Bayesian modelling of the latent human influence field in archaeological urban sediments at Koh Ker, Cambodia

Reconstructing the spatiotemporal trajectories of massive ancient cities remains difficult because archaeological excavation offers temporal depth but limited spatial coverage, while lidar and other survey methods map urban form without chronological depth. Moreover, ancient urban dynamics were multidimensional and indicated by many different variables. Integrating archaeological and environmental proxies indicative of these variables into a single spatiotemporal picture (the urban trajectory) is complicated by heterogeneity in data resolution, distributional form, and measurement structure. To address these limitations, we propose a new framework centred on mapping a theoretical “latent human influence field” (LHIF). This field represents the intensity of past human activity and its effects on the proxies in the sediment archive, which can be estimated by fusing proxy data in rapidly excavated, low-impact trenches. Given sufficient sampling and temporal coregistration, an estimated LHIF promises a coherent multidimensional spacetime view of ancient urbanism. It would fill the inferential gaps between lidar, survey, and excavation by estimating not only where archaeological traces occur, but how strongly human activity shaped local sediments and landscapes — a continuous field of human activity intensity and influence. To explore the potential of this framework, we conducted a pilot study at Koh Ker (Cambodia) using magnetic susceptibility and ceramic abundance data from eight test trenches. We paired continuous magnetic susceptibility scans with ceramic abundance data and fitted a Bayesian hierarchical latent-variable model that treats both proxies as noisy observations of the shared LHIF. This model synthesizes heterogeneous data channels within a unified causal framework and is fully extensible to additional sedimentary proxies, with potential application across archaeological site types and regions.