Spatio-temporal interpolation in archaeogenetics and archaelogy: computational methods for reconstructiong human mobility and cultural history

This dissertation presents empirical methods and software tools for large-scale spatio-temporal meta-analysis in the historical sciences. Conducted between 2019 and 2024 at the Max Planck Institutes for the Science of Human History and Evolutionary Anthropology, the work is interdisciplinary with a particular focus on archaeogenetics and computational archaeology. It consists of four interconnected papers, each addressing challenges and opportunities arising from rapidly growing datasets of archaeological and ancient genomic point observations. The first contribution, Poseidon, introduces a formally specified framework for managing ancient DNA genotype data together with its spatio-temporal context. By integrating standardized data packages, software tools, and public archives, Poseidon establishes reproducible infrastructure for community-based archaeogenetic research. The second contribution, bleiglas, presents an R package for diachronic visualisation of spatio-temporal point data using 3D Voronoi tessellation. It demonstrates how interpolation-based visualisation can reveal large-scale temporal dynamics while preserving information about data density and uncertainty. Building on this conceptual framework, the mobest publication introduces an algorithm to quantify human mobility in Holocene Western Eurasia from ancient genomic data. Using Gaussian process regression to interpolate genetic ancestry fields, mobest derives individual-level, directional mobility estimates that can be aggregated into regional time series and linked to major (pre)historic mobility events. Finally, the unpublished locest manuscript generalises these approaches by implementing Kernel average smoothing-based spatio-temporal interpolation in a flexible software tool applicable beyond archaeogenetics. Through archaeological and historical linguistic case studies, it demonstrates how continuous space-time models enable comparative analysis across data types and disciplines.