Bonobo nest site selection and the importance of predictor scales in primate ecology.

The role of spatial scale in ecological pattern formation such as the geographical distribution of species has been a major theme in research for decades. Much progress has been made on identifying spatial scales of habitat influence on species distribution. Generally, the effect of a predictor variable on a response is evaluated over multiple, discrete spatial scales to identify an optimal scale of influence. However, the idea to identify one optimal scale of predictor influence is misleading. Species-environment relationships across scales are usually sigmoid increasing or decreasing rather than humped-shaped, because environmental conditions are generally highly autocorrelated. Here, we use nest count data on bonobos (Pan paniscus) to build distribution models which simultaneously evaluate the influence of several predictors at multiple spatial scales. More specifically, we used forest structure, availability of fruit trees and terrestrial herbaceous vegetation (THV) to reflect environmental constraints on bonobo ranging, feeding and nesting behaviour, respectively. A large number of models fitted the data equally well and revealed sigmoidal shapes for bonobo-environment relationships across scales. The influence of forest structure increased with distance and became particularly important, when including a neighbourhood of at least 750 m around observation points; for fruit availability and THV, predictor influence decreased with increasing distance and was mainly influential below 600 and 300 m, respectively. There was almost no difference in model fit, when weighing predictor values within the extraction neighbourhood by distance compared to simply taking the arithmetic mean of predictor values. The spatial scale models provide information on bonobo nesting preferences and are useful for the understanding of bonobo ecology and conservation, such as in the context of mitigating the impact of logging. The proposed approach is flexible and easily applicable to a wide range of species, response and predictor variables and over diverse spatial scales and ecological settings.