Fig. 4
From: Quantifying the non-isomorphism of global urban road networks using GNNs and graph kernels

Traditional URN Analysis Methods. (a) Traditional Metrics of Specific URNs. We analyzed 9 traditional metrics and their distributions for 6 specific URNs (as shown in Fig. 1c). In URNs with low regularity, the road network around Burj Khalifa in Dubai exhibits a high proportion of \(circuity \_ avg\), \(degree=3\), and \(degree=5\), while the road network around the Arc de Triomphe in Paris shows a higher proportion of \(degree \le 2\) and \(degree=5\). In high-regularity URNs, the road network in Manhattan, New York, has the highest proportion of \(degree=4\), and the road network around the White House in Washington, D.C., has the highest average degree \(k \_ degree\). (b) Bearing Distribution of 6 Specific URNs. As shown in Fig. 1c, the bearings of URNs around Manhattan, Chicago Loop, and the White House are highly regular, while the bearings of URNs around the Arc de Triomphe, Lujiazui, and Burj Khalifa show varying degrees of diversity. (c) Cosine Similarity Analysis. (d) Spearman Correlation Analysis. In sections c and d, we obtained 9 traditional metrics for the top 30 cities worldwide using \(osmnx.basic \_ stats()\) (same metrics as in panel a). Each city was encoded as a nine-dimensional vector, forming 30 nine-dimensional vectors. Using this data, we performed cosine similarity and Spearman correlation analyses to compare with the graph non-isomorphism achieved through the EdgeCNN.