Eco-Anthropological factors explaining forest patch use by 3 species of wild Atelid monkeys co-existing with a small-scale farming community in Northeastern Costa Rica, Central America

The main conservation risks for wild non-human primates (NHP) in Costa Rica, Mesoamerica, is deforestation and the allocation of lands for agriculture. This is because they result in a mosaic of forest patches that differ in size and ecological properties. NHP, being the vertebrates with the highest risk and rate of extinction, slowly adapt to this rapid environmental change, optimizing their metabolic costs to survive and reproduce. One way to balance these costs is to use forest patches depending on the benefits they provide, such as food, shelter, or social contact. To understand the possible environmental factors that predict the usage of a series of 8 connected forest patches by Ateles geoffroyi, Alouata paliatta, and Sapajus imitator we collected demographic, behavioral, climatological and other environmental data from 2018 until 2021. We used information-theoretic metrics to identify the factors that best explained the presence and behavior of the species of interest in the forest patches studied, and fit the data to a set of models built informed a priori. Using the best explanatory factors, we k-fold cross-validated 9 classifier algorithms to identify the best predictive models for the presence of the monkeys studied and their behavioral patterns given the data. Presence was highest in warmer, more humid days, especially when other groups were present in the same patch. Behavioral patterns were different in each patch; monkeys rested more often when other groups of the same species were present, and foraged more during warmer, more humid days, and smaller groups. Predictive models for the presence of the species studied, trained with the 3 best explanatory factors, reached an accuracy between 70-96%, with Gradient Boost Classifier performing the best. In contrast, behavioral patterns were more unpredictable, with the the algorithms tested only reaching between 43-51% accuracy, the AdaBoost Classifier being the best. Our findings suggest that the usage of the 8 forest patches monitored by the monkeys studied depends on patch characteristics, not related to size nor the presence of a reserve, by the presence of other NHP in the patch and the meteorological conditions. Further work on the ecological characteristics of these patches can clarify the mechanisms modulating behavioral patterns.