Updated July 31, 2013 at 2:45pm:
In response to comments I have received, I have produced charts that show the average speed of operation on parts of the King route. This is at a very fine detail by contrast to the “link time” charts covering route segments that I published in earlier articles. The new material is added at the end of this article.
Original article from July 22, 2013:
The TTC’s Andy Byford has proposed that King Street be reserved for transit vehicles during the AM peak period as a means of improving service quality. In previous articles I have examined service reliability and congestion, as well as the history of transit priority on King Street.
The big issue whenever “congestion” comes up for discussion is that any tactics adopted to improve transit service need to address what is really happening on the street, not an abstract idea that somewhere, sometime, it might be a good idea to have some sort of transit priority. Previous analyses published here show the effects of congestion through charts of “link times” (the time taken by vehicles to travel of specific parts, or links, of a route), but these don’t pinpoint the exact locations or severity of delays.
This article introduces a new type of chart that is intended to make delay locations and times much more obvious as a starting point for discussions of where priority is needed.
Where Are Our Streetcars?
Regular readers of these analyses will be familiar with the time-distance charts of route operations. In June, I published an analysis of one day’s operation on 501 Queen and this includes a time-distance chart of the service. The underlying data for these charts starts out as GPS data from the TTC’s vehicle monitoring system (the same data that feed NextBus in real time). The GPS co-ordinates are mapped to a “flattened” version of the route as if it were a piece of string pulled out straight. In this scheme, one kilometre of travel is 100 units and so 1 unit is 10 metres.
Streetcars that are in motion cover more than 10 metres in the 20-second sampling interval, but if they are stationary, or moving slowly, they will show up at the same location many times. Places where many streetcars tend to spend a lot of time will have more data samples pointing to them than parts of the route where streetcars speed by.
The GPS data provide a snapshot of the system every 20 seconds, and counting the number of observations of a streetcar at each location will show where vehicles spend their time.
Here are the data for Wednesday, May 1, 2013.
Each file contains 19 pages, one for each hour-long block starting at 6:00am and ending at midnight.
The scale across the bottom is in the internal units to which GPS locations have been converted with 0 being just north of Broadview Station and 1270 being just north of Dundas West Station. Each 100-block marks off approximately 1km along the route.
Vertical shaded lines, set to the same scale as the x-axis, show the position of major cross streets.
The vertical scale is chosen so that the data at busy times fill the page.
If you step through the pages, you will see the congestion effects building up and receding through the day.
On the westbound charts, this becomes evident a few hours in. For example, for the period 8-9am, backlogs of cars appear east of University, and the congestion at Spadina (which will be much worse later in the day) start to show up too. The hour 9-10am shows even more congestion because parking is allowed, but service and traffic are still at peak levels. As the day wears on, the congestion east of Spadina worsens, but other locations show up too notably east of The Queensway where the left turn queue slows streetcar operation. Similar, but smaller, effects can be seen at Jameson from left turning traffic headed for the Gardiner Expressway.
By the PM peak, the hour 4-5pm shows the effect of long stop dwell times through downtown including the area east of the core where there is now considerable traffic from George Brown College. The core is particularly congested in 6-7pm, but congestion starts to fall off elsewhere. By 8pm, Spadina remains as the primary location for delay.
Eastbound patterns are, of course, different. Spadina is not as much of a delay location because there is no competition from 510 cars turning at Charlotte Loop. However, Bathurst shows up as a delay location because of construction blocking the curb lane one short block west of the intersection.
Both charts have no data after 11pm for a chunk of the route corresponding to a diversion via Spadina, Queen and Shaw for overhead work.
Other points that show up with high counts include
- Broadview & Dundas West Stations (termini, layover points)
- Both ways on Roncesvalles north of Queensway (busy stop plus crew changes)
- Queen & Broadview Eastbound (same)
Close examination of field conditions at individual stops would reveal more details about operations there. For example, Dundas & Broadview has a long east-west green time during peak periods and this can hold through service on the 504.
The point at issue here is that service delay, manifested as cars being more likely to appear in some locations than others, is not confined to the core area, nor is it as bad as at other times of the day. Conversely, there are some segments of the route where “delay” shows up as a narrow spike in the chart indicating a vehicle likely stopped at a traffic signal or carstop, but not otherwise slowed on the approach. Proposals for transit priority need to address times and locations where they will have an effect on transit service.
For comparative purposes, here are charts of other data.
The charts for Thursday, May 2, 2013 are similar, but not identical to those for the preceding day. Any evaluation of possible improvements must consider day-of-week as well as seasonal effects, not just one day’s data.
The “Week 2” charts combine data from Monday, May 6 through Friday, May 10, 2013 into one chart (note that the scale is adjusted so that the data will fit).
These charts use data from the four Saturdays in May, 2013. The pattern of delays is completely different on Saturday, although Spadina shows up as a particular nuisance.
Updated July 31, 2013
In response to comments I have received asking about graphs of average speed, I developed a variant on the “congestion” charts shown above based on speed rather than on vehicle location. This is done on a fine-grained basis using the same data that generated the first set of charts.
The raw GPS data from TTC vehicles gives the location of each vehicle every 20 seconds, and so there are 180 samples per hour. As explained elsewhere, for analysis purposes the GPS co-ordinates are “flattened” into a single dimension where the route is treated as if it were a straight line. In this co-ordinate system 1 unit is approximately 10 metres. From this, it follows that the change in position between observations can be converted to the speed of a vehicle.
- If a vehicle travels 10 units in one interval, that is 100 metres.
- There are 180 intervals per hour, and the vehicle will have moved at 18,000m or 18km/hr.
Each observation has a location, and if the speeds of all vehicles at that location (that is, within one unit of the flattened co-ordinates) are averaged, the result is the average speed of vehicles observed within that 10m section of the route for whatever period of time is sampled. In this analysis, I have broken the data down by hour as with the congestion charts.
In practice, looking at a single day’s data produces a “thin” collection of data for each hour. The reason is that with almost 1,300 locations on the King route (corresponding to a one-way trip of almost 13km), many locations do not have a car “in them” within one hour’s data. (There are only 180 observations per hour, and with about 40 cars at peak, this yields 7,200 discrete vehicle-location values, less outside of the AM peak. Half of these are not going in the direction of interest, and the remainder of the observations are clustered around stops and terminals.)
I have produced charts for the week of May 6-10, 2013 and for all Saturdays.
These charts complement the information shown in the congestion charts in that where there is congestion (high numbers of observed vehicles) there will be low speed. The difference, however, is that the congestion chart counts vehicles while the speed chart is independent of the number of vehicles observed and shows the general speed of operation over the route.
Stepping through one set of charts one can see the general drop in average speed through the middle of the day and pm peak with a rise again in the evening. The pattern on Saturday is not the same as on weekdays.
(As mentioned with the congestion charts, there were diversions around overhead work west of Shaw for some of the days included and this leads to fewer observations on that section of the route for the weekday chart.)
Notable on the Saturday chart is the low operating speed bothways through the Entertainment District. This effect is somewhat masked in the weekday chart because it does not occur early in the week.