Terrain Analysis and Stochastic modelling for Archaeological site prediction and landscape reconstruction in the Lake Manyara area, Northern Tanzania

Abstract

In this study we focus on paleontological sites in the area of Lake Manyara and the Makuyuni River Basin, Northern Tanzania. This region is known for Middle Pleistocene fossil finds and artefacts. To analyze the spatial distribution of potential paleontological find locations we applied two different methodologies based on statistical mechanics and on boosted regression trees. The first one is able to handle presence-only datasets such as the locations proper. The second approach was used to study the variable importance and to derive information on the related geo-processes for classified paleontological sites. The locations and their spatial distribution were retrieved from literature and collected by own field work over the last years. For the modeling we utilized environmental information such as spatially continuous layers of topography (30 m SRTM DEM), derivatives of topography, vegetation information as well as ASTER multispectral data as predictor variables. The results revealpotential areas where further fossil sites may be located. Moreover, we assessed the processes that are related to sites with specific archaeological evidences. Therefore, the sites were grouped in three categories: i) artefacts sites, ii) fossil sites and iii) mixed sites. We applied boosted regression trees to analyse the processes related to the classified sites. The methodology considers not only site specific characteristics but implicitly also the related pedogenetic and morphogenetic processes. We were able to differentiate between artefact and fossil sites. Moreover, our analyses indicate an influence of transportation processes on the artefacts, whereas deposition of fossils does not seem to involve large scale transportation. Finally, we show that also the landscape can be reconstructed such as the former lake margin.