Find reference reaches drawn on top of one another — check_reference_coincident

Two reaches that occupy the same ground are indistinguishable to anyone reading the fabric: one click returns both, and they appear to be one channel counted twice. The pipeline carries such geometry through faithfully, so the only place to catch it is the input.

The failure is not exact duplication, which is rare and easy to test for. It is near-coincidence: reaches running within a metre of each other for kilometres without sharing a vertex. A strict topological overlap test sees almost nothing there, while a mouse click cannot tell the difference. So this measures the fraction of each reach that lies within tolerance_m of a different reach.

Observed on the CONUS reference at a confluence in VPU 04: four reaches terminating at one point with the same bearing, sharing 2.3 to 5.5 km with each other, three of them from a preliminary source and the fourth a 522,054 km2 river drawn as a single straight 9.5 km segment. Coincident geometry also misleads aggregation, which reads spatial adjacency: those four tributaries were contracted into one flowpath carrying four outlets.

Report-only, and deliberately not wired into run_pipeline(). The geometry belongs to the reference and is not ours to change, and this asks a question about a reference release rather than about a VPU, so it runs once per input rather than once per build.

Usage

check_reference_coincident_geometry(
  reference_gpkg,
  layer = "flowpaths",
  id = "flowpath_id",
  vpuid = NA_character_,
  tolerance_m = 1,
  min_frac = 0.5,
  sample = 5000L,
  strict = FALSE
)

Arguments

reference_gpkg: Path to the reference GeoPackage.
layer: Flowline layer name.
id: Identifier column.
vpuid: Optional VPU to restrict to, when the layer carries vpuid.
tolerance_m: Distance within which two reaches count as coincident. 1 m is three orders of magnitude below the 30 m DEM cell, so it cannot catch channels that are merely close.
min_frac: Fraction of a reach that must lie within tolerance_m of another before it is reported.
sample: Number of reaches to test, or NULL for all. The default keeps the cost bounded on a national reference; the reported rate is an estimate and the message says so.
strict: If TRUE, warn when any are found.

Value

Invisibly, a data frame of id, frac, length_m for the reaches found, most coincident first.