Assessing the role of spatial coordinates in K-means clustering for capturing geometric patterns in archaeological FDEM data

Identifying archaeological features through geophysical data interpretation remains a key challenge in non-invasive prospection. Geophysical surveys are increasingly combined with unsupervised machine learning, particularly K-means clustering, to enhance the detection of buried remains. However, it operates only in attribute space, ignoring the spatial nature of archaeological structures. This study assesses whether adding spatial coordinates in K-means clustering of Frequency Domain ElectroMagnetic data improves the delineation of anthropogenic features. A 2D approach, based only on electromagnetic components, is compared with a 4D configuration including geographic coordinates. FDEM data from the Iron Age site of Torre Galli (Italy) show that adding spatial coordinates does not outperform the 2D approach. Instead, results become more sensitive to normalization and often reflect spatial proximity rather than geophysical contrasts, reducing interpretability. These findings indicate that incorporating spatial coordinates into a standard K-means framework does not systematically improve archaeological feature delineation and increases sensitivity to data preprocessing.