This paper proposes methods for classifying population distribution surfaces according to their topologic characteristics and for measuring overall topologic similarity between two population distribution surfaces. These methods are applied to a comparative study on population distribution surfaces of 20 Japanese cities that range in size from 200, 000 to 400, 000 people.
First, the paper shows a procedure for detecting the topologic structure of a surface. The procedure consists of two steps: i) find critical points (peaks, bottoms, and cols) on a population density surface; and ii) connect the critical points by paths of the steepest ascent or descent. The resulting network is called a “surface network.” The surface network is robust for a broad class of transformations and shows the structural characteristics of the surface. For surface networks, this paper proposes a similarity measure between two surfaces of population distributions, such that two surfaces are structurally similar if their surface networks are identical (isomorphic).
Second, this paper develops a procedure for eliminating trifle peaks and bottoms from a surface network to identify distinctive features of the surface, such as doughnut phenomena. The procedure consists of three steps: i) define the heights and depths of peaks and bottoms; ii) define trifle peaks and bottoms as those for which heights and depths are less than a predetermined threshold; and iii) eliminate the trifle peaks and bottoms in ascending order of their heights and depths.
Third, this paper proposes a method for classifying population density surfaces from which trifle peaks and bottoms are eliminated. Two surfaces are classified into the same category if and only if the surface networks obtained at a given threshold are identical.
Fourth, this paper proposes an overall topologic similarity index between two population distribution surfaces. Although trifle peaks and bottoms are eliminated by the above procedure, there still remains arbitrariness in the choice of the threshold. Therefore this paper regards the threshold as a parameter and proposes a similarity index. The more identical two surface networks are at different thresholds, the higher value the index takes.
Last, the proposed methods are applied to the population density surfaces of 20 middle-sized cities in Japan. From the classification result, it is found that most of the 20 cities have one or two distinct city centers and that cities with more than three centers are exceptional. This paper proposes adistance measure between two population density surfaces that is inversely proportion to the topologic overall similarity, and a distance matrix for the 20 cities is calculated. Multi-dimensional scaling is applied to the matrix so that the interrelationship between the 20 cities can be visually grasped on a plane. From the results, it is revealed that the horizontal axis represents the numbers of the distinctive peaks on the surface networks and the vertical axis represents the complexities of the topologic structures of the surface networks. Furthermore, it is shown that more than half of the cities have similar population distribution structures and that they commonly have one distinct city center.
View full abstract