Satellite-based detection of unique sociocultural communities: the case of Mennonite colonies in Bolivia

Detecting socio-economic characteristics of human populations from satellite imagery remains an emerging area in remote sensing. Prior work has linked satellite data to general socio-economic indicators such as income and population density, but the identification of culturally or religiously distinct communities based solely on spatial land-use configurations has not yet been demonstrated. The purpose of this study is to evaluate whether satellite imagery can be used to detect religious characteristics of human populations through observable land-use patterns. Focusing on Mennonite colonies—known for their distinctive, highly regular settlement structures—we chose a large study area with a complex agricultural history and wide-ranging patterns. Across the 370,000 km2 Santa Cruz Department of Bolivia, we employ 5-m resolution imagery and a ResNet50-based convolutional neural network. We classify 5541 agricultural cells across the entire department as either containing or not containing Mennonite settlements. The model achieves 95% test accuracy (F1 = 0.94), supporting the conclusion that deep learning applied to satellite imagery can effectively identify religious characteristics expressed through spatial organization on the landscape.