GTFS-RT - Departure Times Accuracy
First of all, thanks the maintainers for all the work in building these powerful APIs.
We have been looking to transition our project from `transport.integration.sl.se` APIs to the GTFS datasets. It mostly works great (if not better). The only discrepancy is in the Realtime departure times:
- SL APIs somehow track when the bus actually left the station, returning the departure marked as "Nu" until then;
- The GTFS RT APIs instead remove a departure as soon as the departure time hits zero, even if the bus is still in the process of approaching and then letting people in.
In other works, GTFS RT removes imminent departures too soon, sometimes minutes before the bus even reaches the station.
We wonder if SL APIs use vehicle GPS position to be more precise in their result, and if GTFS RT could in the future mimic the SL RT departures more closely for better accuracy.
Curious if other encountered the same and if we can improve this.
Thanks
We have been looking to transition our project from `transport.integration.sl.se` APIs to the GTFS datasets. It mostly works great (if not better). The only discrepancy is in the Realtime departure times:
- SL APIs somehow track when the bus actually left the station, returning the departure marked as "Nu" until then;
- The GTFS RT APIs instead remove a departure as soon as the departure time hits zero, even if the bus is still in the process of approaching and then letting people in.
In other works, GTFS RT removes imminent departures too soon, sometimes minutes before the bus even reaches the station.
We wonder if SL APIs use vehicle GPS position to be more precise in their result, and if GTFS RT could in the future mimic the SL RT departures more closely for better accuracy.
Curious if other encountered the same and if we can improve this.
Thanks
Följ inlägget
1
följare
The GTFS-RT APIs should return all departures up till 10 minutes in the past. Once a departure/arrival has been observed, uncertainty for that departure/arrival is set to "0".
For example the following stop time update defines an observed arrival and observed departure where the bus was 6 seconds eary and 38 seconds late compared the scheduled date.
stop_time_update {
stop_sequence: 24
arrival {
delay: -6
time: 1754989554
uncertainty: 0
}
departure {
delay: 38
time: 1754989598
uncertainty: 0
}
stop_id: "9022001000177001"
}
Is this compared to a bus which you observed outdoors, compared to the GPS positions on Trafiklab, or compared to the SL API? If the latter is the case, is it possible the SL API is actually a bit slower and reports the departure after it has taken place?
Regards,
Bert
Sorry for the late reply.
I have been conducting more tests, checking both GTFS RT and SL APIs, while pairing observations in the field. What I could conclude is:
- GTFS RT removes a departure as soon as the bus arrives at the station, or even a few seconds _prior_ (probably following the `expected` time of arrival);
- SL APIs maintain the departure even a few seconds _after_ the bus has left the station (keeping it marked as "Nu").
I believe SL APIs are using the actual GPS position to determine if a bus is "AT STOP" or "LEFT THE STATION", which is more accurate but might be updated with a little delay.
My personal conclusion is that both approaches provide very usable data, despite the slightly different nuance in how RT times are provided.
I thought I'd follow up with my findings, just in case it might be of interest.
Thanks!