LAUSR

Abstract Understanding interactions between individual animals and their resources is fundamental to ecology. Agent-Based Models (ABMs) offer an opportunity to study how individuals move given the spatial distribution and characteristics of their resources. When contrasted with empirical individual-resource network data, ABMs can be a powerful method to detect the processes behind observed movement patterns, as they allow for a complete and quantitative analysis of the agent-to-environment relationships. Here we use the small-scale, within-patch movement of bumblebees (Bombus pascuorum) as a case study to demonstrate how ABMs can be combined with network statistics to provide a deeper understanding of the mechanisms behind the interactions between individuals and their resources. We build an ABM that explicitly simulates the influence of distance to the nearest flowering plant (allowing minimal energy expenditure and maximum time spent foraging), plant height and number of flower heads (as a proxy of food availability) on local foraging decisions of bumblebees. The relative importance of these three elements is determined using pattern-oriented modelling (POM), where we confront the network statistics (number of visited plants, number of interactions, nestedness and modularity) of a real B. pascuorum individual-resource network with the emergent patterns of our ABM. We also explore the model results using spatial analysis. The model is able to reproduce the observed network statistics. Despite the complex behaviour of bumblebees, our results show a surprisingly precise match between the structure of the simulated and empirical networks after adjusting a single model parameter controlling the importance of distance to the next plant visited. Our study illustrates the potential of combining field data, ABMs and individual-resource networks for evaluating small-scale, within-patch movement decisions to better understand animal movements in natural habitats. We discuss the benefits of our approach when compared to more classical statistical methods, and its ability to test various scenarios in a new or altered environment.