Statistical seed classifiers of 10 plant families representative of the Mediterranean vascular flora.

The aim of this study was to implement a seed morphometric and colorimetric database in order to develop specific statistical classifiers for 10 families representative of the Mediterranean vascular flora. The 501 analysed accessions consisted of 274 taxa belonging to 161 genera and 52 families. Images were acquired by a flatbed scanner and then elaborated by a macro software program developed for the morpho-colorimetric measurements. The mean seed weight was also calculated for each accession. Statistical classifiers were elaborated for genera and specific and infraspecific taxa discrimination within the following families: Apiaceae, Asteraceae, Boraginaceae, Brassicaceae, Caryophyllaceae, Cistaceae, Fabaceae, Lamiaceae, Poaceae and Scrophulariaceae. Such classifiers, based on Linear Discriminant Analysis and checked by cross validation, showed a performance ranging from 90.4% to 98.5% and from 87.8% to 98.3% at genus and taxon levels respectively. A first application of this technique to gymnosperm seeds was also presented. The performance of correct classification (93.6%) obtained for the genus Juniperus suggests that this technology is also reliable for gymnosperms. Seed weight improved the accuracy of the classifier system, confirming that an extensive database of bio-morphological traits constituted by a large number of seed species may be useful for taxonomic identification.