Deriving trajectories from videos to calibrate and validate agentbased models

The ability to move is an integral part of life, enabling the search for food or the escape from potential dangers. These behaviors are always embedded in socio-ecological systems which provide a context and a variety of stimuli. Agent-based models (ABMs) help to understand the socio-ecological systems by computationally describing the movement and interactions of entities on the basis of theory and analyzing the resulting macro-scale dynamics. Despite ABMs being based on theory, calibration is necessary because (i) for most systems the theory is incomplete, (ii) not all attributes of the entities can be measured, (iii) attributes naturally vary, and (iv) models are an inherent simplification of reality, resulting in aggregation of less relevant processes. In this work, we demonstrate the value of recent developments in image collection and recognition methods for the derivation of observational data on movement as input for the calibration of ABMs. The studied systems are dung beetles and desert ants. Videos are used to derive animal trajectories, which are, in turn, used to derive summary statistics of animal movement for the model (brute force) calibration and validation. While the desert ant study is still at its infancy, the preliminary results of the dung beetle study show that, with the use of the video data, the root mean square error of the movement speed was reduced from 1.6 cm/second to 0.1 cm/second (94%), and of the travelled distances from 362 to 278 cm (23%). Although the direct translation of these methods to ABMs of human systems (topic of this session) is hard, due to differences in e.g. scale, system stationarity, and privacy regulations for the respective entities, we do hope that our work may serve as inspiration for new calibration and validation data and methods of ABMs in general.