Get OSM routes that match shapes, based on geometrical match

Usage

osm_shapes_match_routes(
  gtfs,
  q,
  geometry = TRUE,
  gtfs_match = "route_short_name",
  osm_match = "ref",
  gtfs_osm_match_exact = TRUE,
  log_file = NA,
  osm_file = NULL,
  num_cores = 1,
  osm_stop_order_relaxed = FALSE,
  osm_route_type = "bus",
  metric_crs = 3857
)

Arguments

gtfs: tidygtfs. GTFS feed.
q: osmdata::opq. Overpass query for transit network
geometry: Boolean (Default TRUE). If TRUE, returns sf object with geometry, otherwise, a simple data.frame.
gtfs_match: String (Default route_short_name). routes.txt attribute that identifies routes. Accepted values: route_id, route_short_name, route_long_name.
osm_match: String (Default ref). OSM attribute that identifies routes by matching with gtfs_match. Accepted values: ref, name, gtfs:route_id.
gtfs_osm_match_exact: Boolean (Default TRUE). If TRUE, gtfs and route names are matched strictly. Otherwise, partial string match is considered (all words in gtfs_match must be in osm_match, ignoring case).
log_file: String (Optional). If provided, will log warnings to this file, in addition to the console.
osm_file: character (Optional). Location of OSM extract file with osm.pbf format. Refer to osmextract::oe_download() for more details. If not provided OSM Overpass API is called through osmdata::osmdata_sf().
num_cores: Integer (Default 1). Number of cores to use for parallel computation. Only supported on Unix-like systems (Linux, macOS).
osm_stop_order_relaxed: Boolean (Default FALSE). If TRUE, OSM routes with entry/exit stops not respecting the right order will still be matched (this may indicate OSM data integrity problems). If FALSE, these routes will be ignored.
osm_route_type: character (Default "bus"). OSM route type. Used to query OSM network (e.g., 'bus', 'train').
metric_crs: Integer or character (Default 3857). Projected CRS used to compute shapes and routes lengths and stop-to-stop distances.

Value

A data.frame (sf if geometry=TRUE) with the following columns:

route_id: The route_id attribute from routes.txt file.
shape_id: The shape_id attribute from shapes.txt file.
osm_id: The osm_id attribute from OSM route relation.
distance_diff: The difference, in meters, between GTFS shape and OSM route lengths.
points_diff: The sum of the difference, in meters, between GTFS shape and OSM route start and end points.
stops_diff: The difference between GTFS and OSM routes number of stops.
route_short_name: The route_short_name attribute from routes.txt file.
route_long_name: The route_long_name attribute from routes.txt file.
osm_ref: The ref attribute from OSM route relation.
osm_name: The name attribute from OSM route relation.
geometry: The geometrical data for the OSM route relation.

Details

For each route, matches its trips' shapes with OSM route relations.

The matching algorithm is formulated as follows: Let $R$ be a GTFS route identifier.

1. Filtering and Base Data Selection: Let $\mathcal{O}_R = \{O_1, \dots, O_m\}$ be the set of candidate OSM route relations matching the identifier $R$ (based on gtfs_match and osm_match). If $\mathcal{O}_R$ is empty, route $R$ is skipped. Unless osm_stop_order_relaxed = TRUE, any relation in $\mathcal{O}_R$ with entry/exit stops not in the correct order is discarded. We also retrieve the set of GTFS shapes associated with route $R$, denoted as $\mathcal{S}_R = \{S_1, \dots, S_n\}$.

2. Feature Extraction: For each GTFS shape $S_i \in \mathcal{S}_R$:

Extract the start and end coordinates of its trips' first and last stops: $\text{init}_{GTFS, i}$ and $\text{fin}_{GTFS, i}$.
Compute the shape's total length $L_{GTFS, i}$ and the number of stop times $N_{stops, i}$.

For each candidate OSM route relation $O_j \in \mathcal{O}_R$:

Extract the coordinates of the first and last stops/platforms: $\text{init}_{OSM, j}$ and $\text{fin}_{OSM, j}$.
Compute the relation's geometry length $L_{OSM, j}$ and the number of stop/platform nodes $N_{stops, j}$.

3. Closeness Metric Evaluation: For each GTFS shape $S_i$, we calculate the closeness metric $C(i, j)$ for all candidate OSM routes $O_j \in \mathcal{O}_R$: $$C(i, j) = d(\text{init}_{GTFS, i}, \text{init}_{OSM, j}) + d(\text{fin}_{GTFS, i}, \text{fin}_{OSM, j}) + |L_{GTFS, i} - L_{OSM, j}| + \frac{L_{GTFS, i}}{N_{stops, i}} \cdot |N_{stops, i} - N_{stops, j}|$$

where:

$d(\cdot)$ is the Euclidean distance.
The term $\frac{L_{GTFS, i}}{N_{stops, i}}$ represents the average distance between stops on the GTFS shape, serving as a scale factor for the difference in the number of stops.

Shape $S_i$ is associated with the OSM route $O_{j^*}$ that minimizes the closeness metric: $$j^* = \operatorname{argmin}_{j} C(i, j)$$

4. Conflict Resolution: If multiple GTFS shapes are associated with the same OSM route $O_j$, only the shape $S_i$ that minimizes the closeness metric is retained. The other conflicting shapes are ignored and a warning is triggered.

Be aware that the result might ignore some GTFS routes, in the following cases:

If there is no OSM route relation that matches the GTFS route identifier;
If, for a GTFS route, there is any OSM route relation that has entry/exit stops not respecting the right order (unless osm_stop_order_relaxed is set to TRUE);
If, for the same route, distinct shapes are associated to the same OSM route. In that case, only the shape that minimizes the closeness metric is retained.

If any of these errors occurs, warnings will be thrown at end of the method execution, and those GTFS route will be ignored in the results.

Nevertheless, provided there are enough OSM routes, all the GTFS shapes for each route will necessarily be associated with an OSM one. This might generate wrong results if the topology of routes on OSM does not match the GTFS shapes for that route. Refer to distance_diff, points_diff and stops_diff on the results table to validate the results and identify misassociations.

Examples

if (FALSE) { # \dontrun{
gtfs <- GTFShift::load_feed("gtfs.zip")

q <- opq("Lisbon") |>
  add_osm_feature(key = "route", value = c("bus")) |>
  add_osm_feature(key = "network", value = "Carris", key_exact = TRUE)

# To use OSM API:
shapes_match_routes <- GTFShift::osm_shapes_match_routes(gtfs, q)

# To use a local OSM file:
osm_file <- oe_download("https://download.geofabrik.de/europe/portugal-latest.osm.pbf")
shapes_match_routes <- GTFShift::osm_shapes_match_routes(gtfs, q, osm_file = osm_file)
} # }