Pipeline inputs

A build reads the inputs below. Everything else in the product is derived.

Input	Contents	Required
Reference fabric	GeoPackage with flowpaths and divides	yes
Hydrolocations	GeoPackage of points with hl_reference	yes, or a layer in the reference
Grids	D8 flow-direction and flow-accumulation rasters	yes
Channel table	RiverML parquet (hydraulics_path)	for routing attributes and reach slope
Lakes and lake mask	lake set and reach-to-lake parquet (lakes_path, lake_mask_path)	for the lakes layer

Reference fabric

Layer	Columns
flowpaths	flowpath_id, flowpath_toid, vpuid, lengthkm, hydroseq, source, has_divide, divide_id, and a level-path id (mainstemlp)
divides	divide_id, flowpath_id, vpuid, areasqkm, has_flowpath

The level-path id becomes the product’s mainstem_id. A national reference is sliced by vpuid, and a row whose vpuid names no VPU being built is not used.

run_pipeline() checks the reference before building with check_reference_input(), which catches truncated identifiers and topology that does not resolve:

check_reference_input("reference.gpkg")

Hydrolocations

Hydrolocations are the gages, assimilation and routing points that become POIs. The minimum is POINT geometry and an hl_reference such as nwis-07187000. prepare_hydrolocations() snaps raw points to the reference and assigns flowpath_id, poi_id and linear-reference measures:

library(hydrofabric)
prepare_hydrolocations(
  "gages.gpkg",
  flowpaths = "reference.gpkg",
  mapping   = c(hl_reference = "siteid"),
  out_gpkg  = "hydrolocations.gpkg"
)

Points more than max_dist (1000 m) from a flowpath are dropped with a warning. The build excludes hl_type values named in exclude_hl_types (default "wbin") and bridge hydrolocations (hl_type = "nbi"). check_hydrolocation_input() validates a set against the references before a build.

Grids

The grids are supplied per domain as file paths, a VRT mosaic or GeoTIFF, read through GDAL, and need not share the fabric’s CRS.

Domain	Grids	Resolution
SuperCONUS	NHDPlus V2.1, with HydroSHEDS v2 as fallback	30 m
Alaska	HydroSHEDS v2	1 arc-second
Hawaii	NHDPlus HR	10 m
PRVI	NHDPlus HR	10 m

fallback_fdr_path and fallback_fac_path name an optional second pair, used where the primary grid cannot delineate a divide. Values taken from it are flagged in fac_source.

Channel table and slope

The RiverML channel table supplies the routing attributes and, through its slope_m_m column, the flowpath slope. A separate slope_parquet (id, slope_m_per_m) takes precedence when given.

Recording the inputs

hf_input_manifest() records each input with its size and MD5, and hf_verify_manifest() checks a later run against it:

hf_input_manifest(
  reference_conus = "conus_reference.gpkg",
  fdr_conus       = "conus/fdr.vrt",
  fac_conus       = "conus/fac.vrt",
  out = "inputs_manifest.json")

hf_verify_manifest("inputs_manifest.json", strict = TRUE)

A VRT names tiles stored beside it, so record or check the tiles too. hf_output_manifest() and hf_verify_output_manifest() do the same for the products.