Extending Vancouver's Public Bike Share system: a GIS analysis
METHODS
This project used a multi-criteria evaluation (MCE) in ArcGIS 10.1 to attempt to find the best possible areas into which Vancouver’s PBS might be expanded if it is indeed successful in the launch area of downtown and between Arbutus Street, 12th Avenue and Main Street. After the factors were determined through a literature review (see previous section) and the appropriate data sets obtained, the GIS analysis could commence.
Isolating the study area
The first step in this analysis was to isolate the study area, meaning the portion of Vancouver not included within the currently proposed boundaries of the PBS. The dissemination areas (DAs; the smallest geographical unit at which census data is collected) falling within the current PBS boundaries were selected manually. Those that fell along the boundary were included in the study area if it seemed that the majority of the DA fell outside the Arbutus-12th-Main boundary, and were not included if it seemed the majority of the area of the DA fell inside of it. The selected DAs were then exported to a new layer, and the erase tool was then used to isolate the study area. Finally, each of the factors were clipped using this layer (or, in the case of the DEM, extract by mask was used).
Flowchart of MCE method
The DEM was first converted into a slope surface. The focal statistics tool was then used to find the mean slope value for 0.5km2 around each raster cell, and then to convert each cell into this value. This conversion dulls the influence of cells that have a slope that is unrepresentative of the surrounding area, thus making the slope surface more representative of what might actually influence potential cyclists.
The Vancouver bike paths layer was converted into a raster surface using the Euclidean distance tool. Each cell was assigned a value based on its straight-line distance from the nearest bike path.
The Vancouver greenways layer was also converted to a raster surface using the Euclidean distance tool.
The rapid transit station manipulation was a bit more complicated. Since cycling is most likely within moderate distances of a station (5 km was chosen in this analysis due to the conclusion in Winters et al. that people are more likely to cycle if their destination is within 5 km (2011)), less likely within 1 km, and least likely outside of the 5 km mark, 2 buffers were created around the stations: a 1 km buffer and a 5 km buffer. In order to transform these buffers into one layer each, the dissolve tool was used to merge all of the 1 km buffers into one polygon, all the 5 km buffers into another, and, finally, all the DAs of interest into a third. The dissolved buffers were then clipped to fit the study area. The symmetrical difference tool was used to erase the 5 km polygon from the entire interest area, and then also the 1 km polygon from the 5 km polygon. The merge tool was then used to create one layer from these three mutually exclusive sections. Finally, the layer was converted to a raster surface and the buffer distances reclassified so as to give the areas further than 5 km from a rapid transit station the lowest value, the areas within 1 km of a station a middle value, and the areas in between 1 km and 5 km the highest value.
The first step in finding areas of the city with an age breakdown conducive to cycling was to perform a table join of the 2011 Census data to the 2011 DA boundary layer. Next a new field was created, and using the field calculator it was populated with the total number of people between the ages of 15 and 44 per DA. In another new field, the population density of this new age category was calculated. Finally, the polygon to raster tool was used to create a surface showing the various population densities of the age group most likely to use a PBS.
Similar steps to those above were taken to create surfaces showing the proportion of the population comprised of immigrants and the proportion of the population that is 14 years of age and under.
This process created an inexact, jagged boundary along the currently proposed PBS border, as can be seen in the image to the right. This inexact boundary is a source of uncertainty in the analysis. However, because this analysis is working with census data that corresponds to particular DA boundaries, splitting the DAs that fall along the PBS border and performing the analysis that way would have produced inaccurate results.
Exact study area, illustrating inaccuracy of border between study area and area or Vancouver excluded from analysis
Data Manipulation
Before each of the factors could be normalized into a common scale so that a weighted sum might be performed, there was some manipulation needed to be done to each.
STUDY AREA
Normalization
Each of the seven factors was normalized (converted into a 0-1 scale, with 0 representing areas that are least likely to encourage cycling and 1 representing areas most likely to support a PBS) using the fuzzy membership tool. Each was normalized linearly, but the normalization for the slope, Euclidean distance, and immigrant layers was done inversely.
Weighting
Finally, the weighted sum tool was used to run the MCE. The MCE was run 3 times with the weights shown in the table below. The set of weights on the left corresponds to the research discussed in the introduction, though it is important to note that the ranking and weighting assigned to the variables is only an approximation of the researchers’ conclusions. The set of weights in the middle and on the left are simply for sensitivity analysis - to see if the MCE is sensitive to minor changes in the weightings – with one privileging the mobility factors and the other privileging the social factors.
|
FACTOR |
PROPER WEIGHTS |
ACCESSIBILITY FOCUS |
SOCIAL FOCUS |
|
Slope |
24 |
16 |
10 |
|
Population Density (15-44) |
20 |
4 |
24 |
|
Greenways |
18 |
24 |
6 |
|
Bikeways |
16 |
24 |
6 |
|
Transit Distance |
9 |
24 |
6 |
|
Children |
8 |
4 |
24 |
|
Immigrants |
5 |
4 |
24 |