Visual ODD: A Standardised Visualisation Illustrating the Narrative of Agent-Based Models

Introduction

Reproducibility is the cornerstone of the scientific method. For model-based work, this means that the model used and how it was analysed and applied must be described in such detail that it can be reproduced and the model analysis repeated by others. This should be possible without resorting to the source code that implements the model, as there are simply too many different programming languages. In addition, the reasoning behind key assumptions of the model should be documented. Otherwise it can be difficult to understand why a model behaves in a certain way or how to best interpret its findings.

While mathematically formulated models can be fully documented using the language of mathematics, simulation models consisting of algorithms do not have such a common language. For agent-based models (ABMs), the ODD protocol has therefore been proposed as a general format for documentation (Grimm et al. 2006, 2010, 2020; Polhill et al. 2008; Polhill 2010). ODD has been widely adopted (Grimm et al. 2020; Vincenot 2018), including applications for model types beyond its original scope. The main feature of ODD is that it starts with an "Overview", which only describes the purpose, entities, state variables, scales, and processes of a model, without going into details. This is followed by a section on "Design concepts", which covers 10 different essential aspects of the design of an agent-based model. Finally, the "Details" section specifies all the technical details for initialising and implementing the model, in addition to the input data that may represent environmental drivers.

However, ODD has been criticised for its limited capacity to convey the narrative behind the model’s design. This is because ODD is not designed to tell a story, but to ensure reproducibility. The ODD model descriptions, despite the "Overview" section, are technical documentations and the level of detail required by ODD was perceived as "sometimes excessive" (Daly et al. 2022). To meet the challenge of providing both a quick overview and technical documentation, Grimm et al. (2020) proposed a common format for a "Summary ODD" but this can still be quite lengthy, especially for more complex ABMs.

In addition, as ODD was originally developed with ecological systems in mind, models from other research domains are often not documented following the ODD structure, because some elements, particularly in the "Design Concepts" section, may not be fully applicable to them (however, see Müller et al. 2013). Nevertheless, as interdisciplinarity is increasing, facilitating model comparison across domains and improving understanding among modellers from different backgrounds is of growing importance.

Regardless of the ODD, it has always been a common practice to include a graphical representation of the model to provide a quick overview and understanding of what the model is and does. However, there is no common format for such graphical representations. Consequently, it can require researchers with different backgrounds considerable effort to understand what certain graphical elements should mean (Banitz et al. 2022; Forbes et al. 2023).

Some well-established, standardised diagrams for simulation models exist. Examples comprise flow and causal-loop diagrams (Richardson 1986). However, such diagrams are often not made for individual-level characteristics and processes or are rather complex to apply (Grimm et al. 2010, 2020). Another standardised graphical representation, primarily used in software development, are Unified Modeling Language (UML) diagrams (Breu et al. 1997). They have been utilised for ABMs, for instance, by Bersini (2012), who applied UML to represent a number of classical ABMs. Yet, similar to flow charts, UML diagrams are characterised by dense text detailing the structure and functioning of submodels. While this level of detail can be very useful in terms of model transferability and transparency, it requires a high familiarity with UML standards for understanding and development. Other visual standardisation efforts are currently rather specific to their domain, for example in areas of ecotoxicological risk assessment (Forbes et al. 2023) or hydrology (Wang et al. 2021).

We suggest the "visual ODD" (vODD) as a new visually appealing standardised model overview figure applicable across domains. This could make communication of simulation models even more comprehensible and accessible to various stakeholders involved in decision making. Once the use of a new standard visual format has exceeded a certain threshold so that the relevant scientific community recognises it as a familiar and reliable format, it will shape readers’ expectations about what kind of visual information they expect where and in what detail. Once this is achieved, the vODD along with the ODD can serve as a checklist for authors, which would eventually make the practice of modelling and model use more coherent and efficient. We also suggest that vODDs could be a useful reference even for modellers not currently using ODDs as they require less detail than ODDs, making them more widely applicable and faster to produce.

Grimm et al. (2020) briefly mentioned the use of a graphical representation of the ODD, which was originally developed by Milles et al. (2020) and Rohwäder & Jeltsch (2022). However, they did not explain its rationale in detail, nor did they present a set of examples or provide guidance or even tools for creating such vODDs. Nevertheless, it seems that this sample visualisation has already facilitated the acceptance and use of ODD, in particular for beginners in modelling and publishing (D. Parker, personal communication, March 9, 2023).

Here we aim to elaborate on the initial ideas of visualising ODDs presented in Grimm et al. (2020) by providing guidance on creating vODDs. First, we evaluated the current methods of illustrating simulation models graphically and whether the sample visual ODD, as introduced by Grimm et al. (2020), has been adopted. We reviewed all 248 publications citing the latest ODD update (Grimm et al. 2020) indexed by Web of Science by 15 June 2023 (Appendix D). The majority of publications were associated with life sciences (n = 133), followed by social sciences (n = 60). Among the 248 reviewed publications, 190 featured models, with 171 of them containing ODD documentations. A slightly smaller subset of publications presenting models provided some sort of model visualisation (n = 156, Figure 1A), primarily included in the main text (n = 123). This suggests that most authors already attempt to enhance their models’ comprehension by supplementing their written ODDs with visual representations.

However, our review also demonstrates that current model visualisations, in most cases, do not document all the essential model features, including the purpose of the model, information on the initialisation of the simulation, a definition of the spatial/temporal extent and resolution, the primary submodels, and the outputs observed (Figure 1B). Typically, visualisations take the form of flow charts depicting submodels and their order of execution. To a lesser extent, visualisations include initialisation and observations. Only very few visualisations present extents, resolutions, or the model’s overarching purpose. On average, flow charts and other visualisation types incorporated fewer than two of these model features. In contrast, vODD comparable visualisations captured an average of three model features providing a stronger link between visualisation and written model documentation (Figure 1C). Interestingly, only nine publications included a visualisation similar to the one provided by Grimm et al. (2020): Milles et al. (2020), Lemanski et al. (2021), Sandhu et al. (2022), Szangolies et al. (2022), Rohwäder & Jeltsch (2022), Byer & Reid (2022), Farthing & Lanzas (2021), Ekanayake-Weber & Swedell (2021), Carturan et al. (2020); assembled in Appendix C.

While we can see that these "vODD-similar" visualisations can often convey more information about the model and provide a stronger linkage to the written ODD, they are still diverse and have yet to become a common practice. Here we present a straightforward, accessible and uniform scheme for creating vODDs. In the following sections, we outline the scope and content of a vODD, discuss the applicability of vODDs to more complex models, and provide information on available tools.

Definition and Scope of vODD

To our knowledge, the visual ODD is the first standardised graphical representation for agent-based models universally applicable across multiple research areas. The vODD is a pictorial description of the key elements of the ODD protocol (Grimm et al. 2006, 2010, 2020), including the initialisation of the model and its entities ("Initialisation"), the scheduling of its processes ("Submodels"), the outputs generated by the model ("Observation"), as well as the spatial/temporal resolution and extent ("Scales"). The purpose is to capture the overall narrative of the model with respect to the full cycle of a simulation. The vODD, like the ODD, is primarily designed to report agent-based models, however, it can be adapted to other dynamic modelling approaches (further explored in the section "Applying vODD to non-agent-based models"). The vODD is designed to be a stand-alone figure that provides a comprehensive overview of an ABM and helps define the scope and purpose of a model. However, modellers seeking to replicate and/or extend an existing model will still need to refer to the detailed description. To this end, each element in the visual ODD has a predefined location and uses the same terminology as the corresponding sections in the written ODD, providing a clear link to where more detailed information can be found. We suggest using vODD to represent ABMs in manuscripts, scientific posters, or talks. The vODD could also serve as a graphical abstract for publications, which is a relatively new trend in scientific journals (Agrawal & Ulrich 2023; Krukowski & Goldstein 2023).

The establishment of a standardised visualisation for ABMs serves several purposes. The consistency of predefined sections not only helps the viewer locate information of interest easily but also facilitates conceptual comparisons between ABMs. We suggest that the availability of a uniform template and the use of simple open-source tools (see "Tools and tips for creating visual ODDs") could encourage agent-based modellers to document the model overview and its outputs using graphical visualisations. Providing clear and well-designed visual storytelling of the model could enhance the study’s accessibility, disseminate the model outputs to a broader audience, including non-modellers, and facilitate interdisciplinary discourse.

In general, visualisations are an efficient way to provide information quickly and effectively (Oska et al. 2020). While modellers are fully aware of their model’s components and dynamics, users of the model might encounter challenges in grasping its structure or even its purpose without the aid of a clear visual representation. Furthermore, we experienced how creating diagrams can help with conceptualising models. For these reasons, best modelling practices recommend the use of diagrams and figures to represent a simulation model and communicate its outputs (Forbes et al. 2023; Hall & Virrantaus 2016). This is consistent with the findings of our review, where over 80% of the reviewed modelling studies included a visual representation of their modelling methods in one or multiple figures (Figure 1A). However, commonly used flow charts often lack features like the initialisation and can become quite large and overwhelming when including details of all submodels. Other scholars depicted selective aspects of the model using multiple diagrams, yet, making it challenging to allow for clear links between them. Alternatively, the vODD builds on the advantages of flow charts, which are commonly used by scholars to describe their models, but formalises them and embeds them with more context. The vODD seeks to provide a visually appealing format with minimal text within a single, cohesive figure.

How to Build a Visual ODD

Each part of the vODD is linked to one or more sections of the written ODD and has a predefined and fixed position in the figure (see Figure 2). Certain elements of the model description are mandatory and must be included in the vODD. The baseline vODD consists of three main sections, "Initialisation", "Submodels", and "Observations", represented by connected panels, which together represent the full cycle of model structure and simulation.

The first section, "Initialisation", depicts how a simulation is initialised, including details about entities, their most important state variables, and, if applicable, spatial extents and resolution. The second section, "Submodels", shows the processes and their order of execution, including the temporal resolution (and extent). The final panel, "Observations", presents the information collected from the model. More details on what information to include in each section and where to place it are provided in the following subsections and depicted in Figure 2. Additionally, we provide a vODD template, which can be found in the repository https://github.com/visual-ODD/Templates.git, to help constructing your own visualisation. An example of how to apply the template to an existing model is provided in Figure 3, where we rebuild the original suggestion for vODDs by Grimm et al. (2020) visualising the model of Milles et al. (2020). In comparison to Grimm et al. (2020), we added the temporal and spatial scales and made some slight adaptations in presenting the model’s entities. Figure 3 further shows some additional panels for the "Purpose" and "Scenario" description, which we propose as optional for vODDs.

The vODD can be expanded to include optional elements, such as a "Purpose" line, empirical "Patterns" used, or details on "Scenarios" or simulation experiments (see also Appendix A). While these additions can enhance the comprehension of a model study, they also increase the information load and may not always be necessary. As such, we suggest including these elements selectively, particularly when they support the narrative of the study or strengthen the credibility of the model.

Certain parts of the written ODD have been omitted in the figure for conciseness. For example, details about the "Submodels" are not included. Furthermore, "Design concepts" are not explicitly featured, given their wide variability in applicability across models.

Although the vODD format can serve as a stand-alone visualisation of a model, it is helpful to be familiar with the written ODD model description format (Grimm et al. 2006, 2010, 2020) to construct it effectively. There are two opposing approaches to building a visual ODD: one can start either by first writing the ODD protocol or by creating the visualisation. Each approach has its own advantages. If the written ODD already exists, the different parts can simply be transferred to the visualisation, such as the entities and the submodel names. In contrast, starting with the visual ODD puts more emphasis on creating clear and concise names for the various components, e.g., submodels, to make them as meaningful and self-explanatory as possible. Furthermore, by using the visualisation as an overview, the visual can guide modellers through the often long and tedious writing process of the written ODD. Additionally, constructing the vODD prior to model implementation may assist in the conceptualisation and implementation process, such as in designing the execution order of model processes or selecting outputs for observation. As model development can be an iterative process, ODD protocol and vODD graphic might also be designed simultaneously.

As a general guideline, we suggest maintaining clarity in the visualisation by minimising text and choosing icons and colours carefully (more details in the section "Accessibility tips"). The latter also depends heavily on the context in which the figure is to be presented.

Initialisation

This element illustrates the setup of the model prior to simulation runs, including details on entities and their key state variables. Agent-based models typically have multiple kinds of entities, in which case we recommend structuring this section by the employed entity types (i.e., individuals, collectives, or spatial units). For spatially-explicit models, the initialisation details of the spatial units, such as grid cells, forming the landscape should be provided, including spatial extents and resolution (see "Scales").

It is not necessary to include all processes, parameters, or state variables of an entity. The focus should be on providing the relevant information necessary for the viewer to understand how the model operates and the derivation of its final observations. If certain details from the full model description are omitted, this can be indicated in the figure caption, along with a reference directing readers to where the full information can be found, e.g., a table with all state variables in the full ODD in the supplementary material.

The initialisation may differ among scenarios. However, due to space constraints, this section should provide a generalised setup, while model scenarios or different parameterisations can be optionally presented in a separate panel (see "Optional elements", and example in Figure 3).

Submodels

This vODD element provides an overview of the model’s processes and their order of execution and is therefore closely related to the "Process overview and scheduling" in the ODD. Typically, a simulation model contains several submodels, i.e., algorithms, representing particular processes. Submodels are applied in a defined order or are triggered by certain conditions. We suggest that the names of the submodels are to be placed clockwise around a circle to indicate their repeated scheduling. Submodel names should concisely and clearly describe their purpose and match the names used in the written ODD as closely as possible. Examples may include "Mortality", "Ageing", "Reproduction", "Buying", "Harvesting", "Updating memory", etc. While it is clear that such names may not comprehensively capture the full functioning of the submodels, they provide a quick overview. For in-depth details, viewers can refer to the corresponding section of the written ODD.

The visualisation along a loop may represent the run once per timestep and / or agent (ABMs) and / or entire model run (non-temporal models). Depending on the number of submodels, font size and placement around the circle can be adapted. If, due to a model’s complex dynamics, a single circle is insufficient to present the scheduling of all relevant submodels, multiple circles or alternative visualisations can be used, e.g., to indicate that submodels are run at different temporal resolutions, or that they belong to different entities (see "Visualisation of more complex models" and Figure 4).

To provide additional context and engage viewers, related icons can be provided next to the names of the submodels. Such icons or pictograms are particularly useful for vODDs intended for posters and presentations, but they may also enhance journal publications when used strategically to provide additional insight into submodel functioning. Icons can be sourced from free platforms or custom-crafted (see "Finding graphics, icons, and fonts").

Observations

The "Observations" section of the vODD allows the viewer to quickly grasp the main outputs and hence the intended use of the model. This panel could be used to explicitly state the intended purpose, or it may be clear from the reported output variables. Key output patterns of interest directly derived from the model need to be given here, such as time series of population sizes, growth of individuals, income, proportion of conservative land users, etc. When many outputs exist, grouping them can improve readability, for example, grouping by agent types or for population- and individual-level outputs. Including one or two illustrative figures of what such outputs might look like can provide clarity. However, it is important to note that this section aims to highlight only the most important outputs for addressing the model purpose, not to comprehensively cover the entire model analysis.

Scales

Scales are very important for understanding the scope and dynamics of a model (e.g.,McGill et al. 2015), but are often neglected in model visualisations (Figure 1B). To immediately convey the spatial and temporal scales at which a model operates, these elements are mandatory and now have a fixed position in the vODD. This is a new addition compared to the example figure in Grimm et al. (2020) and has been included in the adjusted Figure 3.

Information about spatial scales is provided in the "Initialisation" section of the vODD, where landscape cells are usually represented as entities. The resolution of individual spatial units and the extent of the entire simulated landscape or environment should be reported for spatial models.

We positioned temporal dynamics in the "Submodels" section. Our suggestion is to report the duration of a time step (i.e., the temporal resolution) within the loop of submodels. The arrows in the circle indicate that the submodels are repeated for several time steps. If submodels are run with different temporal resolutions, this can be illustrated by using subcycles (see examples in Figure 4). The different time steps may then be written in the center of each loop. Once the simulation is complete, outputs can be analysed. The temporal extent of a simulation run, i.e., the number of simulated time steps or any other stop condition, should, hence, be reported between the "Submodels" and "Observations" sections.

Optional elements

We suggest several optional sections that can be added to the basic vODD if beneficial for communicating the model and if space permits (see also Appendix A). Firstly, a clear and short statement of the "Purpose" of the model can be given at the top of the figure. Generally, we would highly recommend to state the purpose of the model in all vODDs. In the basic version of the vODD, the purpose can be included within the "Observations" section or in the figure caption, but to make it even more prominent, it can also be placed directly at the top of the figure as in Figure 3. In this case, the purpose should be conveyed as a short and concise statement, avoiding multiple sentences. It could refer to one or more of the general categories of model purposes: prediction, explanation, description, theoretical exploration, illustration, analogy, and social learning (Edmonds et al. 2019).

In addition, a set of panels for "Patterns" supporting the different phases of model design and evaluation can be included. Such patterns may include data or evidence from relevant literature used for model development or validation, particularly in empirical models. Under the "Initialisation" section, a panel may be used to show patterns relevant to model initialisation, e.g., real landscapes or species distributions. A panel under the "Submodels" section may show patterns from data or literature that are used to define submodel algorithms or their parameters, e.g., mortality rates, movement distributions, home range sizes, etc. A third panel, positioned beneath the "Observations" section, can specifically be used to report patterns used for calibration or validation purposes. To avoid repetition with the "Observations" panel, it should not include results analysis. Including this optional element in a vODD can be useful to increase confidence in a model by showing what information has been incorporated and, in cases, can aid in model comprehension. An example of a vODD including panels for patterns is given in Appendix B (Figure 7). However, due to space limitations, the patterns shown typically represent only a carefully selected subset of the patterns actually used.

Finally, at the bottom of the figure, an additional panel can be added to illustrate the simulated "Scenarios" (see example in Figure 3). This panel provides space to show alternative scenarios (e.g., landscapes, environmental conditions, income distributions). Varying parameterisations of simulation experiments could be presented graphically or even as a table or short statement of the objective of scenarios (e.g., explore the impact of parameters x & y). Thus, this part of the visualisation does not pertain to the model description itself but rather to the definition of the experiments performed with the model.

Visualisation of More Complex Models

Models are as diverse as the conceptual approaches of the modellers themselves, making it essential to recognise that not all models may neatly fit into a standard template as presented in section "How to build a vODD". In this section, we provide alternative approaches for applying the vODD framework for these tricky models and offer related examples.

There are several sources of complexity in agent-based models that may require more detailed vODDs. Examples include comprehensive models that encompass multiple aspects of a single phenomenon (e.g.,Preuss et al. 2022), interdisciplinary or hybrid models (Innocenti et al. 2020), multi-scale hierarchical models that operate at various levels of scale (Kruse et al. 2022), models with sequential temporal processes (Kane et al. 2022), and models with multiple-stage decision-making processes yielding a wide range of outcomes (e.g.,Reinhard et al. 2022).

To address the challenge of visualising complex models, we propose breaking them down into simpler structural components and examining their arrangement within the model cycle. Many complex models feature multiple iterative processes, often referred to as "loops," embedded in their dynamic structures. These loops can take various configurations within the model structure. Figure 4 summarises these arrangements in three different classes and some examples are provided in Appendix B. A) Sequential vODDs include multiple loops that run one after the other in a temporal process and have one-way effects. This format may be useful for hierarchical or "nested" ODDs (sensu Grimm et al. 2020) where model documentation presents complex submodels in high detail (example of a hierarchical vODD found in supplementary Figure 6). B) In cases where specific submodels or processes require more emphasis or operate at different temporal resolutions or entities, a vODD composed of two or more loops can be used, with one enclosed within the other, featuring an interactive feedback process (an example is given in supplementary Figure 7). C) Multiple parallel loops that do not substantially interact with each other can be used to represent models with distinct entities or environments. Examples include two adjacent drainage basins or two sport fields hosting games at the same time. We, however, want to emphasise that this list of possible arrangements of cycles may not be exclusive due to the variety and complexity of model implementations. We encourage modellers to be creative and recommend the adaptation of the composition of model cycles dependent on individual model characteristics.

Although it is evident that the visualisation of submodels will never capture their full details, one option to cope with a very high degree of complexity, may be digital dynamic vODDs. For example in comprehensive models, where a single phenomenon is represented from multiple interdisciplinary aspects, such digital dynamic vODDs could enhance communication by offering additional details and resolution. These formats may incorporate multimedia, animations, zoom capabilities, or interactive and overlay panels with supplementary information. Guidance on Digital Poster formats may be relevant here, such as provided by interactive presentation tools like Prezi (prezi.com) or on the Massachusetts Institute of Technology’s Digital Poster guide (mitcommlab.mit.edu/nse/commkit/digital-poster).

Tools and Tips for Creating Visual ODDs

In the repository https://github.com/visual-ODD/Templates.git, we provide a template for creating a vODD available in both a vector graphic format (.svg) and as a PowerPoint slide (.pptx). For building your own vODD, you can simply select one of these templates, input the necessary information, and customise it using your preferred software. Alternatively, the layout presented in Figure 2 can be used as a guide to rebuild the vODD structure in completely different software depending on your needs and preferences.

In this section, we will explore potential graphic design software options for crafting compelling visual ODDs, provide insights on how to source graphics, icons, and typefaces effectively, and emphasise the vital aspects of ensuring accessibility in visual media.

Software options

When it comes to online graphic design platforms, Canva stands out as a popular choice, renowned for its user-friendly interface and an extensive array of templates (www.canva.com). However, Canva’s primary focus is not on scientific content, so adapting its features to suit the specific needs of a vODD might require some effort. Alternatively, BioRender offers a specialised option primarily tailored to life sciences, and it can be adapted for other domains (www.biorender.com). BioRender offers a comprehensive library of icons and templates which may be useful for vODDs. Another viable choice is Piktochart, which offers customisable templates for crafting infographics along with a user-friendly interface (piktochart.com). Mind the Graph is tailored for scientific infographics with a library of icons, templates, and tools to aid in the design process (mindthegraph.com). It is important to be aware that while some of these online platforms offer free usage, they may require proper attribution, and others may offer premium content locked behind subscriptions.

For desktop applications, Microsoft PowerPoint may be a familiar option, especially for Office users. PowerPoint, and similar presentation software, may not be commonly associated with non-presentation formats, but their familiarity and user-friendly nature make them a viable choice, particularly for individuals lacking graphic design experience. PowerPoint also provides its own libraries for graphics and icons, which may be useful for vODDs. If a free and open-source graphic design software is preferred, Inkscape, a vector graphics editor, offers robust design capabilities, although it may have a steeper learning curve compared to other tools (inkscape.org). It provides precise control for crafting intricate visual ODDs and benefits from a supportive community for learning. Finally, Adobe InDesign, a paid but professional-grade product, excels in handling complex projects combining text, images, and graphics, making it ideal for visually rich content.

Ultimately, the selection of specific software will be a personal choice, influenced by user experience level, accessibility, and specific requirements.

Finding graphics, icons, and typefaces

When creating a visual ODD, the inclusion of icons, symbols, graphics, and non-standard typefaces can significantly enhance the design’s clarity and impact (Murchie & Diomede 2020). To find suitable assets, we recommend using relevant keywords in your online searches. For instance, search terms like "icon," "graphics," or "illustration" along with other specific elements required, e.g., a species, process, or concept, can help discover content that aligns with your model’s theme. Additionally, consider using reputable online platforms that specialise in providing graphics, icons, and typefaces, such as Freepik.com, Iconfinder.com, and fonts.google.com. These platforms offer an extensive array of both free and premium assets to choose from. Also, some of the software options presented in the previous section offer their own libraries of graphics, icons, and typefaces which are mostly free to use. In cases where these built-in resources align with your needs, they can be a convenient and preferable choice that can help streamline the design process. When incorporating graphics obtained from the internet, proper attribution is required to maintain scientific integrity, i.e., the source and creator’s name should be included to give credit where it is due (see example in Figure 2). Authors are expected to adhere to licensing terms specified by the source to avoid copyright issues.

Accessibility tips

To ensure that your vODD is accessible to a broad audience, consider several key factors. First, when it comes to typefaces and fonts, prioritise legibility by using clear, legible typefaces, and avoiding overly decorative styles (Murchie & Diomede 2020). Sans-serif typefaces are often a better choice as letters can appear less crowded (Russell-Minda et al. 2007), examples include Arial, Calibri, Verdana, Trebuchet, Comfortaa, Montserrat, Century Gothic, Roboto, Poppins, and Noto Sans. Font size should be at least 12-14 point or equivalent (The British Dyslexia Association 2024). Additionally, resist the temptation to overcrowd your visuals with text; instead, maintain concise wording to enhance comprehension.

Your colour palette can significantly influence how your vODD is perceived and understood (Crameri et al. 2020; Murchie & Diomede 2020). Choose a palette that is easy on the eyes and provides sufficient contrast between elements. Generally, between three to five colours which complement each other while remaining easy to distinguish should suffice. Colour can and should be used strategically to categorise and group information, set the mood and tone of the piece, navigate the viewer’s eye, signal meaning, and create emphasis. Strategic use of colour can enhance communication, engagement, and information conveyance (Murchie & Diomede 2020).

Furthermore, adhering to colour contrast guidelines, such as those outlined in the "Web Content Accessibility Guidelines" (WCAG) for text and graphics, is essential. These guidelines offer specific contrast levels for different font sizes and scenarios (i.e., AA and AAA compliance). You can reference the WCAG 2.1 guidelines for detailed requirements (www.w3.org/TR/WCAG21). To simplify the process, use online colour contrast checkers, such as www.colourcontrast.cc, that utilise colour hex codes, as many of these tools incorporate WCAG rules, eliminating the need to memorise specific values. In terms of backgrounds, avoid overly complex patterns or photos, and opt for solid-colour or subtle gradients instead.

It is crucial to make your visuals friendly to colourblind individuals by ensuring that your colour choices are accommodating. Colourblindness checks can be performed using online colourblindness simulators, like the Coblis - Color Blindness Simulator, or built-in functionalities in design software like Adobe Illustrator. Additionally, explore colourblind-friendly palettes from sources like ColorBrewer (Harrower & Brewer 2003), Adobe Color Wheel (color.adobe.com/create/color-wheel), or dedicated R packages, such as RColorBrewer (Neuwirth 2022) and viridis (Garnier et al. 2023). Shapes and labels can additionally be incorporated to convey information effectively, minimising reliance solely on colour cues.

Finally, ensure appropriate spacing between text and graphics, avoid text justification, and maintain a 1.5 line spacing whenever feasible to prevent overcrowding. Additionally, increasing the inter-character spacing, when possible, can benefit viewers, particularly those with dyslexia (The British Dyslexia Association 2024).

These comprehensive accessibility tips will help you create a vODD that is inclusive and informative to a diverse audience. More detailed tips can be found in the Dyslexia Friendly Style Guide from the British Dyslexia Association (The British Dyslexia Association 2024).

Applying vODD to Non-Agent-Based Models

The second update of the ODD protocol (Grimm et al. 2020) clarified that only specific elements, primarily encapsulated in the "Design Concepts" section, are unique to ABMs. This distinction has allowed the ODD protocol to be applied to describe non-ABMs, such as matrix models, integral projection models, and ecological-economic models based on difference equations (Grimm et al. 2020).

In this context, we suggest that vODDs can also be beneficial for non-ABMs which lack inherent visual interpretability. By offering a standard yet flexible framework for representation, a vODD can significantly aid in conveying complex interactions and temporal dynamics often found in these models (Christensen & Walters 2004). Importantly, the framework is adaptable enough to encapsulate the unique characteristics that distinguish non-ABMs from their agent-based counterparts. For example, many non-ABMs, such as system dynamics models or integral projection models, incorporate continuous state variables instead of discrete agents (Ellner & Rees 2006). This nuance could be effectively conveyed through the initialisation in the vODD, making it easier for researchers and stakeholders to understand the model’s design and functionality.

While not all non-ABMs may be suitable for depiction via a vODD, particularly models with iterative elements, or loops, seem fitting for our visual framework. These primarily include process-based models with temporal dynamics or event scheduling (Mangel 2015). Such iterative processes often align with the types of complexity that vODD seeks to clarify.

Summary

With this work, we aim to establish a standardised format for model visualisations, particularly for ABMs. The presented vODD layout was specifically designed to capture the fundamental concepts of a model and provide an overview of the full simulation cycle. The provided user-friendly template, along with practical tools and tips, facilitates the creation of such visualisations for each individual model. With our standardised visualisation format we hope to simplify model communication and comparison, and increase model understanding and acceptance among modellers and non-modellers alike. Further, we anticipate that standardised visuals will encourage the reuse of models by establishing a clear link between an attention-grabbing model visualisation and the detailed model description, in the form of a written ODD, needed for reimplementation. Hence, vODDs can serve as an initial, accessible and concise overview of models, suitable for inclusion in manuscripts, presentations, and posters.

References

AGRAWAL, S., & Ulrich, P. (2023). A picture is worth 1000 words: Teaching science communication with graphical abstract assignments. Journal of Microbiology & Biology Education, 24(1), e00208. [doi:10.1128/jmbe.00208-22]

BANITZ, T., Hertz, T., Johansson, L.-G., Lindkvist, E., Martinez Pena, R., Radosavljevic, S., Schluter, M., Wennberg, K., Ylikoski, P., & Grimm, V. (2022). Visualization of causation in social-ecological systems. Ecology & Society, 27(1). [doi:10.5751/es-13030-270131]

BERSINI, H. (2012). Uml for ABM. Journal of Artificial Societies and Social Simulation, 15(1), 9. [doi:10.18564/jasss.1897]

BREU, R., Hinkel, U., Hofmann, C., Klein, C., Paech, B., Rumpe, B., & Thurner, V. (1997). Towards a formalization of the unified modeling language. ECOOP’97 - Object-Oriented Programming: 11th European Conference Jyväskylä, Finland, June 9 - 13, 1997 Proceedings 11 [doi:10.1007/bfb0053386]

BYER, N. W., & Reid, B. N. (2022). The emergence of imperfect philopatry and fidelity in spatially and temporally heterogeneous environments. Ecological Modelling, 468, 109968. [doi:10.1016/j.ecolmodel.2022.109968]

CARTURAN, B. S., Pither, J., Maréchal, J.-P., Bradshaw, C. J., & Parrott, L. (2020). Combining agent-based, trait-based and demographic approaches to model coral-community dynamics. elife, 9, e55993. [doi:10.7554/elife.55993]

CHRISTENSEN, V., & Walters, C. J. (2004). Ecopath with Ecosim: Methods, capabilities and limitations. Ecological Modelling, 172(2–4), 109–139. [doi:10.1016/j.ecolmodel.2003.09.003]

CRAMERI, F., Shephard, G. E., & Heron, P. J. (2020). The misuse of colour in science communication. Nature Communications, 11(1), 5444. [doi:10.1038/s41467-020-19160-7]

DALY, A. J., De Visscher, L., Baetens, J. M., & De Baets, B. (2022). Quo vadis, agent-based modelling tools? Environmental Modelling & Software, 157, 105514. [doi:10.1016/j.envsoft.2022.105514]

EDMONDS, B., Le Page, C., Bithell, M., Chattoe-Brown, E., Grimm, V., Meyer, R., Montañola-Sales, C., Ormerod, P., Root, H., & Squazzoni, F. (2019). Different modelling purposes. Journal of Artificial Societies and Social Simulation, 22(3), 6. [doi:10.18564/jasss.3993]

EKANAYAKE-WEBER, M., & Swedell, L. (2021). An agent-based model of coercive female transfer in a multilevel society. Animal Behaviour, 182, 267–283. [doi:10.1016/j.anbehav.2021.10.004]

ELLNER, S. P., & Rees, M. (2006). Integral projection models for species with complex demography. The American Naturalist, 167(3), 410–428. [doi:10.1086/499438]

FARTHING, T. S., & Lanzas, C. (2021). Assessing the efficacy of interventions to control indoor SARS-Cov-2 transmission: An agent-based modeling approach. Epidemics, 37, 100524. [doi:10.1016/j.epidem.2021.100524]

FORBES, V. E., Accolla, C., Banitz, T., Crouse, K., Galic, N., Grimm, V., Raimondo, S., Schmolke, A., & Vaugeois, M. (2023). Mechanistic population models for ecological risk assessment and decision support: The importance of good conceptual model diagrams. Integrated Environmental Assessment and Management, 20(5), 1566–1574. [doi:10.1002/ieam.4886]

GARNIER, S., Ross, N., Rudis, R., Camargo, A. P., Sciaini, M., & Scherer, C. (2023). viridis(Lite) - Colorblind-Friendly color maps for R. viridis package version 0.6.3. Available at: https://sjmgarnier.github.io/viridis/. [doi:10.32614/cran.package.viridis]

GRIMM, V., Berger, U., Bastiansen, F., Eliassen, S., Ginot, V., Giske, J., Goss-Custard, J., Grand, T., Heinz, S. K., Huse, G., Huth, A., Jepsen, J. U., JNørgensen, C., Mooij, W. M., Müller, B., Pe’er, G., Piou, C., Railsback, S. F., Robbins, A. M., … DeAngelis, D. L. (2006). A standard protocol for describing individual-based and agent-based models. Ecological Modelling, 198(1), 115–126. [doi:10.1016/j.ecolmodel.2006.04.023]

GRIMM, V., Berger, U., DeAngelis, D. L., Polhill, J. G., Giske, J., & Railsback, S. F. (2010). The ODD protocol: A review and first update. Ecological Modelling, 221(23), 2760–2768. [doi:10.1016/j.ecolmodel.2010.08.019]

GRIMM, V., Railsback, S. F., Vincenot, C. E., Berger, U., Gallagher, C., DeAngelis, D. L., Edmonds, B., Ge, J., Giske, J., Groeneveld, J., Johnston, A. S. A., Milles, A., Nabe-Nielsen, J., Polhill, J. G., Radchuk, V., Rohwäder, M. S., Stillman, R. A., Thiele, J. C., & Ayllón, D. (2020). The ODD protocol for describing agent-based and other simulation models: A second update to improve clarity, replication, and structural realism. Journal of Artificial Societies and Social Simulation, 23(2), 7. [doi:10.18564/jasss.4259]

HALL, A., & Virrantaus, K. (2016). Visualizing the workings of agent-based models: Diagrams as a tool for communication and knowledge acquisition. Computers, Environment and Urban Systems, 58, 1–11. [doi:10.1016/j.compenvurbsys.2016.03.002]

HARROWER, M., & Brewer, C. A. (2003). ColorBrewer.org: An online tool for selecting colour schemes for maps. The Cartographic Journal, 40(1), 27–37. [doi:10.1179/000870403235002042]

INNOCENTI, E., Detotto, C., Idda, C., Parker, D. C., & Prunetti, D. (2020). An iterative process to construct an interdisciplinary ABM using MR POTATOHEAD: An application to housing market models in touristic areas. Ecological Complexity, 44, 100882. [doi:10.1016/j.ecocom.2020.100882]

KANE, A., Ayllón, D., O’Sullivan, R. J., McGinnity, P., & Reed, T. E. (2022). Escalating the conflict? Intersex genetic correlations influence adaptation to environmental change in facultatively migratory populations. Evolutionary Applications, 15(5), 773–789. [doi:10.1111/eva.13368]

KRUKOWSKI, R. A., & Goldstein, C. M. (2023). The potential for graphical abstracts to enhance science communication. Translational Behavioral Medicine, 13(12), 891–895. [doi:10.1093/tbm/ibad069]

KRUSE, M., Meyer, C., Schneekloth, F., & Reuter, H. (2022). How artificial potential field algorithms can help to simulate trade-offs in movement behaviour of reef fishes. Frontiers in Marine Science, 9, 1037358. [doi:10.3389/fmars.2022.1037358]

LEMANSKI, N., Silk, M., Fefferman, N., & Udiani, O. (2021). How territoriality reduces disease transmission among social insect colonies. Behavioral Ecology and Sociobiology, 75, 1–13. [doi:10.1007/s00265-021-03095-0]

MANGEL, M. (2015). Stochastic dynamic programming illuminates the link between environment, physiology, and evolution. Bulletin of Mathematical Biology, 77, 857–877. [doi:10.1007/s11538-014-9973-3]

MCGILL, B. J., Dornelas, M., Gotelli, N. J., & Magurran, A. E. (2015). Fifteen forms of biodiversity trend in the anthropocene. Trends in Ecology & Evolution, 30(2), 104–113. [doi:10.1016/j.tree.2014.11.006]

MILLES, A., Dammhahn, M., & Grimm, V. (2020). Intraspecific trait variation in personality-related movement behavior promotes coexistence. Oikos, 129(10), 1441–1454. [doi:10.1111/oik.07431]

MURCHIE, K. J., & Diomede, D. (2020). Fundamentals of graphic design - Essential tools for effective visual science communication. FACETS, 5(1), 409–422. [doi:10.1139/facets-2018-0049]

MÜLLER, B., Bohn, F., Dreßler, G., Groeneveld, J., Klassert, C., Martin, R., Schlüter, M., Schulze, J., Weise, H., & Schwarz, N. (2013). Describing human decisions in agent-based models - ODD + D, an extension of the ODD protocol. Environmental Modelling & Software, 48, 37–48.

NEUWIRTH, E. (2022). RColorBrewer: ColorBrewer palettes. R package version 1.1-3. Available at: https://CRAN.R-project.org/package=RColorBrewer. [doi:10.32614/cran.package.rcolorbrewer]

OSKA, S., Lerma, E., & Topf, J. (2020). A picture is worth a thousand views: A triple crossover trial of visual abstracts to examine their impact on research dissemination. Journal of Medical Internet Research, 22(12), e22327. [doi:10.2196/22327]

POLHILL, J. G. (2010). ODD updated. Journal of Artificial Societies and Social Simulation, 13(4), 9. [doi:10.18564/jasss.1700]

POLHILL, J. G., Parker, D., Brown, D., & Grimm, V. (2008). Using the ODD protocol for describing three agent-based social simulation models of land-use change. Journal of Artificial Societies and Social Simulation, 11(2), 3. [doi:10.18564/jasss.1563]

PREUSS, T. G., Agatz, A., Goussen, B., Roeben, V., Rumkee, J., Zakharova, L., & Thorbek, P. (2022). The BEEHAVEecotox Model - Integrating a mechanistic effect module into the honeybee colony model. Environmental Toxicology and Chemistry, 41(11), 2870–2882. [doi:10.1002/etc.5467]

REINHARD, S., Naranjo, M. A., Polman, N., & Hennen, W. (2022). Modelling choices and social interactions with a threshold public good: Investment decisions in a polder in Bangladesh. Land Use Policy, 113, 105886. [doi:10.1016/j.landusepol.2021.105886]

RICHARDSON, G. P. (1986). Problems with causal-loop diagrams. System Dynamics Review, 2(2), 158–170. [doi:10.1002/sdr.4260020207]

ROHWÄDER, M.-S., & Jeltsch, F. (2022). Foraging personalities modify effects of habitat fragmentation on biodiversity. Oikos, 2022(12), e09056.

RUSSELL-MINDA, E., Jutai, J., Strong, J., Campbell, K., Gold, D., Pretty, L., & Wilmot, L. (2007). The legibility of typefaces for readers with low vision: A research review. Journal of Visual Impairment & Blindness, 101(7), 402–415. [doi:10.1177/0145482x0710100703]

SANDHU, R., Tripp, C., Quon, E., Thedin, R., Lawson, M., Brandes, D., Farmer, C. J., Miller, T. A., Draxl, C., Doubrawa, P., Williams, L., Duerr, A. E., Braham, M. A., & Katzner, T. (2022). Stochastic agent-based model for predicting turbine-scale raptor movements during updraft-subsidized directional flights. Ecological Modelling, 466, 109876. [doi:10.1016/j.ecolmodel.2022.109876]

SZANGOLIES, L., Rohwäder, M.-S., & Jeltsch, F. (2022). Single large and several small habitat patches: A community perspective on their importance for biodiversity. Basic and Applied Ecology, 65, 16–27. [doi:10.1016/j.baae.2022.09.004]

THE British Dyslexia Association. (2024). Dyslexia style guide. Avilable at: https://www.bdadyslexia.org.uk/advice/employers/creating-a-dyslexia-friendly-workplace/dyslexia-friendly-style-guide.

VINCENOT, C. E. (2018). How new concepts become universal scientific approaches: Insights from citation network analysis of agent-based complex systems science. Proceedings of the Royal Society B: Biological Sciences, 285(1874), 20172360. [doi:10.1098/rspb.2017.2360]

WANG, J., Chen, M., Lü, G., Yue, S., Wen, Y., Sheng, Y., & Lu, M. (2021). A construction method of visual conceptual scenario for hydrological conceptual modeling. Environmental Modelling & Software, 145, 105190. [doi:10.1016/j.envsoft.2021.105190]