Learning From Good Maps: A Dataset and Analysis of Layout Patterns

ABSTRACT The layout of thematic maps is fundamental to cartographic design, serving to effectively utilize map space, construct visual balance, and guide the user's visual attention toward key thematic narratives. Despite the long tradition of cartographic design principles, there remains a need for large‐scale, data‐driven studies that explicitly extract and analyze interpretable layout patterns for cartographic evaluation. In this study, we construct a large, high‐quality dataset of 1625 exemplary maps collected from multiple authoritative and creative sources. Then we develop an automated pipeline for detecting cartographic elements and extracting shape of map area using a fine‐tuned YOLOv8‐OBB model and the Segment Anything Model (SAM). Based on these detected elements, we establish a set of quantitative metrics for evaluating the design effectiveness of map layouts, focusing on hierarchy, balance, compactness, and alignment. Statistical analysis, including correlation tests and Scheirer‐Ray‐Hare variance analysis, reveals significant relationships between page layout, map area shape, and layout performance. The results reveal that many widely accepted design conventions, such as the central placement of the map area, a strong tendency toward visually balanced layouts and the typical positioning of titles and legends, are strongly supported by empirical evidence. However, the analysis also exposes notable gaps between conventional practice. For example, although the north arrow is commonly regarded as a standard cartographic element, it appears in only a very small fraction of exemplary maps (1.85%). In addition, the findings uncover substantial flexibility in professional layout strategies, particularly in alignment, where moderate rather than strict geometric alignment predominates. Significant interaction effects further indicate that optimal layout strategies are strongly shape‐dependent, highlighting that cartographic design cannot be reduced to a single universal template. Finally, we also demonstrate how the dataset and framework can support future layout exploration, evaluation, and recommendation.