ABM for Brazilian rock art studies

As part of a research project on computer applications in archaeology, we are proposing to use agent-based modelling for Brazilian rock art studies. Here, we want to analyze the mechanisms of landscape construction, when there are various independent agents. The basic idea is that, when a hunter reaches a rockshelter, a random number below a certain threshold results in a new painting on this particular site. At the end of a certain time lapse, we can study the whole dynamics according to simple rules.


ABM allows us to test a set of hypothesis proposed in the Brazilian literature to explain why some sites are chosen and not others. Brazilian rock art being predominantly an open-air phenomenon, the archaeological literature balances between two tendencies: the first one is purely factual and sees the environment as the main driver for human choices between two individual rockshelters, in terms of geological structure, location or stone quality. Meanwhile the second one is more relativist and considers culture as the main parameter: here, social definition is the motto. For our purpose, we consider the first one defined by [exogenous|global] parameters, and the other as defined by [endogenous|local] criteria.


The three first versions of our model were dedicated to define these variables. Put together in a single model, they allow us to visualize each specific scenarii and compare the results:

  • In the first one, all shelters hold an equally fixed threshold. As such, it means they all offer equal possibilities for the actors. In this case, the landscape is just like a large set of generic sites;
  • In the second one, the threshold is defined exogenously and globally. Practically, a random value is defined for each shelter, within certain limits, once and for all the hunters. This way, the landscape is scaled between “awsomely-fit” and “definitely-not-fit” sites;
  • Finally, in the third one, an endogenous and local random threshold, still within limits, is defined by each individual hunter for each individual shelter. In this scenario, each hunters is able to define what (s)he considers being fit or not, independently of the others.

With netLogo, we can program a whole set of simple rules and let it runs each specific scenario, or even a combination of them. According to the selected hypothesis, a threshold is defined on a base 100 at the beginning of the run. This value has a direct impact on the interaction between hunters and shelters, producing paintings or not, and we can study the evolution of the graphical landscape.

Generally speaking, a low threshold value makes it slow and complicated for hunters to produce new figures, while a high one isn’t so much of a problem. It is important to note that the exact content or meaning of the threshold is not defined. We only consider broad categories of parameters. It means that an exogenous and global threshold isn’t linked to any specific variable, be it rains, animal migration or rock weathering. Similarly, using a cultural threshold doesn’t mean we favour ritual over social hierarchy or anything. The mechanisms we are studying are determined by the origin: the external environment or the social group itself. In any case, we are studying the landscape construction dynamics, and not to particular set of proposals.

With these definitions, we can analize the distribution of paintings in a set of shelters, both quantitatively and qualitatively. Using tests against random movement and random values, we have an opportunity to study if, where and when non-random distributions occur. The results can then be compared to a series of known archaeological contexts.

Running the first tests, we decided to stop the simulation as soon as a shelter reaches 100 paintings. For each hypothesis, the results show different patterns.

  • In the first case, a generic landscape of equal shelters leads to a normal bell-shaped distribution of figures among these. Both the minimum number of motives and the overal mean are high, indicating that hunters were able to make new paintings in all the shelters. The only differences depend on initial positions and movement.
  • In the second case, the shelters are given a randomly set variable, set from the beginning of each run and acknowledged by all hunters. As a result, shelters being given a low threshold have minimal chances of receiving any figures at all. Considering these parameters, we expected the results to be far more clustered around those shelters with high value, regardless the initial positions and movements. Runs showed that the expected dynamics were correct, and the resulting clustering largely depends on the original geographical configuration and distribution.
  • Finally, the third scenario showed mixed situations. In this case, each hunter had an individual cognitive map, through which he was able to attribute a random value for each shelter at the beginning of each run. Considering the whole population, chances were that each shelter had a different threshold for each hunter. Due to this diversity, the model generally resulted in normal bell-shaped distributions similar to the first scenario: the greater the number of hunters, the closest to random the general distribution of paintings on shelter. This can be explained by probability: as each shelter is visited by every hunter many times during a run, the low threshold on one’s cognitive map is balanced by another’s high value. It should be noted, though, that in all these three cases, the construction of landscape through time followed a linear progression.

While rock art is attested in many parts of Brazil, a few larger sets have been defined with more or less precise geographic boundaries. Among these, a group of figurative collective representations has been characterized in the 1980s in an area stretching east of the São Francisco and Parnaíba river valleys to the northeastern Atlantic coastline. Despite such a large region and a cruel lack of data for the intermediary zones, three main clusters have been identified in the Serra da Capivara (Piauí), Seridó (Rio Grande do Norte) and Chapada Diamantina (Bahia). Between them, hundreds of kilometers only show a limited number of poorly documented sites. Yet, even considering the limits of available archaeological data, clustering seems to be an important behaviour.


Using clustering as a marker, the second scenario only was able to show adequate behaviour, when the act of painting is a result of the shelters natural properties. Environmental determinism was particulary influencial in Brazilian archaeology during the 1950s, when a National Program of Archaeological Research (Pronapa) was created under the umbrella of Betty Meggers. Yet, another specific situation could be derived from the results of the third scenario. If we consider that cognitive maps are transmitted from group to group, and from one generation to the other, a conjunction of high random value on specific sites would also be able to create clusters. This very particular mechanism has given birth to the concept of Tradition in the 1960s. According to this idea, an initial definition of basic graphic representation principles was generated, probably in these areas where the oldest cases where attested. It would then have expanded to new regions, through gradual new developments, with population growth.

Such a transmission has been identified in the occurrence of complex figurative scenes envolving the same graphic elements in all the main regional clusters. These emblematic scenes have first been attested in the Serra da Capivara and in the Seridó, and later in the Chapada Diamantina.

As a general mechanism, the Tradition has also been used to map the presence of Nordeste motives on many sites. Yet, if we consider population growth and distance, we should also expect a accompanying growth of social and cultural constraints, limiting the development of new and unforseen characteristics. To this day, there is no archaeological evidence of such burdens as hierarchy or political structures. We may then ask ourselves if another mechanism could have been at work.

The ABM model allows us to propose an alternative perspective, one that wouldn’t be altered by population growth and geographical distance. If we consider the rules set for each hypothesis, they were placed on only two elements of the rock art creation process: shelters and hunters. A third element went dramatically neglected, the motives themselves. A new hypothesis could be created that defines the threshold on the presence of previous paintings. In such a non-linear framework, one particular shelter would be more and more attractive as it gets more and more paintings, as a kind of cumulative effect.


This new hypothesis starts with a minimal threshold for every shelter, in order to allow the probability of the very first motive. The initial conditions would then be very similar to the first hypothesis of complete randomness. As soon as a shelter receives its first paintings, though, it would gain a small benefit added to its threshold value. The larger the number of paintings, the larger the benefit would be, on a regular base. On the model, this was stated through a single rule: the benefit would be equal to the actual number of paintings multiplied by a magnifier between 0.0 and 2.0. A slider was then created to control this value.

The new hypothesis showed interesting results. First, despite initial conditions similar to our most random tests, the magnifier value is able to modify the dynamics. Below 0.5, its effects aren’t sufficient to change the general distribution of paintings in the landscape. Around 0.7, it show a growing clustering of new figures on some specific shelters. Above 1.0, the tendency is inverting, and the benefits of each new paintings become so high that virtually every visit of a hunter in a shelter results in a new figure – even when there is only one single motive on the site. Second, population doesn’t seem to have an effect on the results. In fact, its growth seems to be directly linked to a third effect: the speed at which the landscape is constructed (or at least, at which a first shelter reaches 100 paintings).


Of course, not every run shows the same results. The initial position of both shelters and hunters, as well as the movements of the latter, are important if we want to understand why a specific site receives paintings. Anyway, this was not our objective, as the model cannot be expected to reproduce real situations. What it clearly shows, though, is that a cumulative effect is able to result in clustering, even when hunters have completely different cognitive maps, and when their number grows larger.

The next question is this: if this really were to have been an effective mechanism, what should we then expect to find on archaeological ground?

This work will be presented at the VIth meeting of the Associação Brasileira de Arte Rupestre, on september 14.

Deixe uma resposta

O seu endereço de e-mail não será publicado. Campos obrigatórios são marcados com *