Distribuição de aves florestais endêmicas do Cerrado : ocorrência e abundância na escala local e geográfica

Abstract

The relationship between a species and its environment may result in different distribution patterns, depending on spatial scale. In this regard, habitat and climatic variables can directly affect species distribution and occurrence. An increasing number of studies seek to understand species occupancy as a function of habitat and climatic factors, identifying key variables to which species respond. In the Cerrado region studies with bird species in this area are lacking. The Cerrado is one of the largest tropical savannas in the world and the second-largest vegetation formation in South America. About 72% of bird species are dependent or partially dependent on forest formations and 43% of the 30 endemic species are forest birds. Assessing the relationship between species and their abiotic environment is important to understand how spacial scale, habitat and climatic variables affect organisms distribution and abundance. Thus, considering the lack of information on the Cerrado biome, studies occupancy patterns are needed. Our aim was to understand distribution patterns of Cerrado endemic forest bird species and which factors affect their occurrence in macro and local scale. Also, we assessed and compared estimation methods of species geographic distribution that account for detection probability with methods that do not accunt for this parameter. Our aim in the first chapter was to estimate occupancy and detection probability of five Cerrado endemic forest bird species (Antilophia galeata, Herpsilochmus longirostris, Syndactyla dimidiata, Hylocryptus rectirostris e Basileuterus leucophrys), using single-season occupancy modeling. We evaluated hypotheses related to forest cover (NDVI – Normalized Difference Vegetation Index, and NDVI Standard Deviation) and to the landscape composition (amount of forest, native and non-native surrounding vegetation). We conducted 589 point counts surveys at a total of 310 points in 59 sites within gallery forests during September to December from 2011 to 2013. Model-averaged occupancy estimates (ψ) ranged from 0.33 to 0.75, and detection probability (p) was less than one (0.29-0.71) for all species. Forest cover (NDVI in a 100-m buffer) was an important predictor variable and occupancy probability increased with NDVI values for three bird species. Landscape composition variables had a weak effect on occupancy probabilities. NDVI is an indirect measure of primary production and sites with higher NDVI values might have more resources available to birds. Considering this, in a local scale, the effect of habitat quality on birds’ occupancy may be more important than the amount of available habitat. Our aim in the second chapter was to assess hypotheses related to bird species abundance across the geographic distribution - centroid-periphery and habitat suitability, and to local variables - forest amount and forest cover (NDVI – Normalized Difference Vegetation Index) in a 1km buffer around survey points. We expected to find a negative relationship between species abundance and the distance to the distribution centroid and a positive relationship with forest amount and forest cover. We estimate a suitability index using program MAXENT, considering bioclimatic variables. The mean abundance was 5.22 ± 0.72 to Helmeted Manakin, 2.98 ± 0.88 to White-striped Warbler and 4.62 ± 0.71 to Larged-billed Antwren. We found significant relationships with the variables considered just to White-striped Warbler. Contrary to our expectations, species abundance decreased with increasing forest amount and increased with distance to the distribution centroid. Regarding habitat suitability the relationship with species abundance was positive, abundance increased in more suitable sites. Thus, we found that at least for one bird species, abundance probably increase towards the periphery of the distribution in areas of suitable climatic conditions. Our aim in the third chapter was to understand the effect of detectability on the species distribution modeling and compare approaches that assume p=1 (MAXENT and BIOMOD) with single-season occupancy models that account for imperfect detection. We modeled the geographic distribution of five Cerrado endemic forest birds species (Chapter 1), with different detection (p=0.51 to 0.74) and occupancy probability (ψ=0.28 to 0.89), using all three approaches. We compared the predictive performance between occupancy models and the others by calculating the relative performance of each approach considering the AUC (Area Under the Curve-ROC) and the TSS (True Skill statistic). Contrary to our expectations occupancy models showed the worst predictive performance with a significant difference between these models and the others, considering AUC and TSS accuracy measures. Possible explanations for these results can be the poor quality of the validation data and the choice of the accuracy method to compare among the approaches. Our results showed that different approaches may result in distinct predictions, and it was evident when presence-absence models were compared with presence-only models.