Maximum Entropy Modeling of the Iron Age Settlement Distributions in River Valleys of Turku Region, Southwest Finland

Abstract

Species distribution models (SDM) are predictive modeling tools widely used in analytical biology that have also found applications in archaeological research. They can be used to quickly produce predictive maps for a variety of use cases like conservation and to guide field surveys. Modern SDMs take advantage of advances in computing like machine learning and artificial intelligence to achieve better predictions. In this study Maximum Entropy, or MaxEnt, machine learning SDM algorithm was used to create predictive models of the Iron Age settlement around Turku region in Southwest Finland, focusing on Aurajoki, Savijoki, and Vähäjoki river valleys. MaxEnt is the most popular SDM algorithm, largely due to its ability to create predictions based on presence-only data and consistently good performance. Only open access -data was used, and the selection of variables was based on availability and previous studies. The results show that MaxEnt can create in some cases surprisingly accurate models based on archaeological information, but the results were limited by the quality of existing data. The most influential variable was distance to water, which was the majority contributor whenever present. Even without the variable, the predicted distributions followed the waterways closely due to the influence of other variables. It was concluded that to improve the accuracy of the results the quality of the data should be a major focus. The results should also be tested through field surveys. Additionally, attention should be based on the model conception.