Limnetica 40

Planktonic diatom data sets were compared within different rarity categories to verify their responses using the weighted average (WA) approach. Our hypothesis is to proportionally reach an increased performance for the WA model by reducing the taxa weight according to their categories considering that the use of various cut-off criteria may affect the predictive abilities of different models. The underlying assumption is that WA models are unable to characterize optima and tolerances for low-occurrence taxa; in addition, their overriding may improve the overall model performance. Therefore, we developed forty diatom-training sets for six reservoirs located at two different basins in the Southwest São Paulo, Brazil. 339 diatom taxa were identified and built different models based on their relative abundance values, occurrence frequency, and the data set with no species deletion. The optimum and tolerance per taxon through the WA formula (Zelinka-Marvan weighted averages) was estimated according to their abundance values, in addition to using the pH in the data sets to infer the environmental conditions based on the sample taxonomic composition. First, the procedure using the complete dataset was repeated and subsequently with the down weighting taxa according to their rarity categories. The following procedures were advanced: comparison of predicted and measured pH values via regression analysis, and estimation of the species deletion effects on the predictive ability of the different models in terms of the coefficient of determination values (r2) of the response curves. In contrast to what is expected, since r2 values down weighting rare taxa had negative effect on the transference functions, data sets manipulation had significant influence on WA models performance. However, deleting non-abundant taxa had a positive effect on p values, thus providing robust reconstructive relationship. This work contributes to an improved understanding on diatom ecology, especially in tropical reservoirs, supporting the development of a diatom biological monitoring protocol for the study area.