Introduction

Grape pips are highly polymorphic and have a fundamental role for the taxonomic study within the genus Vitis L. (Rivera et al. 2007) for the distribution and domestication processes of the wild grapevine, as many archaeological discoveries suggest (This et al. 2006). The genotypical heterozygosis and the two reproduction strategies, sexual and clonal, in particular cross pollination by wind for V. vinifera L. ssp. sylvestris (C.C. Gmel) Hegi and self-pollination for V. vinifera L. ssp. vinifera (hereafter simply called V. sylvestris and V. vinifera, respectively), guarantee new combinations of parental alleles and consequently phenotypic variations (This et al. 2006).

The search for and selection of particular phenotypes have been the basis of the domestication process of the wild grapevine, involving, over the years, radical changes both in the biology of grapes, as well as bunch and grape dimensions, and sugar content (Arroyo-García et al. 2006), and in their reproductive system, guaranteeing high production from every individual (Grassi et al. 2003). Also, the seeds were subjected to important modifications due to the domestication processes. The seeds of wild species are small, robust and with a rounded outline or cordate, with short stalks and a flat ventral side with sharp angles and a strongly developed chalaza, while those of cultivated species are large, elongated, oval or pyriform with an elongated stalk (Mangafa and Kotsakis 1996). Nevertheless, many factors determine the shape of the seeds, for instance the number of seeds in each grape, the size of the grape and its ripening (Jacquat and Martinoli 1999; Rivera et al. 2007).

In many archaeobotanical studies, the taxonomic attribution of diaspores has been done by simple morphological observation and comparison with ex situ collections of the seeds. For example, the seeds found in the Roman site of Pamplona in northern Spain, dated back to a.d. 100–300 have been identified as V. vinifera on the basis of optical stereomicroscopic observation (Peña-Chocarro and Zapata-Peña 1996). The seeds found during the palaeobotanical and sedimentological studies in the open-cast lignite mines of Hambach in western Germany, and datable to the late Pliocene, were identified as V. sylvestris on the basis of visual observations (Heumann and Litt 2002), as well as the single example from the archaeological excavations carried out in Monte Trabocchetto, northern Italy, referable to the early Iron Age (Arobba et al. 2003). Also, the Vitis seeds from the archaeological sites of the river Struma in southwest Bulgaria, dated to the early Neolithic, early-middle Neolithic and late Bronze Age, were identified by visual comparison (Popova and Marinova 2007), as well as those from the late Neolithic and Eneolithic or Copper Age site of Hocevaria in Slovenia, were identified as V. sylvestris (Jeraj et al. 2009).

In Sardinia numerous seeds have been recovered at archaeological sites, with various degrees of preservation, with fragmentation and/or carbonization of the teguments. At the Duos Nuraghes site two different kinds of seeds were found, identified as Vitis. Some of them have been described as squat types, with short stalks, typical of V. sylvestris and datable to the late Bronze Age (1400–1000 b.c.), while others showed similarities with those found near the Genna Maria site in central Sardinia and referable to the Iron Age (1000–600 b.c.) (Bakels 2002). Moreover, many more seeds of Vitis are known from Sardinia, for example from the Nuraghe Ortu Còmidu, in the central south of the island and from the Nuraghe Toscono in central Sardinia, both dating to the Punic period, others dating to the Roman period (300 b.c.–a.d. 500) (Bakels 2002). On the occasion of all these archaeological finds, empirical methods were used to identify the seeds, making a comparison or quantitative evaluation difficult if not impossible.

During a recent archaeological excavation in April 2008 close to S’Arrieddu and the village of Cabras, Provincia di Oristano, rich deposits of the pre-Nuragic and Nuragic settlement of Sa Osa were recovered (Usai 2011). Within the site, a structure (Shaft N) was identified and excavated in sandstone sediments to a maximum depth of 4.35 m (Serreli 2011). The shaft contained, in different stratigraphic units, numerous organic materials including wood, charcoal, cork, seeds (mostly of Vitis), animal remains and pottery helpful for dating it to the final and recent Bronze Age (Usai 2011; Sanna 2011). The Vitis seeds were studied using a binocular microscope, highlighting the prevalence of shapes of cultivated varieties (Lovicu et al. 2011).

The use of biometric indices for seed studies has often proved to be of great importance in the understanding of the domestication processes, in taxonomic studies of modern Vitis, as well as for the classification of archaeological remains (Rivera et al. 2007). Mangafa and Kotsakis (1996) used 22 biometric variables and four different algebraic formulae to identify the seeds of Vitis found at the prehistoric sites of Dikili Tash and Toumba Thessaloniki, Greece, as belonging to wild or cultivated species. Their formulae, just for their potentials, have subsequently been used for the study of the seeds of Vitis found at Petra in Jordan, and dated back to 200 b.c.–a.d. 500, allowing recognition of many seeds of wild plants and a certain amount from crops. The data were partially confirmed by the calculation of Stummer’s index (Jacquat and Martinoli 1999). The ratio between stalk length and pip length, used to study five varieties of grapevine (Chasselas, Pinot Noir, Rèze, Amigne and White Humagne), gave different results compared with the Mangafa and Kotsakis formulae, showing that the “sylvestris” type pip morphology could be associated not only with wild grapevines but also with archaic varieties. Consequently, the seed found in Petra could belong to a cultivated grapevine bearing fruits of wild type. This hypothesis seems to be validated by the existence of such varieties not only in Europe but elsewhere (Jacquat and Martinoli 1999).

In southwestern Sardinia, near the Isola di Coltellazzo, various seeds and charred woody materials were found in different states of preservation inside two Phoenician amphorae, referable to the Punic period (600–300 b.c.). Some of these Vitis seeds had an ovoid shape, an evident and lengthened beak and the chalaza situated in the upper part of the seed. According to the morphological features and to the biometric indexes related to the ratio between pip length and pip width, the seeds were assigned to V. vinifera (Marinval and Cassien 2001).

Another study on the identification and grouping of Vitis seeds on the basis of biometric features was carried out on 142 different types of grape, including five taxa of Vitis, 92 cultivars of V. vinifera, 12 feral or wild populations and hybrid rootstock cultivars, measuring 11 morphometric variables by an electronic calliper. The obtained data were elaborated using cluster analysis, placing feral or wild populations and related cultivars in their respective clusters, but missing a cluster of wild European grapevine (Rivera et al. 2007).

In a recent paper, Terral et al. (2010) discussed the potential of morphometric analysis to compare well-preserved archaeological seeds found in southern France and dated back to 100 b.c. with some European modern cultivars and wild individuals, using the Elliptic Fourier Descriptors (EFDs) method. Also, Gong et al. (2010) used digital images to analyze the morphometry of some fossil seeds of Vitis recovered from the Gray Fossil Site in northeastern Tennessee, USA and datable to latest Miocene-earliest Pliocene. On the basis of 11 measured parameters, they placed the seeds in three different morphotaxa.

Using digital images, Bacchetta et al. (2008) characterized seeds of wild plants typical of the Mediterranean basin, implementing statistical classifiers able to discriminate seeds belonging to different genera and species, and achieving promising results. This system was later improved, adding 20 new morphometric and colorimetric features (Mattana et al. 2008). Recently, Grillo et al. (2010) published the results of the use of statistic classifiers, based on morphometric and colorimetric features of seeds, for ten of the most representative families of the Mediterranean vascular flora, confirming the validity of the method.

Currently, this method is fully accepted and utilized, not only for archaeological studies or taxonomic investigations of wild taxa (Bacchetta et al. 2011a, 2011b), but also to study cultivated plants, to compare different varieties, contributing to the cataloguing and conservation in germplasm banks, or allowing the definition of objective parameters for the typifying of particular landraces in the attribution of European trademarks such as protected designation of origin (PDO) and protected geographical indication (PGI) (Grillo et al. 2011; Kiliç et al. 2007; Venora et al. 2007, 2009a, b). All these studies prove that the morphological traits of the seed, such as shape, size and external ornamentations, represent very important diagnostic factors in plant taxonomy studies. Moreover, the increasing availability of seeds of wild plants from archaeological sites and above all those stored in seed banks, emphasizes the importance of seed macro and micro morphology studies in plant taxonomy (Grillo et al. 2010).

This paper proposes an accurate identification approach for recognizing archaeological seeds belonging to the genus Vitis, based on characterization by image analysis. In particular, the aims of this study are:

(1) To assemble a database of morphological parameters and Elliptic Fourier Descriptors to characterize the collected seeds belonging to V. vinifera and V. sylvestris;

(2) To compare the archaeological seeds with the recent seeds of both species based on the established database, to identify the relationships between the archaeological unknown seeds and V. vinifera and V. sylvestris materials, using Linear Discriminant Analysis (LDA).

Materials and methods

Archaeological seed materials

Seeds from two Sardinian archaeological sites were available, and were selected on the basis of preservation state. 790 Selected and waterlogged seeds of Vitis (seven samples of the best 100 seeds, one with only 90 seeds), coming from eight different stratigraphic units between −0.5 and −4.60 m, in shaft N of the pre-Nuragic and Nuragic complex of Sa Osa in central western Sardinia (39°55′16.50′′N; 8°32′46.76′′E), belonging to the Middle and Final Bronze Age (Fig. 1) and dated to 1300–1200 b.c. according to Depalmas (2009), and 1600–1200 b.c. according to Sanges (2010). Six carbonized seeds found inside the Phoenician amphora 78A2, from near Isola di Coltellazzo, Sardinia (38°59′02.00′′N; 9°01′17.78′′E), and dated to 600–300 b.c., were selected too (Fig. 2).

Fig. 1
figure 1

Longitudinal section of the shaft N of the pre-Nuragic and Nuragic complex of Sa Osa, giving dimensions and representative seeds of the eight stratigraphic units

Full size image
Fig. 2
figure 2

Sardinian historical distribution regions of the archaeological sites and studied cultivars and populations of Vitis vinifera L.

Full size image

The selected seed lots were kept in plastic containers of 50 ml in deionised water. Before image acquisition, they were placed on tissue paper to absorb excess water, carefully cleared with a brush, and scanned in sub-lots of ten seeds. After image acquisition, the sample lots were saved in single packets and then grouped in sealed 500 ml bottles in deionised water. Each bottle was stored in the Sardinian Germplasm Bank (BG-SAR) in the dark at 5–10 °C.

Modern seed material

A total of over 4,000 seeds of the five most representative and best preserved populations of V. sylvestris were collected throughout southwest Sardinia from along riverbanks or colluvial sites on hilly humid slopes (Table 1 in ESM, Fig. 2). Fruits/seeds of 37 cultivars from southern and central western Sardinia, two Italian and two French cultivars of V. vinifera were selected from modern populations developed and maintained in the experimental farms of Ussana by the Agenzia per la Ricerca Scientifica della Regione Autonoma Sardegna (AGRIS), in total more than 38,000 diaspores (Table 2 in ESM).

Seed size and shape analysis

Digital images of the modern seed samples were acquired using an Epson GT 15000 flatbed scanner with a digital resolution of 200 dpi and a scanning area not exceeding 1,024 × 1,024 pixels. Image acquisition was performed before drying the seeds at 15 °C to 15 % of R.H. to avoid spurious variation in dimension, shape and colour. Before image acquisition, the scanner was calibrated for colour matching following the protocol of Shahin and Symons (2003), as suggested by Venora et al. (2009a).

Samples consisting of 100 seeds were scanned and used for the digital image analysis. In order to represent the whole variability of each of the modern seed lots, the seed samples were scanned three times, randomly disposing them each time on the flatbed tray. A total of over 42,000 statistical cases were analysed.

The archaeological seeds were scanned differently and they were put on the scanner in organized columns of ten by ten and labelled with unique numbers, to allow their identification during the analysis. Moreover, due to the irregular shape of these seeds, both the dorsal and ventral faces were scanned, in order to consider the whole morphological variability of each sample and contextually to increment the number of statistical cases.

The digital images of the seeds were processed and analysed using the software package KS-400V. 3.0 (Carl Zeiss, Vision, Oberkochen, Germany). A macro specifically developed for the characterization of wild seeds (Bacchetta et al. 2008), and later modified to measure a further 20 morpho-colorimetric seed features (Mattana et al. 2008), was adapted to perform all the analysis procedures automatically, reducing the execution time and contextual mistakes in the analysis process (Grillo et al. 2010). Due to the unsuitable colorimetric features of the archaeological seed lots, only 13 features descriptive of seed size and shape were measured (Table 3 in ESM).

Moreover, because of the few available features descriptive of seed size and shape, the binary images obtained by the segmentation process during the image analysis of the seeds were redefined to 400 dpi to enhance the image definition and apply the EFDs method, to increase the number of discriminant parameters (Bacchetta et al. 2009, 2010). This method allows description of the boundary of the seed projection as an array of complex numbers which correspond to the pixel positions on the seed boundary. So, from the seed apex, defined as the starting point in a Cartesian system, chain codes are generated. A chain code is a lossless compression algorithm for binary images. The basic principle of chain codes is to separately encode each connected component (pixel) in the image. The encoder then moves along the boundary of the image and, at each step, transmits a symbol representing the direction of this movement. This continues until the encoder returns to the starting position. This method is based on separate Fourier decompositions of the incremental changes of the X and Y coordinates as a function of the cumulative length along the boundary (Kuhl and Giardina 1982). Each harmonic (n) corresponds to four coefficients (an, bn, cn and dn) defining the ellipse in the XY plane. The coefficients of the first harmonic, describing the best fitting ellipse of outlines, are used to standardize size (surface area) and to orientate seeds (Terral et al. 2010). According to Terral et al. (2010), about the use of a number of harmonics for an optimal description of seed outlines, in order to minimize the measurement errors and to optimize the efficiency of shape reconstruction, 20 harmonics were used to define the seed boundaries, obtaining a further 80 parameters useful to discriminate between the studied seeds of Vitis (Fig. 3).

Fig. 3
figure 3

Graphical reconstruction of the seed shape contour on the basis of 20 harmonics used to define the seed boundaries by the EFDs

Full size image

Statistical analysis

The obtained data from modern V. sylvestris, V. vinifera and the archaeological Vitis seeds built up a database. The data were statistically elaborated applying the stepwise Linear Discriminant Analysis (LDA) method by using the SPSS software package release 15 (SPSS Inc. 1989–2006), to compare the modern cultivars with the archaeological seeds considered as unidentified cases. This approach is commonly used to classify or identify unknown groups characterized by quantitative and qualitative variables (Fisher 1936, 1940). On the basis of all measured features, the stepwise method identifies and selects the best of them to use for the seed sample identification, using three statistical variables, Tolerance, F-to-enter and F-to-remove. The Tolerance value indicates the proportion of a variable variance not accounted for by other independent variables in the equation. A variable with very low Tolerance values provides little information to a model. F-to-enter and F-to-remove values define the power of each variable in the model and they are useful to describe what happens if a variable is either inserted or removed from the current model (Bacchetta et al. 2010). This method starts with a model that does not include any of the variables. At each step, the variable with the largest F-to-enter value that exceeds the entry criteria chosen (F ≥ 3.84) is added to the model. The variables left out of the analysis at the last step have F-to-enter values smaller than 3.84, so no more are added. The process was automatically stopped when no remaining variables increased the discrimination ability (Venora et al. 2009b).

Finally, a cross-validation procedure was applied to verify the performance of the identification system, testing individual unknown cases and classifying them on the basis of all others (SPSS release 15, SPSS Inc. 1989–2006).

Results

A total of 93 morphological quantitative variables describing seed size and shape were measured and then analysed by stepwise LDA, to implement statistical classifiers able to distinguish the studied cases. For each classifier, the stepwise method chooses between 42 and 64 variables among the 93 available to classify the seed groups. Although the perimeter ratio (P conv/P Crof) was always the first, the second or the third feature selected by the model on the basis of the discriminatory power, showing high values of F-to-remove (data not shown), all the discrimination processes were carried out principally by EFDs. For each implemented classifier, seven or eight EFDs were present among the first ten chosen parameters. Maximum diameter (D max), Feret ratio (D min/D max), Equivalent circular diameter (Ecd) and Convex perimeter (P conv) provided other powerful features for the discrimination model (data not shown, Table 3 in ESM).

Using this model, a preliminary comparison among the seeds of V. vinifera, V. sylvestris and those from the two archaeological sites considered as a unique unknown group was executed, achieving, for the archaeological seeds, grouping percentages of 58.7 and 41.3 % in V. vinifera group and V. sylvestris group, respectively (Table 1).

Table 1 Correct classification percentage between V. vinifera L. ssp. sylvestris (C.C. Gmel) Hegi and V. vinifera L. ssp. vinifera and grouping percentages of archaeological seeds considered as unknown; in parenthesis the number of analysed seeds
Full size table

Afterwards, it was possible to compare the eight seed lots of Vitis from the eight stratigraphic units of the pre-Nuragic and Nuragic complex of Sa Osa, to evaluate the relationship among them (Table 2; Fig. 1). As shown in Table 2, only 29.3 % of the seeds belonging to different stratigraphic levels were correctly identified, and none of the eight stratigraphic units is predominant to the others, so for this reason they were considered as a unique sample lot in the following elaborations.

Table 2 Correct classification percentage of the archaeological seed lots of Vitis from the eight stratigraphic units of the pre-Nuragic and Nuragic complex of Sa Osa; in parenthesis the number of analysed seeds
Full size table

Then, the seeds from the two archaeological sites were compared with the populations of V. sylvestris, achieving an overall cross-validated performance of correct classification of 97.6 % (Table 3). The seeds found in shaft N of the Sa Osa complex were correctly classified in 97.7% of the cases, while 66.7 % of the seeds from the Phoenician amphora 78A2 near Isola di Coltellazzo were correctly attributed. Also, a high percentage of V. sylvestris was identified correctly and only a few seeds were mistaken for archaeological seeds and vice versa.

Table 3 Correct classification percentage between V. vinifera L. ssp. sylvestris (C.C. Gmel) Hegi and archaeological seed lots (Sa Osa, Isola di Coltellazzo); in parenthesis the number of analysed seeds
Full size table

From a more detailed comparison among the seeds from the two archaeological sites and the five populations of V. sylvestris, an overall percentage of correct identification of 82.2 % was achieved (Table 4). The seeds of V. sylvestris populations were distributed into the V. sylvestris group, in which correct classification performances are included between 48.0 and 64.5 %, except the population of Siliqua which showed a percentage of correct identification of 93.0 %; while 98.5 and 50.0 % of the archaeological seed lots from Sa Osa and Isola di Coltellazzo respectively were correctly identified.

Table 4 Correct classification percentage between populations of V. vinifera L. ssp. sylvestris (C.C. Gmel) Hegi and archaeological seed lots (Sa Osa, Isola di Coltellazzo); in parenthesis the number of analysed seeds
Full size table

Comparing the two archaeological seed assemblages with the cultivars of V. vinifera, an overall cross-validated performance of 98.2 % was achieved (Table 5). The archaeological seeds from the Sa Osa complex reached a percentage of correct identification of 79.5 %, in which misidentified seeds were exclusively mistaken for V. vinifera, while the seeds from the Isola di Coltellazzo were correctly classified in 50.0 % of the cases, and also in this case the misclassified seeds were only mistaken for V. vinifera. The correct identification percentage of V. vinifera seeds was very high (99.0 %), with modest numbers of seeds wrongly classified into both archaeological groups.

Table 5 Correct classification percentage between V. vinifera ssp. vinifera and archaeological seed lots (Sa Osa, Isola di Coltellazzo); in parenthesis the numbers of analysed seeds
Full size table

In order to identify a narrower group of V. vinifera to which the archaeological seed lots could be a closer match, a comparison among the seeds of the two archaeological sites and the V. vinifera cultivars clustered for the colour of its grapes was done (Table 6). This approach showed a rather low overall performance (58.8 %), mainly due to the mistakes between the two V. vinifera groups. It is important to highlight the distribution of the misidentified seeds of the two archaeological seed lots. The wrongly classified seeds of the archaeological group of Sa Osa were identified as white grape V. vinifera in 12.7 % of the cases, and as black grape V. vinifera in 4.0 % of the cases, while the misidentified seeds of the archaeological seed lot from Isola di Coltellazzo, although consisting of only three seeds, was classified as white grape V. vinifera in 33.3 % of the cases and as black grape V. vinifera in 16.7 % of the cases.

Table 6 Correct classification percentage between V. vinifera L. ssp. vinifera and archaeological seed lots (Sa Osa, Isola di Coltellazzo) distinguished for grape colour; in parenthesis the numbers of analysed seeds
Full size table

Finally, comparing the two archaeological sites with the seeds of all the cultivar sample lots of V. vinifera individually considered, an overall cross-validated performance of 41.0 % was achieved (Table 7). The seeds from shaft N of the Sa Osa complex reached a percentage of correct identification of 85.2 %. Errors were evenly distributed in more than half of the considered cultivars. Similarly, 50.0 % (three specimens) of the misclassified seeds from the Phoenician amphora were wrongly attributed to three of the V. vinifera cultivars (Grillu, Licronaxiu and Girò di Gonnos). None of the seeds from the two archaeological sites were mistaken for each other. Regarding the identified V. vinifera cultivars, low performances were achieved, ranged between 10.9 % (Licronaxiu bianco) and 80.6 % (Vernaccia), but only ten out of 41 seed lots showed a percentage of correct identification higher than 50.0 %.

Table 7 Correct classification percentage between V. vinifera L. ssp. vinifera cultivars and archaeological seed lots (Isola di Coltellazzo, Sa Osa); in parenthesis the numbers of analysed seeds
Full size table

Discussion and conclusion

Many times morpho-colorimetric seed characterization by image analysis has proved to be a repeatable, reliable and non-destructive method able to accurately identify seeds both of cultivated (Grillo et al. 2011; Venora et al. 2007; Zapotoczny et al. 2008) and of wild plant species (Bacchetta et al. 2008, 2011a, 2011b; Grillo et al. 2010; Mattana et al. 2008).

Because of the non-representative colour features of the archaeological seeds, in this work, only the morphological characterization of seeds was applied to comparison of the archaeological seed lots with cultivars of V. vinifera and wild populations of V. sylvestris. In this, the EFDs proved to be very helpful to discriminate between the studied cases, as proved by many authors studying seeds (Iwata et al. 2010; Ohsawa et al. 1998), as well as other plant anatomical traits (Kawabata et al. 2009; Hâruta 2011; Yoshioka et al. 2007).

From the comparison between the seeds of V. vinifera, V. sylvestris and those from the two archaeological sites, a slight but not explicit similarity of the archaeological seeds and the V. vinifera group was highlighted. Presumably such a result could be due to the great morphological variability within each group, and on the basis of this hypothesis, separate comparisons for V. vinifera and V. sylvestris were subsequently implemented.

Considering the archaeological seeds of Vitis from the eight stratigraphic units of the Sa Osa complex, there are almost no evident differences. Therefore, and according to the studies conducted by the Soprintendenza per i Beni Archeologici per le Province di Cagliari e Oristano (the archaeological authority for the provinces of Cagliari and Oristano) on the different stratigraphic units (Usai 2011; Sanna 2011), in this study the seeds of the eight stratigraphic levels were considered as a unique seed lot when compared with the V. vinifera and V. sylvestris seed lots.

From the comparison of both archaeological seed lots with the wild populations of V. sylvestris, a clear morphological differentiation between them and the wild populations of V. sylvestris was revealed. Similar results are obtained when the seed lots of V. sylvestris were considered as individual populations. A more accurate evaluation of the results shows that the seeds of V. sylvestris populations were widely distributed within the V. sylvestris group. Except the population of Siliqua that showed a good performance of correct identification (93.0 %), the misidentification distribution among the other wild populations of V. sylvestris suggests that the numerical size of each wild population may play an important role. Actually, the population of Siliqua, made up of only 11 individuals, five females and six males, is very small compared with all the others. Consequently, the low intra-population variability of Siliqua could explain the modest morphological diversification of the seeds of this population. In addition, the geographical isolation of the Siliqua population, located far away both from urban areas and from farmlands, in an extreme and wild region, indicates the low probability that intraspecific hybridization phenomena occurred. According to Zecca et al. (2010) wild and domesticated Sardinian grapevines have essentially remained reproductively isolated.

Following the same criteria, the two archaeological seed lots were compared with the seeds of the cultivars of V. vinifera, showing that a certain relationship probably exists between them. Both archaeological seed lots were widely misidentified as V. vinifera, but the seeds of the two archaeological groups were never mistaken for each other. The comparison between the archaeological seeds and the studied cultivars, split into two groups on the basis of the grape colour, gives hints that the archaeological seed lots seem to be closer to the white grape cultivars than to the black ones.

Finally, comparing the two archaeological seed lots with all the studied cultivars individually, it is possible to notice that four of the first five bigger mistakes with the seed lots from the Sa Osa complex are related to white grape cultivars (Cannonau bianco, Codronisca, Mizu and Trebbiano Romagnolo), achieving for them a percentage of misidentification of 6.5 %. A similar consideration can be done on the basis of the misclassified seeds from Isola di Coltellazzo, for which two of the three misidentifications are related to white grape cultivars (Grillu and Licronaxiu).

The results obtained by establishing a general database of morphological features of the genus Vitis and the implementation of a seed classifier for V. vinifera and V. sylvestris species groups, prove once again how image analysis techniques can be considered as a useful tool not only in taxonomic investigations, but also in archaeobotanical studies. Moreover, the adoption of the EFDs as discriminant parameters proves to be very important, especially when colour parameters are not applicable, although these were important when dealing with modern material. Regarding the results, the domestication and selection work by humans, the high degree of intraspecific hybridization of V. vinifera, the low probability of interspecific hybridization phenomena, the annual biological cycle and above all the very long period that separates the ancient from the modern seed lots considered in this study, are very important factors to consider. Surely, all these elements influenced and contributed to determine the evolution of modern cultivars, contextually causing the loss of ones that currently should be very close to the archaeological seeds. This assumption makes the achieved results certainly significant. Moreover, the obtained high level of dissimilarity is plausible also considering the impossibility of using the colorimetric features, which many times proved to be very powerful parameters for the identification process.

The application of image analysis allowed identification of the relationship between the seed lots from the archaeological sites of Sa Osa and Isola di Coltellazzo, and the modern cultivars historically grown close to the archaeological sites, and the wild populations of V. sylvestris collected near the two sites. In particular, the analysis of the archaeological seeds of Vitis from the eight stratigraphic units of shaft N of the pre-Nuragic and Nuragic complex of Sa Osa could show that the seeds are very similar, confirming that the different stratigraphic units belong to the same period, Middle and Final Bronze Age.

Taking into consideration the difference between the two kinds of archaeological seeds, it becomes likely that the archaeological seeds from the Phoenician amphora came from different cultivation areas, far away from Sardinia. If so, the similarity with white grape cultivars should be merely coincidental. A larger database including more cultivars and sample lots may allow a more accurate identification in the future.

Anyway, the greater similarity of the archaeological seeds to V. vinifera cultivars than to V. sylvestris populations, and especially to white grapes rather than black grape cultivars, could prove that in the Campidano in southern Sardinia, white grapes were probably already used at 1600–1200 b.c., explaining that it may be not a chance that white grapes are still traditionally grown today in the Campidano area to produce famed wines, as well as in the Vernaccia.

According to the analysis conducted by Lovicu et al. (2011), the seeds at Sa Osa could be identified as V. vinifera, allowing us to affirm that, between the Middle and Final Bronze Age, very similar varieties of V. vinifera were used to produce wine or to preserve as foodstuffs, although the production, attested in the Nuragic period thanks to the find of a Nuragic wine press in Monastir in southern Sardinia close to Monte Zara, is dated between 900 and 800 b.c., the recent Iron Age (Ugas 1999). Considering that all the previous Sardinian remains of Vitis seeds cited in literature were dated to the Final Bronze Age, and looking at the dating of all the other finds in the Italian peninsula and in the Mediterranean area (Aranguren et al. 2007; Bellini et al. 2008; Buxó and Capdevila 1997; Costantini 1981; Delpino 2007; Forni 2007; Martinoli 2004; McGovern 2003a, McGovern 2003b, 2004; Vaquer et al. 1986), according to Sanges (2010) the seeds from the pre-Nuragic and Nuragic complex of Sa Osa are the oldest remains of Vitis found in Sardinia, the oldest find of domesticated Vitis in Italy and among the most ancient remains of domesticated Vitis found in the Mediterranean area.