Per-location physics gate backed by a hierarchical PIML fit

Upgrades the global-envelope al_physics_gate_piml() to a per-candidate envelope: each candidate's profile is predicted from its own covariates, giving a tighter physical plausibility window driven by the local pedogenic context.

Usage

al_physics_gate_piml_hierarchical(
  hier_fit,
  candidate_covariate_cols = hier_fit$covariate_cols,
  safety_factor = 1.2,
  envelope_depths = c(0, 5, 15, 30, 60)
)

Arguments

hier_fit

A edaphos_piml_hierarchical.

candidate_covariate_cols

Character vector with the column names in the candidate table that correspond to the training covariates (must be the same set, can have different order — the function reorders them).

safety_factor

Numeric >= 1, widening factor on each side.

envelope_depths

Numeric, depths (same units as training) at which the profile is probed to compute min and max for each candidate. Defaults span surface to a deep horizon.

Value

A function suitable for al_query(..., physics_gate = <this>).