Modern plant materialA total of 365 seeds from two cultivars were sampled from the vineyard collection at Ariel University, Israel: the endogenous Israeli variety 292 (Tzuriman S.) from an endogenous collection previously described22,23, and Cabernet Sauvignon from the European varieties collection at Ariel University, Israel.Mature seeds were extracted from ripened grapes from at least three different grapevines. The seeds were washed with water to discard any residual pulp tissue and air-dried for 2 days, then stored in a closed vial until used. Before scanning, each seed was carefully cleaned with brushes and needles from any external tissues coating its crevices, to enable an effective scan of the seed’s morphology.Archeological plant materialA total of 100 grape pips from two archeological sites; Shilo (n = 73) and Beit El (n = 27), were used for scanning and measurements. Below is a description of the archaeological context of the specimens used in the study.Shilo is one of the most important Israelite sites in the land of Israel and is considered the site of the Tabernacle. Grape seeds were identified by dry sifting the sediments in three loci. Locus 8561 (Figure S1 – Online Appendix B) is a layer of soil on a floor in a ‘four-room house’ building, dated to the Iron Age IIb (8th C. BCE), based on the associated pottery. The room was paved with small field stones, and many vessels were discovered on its floor, including many bowls, kraters, and cooking pots. It was therefore interpreted to have been used for food preparation and dining. The building was abandoned during the destruction of the Kingdom of Israel24. Locus 8593 is a favissa pit (Figure S2 – Online Appendix B), dated to the Iron Age IIa (tenth–ninth centuries BCE). The pit was blocked with two separate layers of fieldstones, and a round stone slab was discovered next to it, likely intended to cover its opening. In it, grapevine and olive seeds were discovered.Locus 8583 is also a pit. It is bell-shaped in cross-section and hewn into the rock, interpreted as a storage pit. In its bottom, along with various pottery vessels, wheat, olive, and grapevine seeds were discovered.Beit El The site Khirbet Kfar Murr is located in the Beit El mountains, about 30 km north of Jerusalem. Since 2006, ten excavation seasons have been conducted at the site, in which broad settlement activity has been revealed, dating from the eighth century BC to the eighth century AD. Starting as a small farm or a small village during the Iron Age II, it developed into a small town and reached its peak during the Roman period, being one of the many thriving Jewish settlements in northern Judea.Grape seeds were collected from locus L-6154, a large wine press located at the northeast of the site (Figure S3 – Online Appendix B). The winepress includes a large 7X5 m treading floor, a sedimentation compartment of ca. 1 m2, and a collection pool of ca. 2 m2, 2.6 m deep, with seven steps descending to its bottom and a pip sediment crevice. A simple calculation by the collection pool’s volume gives the capacity of circa 10 m3 of wine in a single batch. Dating by the pottery, the winepress was used until the end of the big Jewish rebellion during the first century AD. The grape pips used in this study were collected by sifting the sediment at the bottom of the collection pool.All grape seeds were dry sifted from the sediment using a 5 mm sift and kept in dry conditions (20% relative humidity, 20–22 °C) until scanning or other analysis.Charring Experiments and Morphometric AnalysisTo identify morphological changes in charred specimens and perform machine-learning-based classification of the archaeological pips, we conducted controlled charring experiments using fresh grape pips of Cabernet Sauvignon and Tzuriman S. 292 grape varieties. Following their extraction from the ripe fruit, and cleaning using a teeth brush and a needle, the fresh grape seed specimens were heat-treated in batches of 20 seeds at a time, at temperatures ranging from 200 to 350 °C with increasing periods between 2 and 8 h in a controlled electric furnace (DFO-240, MRC Ltd. Israel). To simulate actual charring conditions, pips were covered with a thick layer of quartz sea sand to keep low oxygen availability and prevent their burning (oxidation)12,25,26,27.Additional charring experiments were conducted to track and characterize the chemical changes occurring during charring, using FTIR spectroscopy. Three to five seeds of each variety were heat-treated at temperatures ranging from 150 to 300 °C, with increasing time periods between 2 and 24 h, in the same manner described above. Details of the FTIR analysis and the results obtained are summarized in Supplementary file A.S1.Table 1 summarizes the experiment conditions. Note that at 150 °C, there seemed to be no point in running the experiment for 24 h due to almost no difference in morphological parameters between 1, 4, and 8 h of charring, while at 300 °C, the 8 h of treatment already showed highly deformed pips. Out of each charring term, two seeds were selected for FTIR analysis.
Table 1 Summary of charring experiments conditions.Morphometric analysisIn order to quantify the changes in grape seed proportions caused by the charring, we performed a morphometric analysis of each pip before and after charring. We measured the length (black line), the breadth (blue line), and the distance between ventral infolds (red line) on the same line of breadth, as represented in Fig. 1. The selection of the simple morphometric parameters length and width as well as the distance between infolds is due to two considerations, a) preliminary observations we made indicated that the main change following charring of pips is their swelling, dramatically changing these parameters and their ratios. And b) by specifically examining the ratios between width/length and width/DVI (Distance between the Ventral Infolds), we eliminate the differences within a group of seeds from the same variety, caused by different pip sizes. The measurements were conducted using the NIS-Elements D software, whereas the pips were scanned using a Nikon SMZ25 stereomicroscope (Nikon, Tokyo, Japan) equipped with a Nikon DS-Ri2 microscope camera, as described previously15. Such quantification constitutes an important step toward the sortation of archeological seeds in terms of suitability for classification under the proposed methodology.Figure 1Measurements of length (black line), breadth (blue line) and distance between the ventral infolds (DVI—red line).Linear discriminant analysis (LDA) classificationFollowing the presented classification methodology in our previous work15, we first preprocess every stereoscopic pip scan by transforming it into a cloud points representation. In turn, each pixel and its corresponding height value of the scan were scaled by the intrinsic matrix, which consists of the intrinsic parameters of the scanning device. Once we acquired a cloud point for each scanned pip, we formed a data set of cloud points and split it into training and testing datasets. Then, for each data set, we built a similarity matrix. The training set similarity matrix was constructed by applying an Iterative Closest Point28 (ICP) algorithm between every pair of cloud points pips (pip [i] and pip [j]) and populating the matrix with the ICP distance Mean Square Error (MSE) result at index [i,j]. Constructing the testing similarity matrix utilizes the same process, except for the pips cloud points pairs. In this case, the similarity matrix constitutes a combination between the training set and the testing set. Finally, these similarity matrices were used as training and testing sets that were fed into the Linear Discriminant Analysis29,30 (LDA) machine learning model. The LDA model learns the inner and outer classes of the ICP similarities in the training pips, and based on the learned features, it is used to classify the testing pips. In order to achieve proper feature extraction, we normalized the training similarity matrix to zero mean and unit standard deviation, and those parameters were used to normalize the testing matrix as well.In addition, we implemented a tournament methodology described in our previous publication15, conducting multiple classification sessions, while at every session we used different tests and training splits to maximize the approximation of the classification accuracy distribution.Furthermore, an additional normalization factor was used- presenting the scanning pips with equal length while preserving their structural ratio. In such a way, the extracted features were solely based on the morphological differences rather than the physical size.Statistical analysisMorphometric differences following the charring at different times and temperatures were conducted using a Two-way ANOVA test31, followed by the Tukey post hoc test32, using all 20 seeds per treatment. The statistical analysis was conducted using NIS-Elements D software, JMP Pro 15.1.0 Statistical Software (SAS Institute Inc., Cary, NC, USA: https://www.jmp.com/en_us/home.html) to determine the statistical significance of differences between the treatment means at α = 0.05.Plant guideline statementThe experiment with plant and archaeological materials complied with relevant institutional, national, and international guidelines and legislation.
Setting the morphologic quality limits enabling accurate classification of charred archaeological grape seeds
