Archaeological data were collected from publications, databases, unpublished theses and reports. The database, which was made available online, contains 2762 radiocarbon dates from 1023 archaeological sites in lowland South America. These data were analysed using statistical methods for ascertaining spatial patterns in the dispersal of archaeological phenomena - i.e. to understand their origin and speed of advance. Specifically, four archaeological cultures were tested, each related to the diffusion of ceramics and farming and associated with the expansion of a major language family in South America: Saladoid-Barrancoid (Arawak languages), Incised-Punctate (Karib languages), Tupi (Tupi languages) and Una (Macro-Jê languages). Space-time regressions were employed, correlating the earliest date per site with distance from hypothetical origins. The earliest sites were tested as potential origins, but iterations were also performed over all sites. Results showed high correlations and inferred speeds of expansion reasonable for demic diffusion in the case of some cultures, but not in others.
An agent-based model (computer simulation) of population growth and dispersal was developed based on a review of the literature about tropical farmers. Information was gathered on population growth, village size, how often did they move and how far. A range of values for demographic and territorial parameters could then be established. The model was written in Python and based on the architecture of published agent-based models for the expansion of the Neolithic. The model was run for the four archaeological cultures mentioned above, taking into account different simulated start dates and coordinates. For each culture, a comparison was made between simulated arrival times at different regions and the earliest respective radiocarbon dates. In order to arrive at the set of parameters that would best fit the archaeological gradients in radiocarbon dates, genetic algorithms were employed. Even in that case, only two archaeological cultures were found to be adequately modelled as demographic expansions, the first (Saladoid-Barrancoid) showing a speed of expansion comparable to that of the Neolithic in Europe (ca. 1 km/yr).
The previous model did not include a dynamic environment. To specifically test the role of climate change in human dispersal, the Tupi were chosen as a case study. The Tupi are one of the major cultural and linguistic expansions of South America, and one which has repeatedly been linked to climate change/forest expansion. The environment of the model was taken from previously published biome reconstructions based on the HadCM3 climate model. Two scenarios were tested, (1) one in which all land cells can be settled and (2) one in which only tropical moist forest cells can be settled. Simulated arrival times were compared to the empirical radiocarbon record of the Tupi sites. The results showed that including the reconstructed late Holocene vegetation as a constraint to movement greatly improves the match with the empirical radiocarbon record – giving further support to the hypothesis that climate change shaped the dispersal of forest farmers in South America.
