Abstract
Representing dynamic spatial phenomena is a long-standing goal within Geospatial Information Science (GIScience). Geometric positions are fundamental elements of any spatiotemporal process, such that a significant body of related studies has been dedicated to modelling the autonomous changes of geometries over time (movement). There has been a ‘causal revolution’ in computer science, economics, epidemiology, and social sciences, powered by large and complex spatiotemporal data. There is accordingly a need to expand the objectives of movement studies from descriptive models, data mining and pattern recognition to explanations of the underlying causes of movement.
From both a practical and theoretical standpoint, progress in developing approaches for filling the gap of explanatory movement models should be founded on a conceptual model. A short review based on a movement analysis informed examination of advantages and critiques concluded on the need for an integration of two approaches in a causality-led conceptual model for agents: simulation-based (i.e. agent-based modelling) and data-driven (i.e. graphical causal modelling). To achieve this integration, three conceptual levels of abstraction were proposed to frame an agent-based representation of movement decision-making processes: ‘attribute’, (environmental) ‘actor’, and ‘autonomous agent’. These, in combination with three temporal, spatial, and spatiotemporal general forms of observations distinguish nine (3x3) representation typologies of movement data within the agent framework. In addition to these nine representation typologies, there are three levels of cognitive reasoning about spatiotemporal phenomena: ‘association’, ‘intervention’, and ‘counterfactual’. Together, these make for 27 possible types of operation embedded in a conceptual cube with the level of abstraction, type of observation, and degree of cognitive reasoning forming the three axes.
The first application of the conceptual framework was to build and test an agent-based model that was constrained in time and space with clearly articulated rules. The movement of football players was modelled as a design guideline for a vector-agent model. The simulation was submitted to an accreditation process to show how agent-based movement simulations can contribute to our understanding of movement and how they can potentially produce causally relevant evidence.
Informed by the previous approach, an empirically-based implementation was developed to accommodate graphical causal models and their concepts within movement studies (i.e. the ‘association’, ‘intervention’ and ‘counterfactual’ levels of cognitive reasoning in the conceptual cube). This involved a three-level process for inferring quantitative causal evidence from passive observations. A directed acyclic graph was designed and computed on a generated movement dataset of football players to represent the extent to which a correct causal structure of an act of movement can be inferred. The results showed that, given adequate evidence about the probability distribution of the variables involved, graphical causal models could handle the complexity within the causal structure underlying a movement decision-making process.
This thesis has adapted causal thinking in computational movement analysis and put forward a conceptual data-driven agent architecture. A physical vector-agent architecture that shows how an autonomous moving agent, equipped with highly abstracted causal knowledge, compartmentalising its perceptions of the environment and its reasoning for movement, was demonstrated and tested. Issues regarding the implementation of this model in other application domains (e.g. animal movement) were recognised as the next challenge towards causal analysis across movement studies, addressing questions of generalisability and portability. The proposed conceptual model and approach to inferring causality shows one approach to achieving these goals.