Account for facility type and agency when creating depot deadhead trips

[dan-mccabe]

infer_depot_trip_endpoints() in depot_deadhead.py currently matches the start and end location of each point to the nearest point in the Transit_Depot shapefile. However, there are a couple of issues with the existing approach:

• The shapefile includes facility types other than depots, e.g., transit agency headquarters. We should only be using actual depot locations.
• The existing code doesn't filter by agency, so we might be assigning buses to a depot that belongs to another agency.

@YiZoeHe please update the code to filter based on agency and facility type. It looks like this dataset came from the NTD at https://data.transportation.gov/stories/s/gd62-jzra, so you can use the data description there to determine which facility types should be treated as depots, and how to connect the NTD IDs reported in the shapefile with agency names/IDs from GTFS.

[YiZoeHe]

Hi @dan-mccabe , the method for matching bus depot to a block id was adopted from Zhaocai's code. I checked with him, he mentioned that:

• Some transit agencies didn't provide depot information to the NTD. The approach of using depot and headquarters locations to infer transit depot locations was adopted from another NREL VTO transit bus project conducted by Bo Liu.
• The agency names in the NTD data do not match those in GTFS data. He initially attempted to filter by agency, but that method was unsuccessful. The current approach of identifying the nearest depot might be the most efficient and practical solution.

[dan-mccabe]

Thanks for the update @YiZoeHe. Tagging @Willhyliu as well in case he has additional thoughts. After our conversation today and looking at the facility data more, here are some more thoughts:

• I think we have the right general approach of filtering down the facility data to try to infer depot locations for each agency. It would be good to incorporate the source code for that process into this repo as discussed.
• It looks like there are some issues with the current facility processing approach. First, there are many duplicate locations in the FTA_Depot shapefile that I suspect weren't duplicates in the original data. Second, there seem to be issues in how NTD IDs are handled, which introduces some additional facilities. See the screenshots attached: we have both a maintenance facility for agency "00003" and a headquarters for agency "3" next to each other. It seems that this is the same agency and it is appearing twice because of the difference in string formatting. The headquarters should have been omitted because other depot-like facilities are present. I have seen this problem with multiple agencies.

• Filtering by agency name/ID is definitely a tricky issue and my previous suggestion didn't fully recognize that. It would be ideal to account for this, but there are also lots of weird edge cases it will be hard to handle. For example, Sound Transit bus routes are actually operated by other local agencies (e.g. King County Metro, Community Transit). So a bus serving a Sound Transit route will actually be housed at another agency's depot, and filtering down to Sound Transit facilities would not be correct.
• In light of the above point, I would propose that we leave things as-is for now, but add a check to make sure that the depot is not too far away. Then we should provide a clear warning or error message if the distance between a trip and the depot is suspiciously large. The current code doesn't appear to have a check for this, so we could theoretically be assigning a vehicle to a depot for another agency that is 100 miles away. I don't think this is likely to be an issue, but it would be good to have a safeguard.