APPLICATION OF MULTIVARIATE-ANALYSIS FOR THE SELECTION OF CANDIDATES FOR BIOLOGICAL-CONTROL AGENTS

A survey of natural enemies of Senna obtusifolia (Leguminosae: Cesalpi naceae) was made in 46 localities in Brazil. Twenty-one preselected in sect species were used in a multivariate analysis in a search for pote ntial biocontrol agents, based on frequency of occurrence of each spec ies, degree of insect/host association, and damage level. Insect commu nities were organized in a binary matrix (presence/absence of the spec ies); measures of distance and similarities were based on the Jaccard coefficient, and cluster analysis was then conducted. Construction of a dendogram showed similarities among geographically close localities where phenological stages of sicklepod plants were similar. Analysis o f correspondence based on the correlation matrix showed that Eigen val ues of the three first axes explain only 40.1% of the observed variati on. Correlation among the scores produced by species Eigenvectors in e ach locality on axis 1 and the latitude, altitude, and phenological st age showed that only phenological stage was correlated (r = 0.45, P = 0.02). The results indicate that weather will mot be a limiting factor for the establishment of these insect species in an environmental gra dient similar to that of the surveyed area (12 to 22 degrees latitude and 6 to 1200 m altitude). Distribution in the ordinal space of axes 1 and 2 showed that Agrilus oceanicum (Buprestidae), Fundella argentina (Pyralidae), Thyphedanus undulatus (Hesperidae), and Phoebis sennae ( Pieridae) are grouped and separated from the others in axis 2, suggest ing-that these species co-occur with each other. Based on their feedin g guilds, multiple introduction of these species can be considered. Mu ltivariate analysis showed potential to be used as an additional tool to help in the selection of candidates for biocontrol agents. (C) 1996 Academic Press,Inc.