Embedded Files
For a master's course, I developed a new bus optimization index to measure whether buses in any given area are both frequent enough to attract those who drive, and dispersed enough to serve those who don't. Since these are in a zero-sum relationship, it's important to quantify the tradeoff so planners can optimize routes. Which areas have too little coverage, or too much frequency? The answers have serious implications for everything from simple efficiency to social equity and emissions reductions.
Relationship of GTFS tables and columns
Israel is a good case study because it is small enough for the Ministry of Transportation to regulate every bus route in the country. They provide schedule data in the standard GTFS format. These are tables for routes, trips and stops that can be joined using shared columns to calculate where and how often service runs.
I excluded Shabbat and holiday services, which exist in a few places due to a political status quo beyond the scope of ordinary planning decisions. To analyze service distribution within cities, I also excluded intercity routes that would otherwise bias the areas around central stations, even though in practice they serve the entire city.Â
In GIS I drew a 400-m buffer around each route, overlaying their scheduled weekly trip counts on census statistical areas. These are the lowest level for which population, car ownership and other socioeconomic data are available from the Central Bureau of Statistics. Each represents a few blocks or a neighbourhood. However, the census lacks information on job locations which also drive travel demand. Those I obtained again from traffic analysis zones published by the Ministry of Transportation, reallocating them proportionally to the smaller statistical areas.
400-m buffer around Egged route 34 in Jerusalem
The index is a combination of trips "per capita" (including jobs) and trips per square kilometre (within at least one route buffer). Both terms are normalized by the averages for the city, country or whatever other scope of analysis is chosen. Then they are weighted by car ownership, so where ownership is higher the index is based more on trips per capita (relative frequency), and where ownership is lower the index is based more on trips per square kilometre (relative coverage).
The final step is a log2 transformation and residualization against the distance from the centre of each city. These steps help correct for autocorrelation; that is, the unavoidable extent service in one area is influenced by service in adjacent areas.
The results for Tel Aviv, Jerusalem and Haifa are mapped below. Green indicates overservice, brown underservice, and white optimal service.
Anyone familiar with these cities can see the patterns immediately. In Tel Aviv the wealth divide is north-south and in Jerusalem it's west-east, but in both the rich have more service than they need. Only in Haifa is the pattern inverted thanks to the Metronit BRT service in the lower city.
In the chart below, each bar is a statistical area. Those above the midline are overserved and those below are underserved relative to their city. Red indicates areas with a majority of car owners (frequency need) and blue, non-owners (coverage need). Bar widths are proportional to population. 10 is the highest socioeconomic cluster and 1 is the lowest.
Google Sites
Report abuse