Fig. 3: Rationality verification of heritage anchors and the study-area mask.
From: Machine-learning-based reconstruction of Ming-dynasty defensive corridors in Yuxian
a Reliability classes and confidence-weighted kernel density estimation (KDE) of study-area anchors (N = 65). b Delineation of the candidate study area: the 1369–1371 spatial increment of Datong Prefecture boundaries derived from CHGIS is extracted as a proxy mask, and its macro-morphological consistency with modern county boundaries is quantified using the intersection-over-union (IoU). c Local close-up showing the spatial relationship between the proxy boundary and the Great Wall line, major rivers, and mountain passes. d KDE of model positive samples (N = 278). Note: The THAC Datong Prefecture boundaries for 1644 and 1393 are geometrically identical; therefore, (b) displays boundary phases up to 1393 to avoid redundancy.
