Uma nova abordagem para modelagem de distribuição de espécies a partir de registros restritos: aplicação em projeções de padrões futuros de formigas do cerrado

Abstract

The joint effect of changes in climate and on natural environments has triggered the use of species distribution models for conservation planning. However, application of these models has been very limited for species whose geographic information is restricted, a common fact in rare or endemic species, categories which are especially relevant from the point of view of preservation. Insects, an important functional group that compose the largest portion of terrestrial biodiversity, are rarely considered in conservation studies. In this context, the present thesis aims to: 1) evaluate modeling techniques that are less sensitive to the data restriction condition, characterized by both the quantity and the geographic range of the sample; 2) to apply these methodologies to predict the current and future pattern of the species of ants typical of the Cerrado and other types of savanna vegetation of South America. The first goal was achieved by comparing models resulting from the use of restricted data versus those resulting from the use of all data. These comparisons were made for three ant species - which have different geographic patterns and quantities of occurrence records - using two modeling techniques - Maxent and Weights of Evidence. Weights of Evidence is a Bayesian method widely used in environmental modeling, but that has never been successfully applied to modeling species distribution. In order to adapt this method for the application in our available data, I have controlled the number and location of pseudo-absences during the calibration phase. For each species, data restriction conditions comprised between 5 and 27 occurrences, restricted within the Cerrado Biome, while broad data varied between 16 and 47 records distributed throughout South America. The performances of each experiment were assessed by: 1) visual inspection of probability maps; 2) TSS (True skill statistics) evaluated according to different scales. The Weights of Evidence method provided significantly better results than Maxent, especially when the goal was to estimate the distribution of the species outside the range of the calibration area. Then, given the best performance of the Weights of Evidence method, it was applied to model current and future distribution of 12 species of ants typical of savanna vegetation in South America. Estimated distributions have shown that under the worst scenario, these species will have lost between 37% and 88% of their current estimated range. This study suggested that Linepithema cerradense is the least threatened species. On the other hand, Blepheridatta conops and Pheidole cyrtostela might be the most vulnerable ones, with less than 16% of their current estimated distribution to be preserved in 2070. The other species may preserve, by 2070, between 17% and 32% of their current estimated distributions. These species might have, on average, 13% of their current suitable area overlapped with protection zones, a value that can decrease to only 5% by 2070.