EMPD2 reference candidate¶
EMPD2 is one of the eleven contracted source families and remains a candidate pollen reference for the Nordic study. A family contract registers its role, lifecycle, and refusal boundary; it does not admit a row. None of its rows is currently admitted to LRA reconstruction, model tuning, held-out scoring, or a governed public product. The source SQLite file is held in the private source store, outside Git. Its published PANGAEA deposit is marked CC BY 4.0; reuse still needs source attribution and review of the intended join.
The checked-in family lifecycle is raw: missing, normalized: missing,
reviewed: present, and published: missing. Its authority remains
review_required, with publication posture
review_required_not_publication_ready; the compact review evidence does not
promote the locally held SQLite file into a governed raw layer.
The pinned SQLite file is 21,327,872 bytes with SHA-256
5634e85d922ee0741c41a77d081010c9cc1c113966b283bc52e2051a9d3722e2.
The repository's bijux_pollenomics_dev.source_completion.empd2 producer
checks that hash, reads the database without writing it, and preserves a
sample-level reason for every failed count screen. It also binds the separate
EMPD2–Neotoma candidate-identity review by hash. Its results for this source
are:
| Source country | Metadata samples | Pollen count rows | Mechanical count candidates | Unresolved Neotoma overlap leads |
|---|---|---|---|---|
| Sweden | 132 | 3,379 | 131 | 21 |
| Denmark | 20 | 968 | 20 | 0 |
| Norway | 622 | 19,774 | 622 | 66 |
| Finland | 141 | 3,040 | 141 | 25 |
| Total | 915 | 27,161 | 914 | 112 |
These are source-native country labels and database counts, not governed country assignments or publication membership. The four-country source population is useful for coverage accounting, while its admitted scientific and published denominators remain zero.
The mechanical rule requires at least one count row, all rows finite
nonnegative integers, at least one positive row, and source ispercent=0 and
isdigitised=0. One Swedish source count is 49.5 and is retained as a
fractional value with a refusal reason. Mechanical candidates are a screening
denominator, not admitted samples. The 112 overlap leads are unresolved site
or dataset relationships; none establishes that the physical pollen sample is
identical or independent across databases. They must be resolved or grouped
before a split can be frozen.
Danish modern-pollen age leads¶
All 20 Danish metadata samples pass the mechanical count screen, with 968
source count rows and no unresolved EMPD2–Neotoma overlap flag in this pinned
review. In the source SQLite, all 20 carry age-uncertainty code A. Its own
ageuncertainties lookup calls this category “Modern sample” and gives the
range label “0BP to Present”. Their source-native agebp values are -46
(16 samples), -55 (two), -48 (one) and -30 (one). The exact 20 source rows
and the three lookup definitions are retained in
reference-sources/empd2-nfi/empd2/age-lookup.json relative to the private
source store.
Its SHA-256 is c39ebc03ddd038a8f013767fa7745b57ad6c6e602533fd014963bb540241068f.
The EMPD2 source paper
describes Age BP as a numeric sample-age field and age uncertainty as a
categorical field. The A label supports a broad modern-sample description;
it does not supply a verified collection year or a precise age interval for
matching another observation. We preserve the signed native values without
converting them into calendar dates. Taxa/PPE, basin and footprint, count
denominator, source identity, and temporal support still need review before
numerical LRA or a four-country held-out benchmark can use these rows.
Danish historical paired-reference lead¶
The official GEUS record for Nielsen and Odgaard (2005) describes a potentially valuable paper-internal calibration/test design: pollen assemblages and historical land-cover maps for 30 small Danish calibration lakes at about AD 1800, followed by nine distinct test sites. This is a source-recovery lead, not a locally available benchmark cohort. The exact 2005 title and DOI match zero publication rows in the pinned EMPD2 database, so they currently link to zero locally held EMPD2 samples.
Seven broad publication-name matches containing Nielsen are different papers
and cannot establish the missing identity. One of them is the related 2010
Nielsen and Odgaard LRA paper. Its EMPD2 relation reaches exactly one Danish
sample, Nielsen_a1, with native agebp=-48, age-uncertainty category A,
44 pollen-count rows, ispercent=0, and isdigitised=0. That one modern
metadata row does not identify any of the 30 circa-AD-1800 calibration lakes
or nine 2005 test sites.
The bounded source probe preserves the official GEUS page with SHA-256
c370fa01cd8efd25b7f9e0d0faf0af06328c9c38653d75dcef620d18bbf96348.
Its report SHA-256 is
3fede323ba15c6503344e1461c8835b9e6d5a699b08eb9a7801105eb76d6219d;
the separate 2010 metadata report SHA-256 is
1b6a6d53eb5c04e089680c37d4e403f8a25edc0c8c5e77c1752748db631048a4.
Before these publications can contribute an independent paired reference, the
study still needs exact site identities, raw pollen counts, historical-map
values and taxon crosswalk, chronology and footprint support, reuse rights,
and a leakage-safe relation to every locally held sample. The present
admission denominator remains zero historical vegetation–pollen pairs.
No site-level historical-map label or numerical model score was accessed; the
public abstract does disclose its qualitative ERV-versus-PLS RMSE conclusion.
Swedish plot proximity screen¶
The repository's bijux_pollenomics_dev.source_completion.empd2_nfi producer
compares the 132 Swedish EMPD2 candidate coordinates with 43,892 NFI
plot-year rows (6,018 tract/plot keys). It checks the NFI workbook and its
ReadMe-backed SWEREF99TM declaration by hash, transforms only plot geometry,
and measures unrounded WGS84 inverse-geodesic distance. The final tree
measurement columns are not selected, displayed or scored, though the XLSX
parser necessarily streams the whole worksheet XML. The thresholds below are
descriptive source leads, not reviewed ecological pairing distances.
The official SLU NFI calibration guide
declares CC0 1.0 for the downloadable plot subset. It limits the data to
temporary forest plots from 2007–2025, excludes boundary plots, and explicitly
disallows deriving totals or mean values for a geographic area. A pinned copy
of that guide is retained locally with SHA-256
793d90e287b2c686d0007f02497e12e8a72a47e5a45531918d9fc71d78499bc8.
This resolves the stated reuse terms for the supplied plot subset; it does
not turn plot observations into regional vegetation cover.
| Maximum distance | All EMPD2 samples with an NFI row | Mechanical count candidates with an NFI row | All NFI row/sample pairs | Count-candidate row/sample pairs |
|---|---|---|---|---|
| 1 km | 16 / 132 | 16 / 131 | 53 | 53 |
| 5 km | 66 / 132 | 65 / 131 | 696 | 681 |
| 10 km | 96 / 132 | 95 / 131 | 2,846 | 2,821 |
Five of the 16 samples within 1 km carry unresolved EMPD2–Neotoma identity
flags. The one Swedish sample with a fractional count is retained in the
all-metadata 5 and 10 km totals but excluded from the mechanical count
candidate totals. Mechanical count candidacy is still not scientific
admission. Repeated NFI years and nearby plots make row-pair totals larger
than the number of distinct pollen samples. The nearest NFI row is chosen
only by distance; its inventory year does not date or match the pollen
sample. The EMPD2 position datum and uncertainty remain unreviewed. The
source has no explicit collection-year field, and the source age/uncertainty
codes do not establish matching survey support. Spatial proximity also does
not prove that a small NFI forest plot represents a pollen basin or regional
vegetation footprint. The report is a candidate-only local source artifact,
SHA-256 5d35aed59cb5e3b49582976264837fab1c8d4ffa3dc1937e73bf2f04f32e2c66;
zero independent vegetation–pollen pairs are admitted.
Source context for the nearest leads¶
The separate bijux_pollenomics_dev.source_completion.empd2_nfi_provenance
producer checks the official SQLite and the spatial report by SHA-256, then
retains source-native context and publication links for all 16 samples within
1 km. Eight have a linked publication. Their source age-uncertainty codes are
five A, three B, and eight literal NULL; eleven source age values are
missing or source text. The SQLite metadata table has no structured
collection-date column.
One record, Nielsen_b4, has a source note saying the surface sample was
collected in September 2013. Its distance-selected nearest NFI row is
427.12 m away and has survey year 2013. The note is a concrete lead for
source and timing review, not an authenticated match: the pollen is lake
sediment, the small forest plot has different spatial support, and the
EMPD2–Neotoma identity flag is unresolved. The other 15 leads have no
collection-time cue in this source note field. The candidate-only dossier is
retained as a local artifact, SHA-256
bcab3834980f6934143f3a2bf4e2a0155bf3c7024850989de4f0f28f89584c31.
It excludes the structured vegdescription field, but retains unstructured
source notes for manual review; it is not a model-training input.
No vegetation–pollen pair is admitted from it.
The 16-lead reviewer sheet is generated from the
pinned spatial and source dossiers and the official NFI guide. It preserves
the source context and nearest-row facts needed for qualified review without
promoting any lead into a measured vegetation reference.
Source agebp is preserved exactly, including missing and literal NULL
values; it is not converted to an exact collection year. This SQLite release
does not expose a collection-year field. Taxa/PPE, basin support, chronology,
source identity, and compatible count units still require review. EMPD2's
vegetation descriptions and mapped ecosystem classes are contextual labels,
not independently measured vegetation cover. Thus this candidate contributes
zero admitted measured vegetation references and no Nordic accuracy score.
The producer's machine report records the 915 source-native sample rows, country denominators, invalid count values, unresolved overlap flags, source hash, and refusal codes. It can be rerun on the pinned local SQLite file with the pinned 112-row overlap review; both inputs must match their recorded SHA-256 hashes. The next source decision is sample-level identity and timing review, followed by taxa/PPE, basin, rights-for-join and scientific admission. Contracted-family status does not change that sequence or authorize an EMPD2 or EMPD2–NFI publication.