Skip to content

EMPD2–NFI observed vegetation pairing decision

The pair ledger records 16 geographic proximity leads and zero admitted observed vegetation pairs. The machine-readable report binds each refusal to the original EMPD2 SQLite and Swedish NFI workbook SHA-256 identities. This is an outcome-free source review: NFI tree measurement cells were not read, no vegetation reconstruction was evaluated, and no benchmark threshold was frozen.

Source and denominator

The original Swedish subset has 132 EMPD2 metadata samples. The NFI workbook has 43,892 plot-year rows across 6,018 distinct tract/plot keys. WGS84 inverse-geodesic screening after transforming NFI points from SWEREF99 TM gives 53 plot-year/sample combinations within 1 km, across 16 pollen samples. The ledger selects one nearest NFI plot-year per sample, including 14 distinct selected plot-years. Distance is only a lead-generation screen. It does not establish shared time, vegetation footprint, measurement units, or independent validation support.

The original source dossier, spatial screen, EMPD2 SQLite and NFI workbook are preserved in the external private source store. The result records the exact input hashes checked by its producer. The dossier and screen contain exact NFI row locations and are not tracked in Git. The tracked SLU calibration guide is an exact byte copy of its original. The producer checks the private inputs' exact hashes and reads only the NFI workbook's first six identity/coordinate fields to bind every selected plot-year to an original row and recompute each geodesic distance. The source-native NFI date is checked during replay but omitted from the public CSV; pollen agebp is not treated as a collection year. The tracked CSV replaces each NFI row location and key with an opaque SHA-256 identity; it contains no NFI tract, plot, year, date, or coordinates.

PANGAEA's EMPD2 source record identifies the pollen source and CC BY 4.0 terms. The SLU National Forest Inventory calibration guide states CC0 1.0 terms and describes temporary forest plot observations intended for calibration or validation of remote-sensing products. Plot observations cannot by themselves establish regional vegetation area totals or means.

The source measurement schema reads the original workbook's field dictionary and only inventory date, plot area and land-use metadata from its 43,892 plot-year rows. The dictionary describes an undivided circular plot with radius 7 m and area 153.94 m²; every plot-year row has that area. Five tree-group basal-area columns are labeled in m²/ha: pine, contorta, spruce, birch and other deciduous trees. The workbook also contains five land-use codes; their meanings are not assigned by this schema. Inventory dates remain source-native Excel serial values. No basal-area or other tree measurement outcome cell was decoded, and no target taxon composition or minimum benchmark threshold was chosen. Five other workbook headers lack exact field-dictionary names and remain listed in the schema report. To reproduce this bounded schema read:

PYTHONDONTWRITEBYTECODE=1 PYTHONPATH=packages/bijux-pollenomics-dev/src \
  .venv/bin/python -m bijux_pollenomics_dev.source_completion.empd2_nfi_reference.measurement_schema \
  --private-source-store /path/to/private/source-store \
  --output artifacts/jobs/empd2-nfi-pairing/nfi-measurement-schema.json

Current decision

All 16 leads lack an authenticated pollen collection date, opened NFI tree measurement values, a pollen taxon denominator and tree-taxon crosswalk, and a reviewed rule mapping a 153.94 m² plot observation to the pollen basin or reconstruction support. Five leads also have unresolved cross-source physical sample identity; one collection note requires authentication. The benchmark still lacks a qualified independent-source/leakage review and non-null predeclared minimum sample and error thresholds. Each missing field is explicit in the CSV. The 16 rows are refusals for the current benchmark, not 16 negative measured observations.

The bounded decision can be replayed with the original raw holdings and tracked review inputs:

PYTHONDONTWRITEBYTECODE=1 OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 \
  PYTHONPATH=packages/bijux-pollenomics-dev/src:packages/bijux-pollenomics/src \
  .venv/bin/python -m bijux_pollenomics_dev.source_completion.empd2_nfi_pairing \
  --repository . \
  --private-source-store /path/to/private/source-store \
  --output artifacts/jobs/empd2-nfi-pairing/publication

The output destination must be absent. Source-bound replay requires the original SQLite, workbook, dossier and spatial screen in the private source store. The producer checks their identities; default focused tests exercise the tracked guide, public output and refusal contract without them. Its private source-row ledger can be regenerated for review and discarded with the run output. The decision can change only after the missing source facts and scientific review are supplied, then with a new source-bound record; proximity alone cannot promote a lead.