Precompute IDW weights from a shared source grid to hex centroids

One kNN (hex centroids -> source points, great-circle via a unit-sphere kd-tree) whose weights are reused by every model on that source grid. Writes <out_tbl> (hex_id, src_i, w) with k rows per hex, and <out_tbl>_sum (hex_id, w_total) = the IDW denominator (sum of w over the k neighbours, constant per hex regardless of model).

Usage

hex_grid_weights(
  con,
  hex_tbl,
  src,
  k = 8L,
  power = 2,
  where = NULL,
  radius_km = NULL,
  out_tbl = "hex_src_w",
  chunk = 5000000L
)

Arguments

con

DuckDB connection (opened bigint="integer64")

hex_tbl

hex table with hex_id BIGINT

src

data.frame with src_i (integer) + lon,lat source-grid points

k, power

IDW parameters (default k=8, power=2)

where

optional SQL predicate on hex_tbl (e.g. "in_usa") to limit the hexes weighted (a US-only score layer needs only US hexes)

radius_km

optional great-circle cap (km): drop source neighbours beyond this distance so distant points cannot extrapolate. Paired with a full source grid (ocean + absent cells), this reproduces the zero-surround IDW fade — a hex beyond ~1 cell of any present value has only absent (0) neighbours in range and decays to 0. NULL = keep all k neighbours (default).

out_tbl

base name for the weight tables (default "hex_src_w")

chunk

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

Value

invisibly, out_tbl