Interpolate source values onto hex centroids (inverse-distance weighting)

The canonical hex interpolation: for every hexagon in hex_tbl, take the k nearest source points (by great-circle distance, via a unit-sphere kd-tree so it is correct at the poles and antimeridian) and compute the inverse-distance weighted mean (weight = 1 / d^power) of each val_cols column, skipping NA source values per column. The interpolated columns are appended to hex_tbl. For a regular-grid source this reproduces a bilinear-style interpolation; for scattered sources it is classic IDW. This is the ONLY sanctioned way to move values onto hexes — never inherit a containing pixel's value.

Usage

hex_interp_idw(
  con,
  src,
  hex_tbl,
  val_cols,
  k = 8L,
  power = 2,
  lon = "lon",
  lat = "lat",
  chunk = 5000000L
)

Arguments

con

a DuckDB connection (opened bigint = "integer64")

src

a data.frame of source points with lon, lat, and val_cols

hex_tbl

name of the DuckDB hex table (must have hex_id BIGINT)

val_cols

character vector of value columns in src to interpolate

k

number of nearest source points (default 8)

power

IDW power (default 2)

lon, lat

coordinate column names in src (default "lon","lat")

chunk

hex rows processed per kd-tree query batch (default 5e6)

Value

invisibly, hex_tbl