Identifying tropical trees using comparisons with single specimens of species

Abstract

A taxonomic database system CARipé with integrated links to plant images was constructed to characterize the 800 forest tree species recorded for the catchments of the Rio Grandein the south of Minas Gerais state in Brazilusing vegetative morphology. After a description was ready, its integrated system Empar, completed in a few seconds an identification attempt using automatic character correlation. Empar proved operationally successful in identification tests of 62 specimens, of as many species, against 554 reference specimens and species from the ESAL Herbarium. Empar employed a similarity coefficient with the option of a weighting based on character frequency and, in tests, the use of this rarity weighting further improved the performance. Three tree groups based on leaf type and phyllotaxy were analyzed separately. From the start, those with compound leaves received more descriptors and the result was little difference in identification performance between the tree groups. Indications were that experts and non-experts performed characterizations for identification purposes almost equally well. The most important result was that successful identifications were obtained using a reference base of just one specimen description per species.