The purpose of this paper is to study spatial variation of the geometric structure of regional road networks. Assumptions to the model building of the structural variation are as follows. (1) The structure of regional road networks is composed of nodes, links and connectivities. It's form is hierarchical. (2) The spatial pattern of settlements as nodes is given. It is four ordered K7 system on Christaller's space. (3) The major factors influ-encing the structural variation are the size and the technological development level of areas.
The outlines of this heuristic model are that the determinant factor of the structural vari-ation of upper order networks (connected between a upper center and medium centers) is the technological scale, those of medium order (connected between a medium center and lower centers) are the size and the technological scales, and that of lower order (con-nected between a lower center and lowest centers) is only the size. However, the structural variation of lowest order (connected between a lowest center and dispersed settlements) will not be determined by those areal factors.
For the testing of this model, some ordered areas in Japan were selected. And the relationship between road network structure and geographic characteristics of each order areas was investigated by simple correlation analysis. In this model, the road network structure was measured by the number of settlements, the average road length and the β-index. The geographic characteristics of the selected areas were represented by the size and the pereapita annual income.
As a result, in upper order areas, there are significant correlations between the number of upper and medium centers and the percapita annual income (correlation coefficient, 0.8351) and between the β-index and the percapita annual income (correlation coeffi-cient, 0.9106). On the contrary, the average road length is not significantly associated with areal characteristics and is roughly constant. In medium order areas, there are significant correlations between the average road length and the size (correlation coeffi-cient for areas of below 1, 200km², 0.7760), and between the β-index and the percapital annual income (correlation coefficient, 0.8758). The average road length for areas of more than 1, 200km² and the number of medium and lower centers are roughly constant. Also, in lower order areas, there is a significant correlation between the average road length and the size (correlation coefficient for areas of below 120km², 0.7164). The number of lower and lowest centers, the average road length for areas of more than 120km² and the β-index are constant. However, the network structure in lowest order areas has little influence upon these areal characteristics. All elements composing that structure are constant. In addition, the spatial pattern of settlements in the region of Japan is approximately K7 system; the number of settlements is 5 in an upper order area, 8 in a medium order, 5 in a lower order, and 9 in a lowest order.
Thus, the validty of above hypothesized relationshipes was verified. But, this model can not fit to the structural variation of road networks in the marginal regions of Japan due to the spatial pre-patterning of settlements. In spite of the limitation of the study, the writer believes that this model is a useful method for forecasting the structure of regional road networks.