Research project This project investigates early Holocene to Neolithic landscape use in Västerbotten by combining theoretical research, statistical modelling, and high-resolution drone-based LiDAR surveys. By identifying topographical anomalies and verifying them through targeted field inspections, the study aims to uncover new archaeological sites and build a refined, diachronic understanding of human occupation in the Ume River Valley.
This project combines predictive modelling, drone-based LiDAR and advanced deep-learning workflows to analyse prehistoric settlement patterns in Västerbotten. High-resolution point clouds are processed and used to train the convolutional neural networks in order to identify subtle archaeological micro-topographies. Field verification of detected anomalies will refine diachronic models of early Holocene to Neolithic occupations and improve our understanding of postglacial landscape use.
This project investigates the long-term development of prehistoric settlement and landscape use in Västerbotten by integrating predictive modelling, high-resolution drone-based LiDAR, and advanced deep-learning techniques. Northern Sweden, shaped by rapid postglacial uplift and shifting coastlines throughout the early Holocene, presents a highly dynamic environment where archaeological sites can be difficult to detect through traditional methods. The combination of geomorphological complexity, dense forest cover, and limited prior survey coverage makes the region a prime candidate for innovative, technology-driven archaeological prospection.
The research begins with a comprehensive review of environmental history, palaeo-shoreline reconstructions, and known archaeological sites from the early Holocene. This synthesis enables the construction of a predictive framework for prehistoric occupation, identifying the topographic and environmental factors that shaped site location over time. Variables such as elevation relative to past sea levels, slope, water access, resource availability, and mobility routes are integrated into a statistical model that highlights zones with high archaeological potential.
Building on this foundation, the project employs drone-mounted LiDAR to acquire exceptionally detailed three-dimensional data of selected target areas. Unlike national LiDAR datasets, drone-based point clouds provide far higher point density and ground resolution, allowing the detection of subtle micro-topographical features that may correspond to archaeological remains. These may include dwelling terraces, activity platforms, pit structures, sediment extraction zones, or ancient pathways, features that are often invisible in conventional aerial imagery or coarse elevation models.
The processing of LiDAR data is coupled with deep-learning workflows, specifically convolutional neural networks (CNNs), which are trained to recognise and classify archaeological signatures within the landscape. By exposing the network to labelled examples of known anthropogenic and natural features, the system learns to distinguish subtle patterns in microrelief that are otherwise difficult for the human eye to identify. The CNN models automate and accelerate the detection process, increasing survey efficiency and enabling the systematic exploration of large, complex terrains.
Detected anomalies and potential archaeological features are then verified through targeted pedestrian surveys. These ground inspections serve to validate the remote-sensing results, refine interpretations, and document material or structural remains when present. Although the project does not include coring or excavations, non-invasive field verification remains crucial to assess the archaeological relevance of features identified through the deep-learning system.
The integration of these methods will result in a refined diachronic model of human occupation in Västerbotten from the early Holocene. By constructing a detailed archaeological map enriched with newly identified sites, the project will contribute significantly to our understanding of settlement dynamics, mobility strategies, and landscape interaction in northern Sweden. Furthermore, comparisons with postglacial regions where similar environmental transitions occurred will provide a broader context for interpreting patterns of colonisation and land use. Overall, this project represents a major methodological advance, demonstrating how remote sensing, machine learning, and archaeological theory can be combined to address long-standing questions about human adaptation in high-latitude postglacial environments.
