Employees of St. Petersburg University, who are engaged in historical geography, have created a neural network. Based on the features of the landscape of known archaeological sites, it can detect new places of worship that were erected in the South-Eastern Altai in ancient times.

Photo: St. Petersburg State University website

The southeast of Altai is promising for historical and geographical research. Scientists have managed to find here many places of worship from different eras – from the Eneolithic to ethnographic time. This made it possible to study the settlement and migration of people. At the same time, the monuments are distributed over a vast territory of different landscapes – these are mountains and river valleys. There is also a shortage of written sources that describe the life of people in this territory at different times. The results of the study were presented at the international scientific-practical conference “Fifth Landscape and Ecological Readings dedicated to G. E. Grishankov. Nature and Society: Integration Processes”. About it informs university website.

To solve the problem of studying monuments, geographers from St. Petersburg University were able to train a neural network to identify landscape features of known locations of traces of ancient cultures. To create a neural network model, they used various landscape features – absolute height, position relative to watercourses, slope, exposure, sunlight intensity, mountain visibility, and others. The accuracy of object detection by this neural network reaches 80%. It can be improved by expanding knowledge of the studied objects and using new landscape criteria in the analysis.

The settlement of people in ancient times was highly dependent on natural conditions. For example, the tribes of the South-Eastern Altai settled in the river valleys, where rivers of the southeastern and southwestern directions merge, as well as in the river valleys in the intermountain Kurai basin and on the periphery of the Chui basin. Scientists thoroughly studied the local landscape of hundreds of objects discovered there – and taught the neural network to “think” in terms of the environment, as people supposedly did in antiquity. This quite accurately allows us to assume the location of archaeological objects. The geographers carried out the analytical work with the support of a grant from the Russian Foundation for Basic Research.

As a result, the neural network found traces of mounds that have not yet been explored, where human settlements are possible. Such predictions were confirmed by satellite images. In addition, thanks to the analysis carried out by the neural network, one can better understand the principles of the settlement of ancient peoples. So the data will be useful not only to find monuments, but also to interpret them. Scientists also intend to use data from the neural network to conduct field work on scientific expeditions.