Testing the representativity of Palaeolithic site distribution: The role of sampling bias in the european upper and Final Palaeolithic record

Archaeological sites are not distributed evenly throughout the landscape. For the Palaeolithic record, signals derived from the inhomogeneous spatial patterns are used to infer spatial decision-making processes or ecological preferences of our ancestors. However, to date it is still largely unclear how sampling biases affect the large-scale distribution of sites and whether the observable spatial patterns are actually representative of the distribution of humans in the palaeo-landscape. To answer this question, this study assesses the spatial distribution of 4200 Upper and Final Palaeolithic occupations from two different perspectives, i.e., past settlement choice and likelihood of discovery. On the one hand, site distribution is thus examined for settlement-relevant factors such as topography, geology and sedimentology. On the other, discovery-relevant biases, such as recent land cover and building activity are analysed. The comprehensive spatial and statistical assessments show that the actual distribution of sites seems to be most strongly influenced by sampling biases. The assessed environmental variables representing the settlement factors show a far lower statistical association to the distribution of sites. They do, however, still support several common archaeological assumptions. For all approaches using site distribution as input, such as predictive modelling, the results of this study suggests that the sampling bias must be addressed. To this end, we suggest including environmental variables addressing discovery-relevant factors to quantify the potential biases. For further studies on the sampling bias of Upper and Final Palaeolithic sites, we recommend building on this pilot study by adding more occupations to the dataset and more environmental variables to the settlement and discovery factors. Due to the current positive trend in openly available geo-datasets, we see great future potential in this. •New insights into the representativity of the European Palaeolithic record.•The modern land use is the strongest predictor for the presence/absence of sites.•Preferred palaeoenvironmental settings still differ between archaeological classes.•Suggestions on how to tackle the bias include weighting and omission of sites.•The input dataset of 4200 Palaeolithic occupations is made publicly available.