Fig. 4: Movement procedure for badger agents in an ABM,
illustrating two scenarios. The top panel (B and C) shows movement in a
homogeneous habitat, where agents follow step lengths and turning angles
from movement data without considering habitat suitability. The bottom
panel (D and E) shows agents in diverse habitats, selecting the best
option based on Resource Selection Function (RSF) scores, ensuring
adaptive, data-driven movement.
A complementary example is migration, where both static and dynamic data
play critical roles. Static input data can dictate the timing of
seasonal migration based on known patterns or environmental triggers
(e.g., temperature thresholds or day length) and identify suitable
migration endpoints, such as areas with high NDVI or optimal elevation
ranges. During migration, agents access external data to adapt their
routes in response to changing environmental conditions, such as
selecting paths with favourable topography. This integration of static
seasonal triggers with dynamic movement decisions enhances the model’s
ability to simulate realistic migratory behaviours. These examples
showcase how ABMs can combine static parameters, such as
telemetry-derived behaviours or seasonal triggers, with continuous
access to external databases to inform agent decision-making.
Most traditional ABM procedures rely heavily on rule-based or random
behaviours, which can limit the accuracy and ecological validity of
model outputs (DeAngelis & Diaz, 2019; Crevier et al, 2022). By
integrating dual data sources—static input data and continuously
accessed external databases—this method moves beyond such limitations
to enable data-driven procedures. Importantly, these parameters are
drawn from distributions of values derived from field data, introducing
an element of variability that mirrors real-world unpredictability. For
example, step length may range from 5m to 500m, allowing agents to
exhibit diverse and context-sensitive movement behaviours. This inherent
variability is critical for accurately simulating animal movement,
behaviour, and environmental dynamism. This unpredictability within a
data-driven framework is a significant advantage, as it allows for
emergent trends in model outputs that can lead to new discoveries. For
instance, agents may exhibit unexpected patterns of habitat use,
dispersal, or resource selection, revealing insights that are not
immediately apparent from field data alone. These emergent behaviors
enhance the model’s value not just as a predictive tool but as a means
of exploring ecological processes under varied conditions.
Most importantly, because the procedures are grounded in field data,
they are inherently more realistic and ecologically relevant than
rule-based or random approaches. This foundation ensures that the trends
and predictions emerging from the simulations are reflective of
real-world dynamics, making the model outputs more reliable for
informing management decisions and advancing ecological understanding.
Experimentation & Analysis
A well-crafted experimental design must aim to extract data at
critical moments within the simulation, as well as collect cumulative
end-of-model outputs, to provide a comprehensive understanding of the
system under study. One of the most powerful aspects of ABMs lies in
their ability to capture high-resolution data from every single event of
interest—something that is simply unattainable in field ecology. For
instance, in a predator-prey model designed to assess the impact of wolf
density on red deer populations, the model can record detailed
information for every predation event, including the pack ID, pack size,
date, season, habitat, elevation, prey species, prey sex, prey biomass,
and prey density at both territorial and landscape scales. In field
ecology, researchers rely on representative sampling to infer patterns,
but in ABMs, every event is accounted for, allowing for near-complete
data on system dynamics. In addition to capturing fine-scale events,
end-of-simulation reporters play an equally important role in
summarizing broader ecological trends. These outputs might include
overall changes in prey density, the number of natural mortalities, or
metrics related to management interventions, such as the number of
issued hunting licenses.
By integrating both fine-scale event data and cumulative
end-of-simulation outputs, researchers can analyse system dynamics
across multiple scales, offering a nuanced understanding of the
relationships between agents, their behaviours, and their environments.
Once simulation data is generated, robust statistical modeling is
necessary to draw meaningful conclusions. Comprehensive event-level data
collection enhances the capacity of statistical models derived from
ABMs, achieving nearly 100% deviance explained (R² = 1) within the
modelled system. However, it is crucial to note that this does not imply
that the entire real-world system is captured by the model; instead, it
reflects the ability to accurately analyse relationships and patterns
within the model itself, offering valuable insights tailored to specific
management or policy objectives. White et al. (2013) caution against
over-reliance on simple linear analyses, as these can oversimplify
complex systems by focusing only on direct relationships. Multivariate
approaches, particularly flexible statistical models such as Generalized
Additive Models (GAMs), provide a more powerful framework for
interpreting simulation outputs. GAMs allow researchers to incorporate
nonlinear relationships and complex interactions, capturing the full
range of dynamics present in ecological systems. For instance, in the
predator-prey example, prey density might not change linearly with wolf
density due to interactions with other variables such as habitat type,
seasonality, or prey behaviour. By using a multivariate framework,
researchers can disentangle these effects and assess the relative
importance of different predictors. This ensures that the insights
derived from the model are ecologically realistic and biologically
meaningful, rather than constrained by the limitations of simpler
statistical approaches. Such modeling approaches also help to identify
indirect or emergent effects that might otherwise go unnoticed,
providing a richer understanding of system-level processes. By capturing
every event and modeling it comprehensively, ABMs provide a more
detailed and reliable foundation for policy and management
recommendations.
Validation
Validation is a critical final step in this framework, ensuring that
the model outputs align with real-world data and are robust enough to
inform decision-making. Validation relies on independent datasets that
were not used during model parameterisation or calibration. For
instance, in a predator-prey model, independent telemetry data from
wolves or field surveys of prey populations could be used to assess the
accuracy of the model’s predictions. By comparing model outputs to
observed data, researchers can identify discrepancies and refine the
model as necessary. Validation provides a mechanism for building trust
in the model, both within the scientific community and among
stakeholders. A well-validated model gives confidence that its
predictions are not only plausible but also actionable, supporting its
application in management and policy contexts. Ultimately, experimental
design, statistical modeling, and validation form an iterative process
that drives the scientific and practical value of ABMs. Data collected
during simulations, when analysed using sophisticated statistical
approaches, can reveal patterns and dynamics that enhance our
understanding of ecological systems. When validated against real-world
data, these insights gain the credibility needed to inform policy, guide
management strategies, and advance ecological knowledge. By combining
event-level data, multivariate modeling, and rigorous validation, ABMs
become powerful tools for addressing complex ecological challenges in a
systematic and data-driven way.
Evaluation & Feedback
The final and arguably most important step in applied ABM workflow is
the integration of model outputs into a feedback loop that connects
simulation insights with real-world data collection and policy
refinement. All simulation outputs and statistical analyses should be
carefully reviewed with stakeholders and policymakers to ensure they
align with their understanding of the system and address practical
management needs. This collaborative evaluation process is essential for
validating the utility of the model, identifying areas for improvement,
and fostering trust in its recommendations. Unexpected results (for
example, if in a wolf-deer predator-prey model it is found that wolf
packs near wide rivers have a disproportionately greater impact on local
red deer density) can serve as valuable starting points for targeted
field studies or adjustments to annual monitoring programs to ground
truth functional aspects of the model and investigate these emergent
trends in a real-world scenario. Testing these findings in real-world
sites not only provides an opportunity to verify simulation outputs but
also enhances the precision and reliability of future simulations by
addressing data gaps or uncertainties identified in earlier iterations.
This continuous feedback loop transforms the simulation into a dynamic
”living lab,” where model predictions directly inform field data
collection, and field observations, in turn, refine the model. By
incorporating unexpected trends and emergent dynamics into subsequent
monitoring efforts, researchers and managers can systematically test and
validate both the model’s predictions and the ecological processes it
represents. This iterative process allows the model to evolve alongside
the system it simulates, creating a synergistic relationship between
simulation and reality. Over time, this approach increases the
granularity of understanding, reduces uncertainty, and produces
increasingly precise and trusted models. Moreover, the feedback loop
significantly enhances the return on investment for research funding by
doubling the outputs and insights generated across the study system.
Rather than functioning as a standalone tool, the model becomes part of
a broader adaptive management framework, driving both theoretical
knowledge and practical application. The ability to iteratively test
management interventions and policy decisions in both the field and the
simulation provides a unique advantage, offering a deeper and more
nuanced understanding of the study system. Most importantly, as the
model becomes better parameterised with each iteration, its credibility
among stakeholders grows, paving the way for more informed and impactful
management and policy outcomes. In this way, the continuous exchange of
data between the simulation environment and the real-world system
ensures that ABMs remain robust, relevant, and valuable tools for
ecological research and decision-making.