Abstract

The use of seaweed as fuel has been mentioned in ethnographic and historical sources of different coastal regions. Nevertheless, the archaeological record of seaweed burning is still limited to contexts where preservation is exceptional and macroscopic discrimination of charred remains is possible. Here, we evaluate the effectiveness of Raman spectroscopy in discriminating seaweed vs. plant/wood char. Our dataset (N = 92) consists of modern and archaeological seaweed and plant/wood charred remains, including specimens of unknown origin from the Atacama Desert coast, Northern Chile. The charred samples were processed to obtain 13 parameters which were then fed into five supervised machine learning models. The models, built on samples of known origin (seaweed and plant/wood), performed remarkably well in terms of accuracy, kappa, sensitivity, and specificity. The models were used for final predictions on 10 non-identified archaeological charcoals. Our results suggest that Raman spectroscopy combined with machine learning techniques is a robust methodology for discriminating seaweed and plant/wood charred remains in the archaeological record. The predictions on unknown samples confirm that seaweed was used as fuel in a specific funerary ritual in the southern Atacama Desert coast around 5000 cal BP. Furthermore, charred specimens of Lessonia spp. recovered from combustion features in other northern Chile coastal settlements, suggest that seaweed pyrotechnology developed by Atacama Desert coast people is likely a long-term process. As for coastal archaeology, this work encourages new research on seaweed as an alternative/main fuel in coastal deserts and evaluates possible bias for chronologies from coastal archaeological settlements around the globe.