The Levantine Corridor is considered a crossing point to Eurasia and one of the main areas for detecting population flows (and their associated cultural and economic changes) during the Pleistocene. This area could have been closed during the most arid periods, giving rise to processes of population isolation between Africa and Eurasia and intermittent contact between Eurasian human communities [1,2]. 

Zooarchaeological studies of the early Upper Palaeolithic assemblages constitute an important source of knowledge about human subsistence, making them central to the debate on modern behaviour. The Early Upper Palaeolithic sequence in the Levant includes two cultural entities – the Early Ahmarian and the Levantine Aurignacian. This latter is dated to 39-33 ka and is considered a local adaptation of the European Aurignacian techno-complex. In this work, the authors present a zooarchaeological study of the Nahal Rahaf 2 (ca. 35 ka) archaeological site in the southern Judean Desert in Israel [3].

Zooarchaeological data from the early Upper Paleolithic desert regions of the southern Levant are not common due to preservation problems of non-lithic finds. In the case of Nahal Rahaf 2, recent excavation seasons brought to light a stratigraphical sequence composed of very well-preserved archaeological surfaces attributed to the 'Arkov-Divshon' cultural entity, which is associated with the Levantine Aurignacian. 

This study shows age-specific caprine (Capra cf. Capra ibex) hunting on prime adults and a generalized procurement of gazelles (Gazella cf. Gazella gazella), which seem to have been selectively transported to the site and processed for within-bone nutrients. An interesting point to note is that the proportion of goats increases along the stratigraphic sequence, which suggests to the authors a specialization in the economy over time that is inversely related to the occupational intensity of use of the site. 

It is also noteworthy that the materials represent a large sample compared to previous studies from the Upper Paleolithic of the Judean Desert and Negev.

In summary, this manuscript contributes significantly to the study of both the palaeoenvironment and human subsistence strategies in the Upper Palaeolithic and provides another important reference point for evaluating human hunting adaptations in the arid regions of the southern Levant.

References

[1] Bermúdez de Castro, J.-L., Martinon-Torres, M. (2013). A new model for the evolution of the human pleistocene populations of Europe. Quaternary Int. 295, 102-112. https://doi.org/10.1016/j.quaint.2012.02.036

[2] Bar-Yosef, O., Belfer-Cohen, A. (2010). The Levantine Upper Palaeolithic and Epipalaeolithic. In Garcea, E.A.A. (Ed), South-Eastern Mediterranean Peoples Between 130,000 and 10,000 Years Ago. Oxbow Books, pp. 144-167.

[3] Marom, N., Gnezdilov, D. L., Shafir, R., Barzilai, O. and Shemer, M. (2022). Faunal remains from the Upper Paleolithic site of Nahal Rahaf 2 in the southern Judean Desert, Israel. BioRxiv, 2022.05.17.492258, ver. 4 peer-reviewed and recommended by Peer community in Archaeology. https://www.biorxiv.org/content/10.1101/2022.05.17.492258v4

In a recent article, Fumihiro Sakahira and Hiro'omi Tsumura (2023) used social network analysis methods to analyze change in obsidian trade networks in Japan throughout the 13,000-year-long Jomon period. In the paper recommended here (Sakahira and Tsumura 2024), Social Network Analysis of Ancient Japanese Obsidian Artifacts Reflecting Sampling Bias Reduction they revisit that data and describe additional analyses that confirm the robustness of their social network analysis. The data, analysis methods, and substantive conclusions of the two papers overlap; what this new paper adds is a detailed examination of the data and methods, including use of bootstrap analysis to demonstrate the reasonableness of the methods they used to group sites into clusters.

Both papers begin with a large dataset of approximately 21,000 artifacts from more than 250 sites dating to various times throughout the Jomon period. The number of sites and artifacts, varying sample sizes from the sites, as well as the length of the Jomon period, make interpretation of the data challenging. To help make the data easier to interpret and reduce problems with small sample sizes from some sites, the authors assign each site to one of five sub-periods, then define spatial clusters of sites within each period using the DBSCAN algorithm. Sites with at least three other sites within 10 km are joined into clusters, while sites that lack enough close neighbors are left as isolates. Clusters or isolated sites with sample sizes smaller than 30 were dropped, and the remaining sites and clusters became the nodes in the networks formed for each period, using cosine similarities of obsidian assemblages to define the strength of ties between clusters and sites.

The main substantive result of Sakahira and Tsumura’s analysis is the demonstration that, during the Middle Jomon period (5500-4500 cal BP), clusters and isolated sites were much more connected than before or after that period. This is largely due to extensive distribution of obsidian from the Kozu-shima source, located on a small island off the Japanese mainland. Before the Middle Jomon period, Kozu-shima obsidian was mostly found at sites near the coast, but during the Middle Jomon, a trade network developed that took Kozu-shima obsidian far inland. This ended after the Middle Jomon period, and obsidian networks were less densely connected in the late and last Jomon periods.

The methods and conclusions are all previously published (Sakahira and Tsumura 2023). What Sakahira and Tsumura add in Social Network Analysis of Ancient Japanese Obsidian Artifacts Reflecting Sampling Bias Reduction are:

·       an examination of the distribution of cosine similarities between their clusters for each period

·       a similar evaluation of the cosine similarities within each cluster (and among the unclustered sites) for each period

·       bootstrap analyses of the mean cosine similarities and network densities for each time period

These additional analyses demonstrate that the methods used to cluster sites are reasonable, and that the use of spatially defined clusters as nodes (rather than the individual sites within the clusters) works well as a way of reducing bias from small, unrepresentative samples. An alternative way to reduce that bias would be to simply drop small assemblages, but that would mean ignoring data that could usefully contribute to the analysis.

The cosine similarities between clusters show patterns that make sense given the results of the network analysis. The Middle Jomon period has, on average, the highest cosine similarities between clusters, and most cluster pairs have high cosine similarities, consistent with the densely connected, spatially expansive network from that time period. A few cluster pairs in the Middle Jomon have low similarities, apparently representing comparisons including one of the few nodes on the margins on the network that had little or no obsidian from the Kozu-shima source. The other four time periods all show lower average inter-cluster similarities and many cluster pairs have either high or low similarities. This probably reflects the tendency for nearby clusters to have very similar obsidian assemblages to each other and for geographically distant clusters to have dissimilar obsidian assemblages. The pattern is consistent with the less densely connected networks and regionalization shown in the network graphs. Thinking about this pattern makes me want to see a plot of the geographic distances between the clusters against the cosine similarities. There must be a very strong correlation, but it would be interesting to know whether there are any cluster pairs with similarities that deviate markedly from what would be predicted by their geographic separation.

The similarities within clusters are also interesting. For each time period, almost every cluster has a higher average (mean and median) within-cluster similarity than the similarity for unclustered sites, with only two exceptions. This is partial validation of the method used for creating the spatial clusters; sites within the clusters are at least more similar to each other than unclustered sites are, suggesting that grouping them this way was reasonable.

Although Sakahira and Tsumura say little about it, most clusters show quite a wide range of similarities between the site pairs they contain; average within-cluster similarities are relatively high, but many pairs of sites in most clusters appear to have low similarities (the individual values are not reported, but the pattern is clear in boxplots for the first four periods). There may be value in further exploring the occurrence of low site-to-site similarities within clusters. How often are they caused by small sample sizes? Clusters are retained in the analysis if they have a total of at least 30 artifacts, but clusters may contain sites with even smaller sample sizes, and small samples likely account for many of the low similarity values between sites in the same cluster. But is distance between sites in a cluster also a factor? If the most distant sites within a spatially extensive cluster are dissimilar, subdividing the cluster would likely improve the results. Further exploration of these within-cluster site-to-site similarity values might be worth doing, perhaps by plotting the similarities against the size of the smallest sample included in the comparison, as well as by plotting the cosine similarity against the distance between sites. Any low similarity values not attributable to small sample sizes or geographic distance would surely be worth investigating further.

Sakahira and Tsumura also use a bootstrap analysis to simulate, for each time period, mean cosine similarities between clusters and between site pairs without clustering. They also simulate the network density for each time period before and after clustering. These analyses show that, almost always, mean simulated cosine similarities and mean simulated network density are higher after clustering than before. The simulated mean values also match the actual mean values better after clustering than before. This improved match to actual values when the sites are clustered for the bootstrap reinforces the argument that clustering the sites for the network analysis was a reasonable result.

The strength of this paper is that Sakahira and Tsumura return to reevaluate their previously published work, which demonstrated strong patterns through time in the nature and extent of Jomon obsidian trade networks. In the current paper they present further analyses demonstrating that several of their methodological decisions were reasonable and their results are robust. The specific clusters formed with the DBSCAN algorithm may or may not be optimal (which would be unreasonable to expect), but the authors present analyses showing that using spatial clusters does improve their network analysis. Clustering reduces problems with small sample sizes from individual sites and simplifies the network graphs by reducing the number of nodes, which makes the results easier to interpret.

Reference

Sakahira, F. and Tsumura, H. (2023). Tipping Points of Ancient Japanese Jomon Trade Networks from Social Network Analyses of Obsidian Artifacts. Frontiers in Physics 10:1015870. https://doi.org/10.3389/fphy.2022.1015870

Sakahira, F. and Tsumura, H. (2024). Social Network Analysis of Ancient Japanese Obsidian Artifacts Reflecting Sampling Bias Reduction, Zenodo, 10057602, ver. 7 peer-reviewed and recommended by Peer Community in Archaeology. https://doi.org/10.5281/zenodo.7969330