Species composition

The first INV-05 slice ingests a public dominant tree species raster and stores block-level class fractions and a leading species. It is a data-contract layer, not a species distribution model.

Source

The pilot uses the Natural Resources Canada / NTEMS Annual Tree Species (1984-2022) product, layer CA_Forest_Tree_Species_2022:

  • 30 m raster, NAD83 Lambert Conformal Conic;

  • one leading species class per cell, codes 0-37;

  • annual maps; the pilot uses 2022;

  • Open Government Licence - Canada;

  • Hermosilla et al. 2024, Forest Ecology and Management 122313.

The class map carries the raster code, common name, NFI code, and scientific name. Class 0 is non-treed.

Contract

species_composition.csv is long by block and species class:

block_id

Block identifier.

species_code, species_name, nfi_code

Public class identity.

pixel_count, class_fraction

Valid 30 m cells in the block and their fraction.

is_leading

Exactly one leading class per covered block, selected by the largest pixel count.

species_metadata.json records the source id/version, year, native resolution, CRS, licence, citation, full class map, and notes. A block with no valid species pixels is an error; it is not assigned a zero or a default species.

The public product is dominant species only. A second table, species_fractions.csv, carries explicit fraction sources such as VRI species percentages. It is long by block and species code with fraction and is_leading. A block with no species rows is omitted and listed in metadata when --allow-missing is explicit; percentages that do not sum to 100 within one point are an error.

species_fractions_summary.csv reports the leading fraction, non-leading fraction = 1 - leading fraction, and species count. That is the mixedwood/composition measure; no arbitrary binary mixedwood cutoff is applied.

Commands

ws4 species ingest \
  --raster tfl6_species_2022.tif \
  --blocks vri_aoi.geojson \
  --class-map class_map.csv \
  --output species_composition.csv \
  --metadata species_metadata.json \
  --source-version "CA_Tree_Species_Classification_2022" \
  --year 2022 \
  --block-id-field feature_id

ws4 species compare \
  --composition species_composition.csv \
  --metadata species_metadata.json \
  --reference vri_reference.csv \
  --output species_comparison.csv \
  --summary species_comparison_summary.json

The comparison selects the public leading row per block and joins the reference leading code. It reports matches, agreement percentage, and missing public or reference blocks. A mismatch is recorded, not smoothed away.

Pilot

examples/species_tfl6 holds the TFL 6 VRI AOI subset, class map, VRI reference, composition, and comparison. The public 2022 layer uses classes 0, 33, and 35 in the AOI; it did not identify yellow-cedar (11). It gives western redcedar as the leading species for all seven blocks. Against VRI, four of six comparable blocks match (66.7%); block 16624698 is VRI western hemlock and block 16625195 is VRI yellow-cedar. One block has no VRI label and is reported as missing reference. The local agreement is not an accuracy estimate for the national product. VRI fractions show all six labelled blocks are mixed, with non-leading fractions from 0.40 to 0.60. The public dominant layer cannot represent those mixtures.

Caveats

  • A dominant-species raster is not a species-fraction layer.

  • VRI and the public layer have different vintages and label semantics.

  • Agreement is a data-quality check, not an accuracy validation of either product.

  • Mixedwood fractions require a separate source; do not infer them from the leading class.

  • The public layer has national accuracy of 86.1% +/- 0.14%, which is not a guarantee for this local block sample.