Contents lists available at ScienceDirect Journal of Archaeological Science: Reports journal homepage: www.elsevier.com/locate/jasrep Archsphere – A cluster algorithm for archaeological applications Gino Caspari a,b,⁎ , Michael Jendryke c a University of Sydney, Australia b University of Bern, Switzerland c Wuhan University, China ABSTRACT This paper argues that many of the existing cluster algorithms employed by practitioners are too unspecific for archaeological purposes. Based on a large landscape archaeological dataset a cluster algorithm for archae- ological applications is developed. It accounts for shortfalls in generic cluster algorithms like the difficulty to cluster point clouds with varying densities in DBSCAN or the absence of a notion of noise in k-means. The application of the Archsphere algorithm is geared towards archaeological problem sets using readily available data from surveys and excavations as input. The introduced method performs the task of spatially dividing an archaeological dataset of monuments into clusters in a more meaningful way than is possible with standard procedures, effectively setting a solid foundation for a scale-consistent landscape archaeological analysis of monument assemblages. 1. Introduction In summer 2015 a large-scale landscape archaeological survey was conducted in the foothills of the Chinese Altai Mountains in the area of interest (AOI) Heiliutan (黑柳滩) by a team of archaeologists and remote sensing specialists from the Chinese Academy of Sciences, Wuhan University and Hamburg University. The result was a large dataset of funerary and ritual monuments of ancient steppe cultures scattered over 140 km 2 dating from the Early Bronze Age to the Turkic Period and beyond (Caspari et al., 2017). Monuments were dated through a comparative approach matching the morphological charac- teristics with known dated archaeological complexes in the wider region. Presented with an arrangement of several millennia of visible anthropogenic traces which relate and refer to each other, it was of utmost interest to group the data points into clusters for further analysis. We could have just established a subjective solution to this problem as it is very often done by practitioners. However, it seemed to be more interesting to accomplish the clustering by a well-documented process which produces coherent results consistently rather than applying a subjective interpretative approach that is naturally some- what more prone to modifications throughout the process' execution. Grouping data points without a clearly structured methodology carries the risk of jumping between different landscape archaeological scales and inconsistently relating data points which could lead to erroneous conclusions. Also, the application of a model would allow us to mathematically define places of meaning within the landscape. The problem is different from the application of cluster algorithms used to deal with data on a small archaeological scale. Contrary to e.g. accumulations of charcoal or flint in the archaeological record the dataset does not consist of individual events in which the generation of a single datum can hardly be called deliberate. Each monument has been consciously created and its geographical location and the relation to other monuments can be assumed to be the result of an intentional intricate social process. Finding a model for these complex spatial relationships is far from straightforward and any solution obviously is just one of many possible configurations. As every other clustering algorithm, the one proposed here has an explorative character, meaning that there is no explicit solution to the problem. Any presented solutions have to be seen as a possible hypothesis in need of further argumentation, testing, and reasoning, or rejection. However, it is much better adapted to landscape archaeological purposes and archaeological data and we thus expect more meaningful results than from many of the generic cluster algorithms which are currently applied. 2. The dataset The survey which was conducted as a part of the Dzungaria Landscape Project (Caspari et al., 2017) led to the mapping and documentation of almost 1000 previously undocumented structures from the early Bronze Age to the Turkic period. During the Bronze Age (2400–850 BCE) the first scarce traces of human occupation can be found in the area. In the early Bronze Age these consist of small http://dx.doi.org/10.1016/j.jasrep.2017.05.052 Received 14 March 2017; Received in revised form 24 May 2017; Accepted 26 May 2017 ⁎ Corresponding author at: University of Sydney, Australia. E-mail address: gino.caspari@gmail.com (G. Caspari). Journal of Archaeological Science: Reports 14 (2017) 181–188 2352-409X/ © 2017 Elsevier Ltd. All rights reserved. MARK