Methodological Framework – Building Applied ABMs

Effective ecological management requires tools that can adapt to the complexity and variability of real-world systems. To achieve this, we propose a framework for building applied agent-based models (ABMs) that integrate data collection, modeling, and decision-making into an adaptive feedback loop. This approach ensures that ABMs remain relevant and actionable as new data becomes available, enhancing their practical application for conservation and management. The framework is structured to guide the iterative development and use of ABMs in applied ecology. It begins with identifying specific management or policy goals, followed by parameterising the model using diverse field data to reflect real-world systems. The model outputs actionable insights for management decisions and highlights areas where additional data could improve future iterations. This continuous refinement process helps to reduce uncertainty and improve the reliability of model-based recommendations over time. Each component of this framework is detailed in the following sections, including practical guidelines for parameterisation, validation, and application. To illustrate the framework’s utility, we refer to a previously published case study (Murphy et al. 2024) that applies this methodology to explore wildlife hosts’ role in zoonotic disease transmission within a disturbance-driven landscape. While the case study itself is a work apart from this paper, it demonstrates how the framework can be effectively applied in real-world scenarios.Objective ABMs can only be effectively assessed and refined if they are guided by clear, measurable objectives that align with the desired management or policy outcomes (Richardson, 2020). This requires a clear understanding of the purpose the model is intended to serve and tailoring it to provide actionable insights for stakeholders. Setting objectives begins with actively engaging policymakers, wildlife managers, conservation practitioners, and other stakeholders to understand their needs, the context of their decisions, and the type of outputs required (Decker et al. 2012). A key consideration when refining the objective is identifying the available data and aligning it with the desired outputs. For example, an ABM designed to guide predator-prey management might focus on providing year-end prey population indices to set optimal hunting quotas in multi-predator landscapes. By focusing on outputs that directly address the questions at hand, the model remains streamlined, avoiding unnecessary complexity while ensuring relevance to practical decision-making (Murphy et al. 2020). A fundamental objective of these ABMs should not only be to provide immediate management and policy recommendations but to guide future data collection by identifying gaps or uncertainties, such as gaps in the available field data. This feedback loop between modeling and field data ensures that each iteration of the model provides validated recommendations to improve species management within the system while also identifying data limitations that can be addressed in future monitoring efforts.
Data Collection
Data collection to provide high-resolution input data is the foundation of building effective applied ABMs. The quality and resolution of input data determines the model’s ability to emulate complex systems. A well-designed data collection strategy ensures that all available datasets, as outlined in Table 1, are integrated to achieve the model’s specific objectives while maintaining parsimony (Murphy et al. 2020). Collaboration with stakeholders and policymakers during this process ensures the inclusion of all relevant data needed to accurately emulate the system which can foster increased trust in the model’s outputs. Remote sensing data provides critical environmental inputs, enabling the characterisation of landscapes and environmental processes, such as land cover, vegetation indices, and anthropogenic impacts (Skidmore et al. 2021). Telemetry data from GPS and VHF collars captures wildlife movement and behaviour, ensuring realistic agent procedures represented in the model (Hebblewhite and Haydon, 2010). Site-specific spatial monitoring data enhances spatial accuracy, while administrative data incorporates human activity and jurisdictional boundaries, critical for addressing human-wildlife interactions and accurately implementing realistic management and policy interventions in the model. Historical records and grid-based environmental data further enrich the model by allowing for the creation of dynamic patch based sub-models (e.g. population change) which alter agent behaviour and allow for emergent trends as the model progresses forward in time. These datasets collectively allow the model to simulate realistic and dynamic systems, as summarized in Table 1. A crucial component of applied ABMs is obtaining external validation data, which evaluates the model’s performance. Unlike other inputs, validation data must be independent of the datasets used to parameterise the model. However, obtaining such data can be challenging, particularly in regions or for species with limited monitoring efforts. In these cases, proxies be useful alternative data to assess model performance (Murphy et al. 2022). The identification of validation dataset gaps also highlights opportunities for future field data collection, creating a feedback loop between site and simulation that improves our understanding of both systems.
Parameter Databases
The creation of parameter databases bridges the gap between raw input data and the realism of agents and environments in applied ABMs. These analyses transform observational data into distributions of behaviours (e.g., habitat selection, movement speed), enabling agents and environments to access real-world field data to inform their behaviours such that they exhibit data-driven, realistic actions within the model. This approach replaces rule-based procedures with behaviours grounded in empirical data while still allowing for emergent dynamics by drawing parameter values from distributions. Telemetry data from GPS or VHF collars frequently serves as the foundation for these databases (Fig. 1). Using the amt package (Signer et al., 2019), key parameters such as movement speed, turning angles, and step lengths can be calculated to reflect realistic movement patterns. Advanced analyses, such as Kernel Density Estimation (KDE) to estimate home ranges, can be conducted using the adehabitatHR package (Calenge, 2023). Net squared displacement, a metric for quantifying movements such as dispersal or migration, provides additional insights into animal movement at specific points of the year. For habitat selection analysis, Resource Selection Functions (RSFs) can be implemented by correlating GPS points with the availability of environmental data using the glmm function in the lme4 package (Bates, 2010). RSFs compute relative selection probabilities for habitats within the model, which parameterises agents to assess their environment and select the optimum habitat and path in a data-driven way.