Evaluation of a FTIR data pretreatment method for PCA applied to archaeological ceramics

The aim of this work is to explore the potential application of Infrared spectroscopy and chemometric analysis as a tool to differentiate archaeological ceramics. We propose here a new spectral FTIR data pretreatment method for Principal Component Analysis (PCA). Indeed, the selection of a proper algorithm and its application could play an important role in minimizing the manipulation of data. The pretreatment method here proposed considers the data in the spectral region between 400 and 1600 cm-1 since this range takes into account the contribution of the fingerprint of inorganic compounds, below 1000 cm-1, and the carbonate absorption band at about 1400 cm-1. The normalization between 0 and 1 of FTIR data has been followed by the calculation of the second derivative. Indeed, mathematical treatments baseline correction [1-3] results to be affected by excessive modification due to the operator which can influence and modify the PCA results. For this reason, the second derivative of the spectra was considered to be the best choice as it involves a correction in the baseline and an increase of the small differences among the spectra, such as minor shoulders and peaks [4]. Finally, in order to minimize the manipulation of experimental data and to reduce the influence on PCA results, the proposed method does not include the smoothing of data unlike in previous papers. The proposed procedure has been applied to pottery samples: (1) simple and storage ware from Khirbet al-Batrawy (Jordan), III millennium B.C.; (2) black gloss ware from the Phoenician-Punic site of Motya (Trapani, Italy), VI-IV century B.C.; (3) glazed ceramics from the Palatine Hill (Rome, Italy), III-VI century A.D. The results of the test suggest that the method to process statistical data can be applied in the study of archaeological ceramic materials to discriminate between different ceramic productions. Moreover, a series of elaborations with PCA step by step are proposed in order to reduce the number of variables (from more than 1100 to about 200). This allowed to identify the discriminating vibrational bands, thus giving the possibility to explain the separation among samples on the basis of their mineralogical composition.