## Landscapes as Execution Logs: Abductive Computational Rule Discovery from Time-lapsed Observations Geospatial agent-based modeling (ABM) is a computational method built on three primitive agent types that map directly to geospatial data structures: fields (rasters of cells/patches storing continuous spatial properties), particles (vector points as mobile entities), and links (vector polylines and polygons defining relationships and boundaries) (Wilensky & Rand 2015, Crooks et,al 2019 ). Each agent operates according to local rules, making decisions based on its own state and neighboring agents across all three types. These primitives extend naturally to 2D, 3D, and higher-dimensional or time-varying geometries. By grounding fields, particles, and links in geographic data such as topography or infrastructure, these models reveal how local interactions scale up to produce emergent, system-wide patterns critical to understanding spatially complex phenomena. Abduction is a form of reasoning that infers the best explanation for observed phenomena, working backward from outcomes to discover the rules or processes that generated them. Charles Sanders Peirce, who formulated this mode of inference in the late 19th century, described it as reasoning that generates explanatory hypotheses when confronted with facts that require interpretation (Peirce 1931-1958). Unlike deduction, which applies known rules to predict outcomes (if A then B), or induction, which generalizes patterns from repeated observations (many instances of A lead to B, therefore A causes B), abduction starts with surprising or unexplained observations and seeks the simplest set of rules that could have produced them. In the context of geospatial modeling, abduction is particularly appropriate because landscapes present us with complex emergent forms—channels, vegetation patterns, sediment structures—whose generating processes are not immediately visible. Rather than assuming we know the rules beforehand (deduction) or waiting to observe enough repetitions to establish statistical patterns (induction), abduction allows us to treat observed landforms as evidence from which to reverse-engineer the underlying interaction rules and asymmetries that govern landscape evolution. Abductive rule discovery in geospatial agent-based modeling treats the landscape as an active computational process where observed forms serve as execution logs. In this framework, the designer does not begin with a set of known rules. Instead, they use time-lapse data from physical systems, such as the Emriver stream table used in the class or in the field (Cox et al. 2024), to observe real-world trajectories through phase space (Guerin & Douglas 2025a). This empirical data allows for the abduction of ABM primitives and the specific asymmetries governing their interaction. The resulting agent-based models are themselves abducted explanations—hypothetical generative mechanisms inferred from observed landscape dynamics rather than derived from theoretical first principles. This approach supports a computational geospatial curriculum designed for community decision-making by grounding abstract models in observable phenomena ### Intensive Asymmetries as Control Parameters In this model, intensive gradients are expressed as asymmetries within fields, particles, or the topology of links and edges. These asymmetries function as control parameters which determine the behavior and shape of the resulting forms. When these intensive differences reach critical thresholds, they drive the system through phase transitions. The resulting emergent structures are order parameters, or extensive forms, which represent the material registration of underlying forces. By adjusting the temporal intervals of the sensing media, designers can identify the exact points where structural reorganization occurs. This process emphasizes computer-based experiments to explore complex questions across disciplines. ### The Abductive Workflow for ABM Formulation The process of moving from sensing to simulation follows a specific abductive logic: * **Empirical Observation:** Multispectral time-lapse videos of physical systems capture the extensive evolution of forms like channels or vegetation patterns. * **Phase Space Mapping:** The video is treated as a dataset that reveals how the system moves between different states or phases. * **Rule Discovery:** The designer uses AI for abductive rule discovery using pattern inference to explore emergent system behaviors and identify the interaction rules of the primitives (Guerin & Douglas 2025b). For example, transformer-based models can generate Lattice Boltzmann simulations of the stream table, translating observed hydrological patterns into computational fluid dynamics. This methodology reframes the site as an assemblage of space-time structures. It enables the discovery of latent system behaviors that are typically invisible to the human sensorium. The use of LLMs to generate interactive HTML/JavaScript filters accelerates this abductive cycle by lowering the technical threshold for environmental computation. --- ### References Cox, S.J., Manavi, K.M., Guerin, S., Duckworth, S.M. (2024). Developing a Computational Geospatial Curriculum for Community Decision Making. In: Yang, Z., Krejci, C. (eds) Proceedings of the 2023 International Conference of The Computational Social Science Society of the Americas. CSSSA 2023. Springer Proceedings in Complexity. Springer, Cham. Crooks, Andrew T., et al. (2019) Agent-based Modelling and Geographical Information Systems: A Practical Primer. SAGE Publications. Guerin, S., and Douglas, C. (2025a). Stream Vision. https://harvardviz.live/cognitive-landscapes-group/stream-vision.html (Retrieved January 6, 2026) Guerin, S., and Douglas, C. (2025b). Stream Table. https://harvardviz.live/cognitive-landscapes-group/streamtable.html (Retrieved January 6, 2026) Peirce, C.S. (1931-1958). Collected Papers of Charles Sanders Peirce, Volumes I-VI. Harvard University Press. Wilensky, U., & Rand, W. (2015). An Introduction to Agent-Based Modeling: Modeling Natural, Social, and Engineered Complex Systems with NetLogo. MIT Press.