Foraging of Homo ergaster and Australopithecus boisei in East
African environments
Marco A. Janssen
Indiana University
School of Informatics and Center for the study of Institutions, Population, and Environmental
Change
maajanss@indiana.edu
Jeanne M. Sept
Department of Anthropology
Indiana University
sept@indiana.edu
Cameron S. Griffith
Department of Anthropology
Indiana University
casgriff@indiana.edu
Abstract
In this paper we present the initial results of an agent-based model of foraging of hominids. The
model represents foraging activities in a landscape that is based on detailed measurements of
food availability in the modern East African environments. These current landscapes are used as
a model for the environment of the hominids one million years ago. We explore the spatial and
temporal consequences of foraging patterns in different types of semi-arid landscapes and
different types of hominids (Homo ergaster and Australopithecus boisei) who are defined with
different abilities and preferences.
Contact:
Marco A. Janssen
School of Informatics &
Center for the Study of Institutions, Population, and Environmental Change
Indiana University
408 North Indiana Avenue
Bloomington, IN 47408-3799
USA
Voice: (812) 855 5178
Fax: (812) 855 2634
e-mail: maajanss@indiana.edu
Keywords: foraging, hominids, field data
Foraging of Homo ergaster and Australopithecus boisei in East African environments
Marco A. Janssen, Jeanne M. Sept, Cameron S. Griffith
Introduction
This paper reports the initial results of an agent-based model of hominid foraging in a complex dynamic
landscape. Optimal foraging theory argues that foraging behavior is a Darwinian adaptation to search for resources
in a particular environment. Foraging agents make decisions as if they optimize a certain currency, say energy
intake, given the environmental constraints (Pyke, 1984).
What the best foraging choices for a hominid would be depends on nutritional requirements, the cognitive
and communication abilities of the agent, the abilities to make and use tools, the group size, the group dynamics, the
complexity of the landscape, the existence of competitors and predators, etc. We have developed an agent-based
model as a tool to explore the consequences and consistency of different assumptions. The two species we model are
of particular interest to paleoanthropologists because they were sympatric in a number of different habitats in eastern
Africa between 2.0 and 1.3 million years ago, and have distinctive morphological adaptations that have often been
interpreted as evidence of dietary divergence (e.g. Potts, 1998; Wood & Strait, 2004). Australopithecus boisei has
been described as a “megadont,” with teeth, jaws and cranio-facial morphology showing evidence of a diet that
included tough a range of tough plant foods that required crushing and grinding. Whether or not this species
specialized in a diet of low quality plant foods, or was more typically omnivorous and only resorted to such foods in
famine times is a matter of current debate. In contrast to their robust cousins, Homo ergaster, had a relatively small
chewing capacity and lightly built face and jaw, suggesting that its diet would have consisted either of relatively
soft, easy-to-chew foods, and/or foods that were processed with tools before being eaten. Because the larger brains
of H. ergaster would have been costly, metabolically, many paleoanthropologists have suggested that H. ergaster
evolved a dependence on eating significant amounts of meat, in addition to high quality (easily digestible) plant
foods, both of which would have required the use of tools such as stone knives, carrying devices, and digging sticks.
The East African archaeological record during this time span consists of simple stone tools associated with
fossilized remains of animals that have been butchered, and it is generally assumed that H. ergaster was the stone
tool maker who fed on the meat and marrow at these sites. But whether or not A. boisei also could have made tools
and eaten meat is a matter of debate.
Some of the key paleoanthropological questions we address in our model include:
(1) to what extent would differences in chewing abilities and tool use limit access to various food types in the types
of semi-arid landscapes in which these hominids lived?
(2) if these species preferred different types of foods, how would that have influenced differences in their ranging
behavior and frequency of habitat use through time?
(3) as both these hominids existed in several types of semi-arid habitat during periods of climate change – how
would their different morphological and technological traits have led them to respond to the selection pressures in
these habitats in different ways?
Direct evidence of this early phase of human evolution comes from several sources: samples of fossilized
remains of the hominids themselves; associated macrofossils of fauna and flora; archaeological evidence suggesting
where stone tools were made and how they were used, and associated paleoenvironmental indicators such as soil
chemistry and fossil microfauna, or pollen. Since we cannot make direct observations of the foraging behavior of
these extinct species we base our inferences on these different sources of information interpreted within the
comparative framework of evolutionary biology and comparative primate ecology, including observations of
foraging of other primates (Ramos- Fernández, et al. 2004; Goldstone and Ashpole, 2004). Formal models help us to
analyze in a consistent way the consequences of various assumptions.
Earlier work on hominid foraging and agent-based models focused on more cognitively rich agents on a
relative simple landscape of resources (Lake, 2000; Reynolds, 2001, Costopoulos, 2001). Our agents are cognitively
very simple, but they forage on a more complex and empirically-based landscape than related publications. We
assume that selected samples of the current landscape in eastern Africa can be used as a model for the environments
in which hominids were foraging 2.0 to 1.5 million years ago (Sept, 1994). This enables us to use detailed transect
data sampled from modern semi-arid riparian landscapes to create a model landscape which includes the availability
of various food types in space and time. The generated dynamic landscape is populated with agents with simple
foraging related decision rules. We present in this paper some initial results of decision rules for two types of
hominids: H. ergaster and A. boisei foraging in two different semi-arid riparian landscapes (dry and wet).
2
An agent-based model of foraging of hominids
The landscape the agents forage on is based on samples of Kenyan habitats analogous to sites where early hominids
lived 2.0-1.5 million years ago (Sept 1984, 1994). We describe now in formal terms the landscape dynamics and the
rules the agents use for foraging. Consider a population of NA agents in a landscape of N1 x N2 regular cells. These
cells represent areas of 100m by 100m. A cell ij represent one of different types of land cover Lij, and contains
different amount of units NUijf for various types of food sources F.
Landscape dynamics
The landscape consists of three zones besides the river which cross the landscape (Figure 1): the channel and
margins, the flood plain, and the unflooded zone. Detailed transect data from Sept (1984) are used to estimate
density of vegetation per hectare (Eberhardt, 1978). Using the average density estimates of vegetation for the
different land cover, we populated the landscapes with vegetation per ha by a stochastic process.
We distinguish two landscapes: Voi and Turkana. The Turkana area today is relatively drier than the Voi
landscape, and the vegetation structure and composition in the two regions reflects this difference. We include about
30 different types of food items in the landscape, available in different periods of the year. The limited space
available for this paper does not allow us to provide the detailed information, but this will be reported in a separate
publication.
The landscape is updated every simulated day, and consists of four different seasons. The first season, from
March to May has the main rains. The second season, from June to August, is mainly dry. The third season, from
September to November, is dry with short periods of rain, and the forth season, from December to February, is
mainly dry. For each season we have an estimate on the availability of every food item. To calculate the available
food in a cell, we calculate the increase and decline of the availability of food. The potential amount of food is
assumed to increase during the beginning of the season (growth), and declines in the second half (decay, consumed
by other species). The actual available amount of food available is the difference between the potential and the
amount that is consumed of that food item. The availability of vegetation is used to calculate the kcal available in the
landscape by using estimates on the amount of food items (berries, seeds) per unit of bush, tree, etc, and the amount
of kcal per food item.
In some simulations for H. ergaster we include meat on the menu (Wood & Strait, 2004). Since A. boisei is
often assumed not to eat meat, we have assumed they are vegetarian for this initial version of the model. In the
simulations where meat is included in the menu, carcasses of different sizes are placed randomly in the landscape.
Our estimates of carcass density are based on modern data collected in the Serengeti and Ngorongo ecosystems of
Tanzania and the Galana and Kulau ranches of the Tsavo East National Park in Kenya (Blumenshine 1986,
Dominguez-Rodrigo 1996). The availability of a carcass declines rapidly over time, in order to simulate
consumption by predators like lions and hyenas. Following Blumenschine and Dominguez-Rodrigo, we assume
competition for carcasses is higher in the unflooded area (more open area) than in the more forested area around the
river channel.
Figure 1: Stylized landcover of Voi (left) and Turkana (right): blue cells are channel and margin, green cells are floodplain, light
brown is unflooded. Both areas are 4 by 10km.
Agents
An agent looks for food during a day until one of the following three conditions is met.
- The agent’s stomach is full, (5kg gut capacity for all agents).
- The agent has consumed a minimum level of kcal (3000 kcal for H. ergaster and 2500 kcal for A. boisei).
- The agent has spent a maximum amount of hours that day on foraging. If food is scarce and agents move
around to search for food, we assume they stop after 12 hours. Included in the model is time spent by
agents on the various actions during foraging, i.e. harvesting food items, processing of the food, traveling,
etc.
3
Each day agents are randomly selected to be updated. Each update consists of consuming food items or, in case of
not finding food, one random movement to another cell. The searching process of an agent consists of defining the
target, moving to the target, potentially encountering other food items, and handling the food item.
An agent has limited vision. It can see certain visible, “canopy” food resources in the cell in which it stands
with probability vfC. The agent can also see canopy food resources in other neighboring cells. We assume that the
agent can see food items in the eight surrounding cells (the Moore neighborhood). The food items in the four
neighboring cells that are adjacent to the cell are assumed to have a probability being seen of vfA. Food items in the
four neighboring cells at the corners of the cell are assumed to be identified with probability vfN. The probabilities
in adjacent cells and corner cells may be different since parts of the corner cells are further away from an agent in
the center cell, compared to the adjacent cell, and therefore less visible.
An agent thus selected for action will first defines a target to approach. An agent looks for vegetation
bearing food of interest in the Moore neighborhood. A bush/tree/plant is spot when a random number between 0 and
1 is larger than (1–v )M, where v is the probability that vegetation can be seen from the current position of the agent,
and M the number of plants/bushes/trees. The preference rating of spotted food items depend on the kcal per gram
per minute handling rate.
The agent will then move toward its chosen target, but can encounter a more desirable food item which
could not have been seen from the original position (such as low-lying squash plants or small berry bushes). The
suite of food items available for a particular cell is checked for whether it is probabilistically encountered by the
agent on its way to its target. If it encounters a food item, the agent stops and consumes the encountered food item.
To calculate the probability of encountering we use the average number of vegetation per transect of 100 meter.
When we include meat we assume that carcasses can be spotted from a greater distance (500m) in the
unflooded area (due to the presence of circling vultures). When a carcass is spotted, agents will always go to it. They
move to it, not as an individual, but by a sample of the group (to compete with predators) and for simplicity’s sake
we assume that the available meat is shared among all agents in the group.
For this model we assume that agents sleep in a group, and return to a camp/nesting site by default until
food availability around the sleeping site is depleted (the average consumed kcal is smaller then the minimum
required amount of kcal), at which point the sleeping site is moved to a new nesting location, where agents of the
group found the most food during that day. Such behavior is not a realistic reflection of primate nesting behavior
today, but a useful starting point for our model. Primates such as baboons and other open-country monkeys will
often forage and nest in groups for safety, while larger apes, such as chimpanzees, are often forced to forage and
nest independently when food is scarce, though they will share feeding sites and seek to sleep in trees near each
other when food availability makes that feasible.
Model experiments
We present some initial results of a group of 20 agents for simulations of a 100 year period. Basic experiments
include four simulations, with one of each type of hominid on each type of landscape. Figure 2 shows that agents
have difficulty meeting their basic energy requirements from the available plant foods alone. Especially in the harsh
fourth season there is a problem for the agents to sustain themselves. The agents either reach their gut capacity with
low-quality, fibrous plant foods, and/or run out of time searching for higher quality foods before they have enough
food. It is not the availability of food itself, but the quality that leads to constraints.
Figure 3 shows the travel distance of H. ergastus in 2 different landscapes in the four different seasons. The
power law distribution of travel distance is similar to empirical observation of spider monkeys (Ramos- Fernández,
et al. 2004). Most of the days, the agents travel only a small distance, but in some days they travel a lot.
There are differences between the seasons which affect agent behavior. Season 4 is the harshest in the Voi
landscape, leading to large distances of travel every day. The first 2 seasons are more resource-rich leading to less
travel. The Turkana landscape has fewer differences between the seasons, in term of travel distances. Only in the
second season there is more travel. Interestingly, travel is not more in season 3, when agents do not meet their
required kcal.
When we include meat in the menu of H. ergastus the agents meet their required kcal (Figure 4). The
agents also will travel more since they go after the carcasses and see opportunities on larger distances. Where do the
hominids forage? Agents forage at higher density in the unflooded area of Voi and in the channel and margins of
Turkana. There are small differences between the two types of hominids caused by the different preferences (Figure
5).
4
3500 3000
3000 2500
seeds seeds
2500
kcal per agent per day
kcal per agent per day
root root
2000
pod pod
2000 leaf leaf
fruit 1500 fruit
bud bud
1500
blossom blossom
berries 1000 berries
1000
500
500
0 0
1 2 3 4 1 2 3 4
season season
3000
3500
3000 2500
seeds
seeds
kcal per agent per day
2500 root
kcal per agent per day
root 2000
pod
pod
2000 leaf
leaf
fruit 1500 fruit
bud bud
1500
blossom blossom
1000 berries
berries
1000
500
500
0 0
1 2 3 4 1 2 3 4
season season
Figure 2: Menu of hominids aggregated into 8 types of food resources. The upper results are from Voi, the lower from Turkana.
The left is from the H. ergastus, the right is from A. boisei.
1 1
1 11 21 31 41 51 61 71 81 91 101 1 11 21 31 41 51 61 71 81 91 101
0.1 0.1
1 1
2 2
0.01 3 3
0.01
4 4
frequency
frequency
0.001 0.001
0.0001 0.0001
0.00001 0.00001
0.000001 0.000001
travel distance per day travel distance per day
Figure 3: Distribution of travel distance (number of cells per day) per agent per day for the 100 year simulations of H. ergastus.
Left is the Voi landscape. Right is the Turkana landscape.
3500 20
3000
15
2500
kcal per agent per day
travel per day (100m)
2000
voi-h 10
1500 voi-h-meat
voi-a voi-h
turk-h voi-h-meat
1000
turk-h-meat voi-a
5
turk-a turk-h
500
turk-h-meat
turk-a
0
0
1 2 3 4
1 2 3 4
season season
Figure 4: Left figure shows the average kcal per day for the different simulations. Right figure shows the average travel distance
per day for the different simulations.
5
5
4.5 Unflooded
Flooded
4
kcal per agent per day per ha
Channel Margin
3.5
2.5
1.5
0.5
0
voi-Homo voi-Homo voi-Austra turkana-Homo turkana-Homo turkana-Austra
(meat) (meat)
case
Figure 5: Average kcal per ha per agent per day for the different types of land cover in the 6 different simulations.
Discussion
We presented the initial results of the agent-based model of foraging of hominids in a complex dynamic landscape.
Our main focus so far has been to create an empirically based landscape that covers the complex environment
wherein the hominids where foraging. The agents for now are immortal clones with simple decision rules. Future
work will explore more elaborate cognitive and social processes like decisions when to go after which type of food,
the inclusion of spatial memory, and having the agents take on different roles in their group (age and gender
differences, food sharing and provisioning), including the role of tools. We also want to use this framework in the
longer term to under which circumstances what kind of agents emerge when we include evolutionary processes.
References
[Blumenschine 1986] Blumenschine, R.J., 1986, “Early Hominid Scavenging Opportunities: Implications of Carcass
Availability in the Serengeti and Ngorongoro Ecosystems.” Bar International Series 283, Oxford.
[Costopoulos, 2001] Costopoulos, A., 2001, “Evaluating the impact of increasing memory on agent behaviour:
Adaptive patterns in an agent based simulation of subsistence,” Journal of Artificial Societies and Social
Simulation 4(4): <http://www.soc.surrey.ac.uk/JASSS/4/4/7.html>
[Dominguez-Rodrigo 1996] Dominguez-Rodrigo, M., 1996, “A landscape study of bone conservation in the Galana
and Kulalu (Kenya) ecosystem,” Origini 20: 17-38.
[Eberhardt, 1978] Eberhardt, L.L., 1978, “Transect methods for population studies” Journal of Wildlife Management
42(1):1-31.
[Goldstone & Ashpole, 2004] Goldstone, R.L., B.C. Ashpole 2004, “Human foraging behavior in a virtual
environment”, Psychonomic Bulletin & Review 11, 508-514.
[Lake, 2000] Lake, M.W., 2000, “MAGICAL computer simulation of Mesolithic foraging,” in T.A. Kohler & G.J.
Gumerman, editors, Dynamics in Human and Primate Societies: Agent-Based Modelling of Social and Spatial
Processes, Oxford University Press, New York, pp. 107-143.
[Pyke, 1984] Pyke, G.H. 1984, “Optimal foraging theory – a critical review”, Annual Review of Ecology and
Systematics 15: 523-575
[Ramos- Fernández, et al. 2004] Ramos-Fernández, G., J.L. Mateos, O. Miramontes, G. Cocho, H. Larralde, & B.
Ayala-Orozco, 2004, “Lévy walk patterns in the foraging movements of spider monkeys (Ateles geoffroyi),”
Behavioral Ecology and Sociobiology 55:223–230
[Reynolds, Whallon & Goodhall, 2001] Reynolds, R., R. Whallon & S. Goodhall, 2001, “Transmission Of Cultural
Traits By Emulation: An Agent-Based Model Of Group Foraging Behavior”, Journal of Memetics, 4.
http://jom-emit.cfpm.org/2001/vol4/reynolds_r&al.html
[Sept, 1984] Sept, J.M. 1984, “Plants and early hominids in east Africa: a study of vegetation in situations
comparable to early archaeological site locations”, PhD dissertation, University of California, Berkeley,
[Sept, 1994] Sept, J.M. 1994, “Beyond bones: archaeological sites, early hominid subsistence, and the costs and
benefits of exploiting wild food in east African riverine landscape”, Journal of Human Evolution 27: 295-320.
[Potts, 1998] Potts, R. 1998, “Environmental hypotheses of hominid evolution”, Yearbook of Physical Anthropology
41: 93-136.
[Wood & Strait, 2004] Wood, B.& D. Strait 2004, “Patterns of resource use in early Homo and Paranthropus”,
Journal of Human Evolution 46:119-162.
