Bastian Germundsson1,3, Anders Pihl2 and Kim H. Esbensen3,4
1Department of Geosciences and Natural Resource Management (IGN), Copenhagen University, Denmark. E-mail: [email protected]
2Bornholm Museum, Rønne, Denmark Email: [email protected]
3Geological Survey of Denmark and Greenland (GEUS). Copenhagen. Denmark. Email: [email protected] Phone: +45 20214525
4ACABS Research Group, Aalborg University, Denmark
In archaeology it is of interest to ascertain whether a particular Bronze-age field has been cultivated or not based on traditional archaeological evidences, but these often deal with one chemical element only, Phosphorous. We here augment this endeavour to include multi-element geochemistry characteristics. A pilot study sampling campaign was carried out (2014) on the island of Bornholm with the objective to discriminate between well-documented cultivated and un-cultivated Bronze-age agricultural fields based on multivariate data analysis (chemometrics) of soil chemistry (metal concentrations, ICP-MS). All samples originate from the same soil depth corresponding to the paleo-cultivated layer, or the equivalent depth in uncultivated fields. The experimental design focused on proper field sampling (Theory of Sampling), including replicate sampling at two levels. Applying Principal Component Analysis (PCA), the first three components corresponds to 68 % of the most discriminative variance in the 15 variable/41 sample array. The first and third PC-component reveals a complete discrimination of un-cultivated vs.3 cultivated fields; it is likely that general soil chemistry features are compensated for by the second component in the PCA solution. We present the specifics pertaining to the field sampling procedure, including the hierarchical two-level experimental design, which allow assessment of the local vs. field-wide heterogeneities in order better to understand the successful discrimination achieved. Five elements appear to be particularly involved in the discrimination [P, Fe, Mn, Zn, Pb], currently undergoing paleo-agricultural/geochemical interpretation. Based on these first results we plan a full test-set validation campaign in 2015 which will be the ultimate performance test for this type of archeometric discrimination. This contribution illustrates the versatility and power of multivariate data analysis (chemometrics) applied to data with a substantial proportion of potential sampling errors, in need of effective management (TOS).
Publication date: 9 June 2015