Noisy data and distribution maps: the example of Phylan semicostatus Mulsant and Rey, 1854 (Coleoptera, Tenebrionidae) from Serra de Tramuntana (Mallorca, Western Mediterranean)

Abstract

Distribution maps are key tools for environmental management and biogeographic analyses. However, success in predicting spatial distribution is limited when using noisy presence/absence data sets. Both false absences and presences can be related with local departures from equilibrium (for example, temporary extinctions or unsuccessful colonisations). Moreover, false absences can arise from limited sampling effort. Here we explore an analytical strategy to get additional information on the presence/absence pattern of one target species from the presence/absence of all other species in the community. The logic is simple: the target species should display higher probability of presence at a site if a sample from this site is faunistically very close to the samples from other sites where the species occurs. Therefore, we first model presence/absence of the target species as a function of between-sample faunistic similarity. Second, the observed data for the target species are readjusted as a function of the expected probability of presence: current presences at sites with extreme low probability of presence are interpreted as unstable presences, and are recoded as absences. Seemingly, absences at sites with high probability of presence are interpreted as false absences, and are recoded as presences. In the experimental case presented herein, the recoding procedure is based on the presence/absence of 174 species, covering a broad taxonomic scope (snails, beetles, spiders and isopods). 1 km² distribution maps of presence/absence of the endemic beetle Phylan semicostatus were modelled from these recoded data. Mapping is done using GARP based on four environmental explanatory variables. These maps seem to be more stable and less prone to fail in predicting presence than those derived directly from the observed data.