Japanese early-modern woodblock prints depicting pastoral views of the countryside, so-called meisho-e (images of famous places), are often defined today as landscapes (fūkei). However, the notion of fūkei is a modern cultural translation, which obscures specificities of Japanese visual culture, and intricacies of early modern spatiality or a socially produced space. To uncover these characteristics and provide a more nuanced understanding of meisho-e prints, we have engaged in a macroanalytical study of relationships between places depicted in prints and actual topography, aided by computational technologies rooted in Natural Language Processing (NLP). In our prior work, we experimented with automated harvesting of geospatial data from image-content-related inscriptions on two hundred prints. In this follow-up work, we undertake a large-scale automated mapping of meisho and we study the geographical distribution of sites featured in these prints. We explore two different computational paths, one using deep learning and one based on digital gazetteers, and reflect on the challenges and benefits of the applied computational approaches. We improve the former, which was the state-of-the-art, using pre-training, and we show that the latter is beneficial in terms of mapping. Finally, by using automatically extracted place-name entities, we undertake an analysis of prints over space and time. We release our code and the dataset for public use: https://github.com/Connalia/ai-jan-art.