A Data-Driven Method for Improving Historical Maps' Positional Accuracy

Positional accuracy improvement (PAI) of historical maps involves correcting their inherent geometric distortions, which often limit their usability in modern applications. Although resurveying an entire map provides the most accurate solution for PAI, it is costly, time-consuming, and often impossible. This study proposes a cost-effective, alternative data-driven method, using Generalized Additive Models (GAMs) to enhance positional accuracy, without requiring complete resurveying. GAMs were utilized within a statistical learning framework to: (a) identify spatial patterns of geometric errors; (b) systematically correct these errors; and (c) evaluate their generalization ability via spatial cross-validation. The method was tested on a 1925, urban cadastral, map from Greece, which was georeferenced and vectorized for the first time. A dataset of 2,287 homologous points from modern land surveying was compiled. To simulate sparce data conditions, 10% of these points were used for training and 90% for testing. The results revealed significant spatial structures in positional errors, and cross-validation confirmed the model’s robustness. When applied to the test set, the fitted model doubled the spatial accuracy of the 1925 map, meeting the modern geometric standards set by the Greek National Cadastre. These findings demonstrate the method’s effectiveness and potential to enhance historical maps and other geospatial datasets that are facing similar issues across diverse research areas, such as urban planning and environmental history.