Approaches for comparing the structural characteristics of different representations of transportation systems
Collection and monitoring of locational information has considerably increased over the last decade as the value of location-based data has been realized by planning organizations and agencies. Global Positioning Systems (GPS) have played a major role in the acquisition and widespread use of this information. However, there are known problems with GPS that can lead to misrepresentation or conflicting representations of georeferenced phenomena. As a result, it is unlikely that the location of a feature reported by a GPS will perfectly align with its location as reported by a GPS at another time. Also, given that other geographically referenced datasets (i.e., roads, buildings, land use/ land cover) vary in their positional accuracy, the likelihood of a perfect correspondence among layers of geographic information is even lower. As the use of geographic information becomes more widespread, it is common for multiple representations of the same set of transportation features to be available from different sources. For example, a road dataset may be maintained at the federal level while state and municipal agencies may maintain their own road databases. Road datasets may also be created as part of specific infrastructure research projects, created by commercial data vendors, or created as part of open-source data collection campaigns (i.e., OpenStreetMap). However, it is often the case that while the features rendered by different sources may be the same, the geographic placement and representation of the features can be different. For example, in some cases, the lines representing a road segment in two datasets may be very close. In other cases, the segments may be very far from one another or even absent in one or both datasets. Even when two segments may appear geometrically similar, it cannot be implied that they can used in the same way. That is, if their topology is different, their role within the network itself can also differ. For example, an intersection in one network may not permit the same types of movement in another given difference in how the features are connected. To this end, this research seeks to investigate alternative approaches for assessing the correspondence among different representations of a transportation system. Three different methodologies for accomplishing this are detailed and implemented for comparison. The first method is a spatial quadrant analysis in which different road systems are overlaid with a grid of systematically sized/spaced polygons (i.e., 500m x 500m, 1,000m x 1,000m, and 1,500m x 1,500m) in a GIS. The length of roads segments within each polygon can then be summarized, facilitating comparison among the different road representations. The second method compares the differences among different road representations by attempting to match paths in one road dataset to paths in another road dataset. This new methodology moves beyond the traditional matching of individual components (i.e., arcs/nodes) and better permits assessment of differences in network topology. The third approach involves matching GPS data representing the movement of fleet vehicles that regularly traverse a road system to different representation of the transportation system. This new approach permits comparison of alternative representations of networks with a different data source, one which can vary based upon the spatial and temporal characteristics of the GPS and vehicle movements. The three methods of comparison are applied to assess spatial correspondence between three different representations of an actual transportation system (referenced here as Net A, Net B, and Net C. The results of the quadrant analysis show that there are extreme changes in absolute aggregate length among the three road datasets. These extreme differences highlight the extent to which networks prepared by different agencies can vary geometrically. Network-path based method results showed that the average distance between path vertices in Net A and assigned arcs in Net B grouped into two distance groups (<=10m, and > 10m). 97.5% of the assigned arcs in Net B are within 10 meters of Net A path vertices, and only 2.5% are beyond this distance. Similar observation occurred when comparing Net B to Net A (i.e., most of records are within 10m distance). Comparing the computed distance between Net A path vertices and the assigned arcs in Net C shows that 99.2% of the records are within 10m, and only 0.8% of the arcs have a greater distance than 10m. Comparing Net C to A share the same results with previous comparison (i.e., Net A to C). Third group of Nets comparison is comparing Net B path vertices to assigned arcs in Net C, and vice versa. For both comparisons approximately 98% of the assigned arcs are within 10m of the original test path vertices. The Networks correspondence to GPS records is evaluated and measured in terms of the average positional error which categorized. Net A average positional error ranges between a minumum of 0.013m and a maxium of 19.682m for the 5,341 covered GPS points. Approximately 94.93% of the total GPS points were on average of 10m from their covering arc, with the remaining 5.07% being on average more than 10m from the covering arc. The average positional error for Net B ranges from a minimum average distance of 0.012m to a maximum average of 19.996m for a total of 4,694 covered GPS records. Approximately 96.51% of the total GPS points were on average of 10m from their covering arc, with the remaining 3.49% being on average more than 10m from the covering arc. Network C average ranges between a minimum of 0.002763m and a maximum of 19.952m for the 4,472 covered GPS records. Approximately 96.31% of the total GPS points were on average 10m from their covering arc, with the remaining 3.69% being on average more than 10m from the covering arc. These results indicate a good correspondence between the GPS points and the assigned arcs in the three tested Nets regardless of the inherent differences among these networks. In this dissertation, GPS records obtained from service fleet vehicles are used to demonstrate this methodology.