Skip to content

Lilla Gloppsjön observed-pollen transfer

The protocol fixes Neotoma/EPD site 3234 at 14.62333° E, 59.80167° N, its LGLOPPSJ core 4071, and pollen dataset 4200 (dataset DOI). This is a lake-center distance prediction with the existing 16 literal pollen labels, 500-cal-BP point-age bins, and fixed training mean, nearest-site, and inverse-distance-squared baselines. The prediction freeze contains no target pollen composition; scoring is a separate stage below. The source site is distinct from the four earlier Swedish targets. A same-name LandClim point is 0.396271 km away, but the governed site comparison does not establish physical identity or permit its radius to be transferred.

The single source core has 86 samples. The protocol names all 86 sample IDs, the exact 84 comparable calibrated sample-age claim IDs selected from chronology neotoma:chronology:4071:29939, and the two refused claims. The refused samples have negative younger bounds. The selected point ages span 94–11,082 cal BP; all 84 have source younger and older bounds, whose confidence level is not stated. The calibrated chronology is the Giesecke et al. 2014 MADCAP smooth spline using IntCal09 and 14 source controls. The collection-unit reference points instead to radiocarbon-year chronology 29938; the protocol explicitly refuses that chronology for calibrated bins.

The prepared metadata and input manifest lock the allowed metadata read set and reuse the previously governed deterministic gzip profile of 84 distinct training sites. It contains no site 3234, dataset 4200, or target site-name alias. The historical source-validation screen that made that training profile parsed the full Neotoma observation partitions, including this future target, but selected only 81 DK/FI/NO and three other Swedish physical-site profiles for training. This study does not open target taxon observations or count outcomes during prediction preparation. That historical parsing limits an untouched-lockbox claim.

The prediction file contains 69 prediction rows: all three fixed methods for each of 23 age-candidate bins. Every bin has at least two positive-count training physical sites (38–73 across bins), so the fixed training-support rule refused zero bins. These are training-site support figures; target pollen-positive support was unknown at the freeze. The deterministic freeze is SHA-256 bc89f334d95b727d2a051efaa794a131f1d13eee0613f1c32f6e6ace67a6e997; the predictions SHA-256 is 844db7dd47a93eba5d0630719df992d211ced59464e917314e2b5ff131d36a17. The frozen primary method is inverse distance squared. No method or support rule has been selected using this target's pollen outcomes.

The separate committed scoring protocol, SHA-256 e3ebc64b8804a6583d72d542beb9550f7046d735f92f2a6817a60eb18f545e51, fixed the source read order, count denominator, equal-bin MAE and Hellinger, paired support, and refusal handling before target outcomes were interpreted. The scorer verified the committed freeze, prepared metadata, training receipt, relational manifest, and every recorded metadata and observation part hash before parsing target counts. The sample-count table, bin coverage, scores, and machine report preserve the result.

All 86 source samples have admitted direct pollen count rows. The 84 age-selected samples contribute 123,384 native NISP, of which 54,929 are on the exact 16-label axis and 68,455 are outside it. The two age-refused samples remain visible with 3,643 native, 709 closed-axis, and 2,934 outside-axis NISP. Across all 86 samples, native NISP is 127,027, closed-axis NISP 55,638 (43.80%), and outside-axis NISP 71,389. These are observed pollen count denominators, not vegetation fractions. The source has 1,946 admitted direct count rows and 579 other target observation rows excluded by the declared count rule.

All 23 candidate bins had positive target closed-axis counts and were scored by all three frozen methods on identical support. There were zero missing, zero-axis, training-support-refused, or failed bins. Mean absolute error and Hellinger are equal-bin means for this single site:

Fixed method Scored bins MAE Hellinger
Inverse distance squared, frozen primary 23 0.017295 0.237576
Nearest site 23 0.020000 0.228724
Training mean 23 0.016272 0.256392

The frozen primary method is not the lowest on either metric here: training mean has the lowest MAE, and nearest site has the lowest Hellinger. This one site is descriptive development evidence and does not justify changing the fixed method. Source age bounds are not propagated through point-age bins.

The contained-age sensitivity reuses these frozen predictions and scores the 20 bins supported by samples whose full reported age bounds stay within their point-age bin. It preserves all source counts and the three unsupported-bin refusals separately from this 23-bin result.

The retained primary-publication response identifies Almquist-Jacobson, H. (1994), Interaction of the Holocene climate, water balance, vegetation, fire, and cultural land-use in the Swedish Borderland, Lund University doctoral dissertation, LUNDQUA Thesis 30:1–82. The provider response is bijux-pollenomics-data/neotoma/publication-links/2026-09-29/responses/dataset-4200-publications.json at SHA-256 1a13124e576effcbd1f95d5db4b9b7cbda65baf92ab679381b161f209319d471. Neotoma's 2025 Data Use and Embargo Policy permits available data under CC BY 4.0 with Neotoma, EPD, original-investigator, primary-publication and dataset-DOI attribution. The relational source manifest is SHA-256 6f6777a40b67bd0704eb1a982f19e6d451cad0d6e4a41dfebf5e3f28be33105b.

Replay from the repository root into absent directories under artifacts/; the external Neotoma publication-link source store must be present at the workspace sibling path. Preparation and prediction do not interpret target observations; the final command verifies the source read set before scoring:

PYTHONDONTWRITEBYTECODE=1 OMP_NUM_THREADS=1 OPENBLAS_NUM_THREADS=1 \
  MKL_NUM_THREADS=1 NUMEXPR_NUM_THREADS=1 VECLIB_MAXIMUM_THREADS=1 \
  .venv/bin/python -m \
  bijux_pollenomics_dev.source_completion.lilla_gloppsjon_direct_pollen prepare \
  --repository . --output artifacts/jobs/lilla-gloppsjon-freeze/inputs
PYTHONDONTWRITEBYTECODE=1 OMP_NUM_THREADS=1 OPENBLAS_NUM_THREADS=1 \
  MKL_NUM_THREADS=1 NUMEXPR_NUM_THREADS=1 VECLIB_MAXIMUM_THREADS=1 \
  .venv/bin/python -m \
  bijux_pollenomics_dev.source_completion.lilla_gloppsjon_direct_pollen predict \
  --repository . --prepared artifacts/jobs/lilla-gloppsjon-freeze/inputs \
  --output artifacts/jobs/lilla-gloppsjon-freeze/frozen
shasum -a 256 artifacts/jobs/lilla-gloppsjon-freeze/inputs/*.json \
  artifacts/jobs/lilla-gloppsjon-freeze/frozen/*.json
PYTHONDONTWRITEBYTECODE=1 OMP_NUM_THREADS=1 OPENBLAS_NUM_THREADS=1 \
  MKL_NUM_THREADS=1 NUMEXPR_NUM_THREADS=1 VECLIB_MAXIMUM_THREADS=1 \
  .venv/bin/python -m \
  bijux_pollenomics_dev.source_completion.lilla_gloppsjon_direct_pollen_scoring \
  --repository . --prepared artifacts/jobs/lilla-gloppsjon-freeze/inputs \
  --frozen artifacts/jobs/lilla-gloppsjon-freeze/frozen \
  --freeze-sha256 bc89f334d95b727d2a051efaa794a131f1d13eee0613f1c32f6e6ace67a6e997 \
  --output artifacts/jobs/lilla-gloppsjon-freeze/scored
shasum -a 256 artifacts/jobs/lilla-gloppsjon-freeze/scored/*

This is a new-site direct observed-pollen development result with a prior all-part source-validation parse. It is neither an untouched independent confirmation nor LRA, measured-vegetation accuracy, basin-buffer validation, or broad Swedish generalization.