The complementarity of luminescence dating methods illustrated on the Mousterian sequence of the Roc de Marsal: A series of reindeer-dominated, Quina Mousterian layers dated to MIS 3

Located in southwest France, Roc de Marsal is a cave with a rich Mousterian stratigraphic sequence. The lower part of the sequence (Layers 9–5) are characterized by assemblages dominated by Levallois lithic technology associated with composite faunal spectra (including red deer, roe deer and reindeer) that shows a gradual increase in the frequency of reindeer. The top of the sequence (Layers 4–2) are characterised instead by Quina lithic technology (both in terms of technology and typology) with the faunal remains dominated by reindeer. Roc de Marsal thus provides a very interesting case study to place behavioural changes in a context of changing climates and environments in western Europe during the late Pleistocene. To link the occupations at Roc de Marsal with global and regional climatic conditions known independently, a robust chronology is needed. With this aim in mind, we applied three luminescence dating methods (TL, OSL and IRSL) on different minerals (flint, quartz and K-feldspar extracts). Here the results of two of these methods are presented in detail (TL and OSL) and compared with preliminary IRSL data. At Roc de Marsal, a comparison of methods was necessary to overcome a complex sedimentary history, with very heterogeneous dose rate distributions, both at the beta (mm) and gamma (dm) dose rate scales. The results indicate that the lower Levallois layers are dated to ∼65–70 ka, while overlying Quina layers are dated to ∼49 ka. These ages for the lower layers fit well with some models that place mixed faunal assemblages in the initial MIS 4; however, while the Quina ages overlap with several other Quina assemblages from the region, they place the reindeer dominated fauna well after the peak cold of MIS 4 and suggest a more extended and complex period of contemporaneous lithic techno-complexes than posited by some current models.