Butterfly species richness in mainland Portugal: predictive models of geographic distribution patterns

Abstract

A three‐step protocol described elsewhere is used to obtain a map of butterfly species density in Portugal on a 50×50 km grid. First, all available faunistic information was compiled and analysed to explore the historic patterns of butterfly sampling in Portugal, and to determine which grid cells are sufficiently prospected to produce reliable estimates of species richness. Then, we relate the estimated species richness scores from these areas to a set of environmental and spatial variables by means of General Linear Models, obtaining a function to extrapolate of species density scores to the rest of Portugal. Finally, the model is validated, results explored and outliers identified and deleted. Any spatial autocorrelation remaining in the residuals is examined. Lastly, model parameters are recalculated in absence of deleted outliers, and the resulting function is used to predict species richness scores throughout mainland Portugal. A highly‐predictive function based on some variables previously related to butterfly composition at macro‐scale, such as number of sunny days per year, temperature or environmental heterogeneity, was obtained. However, in Portugal those variables are highly spatially structured along a steep latitudinal gradient, leading to difficulty in ascertaining if the latitudinal gradient detected by our analysis is due to macroecological or historic effects. Information on European and Iberian butterfly assemblages and causal processes are discussed in the light of the patterns observed. Then, previous information obtained on Portuguese scarabs is added to identify conservation areas, biogeographically important for both insect groups. Finally, the main drawbacks and advantages of this approach to mapping biodiversity for conservation are discussed briefly.

A synecological framework for systematic conservation planning

“…For example, as Atlas data has improved, many predictions developed from old Atlases agree with the data presented in new ones (Dennis and Shreeve 2003). Biodiversity is now modelled mainly from: i) single species distribution data, one-by-one (autoecology, see Guisan and Zimmermann 2000;Scott et al 2002;Ferrier et al 2002a;Pearson and Dawson 2003;Peterson et al 2004;Soberón and Peterson 2005;Araújo and Guisan 2006); or ii) composite biodiversity variables, such as species richness, assemblage composition, endemism, rarity, and others, assumed to be biodiversity surrogates for monophyletic groups (synecology; see Lobo and Martín-Piera 2002;Hortal et al 2001Hortal et al , 2003Hortal et al , 2004Ferrier 2002;Ferrier et al 2002b;Ferrier and Guisan 2006).…”

“…In the same way, we also believe that the 'first predicting and then assembling' strategy would result in erroneous pictures of the distribution of biodiversity due to the effect of these two topics, underperforming the results obtained by assembling first, and then predicting. On the contrary, we hypothesize that both the aggregation of errors and the lack of representation of rare species can be overcome by using the first strategy (assemble first, predict later) if two steps are previously added to the modelling protocol: a sampling effort assessment to identify the well-sampledenough areas (and discard those with unreliable inventories), and the extrapolation (when possible) of the scores of the synecological variables to diminish the effects of incomplete inventories (i.e., checklists are unlikely to be complete even at well-sampled areas) in these areas (see Hortal et al 2004).…”

“…As these variables are based on data from many species in each territory unit, sampling effort assessment is easier. Species accumulation curves (Soberón and Llorente 1993;Colwell and Coddington 1994) can help to identify well-sampled areas for a given group of species (Lobo and Martín-Piera 2002;Hortal et al 2001Hortal et al , 2004Martín-Piera and Lobo 2003;Jiménez-Valverde and Hortal 2003). They can also yield estimates of richness scores for each territory unit, from information restricted to that unit (in addition to the many other species richness estimators available; see, e.g., Soberón and Llorente 1993;Colwell and Coddington 1994;Chazdon et al 1998;Gotelli and Colwell 2001;Chiarucci et al 2003;Brose et al 2003).…”

“…Reliability of models, built from more complete and robust data, is improved too. Interestingly, there is a reduction, or even elimination, of sampling bias in scores from models developed from asymptoticallyestimated richness values (see Hortal et al 2004). Moreover, the set of species considered is enlarged by the inclusion of rare species in the estimates of biodiversity surrogate scores, which become a major part of final values.…”

Assessment of richness estimation methods on macroinvertebrate communities of mountain ponds in Castilla y León (Spain)

“…This ideal curve represents an unbiased description of the sampling process, where possible effects due to the order by which the samples have been taken or listed are removed by randomizing their order of entrance in the curve. We used the software Statistica 6.0 to fit each function to the data and calculate the asymptote value from the obtained parameters (Sobero´n and Llorente, 1993;Hortal et al, 2004). These accumulation functions are able to predict estimate richness when they are close to the asymptote.…”