- On the identification of optimal plant functional types - 631 Journal of Vegetation Science 10: 631-640, 1999 © IAVS; Opulus Press Uppsala. Printed in Sweden Abstract. The study of vegetation response to environmental change on a global scale cannot rely on species because most plant species have geographically limited distributions. To allow ecological predictions beyond the scale of the floristic region, models have to rely on vegetation descriptions using plant types other than the species. The crucial problem is how to define the types. Since types are described by traits, the problem translates into one of optimal trait selection. The best plant traits are those that when used to define plant types optimize the perception of association between vegetation and environmental (e.g., climate, disturbance) variation. I con- sider trait selection as a two-step procedure. The first step is the selection of a larger trait set based on past experience and known practicality, which is used for community description. The second step, for which the paper describes new methods, is accomplished on the data analytically by suitable computer algorithms that can find the optimal subset among the pre- selected traits. This subset defines optimal plant functional types (PFTs). The methods involve a fuzzy set approach and community description by plant types. The optimization algo- rithms described are tested with data from plant communities in South and North America. The utility of the approach in the evaluation of convergence of phylogenetically distant plant communities is discussed. Keywords: Chaco; Climate; Convergence; Fuzzy set; Leaf functional type; Monte; Optimal PFT set; Sonoran Desert. Nomenclature: Prado (1993). Introduction To describe communities we need a taxonomy to dissect the assemblage of organisms into populations. We take these populations as community components. Pioneer studies in vegetation recognized plant types by morphology and function (see Du Rietz 1931 and refer- ences therein). Vegetation science has been essentially based on species, but this view has been criticized (e.g. Grime 1979; Ghiselin 1987). It is clear why we need vegetation descriptions with plant types defined by traits and not only by species. On a large scale, predictions based on plant species are geographically bound (Wood- ward & Cramer 1996). On a small scale, species are in On the identification of optimal plant functional types Pillar, Valério DePatta Departamento de Ecologia, Universidade Federal do Rio Grande do Sul, Porto Alegre, 91540-000, Brazil; Fax +55513191568; E-mail vpillar@ecologia.ufrgs.br some instances so broad and variable that by describing communities by species composition we may not per- ceive relevant patterns occurring below the resolving power of species (Díaz et al. 1992). The problem has been studied in connection with the IGBP (Steffen et al. 1992), where these types are designated as ‘plant func- tional types’ (PFTs). It is well known that regions of the world with similar climates tend to support structurally similar veg- etation (e.g. Naveh 1967). This led to the proposition that community evolution is convergent (e.g. Mooney & Dunn 1970; Barbour & Diaz 1973; Orians & Solbrig 1977a), but suitable analytical techniques were lacking. The existence of PFTs with global validity is particu- larly grounded on this hypothesis. There are different approaches to analyse vegeta- tion data based on traits or plant types. A prevalent analytical scheme (e.g. Feoli & Scimone 1984; Díaz et al. 1992; Díaz & Cabido 1997) is the multiplication of the matrix of traits by species by the matrix holding presences/absences or performances of the species in the sampled sites, yielding a matrix of traits by sites. The latter is then analysed by conventional multivariate meth- ods. A limitation of this approach is that the traits must be binary or quantitative. A trait which is qualitative multistate may be coded into as many binary traits as the number of states, but then problems of unequal weight- ing and non-independence among traits may arise (Feoli & Scimone 1984; Digby & Kempton 1987). PFTs may be groups of species defined by cluster analysis of the matrix of species by traits. We may instead define plant types as a combination of trait states, in which case the classification is monothetic. Orlóci (1991) used the term ‘character set type’ for trait combinations. He suggested accordingly an analytical scheme in which the traits are considered hierarchically based on a data matrix of character set types by sites (see Orlóci & Orlóci 1985; Orlóci et al. 1986 and the review by Pillar & Orlóci 1993a). The analysis may adopt a fuzzy set approach (Pillar & Orlóci 1991). The crucial problem, however, is how to select the traits so that the PFTs will likely be ‘functional’. The traits must be observable expressions of forms or behaviors