城市点与路网集成的小比例尺制图综合

The next stage is to generalize road network. Since we have a set of generalized city points they can be used in generalization process. Here we applied a following strategy:

1. Build the graph using the source road data. Introduce 3 levels of hierarchy in the network to make the highways preferable for routing.

2. Derive the routes between every pair of generalized points. In fact the number of pairs can be limited to a few without altering the result.

3. Select edges, which are included in one or more routes.

4. Reconstruct road network using selected edges.

5. Simplify roads geometrically.

This algorithm was also implemented using ArcGIS Model Builder and ArcGIS Network Analyst Extension. For geometric simplification of roads we used bend simplify method (Wang and Muller 1998). Example of application of algorithm is presented on Figure 2.

Figure 2. Result of road network generalization using selected cities. Some points are located on isolated parts of the network and some are not reached by the roads in source dataset.

Our approach was apllied to city points and road generalization from 1:1 000 000 scale to 1:10 000 000 for Far East region of Russia which is well known for its non-uniform settlement distribution (Figure 3). The number of city points was reduced from 4703 to 184 and the number of road segments was reduced from 8414 to 455. For comparison of the results we prepared a map derived by attribute selection, which is the common practice in modern web services (Figure 4).

Our results show high degree of reconciliation between city and road layers that is close to manual generalization. However there are many ways in which algorithm can be improved. The first is to extend the variety of factors that influence the importance of objects. The next is to embed road network pattern recognition and detection of nodal points that comprise the distinctive features of this pattern. Another suggestion is to implement collaborative algorithm that generalizes points and roads together instead of generalizing them in two steps. In current version of algorithm the point generalization step is fully independent from road generalization. We plan to improve the methodology and estimate its applicability in multiscale basemapping.