1. Introduction
The daily lives of hominin foragers, regardless of their species, geographical location, or chronological framework, has been primarily influenced by their surrounding ecosystems and the availability of resources in both space and time (Foley, 1986; Winterhalder and Smith, 2000). In most environments, a diverse array of animals and plants may have been successfully acquired, processed, and consumed (Robson and Kaplan, 2006). However, hominins in any given area rarely targeted all available resources; instead, they focused on acquiring and consuming a specific subset of them (Boone, 2002). They developed specific subsistence strategies that involved specific actions to meet their basic survival needs, including food, clothing, and shelter (Huguet et al., 2013). The environmental conditions they encountered, together with their available subsistence strategies and associated technology, played a crucial role in determining how their groups foraged and moved throughout the year, as well as their level of interaction with other groups (Nettle et al., 2013). But local dynamics of early hominin groups are still not very well known. Due to the open-air nature of many Early and Middle Pleistocene sites, important aspects are poorly preserved (Pineda and Saladié, 2022), making it difficult or impossible to connect faunal materials with found lithics (Pineda et al., 2024). Many occupations may be undiscovered or lost to taphonomic processes, resulting in the lack or inaccessibility of crucial data. As a result, subsistence behaviour or local behaviour of early hominins has mostly been reconstructed using ethnographic analogues, as identifying behavioural practices can be challenging to discern in older assemblages (Shultz et al., 2012).
The LATEurope ERC project aims to identify the reasons behind the seemingly later occupation of Western Europe (ca 1.4–1.2 million years ago) in comparison to Asia (ca 2.1 million years ago) (Moncel, 2010; Leroy et al., 2011). The modelling framework addresses this complex and extensive question by combining different modelling techniques (i.e. environmental modelling, niche modelling, cellular automata and agent-based modelling). Each of these techniques focusses on a different part of the system at different levels of resolution (Moncel et al., 2025). In this paper, we focus on the role of hominin-environment interactions at the local level and how they can influence mobility patterns and home range. We want to approach the local dynamics behind the first “Out of Africa” event by analysing how variability in subsistence behaviour shaped the spatial behaviour of hominin groups, influences population densities and their long-term viability. By employing ABMs we can study underlying processes that are no longer observable or reproducible (Breitenecker et al., 2014) and generate data which is inaccessible otherwise. This approach is particularly well-suited to examining hominin mobility and spatial behaviour as it involves dynamic ecological and social processes (Griffith et al., 2010).
We are planning to create a model which allows us to study how different environmental conditions and subsistence strategies affect hominin-environment interactions, i.e. how successfully can they exploit the different resources and adapt to the changes in resource availability throughout a year in glacial versus interglacial conditions? We are mostly interested in how groups adapt their spatial movement as a reaction to changes in the available resources and subsistence behaviour as the observable home-ranges will inform us about the potential population density of the given scenario (Mandryk, 1993; Pearce, 2014). Such a model will have to operate on a local scale, i.e. the smallest spatial unit is one square kilometer or smaller and the agents act on an hourly basis or within shorter time periods, to both accurately represent a range of resources and their spatial availability, such as different plants and animals, and the behaviours of the agents like foraging or interactions with other agents. Each of these resources will vary in its spatial and temporal availability and foraging-related attributes, such as search and handling times, allowing for different return rates when exploited by the agents. The availability of these resources will also change throughout the year to represent different seasons, by adjusting the available resources every month.
This review paper aims to assess existing agent-based models investigating hominin foraging behaviour and identify general trends and key decisions made during the development of these models. On the basis of this assessment, we outline a structural design for an ABM which permits us to study the interrelations between environmental dynamics, subsistence patterns, social cooperation and mobility behaviour, as well as their impact on population dynamics.
2. Material and Methods
We followed the State-of-the-Art review methodology described by Barry et al. (2022). Data were retrieved from the Web of Science database, one of the major search engines for scientific sources. We first searched for targeted keywords and regarding the used method (“agent-based model”, “agent-based modelling”, “modelling”) AND our main two design concepts we wanted to focus on (“foraging”, “subsistence behaviour”). We then performed a second search round using the same keywords for method AND adding additional keywords identified within the papers found during the first search, such as “mobility”, “behaviour model”, “central-place-foraging” and “optimal-foraging theory”. Furthermore, a final third search was performed focusing on the desired timeframe and target entities by again combining the method AND the following keywords “archaeology”, “hominin”, “human”, “hunter-gatherer”, “palaeolithic”, “Middle stone age”, “Pleistocene” to detect any peculiar publications missing any of the previously used design concepts.
We retrieved a total of 15 models. We summarised their main features using the same expressions used during the literature search but furthermore added a set of more technological aspects (agent-levels, movement model, sensing, decision making, responses, large-scale connectivity) inspired by the ODD+D Protocol (Grimm et al., 2010; Müller et al., 2013; Grimm et al., 2020), which is a convenient framework for consistently describing the properties of ABMs.
Finally, we classified the models into three groups according to the main dynamic they represent. We established three categories (i.e., foraging models, central-place models and social interaction models) that specify the aspects and dynamics each model focuses on (Table 1). We only analysed models addressing the process of foraging; however, the complexity of both the environment and the associated foraging activities—such as gathering, hunting or fishing—is represented with varying levels of detail. Foraging models concentrate on resource acquisition processes, incorporating different activities like gathering, hunting, or scavenging within often complex environments that offer a variety of resources. These models examine the return rates of agents and which habitats and resources they exploit. In contrast, central-place models focus on the dynamics of a group and their camp location, including how the environmental conditions affect the frequency of camp relocations as described by Kelly (2013). Finally, we identified two social-interaction models that emphasise decision-making behaviour, particularly in relation to the outcomes of cooperative actions.
Table 1
Overview of foraging models: Summary of all analysed foraging models and their properties.
| SOURCE | NONAKA AND HOLME (2007) | GRIFFITH ET AL. (2010) | JANSSEN AND HILL (2014) | WREN ET AL. (2018) | WREN ET AL. (2020) | GRAVEL-MIGUEL ET AL. (2022) | RODRíGUEZ ET AL. (2023) | KOPELS AND ULLAH (2024) | SEURU ET AL. (2024) |
|---|---|---|---|---|---|---|---|---|---|
| Model ID | F1 | F2 | F3 | F4 | F5 | F6 | F7 | F8 | F9 |
| Keywords | – | Paleoecology, Agent-based models, Human evolution | Optimal foraging theory, Agent-based modelling | agent-based modelling, optimal-foraging theory, middle stone age, marine foraging | Holocene, Paleogeography, Southern Africa, Data treatment, Data analysis, Agent-based modeling | Agent-based model, Coastal adaptation, Middle stone age, South Africa, Paleoscape model, Hunter-gatherers | – | Megafauna extinction, anthropogenic impacts, social-ecological systems, complex adaptive systems, agent-based modeling, South Africa | Agent-based model Social division of labor European rabbit Net-hunting Optimal Foraging Theory Upper Palaeolithic Iberia |
| Timeframe | Gatherer | Plio-Pleistocene hominids | Ache Hunter-gatherer | Middle Stone Age | Holocene | pre-agricultural Holocene | Late Pleistocene | Late Quaternary | Last Glacial Maximum |
| Spatial scale | – | 8,000 cells with each cell 10,000 m2 | 58,408 cells with each cell 10,000 m2 | 60,000 cells with each cell 10,000 m2 | 421,200 cells with each cell 10,000 m2 | 60,000 cells with each cell 10,000 m2 | 2,691 cells with each cell 1 km2 | 44,521 cells | 2,640,000 cells with each cell 0.01 km2 |
| Time scale | 100,000 ticks | 262,800 ticks per simulation, with each tick representing one minute. | Tick = 5 minutes, 100 years model run | Each tick represents 1 day | Each tick represents 1 day | 365 ticks per simulation, with each tick representing one foraging day of varying length. | 9,000 ticks per simulation, with each tick representing one hour. | 73,000 ticks with each tick representing a day | 18,250 ticks with each tick representing a day |
| Spatial explicitness | Abstract environment | Two maps representing the Voi and Turkana regions. Each cell is classified as either channel, flooded, or unflooded. | Map of the Mbaracyu Forest Reserve, divided into cells representing one of seven major vegetation classes. | A map depicting a section of South Africa, divided into 14 terrestrial and coastal habitats. | A map depicting a section of South Africa, divided into 14 terrestrial and coastal habitats. | A map of the South Cape of South Africa featuring fourteen different habitat types, based on the reconstructed pre-modern distribution of terrestrial vegetation and habitats in South Africa. | Abstract homogeneous landscape. | Abstract map featuring cells which are either vegetated or vegetated. | A map of north-eastern Iberia featuring elevation and habitat suitability. |
| Resource complexity | A general resource that varies in density depending on the scenario and regenerates over time. | Cells contain edible plants that regrow during the growth season. Each topographic zone has an associated daily probability of carcass appearance. | 26 prey species with varying hunting-related attributes (e.g., pursuit time, weight) and individual population dynamics. | Each cell in the map provides a caloric return rates of harvesting, time required to harvest, current state of depletion, and time until replenishment, which vary based on habitat type and is influenced by seasonal and tidal cycles. | Each cell is assigned associated variables relating to the caloric return rates of resources, time required for resource exploitation, current state of resource depletion, and time until replenishment based on habitat type and is influenced by seasonal and tidal cycles. | Each habitat type includes estimated caloric returns from plants or shellfish, time required to gather these resources, and density of animal prey. The environment experiences seasonal and tidal changes. | Predators move randomly throughout the environment and produce carcasses stochastically. The frequency of carcass production and the nutrient content (energy) vary by predator species, based on reconstructed data. | Cells can once be exploited by grazing animal agents. Grazed cells regrow after a specified number of time steps. Animal agents provide energy units. | Each cell has information about the number of individuals per cell, and information about each prey type including: its energetic value in kilocalories, the mean handling time to acquire it and its breeding rate |
| Agent Levels | Individual forager | Individual hominids | Individual hunters and Camps | Agents represent individual foragers or camps | Agents represent individual foragers or camps | Band agents represent groups of foragers. | Agents represent groups of hominins, predators, or scavenging species. | Band agents representing groups of foragers. | Individual humans and Camps |
| Foraging strategies | An agent that consumes the general resource by following an optimal foraging strategy. | Individual hominids consume plants and animals based on their daily caloric needs and dietary restrictions, following an optimal foraging approach. The energy gained depends on local searching processes and methods used by the hominids. | Individual hunters operate from camps that focus solely on hunting. Prey are ranked by “profitability” based on expected meat yield per hour of pursuit (kg). Only prey types with profitability above recent mean hunting returns are targeted. | Individual foragers gather plants and shellfish using an optimal foraging approach to meet their daily energy needs. | Individual foragers gather or hunt using an optimal foraging approach to meet their daily energy needs. | Foraging agents must meet a daily energy demand through gathering plants, shellfish, and hunting mammals using an optimal foraging approach; 30% are designated as hunting agents, with pursuit time varying by prey type. | Hominin agents must meet a daily energy requirement. They only scavenge and always target the nearest available carcass when necessary. | Hunting | Hunting and catching at warrens. |
| Movement model | The forager can either travel between resource patches or forage within a single patch. | Hominid agents can travel to any of their 8 neighbouring cells at equal cost, taking one timestep to move to a neighbouring cell. | Camps relocate based on specific scenarios (1, 2, 4, 8, or 16 days). Foragers spend their days acquiring resources and moving to new camp locations. | Every Day foragers move in the direction of the new camp location while using the available time to gather resources. | One randomly chosen hunter picks a direction all hunters follow. | Individual foraging agents navigate the landscape and may relocate their camp if their caloric threshold is not met. | Hominins move at a speed of 5 km/h. | Direct movement towards the closest, highest-ranked prey animal within their foraging radius. Without nearby animals the agents perform random-walk. | Direct movement towards the best habitable cell within their mobility range. |
| Sensing | The forager has awareness of the resources available in their immediate foraging area. | Hominid agents can assess the energetic return of all plants in their vicinity and detect all carcasses within their detection range. | Agents are only aware of the conditions in the cell they occupy. | Agents can assess expected caloric returns from patches within a specified radius; for coastal patches, this evaluation can extend over greater distances. Camps can predict the future availability of resources. | Gatherers can assess the current return rate of patches within a specified radius and coastal patches even if they are outside of the radius. Hunters can assess the probability of encountering specific species per habitat type. | Agents can possess knowledge about the state of all cells or only within a specific vision radius regarding future resource availability. They select the cell that offers the highest net caloric return based on acquired energy. | Hominins can sense available carcasses within their range of vision. | The agents can assess the estimated energetic return of all prey animals within their foraging radius (6 or 12 cells) | The agents can assess the habitat suitability of all cells within their mobility range. The mobility range depends on the movement speed and the available time. |
| Ingroup interactions | – | Agents may choose to nest together as a cooperative strategy. Group-nesting agents may relocate if several members have not met their caloric demands. | Hunters within the same camp stay close to each other and may cooperate during hunting encounters. | – | Hunters follow a common direction. | Cooperation allows foragers to compete effectively against other scavengers. | – | Individual agents share acquired resources in the camp. | |
| Group decision making | – | When relocating, they follow the agent that previously acquired the most food, especially after failing to meet energy demands over a specified period. | New camp locations are chosen randomly from cells located at least 2 km away from the current position. | The group moves strategically to maximize caloric returns. | The camps move if they failed to acquire the needed calories over the last seven days. | Agents chose the Cell with the highest net calory return | Hominin groups will move to the closest cell containing a carcass if they require new energy. | Agents always target the closest, highest-ranked prey animal. | The group moves if the caloric needs of its 25 members have not been met over the last seven days. |
| Population dynamics | – | – | – | Population size can be determined at the start of each simulation run. | Population size can be determined at the start of each simulation run. | Agents may die if they fail to acquire sufficient energy but do not reproduce. | – | – | |
| Intergroup interactions | – | – | Hunters from different camps do not interact with each other. | – | – | – | Competition for resources occurs; if multiple agents attempt to exploit the same carcass, the species size or group size determines which agent succeeds. | – | – |
| Responses | The energy gained and the average return per unit of time spent foraging are tracked. | Records include calories consumed per resource and season, as well as idle time spent in camp and the type of vegetation inhabited. | Average return in kg, percentage of time spent searching, percentage of days without a catch, and prey composition. | Average caloric returns, days without food, and ratios of different food types. Key mobility factors include frequency of camp movement, distance travelled per camp or forager, types of vegetation occupied, and time spent near the coast. | Average caloric returns, days without food, and ratios of different food types. Key mobility factors include frequency of camp movement, distance travelled per camp or forager, types of vegetation occupied, and time spent near the coast. | Metrics include average caloric return per time spent foraging, calories consumed per resource, and seasonal variations in resource availability. | Survival success is measured by the total population at the end of the simulation, | Number of prey-items taken per 10 ticks, Proportion of grazed to ungrazed grass, Forager energy and Number of animals | Daily energetic intake per human and the number of individuals hunted from each prey type. |
| Large-scale connectivity | – | Illustrates changes in land usage based on environmental conditions and access to tools. | Suggested population sizes are provided based on the characteristics of the studied area. | Suggested population sizes are provided based on the characteristics of the studied area. | Small but productive coastal habitats are often reoccupied over extended periods. | The results indicate that passive scavenging may be a highly effective strategy for early Pleistocene hominins in Europe. | – | – | |
| Open-Source Code | No | – | https://www.comses.net/codebases/3902/releases/1.1.0/ | https://www.comses.net/codebases/5356/releases/1.0.0/ | https://www.comses.net/codebases/2d6a597a-76af-4ee8-bf90-0ba8c531b686/releases/1.0.0/ | https://www.comses.net/codebases/7bbb91d3-e455-4afd-82b2-a62c94ed1aef/releases/1.0.0/ | https://figshare.com/articles/code/SCAVCOMP-ABM_A_computer_model_to_simulate_competition_among_scavengers/22716427 | https://github.com/isaacullah/MegafaunaHuntingPressure/tree/main | https://www.comses.net/codebases/584b56bd-13b4-4896-9d4b-336777cf2437/releases/2.0.0/ |
| Programming Language | – | Java | Netlogo 5.0.3 | NetLogo 5.3.1 | Netlogo 6.0.3 | NetLogo v. 6.1.1 | Net Logo 6.2.2 | NetLogo 6.3.0 | NetLogo 6.0.3 |
3. Results
3.1 Foraging models
The majority of the analysed models, nine from the fourteen models found in total, take the straightforward approach to studying hominin subsistence behaviour by focusing directly on the foraging aspects.
While this approach focuses on the process of exploiting resources, some foraging models utilise a simple, abstract environment with a generic resource distributed throughout to study how foragers react to varying levels of resource accumulations (Nonaka and Holme, 2007, from now on addressed as F2, see Table 1). This approach is still applied (Romanowska et al., 2021), as specific spatial and temporal scenarios can be created within foraging models by combining a more abstract representation of an environment with area-specific values for the type and density of available resources (e.g., Rodríguez et al., 2023, from now on addressed as F7, and Kopels and Ullah, 2024, from now on addressed as F8 see Table 1). However, there has been a tendency towards using geographic maps and highly detailed environmental features to create more intricate scenarios (Griffith et al., 2010, from now on addressed as F2, see Table 1; Janssen and Hill, 2014, from now on addressed as F3; Wren et al., 2018, from now on addressed as F4; Wren et al., 2020, from now on addressed as F5; Gravel-Miguel et al., 2022, from now on addressed as F6; Seuru et al., 2024, from now on addressed as F9).
Four of these models (F3, F4, F5, F6) are similar in their code and functionality as they are variants of the Ache model (F3), but we address them as independent models as they vary in the included dynamics and behaviour. The environment of more recent foraging models comprises several different vegetation types that influence both the type and quantity of available resources (F2, F3, F4, F5, F6). In some of these (F2, F3, F4, F5, F6), the environment changes over time reflecting, for example, annual cycles, and the regeneration of resources is linked to seasonality. Larger animals, when included in the model, are often represented with specific densities and vary in their hunting-related characteristics (F3, F5, F6, F9). Whenever a more complex environment is created, the distribution and available amount of resources are grounded in empirical research and additionally incorporate theoretical models to describe broader processes (F3, F4, F5, F6, F7).
Agents in foraging models often represent individual adult foragers (F1, F2, F9), which may belong to specific groups and move and nest together (F2, F3, F4, F5, F6, F9). When these individual agents organise into groups, models often introduce an additional type of agent, referred to as a “group agent”, to facilitate processes related to collective decision-making (F3, F4, F5, F6). The organisation of these individual agents into groups resembles CPF but in contrast to specifically central-place models (see 3.2 Central-Place Models) the rules regarding behaviour and movement do not follow the dynamics of logistical foraging trips and groupwide residential moves, as described by Kelly (2013). Foraging models may also have agents representing groups of hominins of varying sizes without incorporating individual foragers (F7, F8).
As the focus of foraging models are the interactions between agents and their environments, they detail the processes involved in resource acquisition. Within foraging models, two subcategories can be identified based on the complexity of the acquisition process and how foragers decide which resource to target (Romanowska et al., 2021). In patch-choice models the agents choose patches based on the average return per patch and how far they are away (See F1, F4, F6, F9). In prey-choice models or diet-breadth models the foragers also include additional costs and risks to identify the best cell or resource to target (See F2, F3, F5, F7, F8). To be able to evaluate the best cell to exploit, agents are able to perceive their surroundings; they typically have knowledge of where resources are located, at least within their immediate foraging area (F1, F2, F8) and often beyond, allowing them to move toward areas with more or better resources over time. This long-range awareness is usually limited to several kilometres in the case of a spatially explicit map (F6, F7) but may also extend to predicting future resource availability. Such foresight enables foragers to arrive at locations just in time for when the new resources become available (F4, F5, F6).
We will first address the processes related to foraging processes (logistical movement) and then how the group moves around the landscape (residential movement). In some foraging models (F1, F2, F4, F5, F6, F8) the agents utilise OFT which allows them to maximise resource acquisition by evaluating the potential returns from different cells during their decision-making processes, considering both the available resources and their distance from their current location. In four foraging models focused specifically on meat acquisition; agents adopt different strategies during their decision-making process. They either search for prey at their current location while moving toward a randomly selected spot or direction (F3, F5), consistently head toward the nearest location with an available carcass (F7) or follow a habitat suitability map without including the occurrence of prey in their decision-making process (F9).
The energy that agents acquire depends not only on the available resources but also on various factors such as local search processes and methods used to gather plants (F2), as well as the time required to pursue moving prey while hunting (F3, F5, F6, F9). The agents often forage until they reach a certain level of resource or energy that meets their daily demands (F2, F4, F5, F6, F7, F8, F9). However, they may also engage in continuous foraging (F1), particularly when hunting (F3, F6).
The decision-making process involved in residential moves typically requires an assessment of the available resources at potential locations. Consequently, groups often avoid areas that have already been exploited, either by themselves or by other groups (F2, F3, F4, F5, F6, F7). The agents’ knowledge of their environment varies, if they can predict the availability of resources, they can select a new location based on both current resources and those anticipated to become available in the future (F4, F5). Once a new location is chosen, the group typically moves directly there and resumes foraging. However, some models allow agents to deviate from their direct path to acquire resources along the way (F3, F4, F5, F6).
When implemented, interactions between agents are focused on those related to foraging behaviour. While one foraging model features a single individual forager as the sole entity (F1), most foraging models include multiple agents that often belong to a larger group. In terms of foraging dynamics, cooperation among foragers can be established as the default strategy (F2) or as a necessary condition for success when employing strategies such as scavenging (F7). Alternatively, cooperation among group members may present certain advantages and costs for the foragers (F3). However, how agents communicate and organise themselves within the group—whether for collective foraging or relocation—is represented in a highly simplified way in foraging models (F2, F3, F4, F5, F6, F7, F8, F9). Often, a group agent is introduced as the entity responsible for representing and executing processes related to collective decision making, aiming to maximise benefits for the group by following the same OFT approach used during the foraging process (F3, F4, F5, F6). An alternative approach is demonstrated in Griffith et al.’s “HOMINIDS” ABM from 2010, which does not employ a group agent. Instead, individual foragers directly participate in the group’s decision-making process by advocating for relocation if they fail to acquire sufficient resources (F2).
The success of the agents in foraging within a given environment, as well as the effectiveness of accessible strategies, is measured by counting the energy acquired and calculating the average return on time spent foraging (F1, F4, F5, F6). Additionally, researchers can track how often agents fail to gather the necessary resources (F2, F3). However, even while failing to acquire sufficient resources the agents continue to move without facing negative consequences, as most models do not account for death—whether from hunger, predators, or other causes. Only one foraging model includes agent mortality—with death occurring if agents completely deplete their energy reserves—using the number of surviving agents as an indicator of environmental suitability and the adaptability of agents in exploiting available resources (F7). Furthermore, foraging models also document the types of resources consumed by agents (F2, F3, F4, F5, F6, F9) and examine how this consumption correlates with their movement by analysing which types of vegetation they inhabit (F2, F4, F5).
3.2 Central-place models
We categorised three models as central-place models as they emphasise the movement of foraging groups engaged in CPF, as described by Kelly (2013). In contrast to foraging models which focus primarily on the foraging process and its success, these central-place models examine how group movement varies with environmental conditions and how these changes may influence interactions between distinct groups.
The environment of central-place models is often more simplified as compared to foraging models and is created using a limited set of straightforward parameters. These parameters primarily dictate the number of cells containing resources and their distribution (Premo, 2012, from now on addressed as CP1, see Table 2; Premo, 2015, from now on addressed as CP2; Sikk and Caruso, 2020, from now on addressed as CP3). The resources are represented as a single generic resource with no further detailed attributes, but they are dynamic in that their occurrence changes over time (CP1, CP2, CP3). All cells may contain the same amount of resources (CP2, CP3), or only a small subset of cells may hold any resources at all (CP1).
Table 2
Overview of central-place models: Summary of all analysed central-place models and their properties.
| SOURCE | PREMO (2012) | PREMO (2015) | SIKK AND CARUSO (2020) |
|---|---|---|---|
| Model ID | CP1 | CP2 | CP3 |
| Keywords | Foraging; Human evolution; Hunter-gatherer Mobility; Simulation; Spatial model | Human evolution; Hunter-gatherers; Metapopulation model, Mobility, Modern human behavior | agent-based model, hunter-gatherers, central place foraging, mobility, settlement choice |
| Timeframe | Late Pleistocene | Modern Human/Central-Place Foragers | General Hunter-gatherers |
| Spatial scale | 40,000 cells | 62,500 cells | 10,000 cells with each cell 1 km2 |
| Time scale | 5,000 ticks per simulation | 2,000 ticks per simulation, with each tick representing an abstract period required for social learning. | 80 ticks per simulation, with each tick representing one week. |
| Spatial explicitness | Abstract map featuring a randomly selected subset of cells that provide resources | An abstract map featuring cells that provide resources. | An abstract map featuring cells that provide varying amounts of potential energy based on available technology and social organisation. |
| Resource complexity | One generic resource that regenerates every 1,000-time steps | One generic resource that regenerates every 800-time steps. | One generic resource that regenerates every 4 ticks at a specified rate. |
| Agent Levels | Each agent represents a foraging group | Each agent represents a foraging group. | Each agent represents a residential unit consisting of 20 individuals. |
| Foraging strategies | Agents completely exploit one cell per time step | Agents completely exploit one cell per time step. | Agents exploit cells based on their population size. |
| Movement model | Agents can gather food within their effective foraging radius; residential camps move a distance of 2re + 1 | Agents can procure food within their effective foraging radius. Follows a correlated random walk with a step length of 2 × Log + 1. | Agents procure food from surrounding areas following the optimal foraging approach. Agents will relocate if the time required to meet their energy needs exceeds the time needed to move to a new site, factoring in relocation costs. |
| Sensing | Foragers can only detect resources within their designated foraging range | Foragers can only detect resources within their foraging range. | Agents can only detect the energy levels and net energy returns of cells within their residential range. |
| Ingroup interactions | Group activities are coordinated by the group agent to reflect collective decision-making | Group activities are conducted by the group agent to represent collective decision-making. | Active foragers influence the decision-making process regarding resource acquisition. |
| Group decision making | – | – | Agents select new camp locations based on the highest maximum energy return available in nearby cells. |
| Population dynamics | – | – | – |
| Intergroup interactions | – | Cultural variances can be transmitted between groups when they are within the interaction radius. | – |
| Responses | Residential moves and interactions occur between two agents | Residential moves, cultural variance and time until fixation of one culture | Residential moves per year and the duration of logistical forays during each residential stay (averaged across all agents). |
| Large-scale connectivity | A population primarily employing a logistical mobility strategy exhibits fewer intergroup interactions | Populations primarily using logistical mobility strategies may exhibit higher cultural diversity due to rare intergroup cultural transmissions. | Resource distribution impacts mobility: however, settlement locations are influenced by factors beyond just available energy. |
| Open-Source Code | https://www.comses.net/codebases/3582/releases/1.0.0/ | https://www.comses.net/codebases/5038/releases/1.0.0/ | |
| Programming Language | NetLogo 5.0.2 | NetLogo 5.0.2 | NetLogo 6.04 |
The agents represent groups of hominins with varying sizes (CP1, CP2, CP3). In two of the analysed central-place models (CP1, CP2) agents lack information about resource availability beyond their given effective foraging range, defined as the area in which the expected return is higher or equal to the required return rates based on their energy demand and available time (Kelly, 2013). In contrast, in one central-place model (CP3) the agents use an OFT approach to evaluate all resources within their residential range, which is twice the effective foraging range, which allows them to choose the optimal location for their next residential camp. Given the simplicity of the environment, the process of acquiring resources is implemented with low complexity. However, even in such a straightforward setting, processes like local searching and various methods of resource acquisition can still be incorporated (CP3).
The decision to move the residential camp is influenced by the depletion of resources in the current location (CP1, CP2, CP3), but it may also consider the potential resources at a new location while weighing the costs of moving (CP3). It is crucial for the group to maintain a certain distance from the old camp, which depends on the distances travelled during foraging activities, so as to avoid the previously exploited area (CP1, CP2, CP3).
In terms of agent interactions, direct interactions between individual foragers do not occur explicitly in central-place foraging models as the agents represent a group of foragers. In terms of groupwise activities, the group agents embody and execute all processes related to collective decision making aimed at maximising benefits for all group members (CP1, CP2). This process has also been described as being due to the active foragers within the group who influence the decision-making process (CP3). The different groups do affect each other by targeting the same resources and may also exchange cultural traits if located close to one another (CP2).
Central-place models observe the movement pattern of the group mainly by counting the number of residential moves per year (CP1, CP2, CP3). Additionally, for further insight the mean length of these residential moves and the mean length of logistical moves can be observed (CP3). This allows central-place models to examine how resource availability generally affects the mobility patterns of foraging groups (CP1, CP2, CP3) and how these patterns change based on access to tools or skills that facilitate effective resource exploitation (CP3). Furthermore, central-place models can illustrate how the frequency of inter-group interactions varies under different environmental conditions (CP1, CP2).
3.3 Social-interaction models
The third type of model we identified focuses on social decision making in cooperative behaviour, which is often overlooked by both foraging and central-place models. These models provide agents with various strategies for how they cooperate and share resources.
Focusing on agent-agent interactions, the environment of these social-interaction models varies from one model to another, as each is structured individually to facilitate the specific aspects of interaction that the model aims to explore. The environment may be represented abstractly, including only certain elements (Santos et al., 2015 from now on addressed as SI1, and Pereda et al., 2017 from now on addressed as SI2, see Table 3), or it may utilise a geographical map along with climate data, such as temperature, to create a scenario that is spatially and temporally explicit (Coto-Sarmiento et al., 2023 from now on addressed as SI3). Since social-interaction models focus on the social behaviour of foragers during resource acquisition or the sharing of resources afterwards, these resources are characterised by specific attributes; for instance, they have been implemented as occurring randomly (SI2) and sparsely but then offer large quantities of resources (SI1), or their distribution is based on archaeological sites following the assumption that these areas will have offered resources during the studied period (SI3).
Table 3
Overview of social-interaction models: Summary of all analysed social-interaction models and their properties.
| SOURCE | SANTOS ET AL. (2015) | PEREDA ET AL. (2017) | COTO-SARMIENTO ET AL. (2023) |
|---|---|---|---|
| Model ID | SI1 | SI2 | SI3 |
| Keywords | Whales, Meat, Social influence, Random Walk, Trees, Agent-based modelling, Animal sociality, Foraging | – | Pleistocene, Central Asia, behavioural adaptations, evolutionary model, human cooperation, Agent-Based Model |
| Timeframe | Human hunter-gatherer | Human hunter-gatherer | Hominins during the Pleistocene |
| Spatial scale | 40,401 cells | 256 cells | 175,000 cells with each cell 1 km2 |
| Time scale | 100,000 ticks per simulation, with each tick representing an abstract period of days, weeks, or months during which a whale may beach. | 5,000 ticks per simulation, with each tick representing an abstract period between a new generation | 1,800 ticks per simulation, with each tick representing one month. |
| Spatial explicitness | An abstract water landscape featuring additional land cells surrounded by beach cells. | An abstract space in which foragers find resources depending on the chance to find resources. | The area surrounding the Altai Mountains and the Tian Shan region is mapped. Each cell is characterized by either water or mountains and contains a certain amount of resources. |
| Resource complexity | Randomly selected beach cells may contain beached whales. | Acquiring resources sorely depends on the chosen probability of finding resources. Each found resource provides one unit of energy. | All patches provide a general resource that is randomly distributed within a range of 0 to 50. Some cells offer higher resource amounts that regenerate over time (referred to as “attractor places”). |
| Agent Levels | Agents represent households or canoes that forage for resources. | Individual foragers | Individual foragers |
| Foraging strategies | Foragers exclusively exploit beached whales as their resource. | Agents have a chance to find an abstract resource. | Individual foragers acquire resources, with the amount consumed varying based on their cooperative strategy. |
| Movement model | Agents utilize random walk or Lévy flight movement, covering distances between 1 and 13 cells. | Agents do not move | Foragers will migrate if they cannot find available resources within a specified period or if the resource levels in their current patches fall below 20%. |
| Sensing | Agents can sense beached whales within a specified vision range. | Agents sense if other agents have not acquired any resources. | Agents can sense nearby resources after becoming familiar with their surroundings. Environmental knowledge is shared among agents. |
| Ingroup interactions | – | Successful foragers share resources either with the one unsuccessful forager that has previously shared the most resources or with a random unsuccessful forager. | – |
| Group decision making | – | – | – |
| Population dynamics | – | – | Agents remaining in unsafe locations for extended periods are at risk of hypothermia and eventual death. Agents age and die based on an average mortality rate, with a reproduction rate of 10%. |
| Intergroup interactions | Agents have the option to cooperate with other agents by sharing information about beached whales, which rewards them with reputation points. | – | During difficult periods, cooperating agents ration and share resources to support one another. Agents can choose to consume fewer resources and collaborate with other cooperators to establish new attractor places. |
| Responses | The average number of cooperating agents within the population is tracked. | Mean energy of all Agents and the composition of the sharing behaviour. | The number of cooperators and surviving agents is tracked. Cooperative behaviour increases the likelihood of survival during harsh environmental conditions. |
| Large-scale connectivity | Aggregation events, such as the presence of beached whales, promote intergroup interactions and increase cooperation within the metapopulation. | – | During harsh environmental conditions cooperative behaviour increases the chance of individuals surviving |
| Open-Source Code | https://www.comses.net/codebases/4249/releases/1.0.0/ | https://www.openabm.org/model/5287/. | https://doi.org/10.17605/OSF.IO/JM3ZY |
| Programming Language | NetLogo 5.0.5 | NetLogo 5.3 | NetLogo 6.2.2 |
Like the environment, the agents can take various forms, ranging from groups of hominins of varied sizes (SI1) to individual foragers (SI2, SI3). They are capable of sensing resources within a limited area and moving towards them. This process is explicitly implemented in these social-interaction models as cooperative processes, such as the acquisition and sharing of environmental knowledge among the agents (SI1, SI3). A unique feature of social-interaction models is the availability of multiple options for the agents on how to interact with one another depending on the agent’s characteristics or environmental condition. For example, agents may choose to share information about available resources in exchange for reputation (SI1), depending on how many resources the other agent shared before (SI2), or they might decide how to distribute or ration resources during challenging times (SI3).
4. Discussion
Summarising our analysis, we found that a solely foraging-focused approach is particularly effective at examining how foraging success varies with different environmental and behavioural scenarios. Foraging models focus on simulating a complex environment that agents exploit using varying strategies like gathering, scavenging, or hunting while non-foraging-related interactions among agents are simplified. Foraging models can closely replicate ethnographically observed foraging returns (F3, F4). They allow to test how changes in the agents’ diet (F2), group size (F3, F7), cooperative behaviour (F3, F9) and planning (F4) affect the foraging success. This experimental approach can also be extended to analyse how variation in overarching dynamics such as environmental structure (F1) or population size (F4) affects foraging success. When utilising a detailed environmental reconstruction, foraging models can predict potential diet (F4), land use (F2, F6) and population size (F5). Foraging models have also been used to identify areas of heightened archaeological visibility (F2), to study the role of hunting humans on extinction events (F8) or to validate mathematical theories such as the Mean-Value theory (F1).
However, implementing various subsistence strategies simultaneously in one model seems to prove quite challenging, so agents in these models rarely acquire both plant materials and meat. Only the model by Wren et al. (2020), which explores human foraging in coastal South Africa, has the foragers acquire both plants, shellfish and hunting meat. Depending on the targeted resource, the foragers move differently across the landscape with gathering foragers targeting specific cells while hunting foragers move in environments with a high chance of encountering prey (Wren et al., 2020). Furthermore, agents often lack the ability to react dynamically to changes in resource availability by adapting their logistical and residential mobility. This is due to the often very simplified implementation of in-group organisation and the lack of different options for the foragers to interact and cooperate with one another. In none of the analysed foraging models changes in the foraging access impact the agents. They neither adapt their movement pattern during periods of insufficient resource acquisition or die if they are unable to increase their resource intake.
The central-place approach addresses how foragers move across a landscape and react to different or changing environmental conditions. CPF Models provide a solid structure for experiments as they are built upon clearly defined rules regarding how agents move across the landscape. The models effectively illustrate how resource density and access to resources influence group behaviour, demonstrating that deteriorating conditions lead to increased mobility in search of new resources (CP3). Existing central-place models illustrate successfully how declining conditions lead to increased mobility in search of resources. A model focused on the dynamics related to CPF can also include more detailed foraging behaviour as shown in the “OFTpatterns” ABM presented in Sikk and Caruso (2020). Alternatively, foraging-focused models can also create mobility patterns similar to those proposed by the central-place approach by introducing simplified aspects of CPF (Wren et al., 2018). Nevertheless, the central-place models we studied do not account for the processes that drive the performance of CPF, including the necessary communication and organisation among foragers.
The significance of this social behaviour is emphasised in social-interaction models, which illustrate how specific environmental conditions can facilitate social interactions or are essential for survival. The “WWHW” ABM by Santos et al. (2015) illustrates how sporadic events of extensive resource availability like the beaching of a whale increase inter-group interactions and increase the cooperation between a usually widespread metapopulation as they experience momentary aggregation. When applied on larger spatial and temporal scale, as demonstrated in the “PaleoCOOP” ABM developed by Coto-Sarmiento et al. (2023), which features monthly time steps and an overall runtime of 150 years, the effects of these cooperative decisions can be observed over an extended period and analysed alongside population dynamics. Consequently, changes in survival rates can be directly measured by tracking the number of surviving agents (SI3). However, social interaction models create specific environmental conditions to study the specific interactions in which they are interested. Therefore, developing a range of social behaviour scenarios with different forms of exchange of resources, tools or knowledge and measuring how they interact with different foraging-related processes and affect the group will pose a significant challenge. This is particularly the case given that the structure of the general environment in spatially explicit foraging models is often specifically tailored to the chosen area and objective.
The chosen modelling approach has a large effect on the challenges that need to be addressed during the creation of the model. But whenever the chosen approach involves a detailed environment, a large portion of the model development needs to address the acquisition of all the necessary environmental data. This can be especially difficult when studying the past, as available environmental reconstructions are often limited to specific periods and regions. Most comprehensive environmental reconstructions of the past focus on well-studied sites but still lack crucial information about the types of resources available to hominin foragers in those environments. Reconstructing resource availability typically requires extensive additional research and the use of contemporary reference models. As a result, many small-scale models with highly detailed environments are set in the present day (see Janssen & Hill, 2014; Gravel-Miguel et al., 2022) or have made significant efforts to recreate conditions from earlier periods (see Griffith et al., 2010; Wren et al., 2018; Wren et al., 2020).
Furthermore, the large variety in approaches chosen to represent environmental conditions in the analysed models showcases the large complexity of environmental dynamics and how they affect which, when and where resources are accessible for hominin foragers. Depending on the targeted location and time period, a model may have to include very specific environmental dynamics to accurately describe the important processes, such as whales stranding in an unpredictable pattern (Santos et al., 2015), which may not occur at all even in a nearby region. Even on a smaller scale, the location and time period may drastically change the availability and structure of resources and how accessible they are for hominin foragers (Kelly, 1983). Identifying the resource accessibility of a certain region or type of vegetation may be difficult or impossible to obtain for older periods or less frequently studied areas.
A similar problem arises when trying to model behaviour ascribed to early hominins that differs from, or is not visible at all in, recent hunter-gatherer societies. Potential early hominin behaviour is often heavily influenced by ethnological data from recent hunter-gatherer societies (French, 2018) but modelling other non-modern hominins requires not only assessing which tools they may not yet have access to (Pobiner, 2020) but also how this would affect foraging-related activities. But concerns regarding the usage of ethnographic analogies go deeper (Currie, 2016), as there is still a discussion regarding the cognitive capabilities of early hominins to perceive and exploit the environment in the same way as modern hunter-gatherers do (Schuppli et al., 2016; Wynn and Coolidge, 2016; Migliano and Vinicius, 2021).
5. Conclusion
For the model we plan to develop, applying a CPF approach seems most promising as it will provide a good framework for analysing how the mobility of a hunter-gatherer group and the occupied space change depending on the behavioural and environmental conditions. We also need to include an OFT approach as we want to compare different environments and how the difference in resource availability and accessibility affect the foraging success. This combination of CPF with more detailed foraging behaviour has already been done in the “OFTpatterns” ABM presented in Sikk and Caruso (2020). The model allows us to study the dynamics between resource availability and foragers’ mobility patterns, but the environment is an abstract representation of a heterogeneous environment. The current level of detail in the environment and how the agents interact with it does not suffice for the kind of study we would like to perform. Therefore, a combination of a detailed environment and CPF will be a novel approach.
The resulting model would be able to represent hominins dynamically foraging within a complex environment covering around 10,000 square kilometres and acquiring both plants and meat using different strategies like catching, scavenging or hunting. Such a model would allow us to understand how foraging success and mobility strategies changed across time and space. The occupied home range together with the number of members in the group can be used to calculate population densities. In the proposed model, scenarios with insufficient resources or with the foragers lacking the abilities to exploit the available resources would force the foragers to exploit a larger area, resulting in lower population densities. If the population density fell below a threshold of 0.005 ind/km², the region most likely would not support a population in the long term (Wobst, 1974; Mandryk, 1993). Implementing other metrics to identify unsuitable conditions are of course also possible. Subsequently, when combining our model with a detailed environmental reconstruction including reconstructed plant availability, based on climate reconstructions, and reconstructed fauna composition, based on fossil remains, we expect to gain new insight into how local dynamics in resource availability may have shaped the spatio-temporal pattern of the population of Western Europe at a temporal scale invisible in the archaeological record.
Eventually, the outcomes and knowledge of the interplaying dynamics we gain from a local-scale model as the one we described in the introduction could be used to parameterise further agent-based models. The LATEurope multi-scale modelling approach has the goal to study the nested structure of the complex adaptive system behind the first hominin occupation of Europe (Cucart-Mora et al., 2026). While the proposed subsistence model will cover local-scale dynamics on a yearly basis, the observed home ranges will inform other models on a meso and macro scale (sensu Cucart-Mora et al., 2026). This approach allows for agents within a continent-scale model to behave differently according to the environment they occupy with the option to simplify the implementation of local foraging dynamics, relying on the data generated by smaller-scale models. Furthermore, we will also be able to infer population densities from such a local-scale model and compare the results with population dynamics visible in a continent-scale simulation.
Competing Interests
The authors have no competing interests to declare.