Department of Physics, Chemistry and Biology (IFM)

Of the total number of butterflies in traps, 2.7% remained unidentified (escaping during handling, or damaged body parts like wings), and they were excluded from analyses.

Species accumulation curves by trap as well as over time were done using EstimateS Ver 7.51, using the number of species observed (Mau Tao index) and the number of species expected (Chaos 2 index). Chao 2 was chosen as nonparametric estimator as it performs well on small samples (Colwell & Coddington, 1994) and has been considered as the least biased estimator dealing with total species richness (Bruno & Moore 2005). These analyses were used to evaluate the success of the sampling and to estimate the total richness in the area.

Multivariate statistical analyses were performed with the CANOCO 4.5 software package (ter Braak and Smilauer, 2002) using multivariate methods based on linear assumptions, as the beta-diversity in the data was relatively low (following recommendations by Leps and Smilauer, 2003). Species data were transformed (log₁₀(x+1)) in order to minimize the impact of very abundant species.

A first principal component analysis (PCA) compared the species composition in traps in understory and gap, with the light level included as a passive environmental variable.

A strict test of our assumption about the importance of gap size needs to consider the pair-wise nature of our data. We therefore used c ² distances within each pair of site, and the hypothesis that this measure of dissimilarity in composition would increase with gap size (linear regression). c ² distances were based on transformed species data (log₁₀(x+1)) and gap sizes were square-root transformed.

A second PCA involved only data from gaps, to assess the influence of gap size and including passive environmental variables describing vegetation structure (density, height, and diameter of vines and woody plants).

To further highlight the differences between gaps and understory traps, a partial redundancy analysis (pRDA) was conducted. This involved gap/understory as the only (categorical) environmental variable and trap-pair identity as a number of categorical covariables (thereby eliminating some of the potential spatial variation in the data). The statistical significance of the model was assessed in a Monte Carlo permutation test (9999 permutations). Furthermore, the model differentiated gap vs. understory specialist species. Finally, the species scores in the pRDA were correlated with the morphological data (log transformed), using the average of records per species.