Authors

Matt Grove¹, Lucy Timbrell^1,2, James Blinkhorn^2,3

¹Department of Archaeology, Classics and Egyptology, University of Liverpool, 12-14 Abercromby Square, Liverpool, L69 7WZ, United Kingdom

²Pan-African Evolution Research Group, Max Planck Institute for Geoanthropology, Kalhaischestraße 10, Jena, 07745, Germany

³Centre for Quaternary Research, Department of Geography, Royal Holloway, University of London, Egham, TW20 0EX, United Kingdom

Abstract

Introduction

There is a considerable history of research into the relationship between palaeoenvironments and hominin habitation. In recent years, this has been developed further via the use of various forms of species distribution models (SDMs) which relate hominin presences in prehistoric landscapes to specific palaeoenvironmental variables and allow for the projection of prehistoric distributions that extend beyond the known locations of archaeological sites. Such models have proven useful in numerous modelling scenarios but have thus far been restricted to predicting just hominin presences or absences (or the probability of hominin presence), without finer-grained predictions concerning climate-related variation in behaviour that can be compared directly to the archaeological record. In this paper we develop a method, based on the random forest algorithm, that allows us to predict not only the palaeoenvironmental conditions conducive to hominin habitation but also the variation in assemblage composition that occurs under different environmental conditions.

Data

Archaeological data are drawn from an expanded version of the Eastern African Lithics Database developed over the past five years. This database collates all published eastern African MSA and LSA assemblages from 300ka to 20ka using a common typology consisting of the presence or absence of 17 technologies, providing an unparalleled opportunity to compare diverse assemblages on a consistent basis. Palaeoenvironmental data are drawn from simulations developed by Manica and colleagues which provide numerous bioclimatic variables at a 0.5-degree spatial resolution and a 1,000-year chronological resolution across the study area.

Methods

The initial stage of the process involves the production of a standard presence / absence SDM via a weighted average of multiple runs of a random forest (RF) algorithm with random selection of geographically and environmentally stratified pseudo-absences. The result of this modelling stage is then reduced to a single outer contour encompassing 95% of the total predicted inhabitable area; such contours are calculated and saved for each time-slice in the palaeoenvironmental raster stack encompassing the spatio-temporal extent of the archaeological database.

SDMs for individual technologies are also produced using RF. As the database consists of records of the presence or absence of 17 technologies across 194 assemblages, this stage of the process benefits from the use of true absences (i.e., the locations and dates of assemblages that do not contain a given technology). Weightings are applied to equalise the effects of presences and absences on the models, with a k-fold cross-validation procedure employed to test the validity of results against the archaeological data, and probabilities of the occurrence of each technology are then projected within the outer (95%) presence contour. This procedure is repeated for each time-slice and for each of the technologies in the archaeological database, producing probability of presence estimates for each of the 17 technologies at 1,000-year and 0.5-degree resolution throughout the spatio-temporal span of the database.

Probabilities of occurrence for each of the technologies in the database are then aggregated into toolkits, enabling examination of 1) predicted changes in toolkit composition through time for any given location and 2) predicted changes in toolkit composition across space for any given time-slice.

1. Allows us to compare temporal trajectories in different spatial locations via scaling statistics and is used to focus on the nature of the MSA to LSA transition. This transition occurs at different times in different regions of eastern Africa, but this method allows us to examine whether the transition is of a similar structure in each region. Results are validated by comparing estimated trajectories to those documented empirically at sites that have multiple levels spanning the MSA/LSA transition, including Panga ya Saidi, Mumba, Nasera, and Kisese II.
2. Allows us to examine spatial clustering via a modified Wombling procedure, delineating areas that are marked by consistent technological constellations and the locations of the boundaries between those areas. By examining multiple sequential time-slices, we examine how technological boundaries shift through time, corresponding to the geographical expansion and contraction of particular toolkits. This latter analysis provides a means of estimating the degree of spatial structure in material culture and will focus on reported differences between the material culture of the Lake Victoria, Rift Valley, and Coastal areas.

Results and Discussion

Results show a series of varying technological constellations both through time and across space. Boundaries between spatial technological clusters correspond broadly to different environmental regimes, but with a number of key sites occurring close to ecotones and therefore equipped with toolkits for exploiting multiple resource bases. Chronological analyses show key similarities and differences between the MSA to LSA transitions in different regions and at different times and reiterate the major technological distinctions between MSA and LSA technologies identified in previous analyses.

These results are discussed in relation to the view that Homo sapiens populations within Africa were highly structured, with genetic and cultural exchange marked by periodic fragmentation and coalescence of subpopulations, concordant with shifting ecological boundaries. We discuss the distinct archaeological signatures of indigenous innovation as opposed to cultural transmission between regions and highlight the relationship between ecotones and cultural boundaries. We also assess the nature of the MSA to LSA transition and discuss the extent to which different regional records of the transition, despite having different chronological onsets and durations, can be regarded as homologous in terms of their technological trajectories. The focus on projections of individual technologies both across space and through time allows us to build composite SDMs that reveal not only the regions / areas in which prehistoric Homo sapiens populations are likely to have lived, but also the variation in the toolkits that they are likely to have employed.