Beyond site-specific archaeological prospection: Spectral-geometry reveals cross-site cropmark signals in satellite data
- 1Department of Civil Engineering and Geomatics, Cyprus University of Technology , Limassol, CY-3036, Cyprus
Abstract
Remote sensing has long supported archaeological prospection, mainly through the detection of cropmarks, a key archaeological proxy expressed as spectral vegetation anomalies over subsurface features. Yet, cropmark detection is still assumed in the literature to be highly site-specific, governed by local agronomic and environmental conditions. Using multispectral Sentinel-2 data from three European archaeological sites, we show that models trained in one landscape can recover meaningful archaeological proxy signals in other locations. By relying strictly on satellite spectral signatures for training, our models achieve measurable cross-location transferability without incorporating ancillary environmental data. We find that cross-site performance arises not from geographic proximity but from the spectral-geometric similarity of vegetation responses, and that transferability is strongly directional. We quantify this relationship through a Directional Strength Score (DSS), derived from monthly subcluster distances under multiple norms, which captures directional geometric similarity and correlates strongly with machine-learning performance. To keep the analysis grounded in the underlying spectral signal, we relied on architecturally restrained nonlinear classifiers (SVM and constrained XGBoost) and used NDVI-identified phenophases to align training and evaluation with comparable seasonal windows. Taken together, these results demonstrate that partial, spectrally mediated cross-location cropmark detection is feasible and that the underlying vegetation responses exhibit stable spectral structure, with spectral geometry offering a clear interpretive lens on why these signals can transfer across sites. Leveraging globally harmonized satellite data, this approach offers a scalable complement to uneven historical aerial archives and supports systematic regional archaeological prioritization to be deployed within integrated archaeological workflows.
Related
- Optimizing spectral-matching component separation for measuring CMB B-mode polarization with a satellite mission
- χ‑Spectral Geometry of Cosmological Mass Renormalization: Operator‑Topological Reconstruction of the Perimeter Institute "Dark Force" Effect
- Operator‑Spectral Geometry of Rotational Gravity: χ‑Modular Curvature, Spectral Flow, and the Precision Lense–Thirring Measurement of LARES‑2
- Topological Operator Geometry of ZEBTS: Renormalizability, Quantum Consistency, and the Spectral Theory of Dark Matter
- χ‑Spectral Geometry of Reality: Operator‑Geometric Origin of Spacetime, Curvature, and Gravity
BibTeX
@article{gravanis2026beyond,
title = {{Beyond site-specific archaeological prospection: Spectral-geometry reveals cross-site cropmark signals in satellite data}},
author = {Elias Gravanis and Athos Agapiou},
journal = {PNAS Nexus},
year = {2026},
month = oct,
doi = {10.1093/pnasnexus/pgag336},
url = {https://academic.oup.com/pnasnexus/advance-article/doi/10.1093/pnasnexus/pgag336/8907313},
note = {CC BY-NC 4.0},
}
RIS
TY - JOUR
TI - Beyond site-specific archaeological prospection: Spectral-geometry reveals cross-site cropmark signals in satellite data
AU - Elias Gravanis
AU - Athos Agapiou
PY - 2026
DA - 2026/10/10
JO - PNAS Nexus
DO - 10.1093/pnasnexus/pgag336
UR - https://academic.oup.com/pnasnexus/advance-article/doi/10.1093/pnasnexus/pgag336/8907313
AB - Remote sensing has long supported archaeological prospection, mainly through the detection of cropmarks, a key archaeological proxy expressed as spectral vegetation anomalies over subsurface features. Yet, cropmark detection is still assumed in the literature to be highly site-specific, governed by local agronomic and environmental conditions. Using multispectral Sentinel-2 data from three European archaeological sites, we show that models trained in one landscape can recover meaningful archaeological proxy signals in other locations. By relying strictly on satellite spectral signatures for training, our models achieve measurable cross-location transferability without incorporating ancillary environmental data. We find that cross-site performance arises not from geographic proximity but from the spectral-geometric similarity of vegetation responses, and that transferability is strongly directional. We quantify this relationship through a Directional Strength Score (DSS), derived from monthly subcluster distances under multiple norms, which captures directional geometric similarity and correlates strongly with machine-learning performance. To keep the analysis grounded in the underlying spectral signal, we relied on architecturally restrained nonlinear classifiers (SVM and constrained XGBoost) and used NDVI-identified phenophases to align training and evaluation with comparable seasonal windows. Taken together, these results demonstrate that partial, spectrally mediated cross-location cropmark detection is feasible and that the underlying vegetation responses exhibit stable spectral structure, with spectral geometry offering a clear interpretive lens on why these signals can transfer across sites. Leveraging globally harmonized satellite data, this approach offers a scalable complement to uneven historical aerial archives and supports systematic regional archaeological prioritization to be deployed within integrated archaeological workflows.
SN - 2752-6542
ER -