LAUSR

Abstract Bees are widely recognized as biomonitors of the environment. However, simple, low cost indicators based on bee activities (other than mortality) that can be used to monitor agroecosystems’ environmental status have not been reported yet. Biological and chemical indicators were evaluated using Support Vector Machine (SVM) models. The objective was to classify environments into the categories of affected and non (or minimally) affected by pesticides presence. SVM models based on simple population and brood area estimations provided 57% accuracy in classifications. Models based on variables constructed using chemical analysis results (the number of pesticides detected in each sample; the total sum of toxic units, related to bees and mammals; and the sum of field Environmental Impact Quotient, EIQ, of each detected pesticide) provided 100% accuracy in classifications. This fact can be used to obtain biological indicators from a wide number of sites and apply the SVM model to indicate risky seasons and ecosystems where further chemical indicators should be studied. As the input variables are totally explicit and can be changed or updated, the approach followed in this study using SVM provides a stable basis to improve the development of indicators for agricultural production systems biomonitoring.