Google Scholar CrossRef Search ADS   Leatherman S.P., 1987. Coastal geomorphological applications of ground-penetrating radar, J. Coast. Res. , 3( 3), 397– 399. Lorenz R.D., Zimbelman J.R., 2014. Dune Worlds. Springer Praxis Books, Berlin/Heidelberg, 308 pp. Love P.L., Simaan M., 1985. Segmentation of a seismic section using image processing and artificial intelligence techniques, Pattern Recog. , 18( 6), 409– 419. https://doi.org/10.1016/0031-3203(85)90011-1 Google Scholar CrossRef Search ADS   McClymont A.F. et al.  , 2008. Visualization of active faults using geometric attributes of 3D GPR data: an example from the Alpine Fault Zone, New Zealand, Geophysics , 73( 2), B11– B23. https://doi.org/10.1190/1.2825408 Google Scholar CrossRef Search ADS   McGourty J., Wilson P., 2000. Investigating the internal structure of Holocene coastal sand dunes using ground-penetrating radar: example from the north coast of Northern Ireland, Proceedings of SPIE–The International Society for Optical Engineering, 4084 , pp. 14– 19. Mitchum R.M. Jr, Vail P.R., Sangree J.B., 1977. Seismic stratigraphy and global changes of sea level, part 6: stratigraphic Interpretation of seismic reflection patterns in depositional sequences, Geophys. Res. Lett. , 165( 22), 117– 133. Moore L.J., Jol H.M., Kruse S., Vanderburgh S., Kaminsky G.M., 2004. Annual Layers Revealed by GPR in the Subsurface of a Prograding Coastal Barrier, Southwest Washington, U.S.A., J. Sediment. Res. , 74( 5), 690– 696. https://doi.org/10.1306/021604740690 Google Scholar CrossRef Search ADS   Moysey S., Knight R.J., Jol H.M., 2006. Texture-based classification of ground-penetrating radar images, Geophysics , 71( 6), K111– K118. https://doi.org/10.1190/1.2356114 Google Scholar CrossRef Search ADS   Neal A., 2004. Ground-penetrating radar and its use in sedimentology: principles, problems and progress. Earth-Sci. Rev. , 66( 3–4), 261– 330. https://doi.org/10.1016/j.earscirev.2004.01.004 Google Scholar CrossRef Search ADS   Neal A., Roberts C.L., 2000. Applications of ground-penetrating radar (GPR) to sedimentological, geomorphological and geoarchaeological studies in coastal environments, in Pye K, Allen J.R.L. eds, Coastal and Estuarine Environments . Sedimentology, Geomorphology and Geoarchaeology , Vol. 175, pp. 139– 171. Geological Society Special Publication. Google Scholar CrossRef Search ADS   Nielsen S.T., Clemmensen L.B., Andreasen F., 1995. The middle and late Holocene barrier spit system at Vejers, Denmark: structure and development, Bull. geol. Soc. Denmark , 42( 1), 105– 119. Nobes D.C., Jol H.M., 2015. Enhancing form and structure: complex attributes as aids for ground penetrating radar interpretation, in Near-Surface Asia Pacific Conference ,, Waikoloa, Hawaii, pp. 312– 315, Society of Exploration Geophysicists, Australian Society of Exploration Geophysicists, Chinese Geophysical Society, Korean Society of Earth and Exploration Geophysicists, and Society of Exploration Geophysicists of Japan. Nobes D.C., Jol H.M., Duffy B., 2016. Geophysical imaging of disrupted coastal dune stratigraphy and possible mechanisms, Haast, South Westland, New Zealand, N. Z. J. Geol. Geophys. , 59( 3), 426– 435. https://doi.org/10.1080/00288306.2016.1168455 Google Scholar CrossRef Search ADS   Oliver T.S., Woodroffe C.D., 2016. Chronology, morphology and GPR-imaged internal structure of the Callala Beach prograded barrier in Southeastern Australia, J. Coast. Res. , 75( suppl.1), 318– 322. https://doi.org/10.2112/SI75-064.1 Google Scholar CrossRef Search ADS   Orlando L., 2002. Detection and analysis of LNAPL using the instantaneous amplitude and frequency of ground-penetrating radar data, Geophys Prospect. , 50( 1), 27– 41. https://doi.org/10.1046/j.1365-2478.2002.00288.x Google Scholar CrossRef Search ADS   Palmerini V., Ulzega A., 1969. Sedimentologia e Geomorfologia del settore costiero tra la foce del Rio Piscinas e Capo Pecora (Sardegna sud-occidentale). Rendiconti del Seminario della Facoltà di Scienze dell’Università di Cagliari , Vol. 39, pp. 1– 38. Reading H.G., 1996. Sedimentary Environments: Processes, Facies and Stratigraphy . Blackwell Publishing, Oxford, 687 pp. Rockett G.C., Barboza E.G., Rosa M.L.C., 2016. Ground penetrating radar applied to the characterization of the Itapeva Dunefield, Torres, Brazil. J. Coast. Res. , 75 ( suppl. 1), 323– 327. https://doi.org/10.2112/SI75-065.1 Google Scholar CrossRef Search ADS   Santalla I.R., García M.J.S., Montes I.M., Ortiz D.G., Crespo T.M., Raventos J.S., 2009. Internal structure of the aeolian sand dunes of El Fangar spit, Ebro Delta (Tarragona, Spain). Geomorphology , 104( 3–4), 238– 252. https://doi.org/10.1016/j.geomorph.2008.08.017 Google Scholar CrossRef Search ADS   Schenk C.J., Gautier D.L., Olhoeft G.R., Lucius J.E., 1993. Internal structure of an aeolian dune using ground-penetrating radar, in Pye K, Lancaster N., eds, Aeolian Sediments , pp. 61– 69. Blackwell Publishing Ltd., Oxford, UK. Google Scholar CrossRef Search ADS   Soh L.-K., Tsatsoulis C., 1999. Texture analysis of SAR sea ice imagery using gray level co-occurrence matrices, IEEE Trans. Geosci. Remote Sens. , 37( 2), 780– 795. https://doi.org/10.1109/36.752194 Google Scholar CrossRef Search ADS   Tercier P., Knight R., Jol H., 2000. A comparison of the correlation structure in GPR images of deltaic and barrier?spit depositional environments, Geophysics , 65( 4), 1142– 1153. https://doi.org/10.1190/1.1444807 Google Scholar CrossRef Search ADS   Tillmann T., 2014. Landscape development of Amrum's west coast (Southern North Sea): GPR and sedimentology, in 15th International Conference on Ground Penetrating Radar (GPR) , pp. 250– 254, IEEE. Tronicke J., Villamor P., Green A.G., 2006. Detailed shallow geometry and vertical displacement estimates of the Maleme Fault Zone, New Zealand, using 2D and 3D georadar. Near Surf. Geophys. , 4( 3), 155– 161. Van Dam R.L., Schlager W., 2000. Identifying causes of ground-penetrating radar reflections using time-domain reflectometry and sedimentological analyses, Sedimentology , 47( 2), 435– 449. https://doi.org/10.1046/j.1365-3091.2000.00304.x Google Scholar CrossRef Search ADS   Van Heteren S., Fitzgerald D.M., Barber D.C., Kelley J.T., Belknap D.F., 1996. Volumetric Analysis of a New England Barrier System Using Ground-Penetrating-Radar and Coring Techniques, J. Geol. , 104( 4), 471– 483. https://doi.org/10.1086/629840 Google Scholar CrossRef Search ADS   Van Heteren S.V., Fitzgerald D.M., Mckinlay P.A., Buynevich I.V., 1998. Radar facies of paraglacial barrier systems: coastal New England, USA, Sedimentology , 45( 1), 181– 200. https://doi.org/10.1046/j.1365-3091.1998.00150.x Google Scholar CrossRef Search ADS   Van Overmeeren R.A., 1998. Radar facies of unconsolidated sediments in The Netherlands: a radar stratigraphy interpretation method for hydrogeology, J. appl. geophys. , 40( 1–3), 1– 18. https://doi.org/10.1016/S0926-9851(97)00033-5 Google Scholar CrossRef Search ADS   West B.P., May S.R., Eastwood J.E., Rossen C., 2002. Interactive seismic facies classification using textural attributes and neural networks, Leading Edge , 21( 10), 1042– 1049. https://doi.org/10.1190/1.1518444 Google Scholar CrossRef Search ADS   Zhao W., Forte E., Colucci R.R., Pipan M., 2016a. High-resolution glacier imaging and characterization by means of GPR attribute analysis, Geophys. J. Int. , 206( 2), 1366– 1374. https://doi.org/10.1093/gji/ggw208 Google Scholar CrossRef Search ADS   Zhao W., Forte E., Pipan M., 2016b. Texture attribute analysis of GPR data for archaeological prospection, Pure appl. Geophys. , 173( 8), 2737– 2751. https://doi.org/10.1007/s00024-016-1355-3 Google Scholar CrossRef Search ADS   Zhao W., Forte E., Pipan M., Tian G., 2013. Ground penetrating radar (GPR) attribute analysis for archaeological prospection, J. appl. geophys. , 97, 107– 117. https://doi.org/10.1016/j.jappgeo.2013.04.010 Google Scholar CrossRef Search ADS   APPENDIX: TEXTURAL ATTRIBUTES Textural analysis considers an ensemble of traces as an image with the goal to mathematically describe and quantify the distribution of values (pixels) within a portion of data. This means computing the spatial organization of reflections (or any other event) in terms of their continuity, smoothness, coherence, extracting and highlighting the global ‘signature’ of the analysed region. Following Haralick et al. (1973), a 2-D digitized image (I), including in this category any seismic or GPR profile, is simply a 2-D array or ordered values referred to as time (usually a two-way traveltime) and spatial samples (or pixels/voxels), when considering the two image axes. If R(x) = {1, 2, …, Nx} and C(y) = {1, 2, …, Ny} are the x and y spatial domains (often reported as rows and columns, respectively), then R(x) × C(y) is the set of resolution cells and (I) is a function assigning grey-tone values G ∈ {1, 2, …, Ng} to each singular resolution cell. (G) is called GLCM and it is a measure of how often different combinations of pixel amplitude values occur in (I) being Ng the quantized grey tones. Because typically two samples are compared, GLCM is referred to as a second-order texture classification method, but the previous consideration is still valid for any data dimension. We can assume that the texture information in (I) is represented by the ‘global’ spatial relationships among grey levels. So, this information is specified by a matrix of the relative frequencies (i.e. probabilities) Pi, j of two neighbouring resolution cells, separated by a distance d and having one grey tone i and the other grey tone j. Pi, j matrices are therefore function of the distance between neighbouring resolution cells and angular relationships between them. If angles ϑ are quantized by 45° steps, there are horizontal and vertical GLCM directions analysis, respectively, for angles equal to 0° and 90° and two diagonal directions for angles equal to 45° and 135°. So it is possible to calculate four relative frequency matrices P(i, j, d, ϑ) for the previous described angle values. If N ∈ {R(x) × C(y)} is the number of elements in the considered set defined by N[(k, l); (m, n)], then P(i, j, d, ϑ) for ϑ = 0°; 90°; 45°, 135° are respectively:   \begin{eqnarray*} P \left( {0^\circ } \right) &=& \ N\{ ( {k,l;m,n} )\ \times ( {R( x ) \times C( y)} )| {k - m}\nonumber\\ & =& { 0;\ | {l - n} | = \ d;I ( {k,l} ) = \ i,I ( {m,n} ) = \ j} \} \end{eqnarray*}   \begin{eqnarray*} P ( {90^\circ } ) &=& \ N\{ ( {k,l;m,n} ) \times ( {R( x ) \times C( y )} )| | {k - m}|\nonumber\\ &=& d;l - n\ = \ 0;I ( {k,l} ) = \ i,I ( {m,n} ) = \ j \} \end{eqnarray*}   \begin{eqnarray*} &&{P ( {45^\circ } ) = \ N\Bigg\{ ( {k,l;m,n} ) \times ( {R( x ) \times C( y )} )\Bigg|}\\ &&{\begin{array}{@{}*{1}{c}@{}} {( {k - m\ = \ d;l - n\ = \ - d} )}\\ \quad{{\rm or}\ ( {k - m\ = \ - d;l - n\ = \ d} );I ( {k,l} ) = \ i,I ( {m,n} ) = \ j} \end{array}} \Bigg\} \end{eqnarray*}   \begin{eqnarray*} &&{P ( {135^\circ } ) = N\Bigg\{ ( {k,l;m,n} ) \times ( {R( x ) \times C( y )} )\Bigg|}\\ &&{\begin{array}{@{}*{1}{c}@{}} {( {k - m\ = \ d;l - n\ = \ d} )}\\ \quad{{\rm {or}}\ ( {k - m\ = \ - d;l - n\ = \ - d} );I ( {k,l} ) = \ i,I ( {m,n} ) = \ j} \end{array} \Bigg\}} \end{eqnarray*} From the initial assumption that all the texture information of (I) is contained in the GLCM, it is possible to statistically derive several textural attributes. Haralick et al. (1973) proposed 14 attributes, while Soh & Tsatsoulis (1999) developed 10 additional quantities. All these attributes are usually assigned in the central point of each analysis running window, and repeating the calculation for all the subregion selected it is possible to obtain an attribute image (I'), R(x) × C(y) wide. The mostly used attributes for seismic data analysis are defined by the following equations, in which Ng is the number of distinct grey levels considered in the quantized image and the GLCM are normalized. For each of them, we also give a brief qualitative description and the maximum possible range (Rg).   \begin{equation*} Contrast\!:{C_{i,j}} = \mathop \sum \limits_{i,j}^{{N_g}} {\left| {i,j} \right|^2}\ p\left( {i,j} \right) \end{equation*} It returns a measurement if the intensity contrast between a pixel and its neighbour; (Rg) = [0, ((R(x) × C(y)) − 1)2]. Ci, j is 0 for a constant amplitude image.   \begin{equation*} Energy\!:{\rm{\ }}{E_{i,j}} = \mathop \sum \limits_{i,j}^{{N_g}} p{\left( {i,j} \right)^2} \end{equation*} (Rg) = [0, 1]. Ei, j is 1 for a constant amplitude image.   \begin{equation*} Homogeneity\!:{H_{i,j}} = \mathop \sum \limits_{i,j}^{{N_g}} \frac{{p\left( {i,j} \right)}}{{1 + \left| {i - j} \right|}} \end{equation*} It returns a measurement of the closeness of the distribution of the elements in GLCM as compared with the GLCM diagonal; (Rg) = [0, 1]. Hi, j is 1 for a diagonal GLCM.   \begin{equation*} Dissimilarity\!:{D_{i,j}} = \mathop \sum \limits_{i,j}^{{N_g}} \left| {i - j} \right|\ p\left( {i,j} \right) \end{equation*} It is a measurement that defines the variation of grey-level pairs in an image. It is the closest to Ci, j with adifference in the weight because Ci, j unlike Di, j grows quadratically. (Rg) = [0, 1].   \begin{equation*} Entropy\!:{T_{i,j}} = \mathop \sum \limits_{i,j}^{{N_g}} p\left( {i,j} \right){\rm{log}}\ p\left( {i,j} \right) \end{equation*} It is a measurement of spatial disorder. A completely random distribution would have very high entropy, while a constant amplitude image would have an entropy value of 0. Seismic textural attributes were first introduced into geophysical exploration by Love &Simaan (1985), which fist highlighted some of the possible advantages for data analysis and interpretation. For seismic and GPR data, the number of grey levels typically considered is between 4 Bits (i.e. 16 levels) and 8 Bits (i.e. 256 levels, West et al.2002; Eichkitz et al.2015), but more accurate (and time-consuming) analysis can be performed with higher sampled dynamic ranges. The previous calculations refer to amplitude seismic (or GPR) section, however it is possible to apply all the algorithms to virtually any input quantity including, for instance, smoothed versions of the original data, frequency- or phase-related attributes, data obtained after the application of edge detection techniques, and so on. © The Author(s) 2017. Published by Oxford University Press on behalf of The Royal Astronomical Society.

Journal

Geophysical Journal InternationalOxford University Press

Published: Apr 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off