fork default: LCCClassesToReplaceNNMethod = 'nearestWeighted'; require LandR (>= 1.2.0.9005)

[achubaty]

Fork-side counterpart of PredictiveEcology#102, which adds the LCCClassesToReplaceNNMethod parameter upstream.

Depends on PredictiveEcology/LandR#197 — merge that first.

Deliberate divergence from upstream

Same parameter, different default:

	default	why
upstream (PredictiveEcology#102)	"nearestRandom"	preserves the stochastic allocation the module has always had
this fork	"nearestWeighted"	the projects consuming this fork (LandWeb et al.) want reproducibility without seed management, and want a cropped development subset to agree with the full-extent run

"nearestWeighted" keys the draw on the pixel's ground position rather than the RNG. It is deterministic, needs no set.seed(), is stable under Cache(), and — because the key is the cell centre rather than the cell index — a grid-aligned crop of the study area gives the same answer as the full extent. The abundance weighting is identical to "nearestRandom"; only the source of the draw differs.

Setting it as the fork default rather than in each project's parameter config means projects get the intended behaviour without having to remember to pass it.

Also here, and not upstream

The reqdPkgs floor moves to LandR (>= 1.2.0.9005) for a second reason that only applies to this fork: LandR 1.2.0.9004 dropped the newPossLCC column that this fork's write-back block uses.

if (!is.null(newLCCClasses$newPossLCC)) {
  rstLCCAdj[newLCCClasses$pixelIndex] <- newLCCClasses$newPossLCC
}

That is.null() guard silently stopped firing under 1.2.0.9004, leaving rstLCCAdj — and therefore ecoregionMap — showing the un-replaced classes. LandR#197 restores the column; the comment now records the gap so the floor isn't lowered later without noticing. Upstream doesn't carry this block, so PredictiveEcology#102 doesn't need the note.

Testing

• Module metadata parses; SpaDES.core::moduleParams() registers the parameter, moduleMetadata() reports the bumped floor and version = 1.5.13.
• Not yet run end-to-end: LandWeb's renv still pins LandR 1.2.0.9004, so the lockfile needs updating once LandR#197 merges before a dataPrep run will exercise this.

Related

• new parameter LCCClassesToReplaceNNMethod; require LandR (>= 1.2.0.9005) PredictiveEcology/Biomass_borealDataPrep#102 — the upstream parameter addition
• convertUnwantedLCC(method = 'nearestRandom') under Cache(): one draw replayed across replicates PredictiveEcology/Biomass_borealDataPrep#101 — Cache()/RNG interaction; affects the upstream "nearestRandom" default, not this fork's "nearestWeighted"
• convertUnwantedLCC(): remove the lowest-class tie-break; weight allocation by local abundance PredictiveEcology/LandR#197 — the method argument itself

🤖 Generated with Claude Code