Research Request - Spatial accuracy metric

[tiffanychu90]

Complete the below when receiving a research request, and continue to add to this issue as you receive additional details and produce deliverables. Be sure to also add the appropriate project-level label to this issue (eg gtfs-rt, DLA).

Research Question

Single sentence description: Implement a version of Prof Greg Newmark's spatial accuracy metric, which looks at how many vehicle positions were within certain distance of the shape's geometry. Metric is % of vp produced with x meters of shape.

Detailed description:
Let's keep our 35 meters buffer. The Newmark paper lists 14 m as roughly the 90th percentile distance for the sample of 5 operators. From his other papers, 12 m is roughly enough for local routes, but 50 m is needed for routes traveling on freeways. We'll stick with the 35 m we use for our segment speeds.

• Epic - GTFS Schedule to RT Metrics #824

How will this research be used?

A version of this metric should probably make it to fct_observed_trips. Right now, we need to join the trip's associated shape geometry and buffer it, then do a count.

Metrics

• for a trip, count the number of raw vp
• for a trip, count the number of vp that intersect with a buffered shape
• calculate percent of "spatially accurate" vp

Data sources

• Cal-ITP data sources:
• GCS bucket: rt_vs_schedule
• vp_{analysis_date} -- raw vp
• scheduled shapes

Deliverables:

Script. Let's use the dask cluster if we can, there aren't many functions, if at all, that rely on segment_speed_utils or shared_utils, and it's big because we're doing it on the entire raw vp dataset.