Filter Variants: Optimization metrics and filtered station observations for winter wheat (Triticum aestivum, DWD Plant ID 202) in Germany

What's new in v1.6.0

Rebuild of the plant 202 filter-variants deposit against PhenoPhaseR v1.9.0. A minor, metadata-only version: the OPT_* optimisation tables and the SHP_* filtered station shapefiles are byte-for-byte unchanged from v1.5.0. What changes is how the accuracy metrics are described in ro-crate-metadata.json, and the removal of an unverifiable crop crosswalk.

What changed

• Honest validation basis on every accuracy metric. Each per-(phase, year) quality measurement now records the regime that produced it. The RMSE / MAE / correlation of this filter variant come from critical_doy_determination() and are in-sample (resubstitution) scores: the temperature-sum threshold is derived from the same stations on which it is then scored, T_SUMS encodes each station's observed date (so the calibration target enters the threshold), outliers are removed before scoring, and the reported quantile is the one that minimises MAE on that same data. They are therefore optimistically biased and are not estimates of out-of-sample predictive skill — a caveat now stated explicitly on every such measurement, following the pipeline's own critical_doy_determination() header.
• In-sample metric tokens. These metrics are now emitted under dedicated tokens is_rmse / is_mae / is_cor (in-sample), replacing an earlier labelling that carried a hold-out wording over from the PHASE crate. Metric descriptions are drawn from a single shared registry, so the same measure can no longer be described differently across the filter-variant and PHASE crates.
• Removed: unverifiable crop → Wikidata crosswalk. The skos:exactMatch link from the crop concept to Wikidata has been removed. AGROVOC publishes no Wikidata match for these crop concepts (they link to USDA NAL, the World Bank thesaurus, and DBpedia), and the previously embedded Wikidata QID was unverifiable and, on checking, incorrect. The verified AGROVOC concept URI is retained as the sole, authoritative subject anchor.

Unchanged from the previous version

The OPT_* / SHP_* data files (byte-identical), the ISO 19157-1:2023 anchoring of the accuracy metrics (iso19157:QuantitativeAttributeAccuracy via dqv:inDimension), the provisional fairagrodq: namespace declaration, the PROV-O provenance chain, the DCAT 3 / Dublin Core descriptors, and the RO-Crate 1.2 structure. As before, this deposit carries only accuracy / evidence metrics, so it emits no spatial-uncertainty dimensions and no expression principles.

Coverage unchanged: plant 202 (DWD/JKI code, Triticum aestivum), phenological phases 10, 12, 15, 18, 19, 21, 24, years 1993–2025. Direct input to spatial_interpolation.R (Step 7); the final PHASE entry-date COGs are at https://doi.org/10.5281/zenodo.19571847.