What's new in v1.7.0 Rebuild of the winter wheat PHASE deposit against PhenoPhaseR v1.7.2. This is a minor version on two counts, both additive: it adds the 2025 data year (the prior v1.5.0 record covered through 2024) and the new k-fold cross-validated PICP/MPIW calibration of the BSE uncertainty layer. The existing years and rasters are otherwise unchanged. What changed ro-crate-metadata.json carries the verified AGROVOC subject anchors (reinstated in PhenoPhaseR v1.7.0). Earlier hand-typed AGROVOC URIs in _crop_specs.R were found, on resolution against agrovoc.fao.org, to point at the wrong concepts for seven of ten terms; every URI is now corrected and confirmed, and a build-time guard (verify_agrovoc_uris()) resolves each against the live catalogue and fails the build on any mismatch. Anchored to the verified AGROVOC concept c_8412 (Triticum aestivum). Honest DQV anchoring. Quality measures no longer mint iso19157: IRIs (no such concepts exist in ISO 19157). schema:propertyID carries the bare measure token, the fabricated skos:closeMatch is removed, and the ISO link is made only at the genuine dimension-class level via dqv:inDimension (DQ_ThematicAccuracy, DQ_UsabilityElement, DQ_CompletenessOmission, DQ_LogicalConsistency). The RO-Crate profile claim is Process Run Crate (w3id.org/ro/wfrun/process/0.5); the CreateAction with instrument pointing to the PhenoPhaseR software DOI satisfies it, with a Profile Crate entity in @graph. DQV completeness fractions are stored exactly (e.g. 32/33 = 0.96970) via the jsonlite digits parameter. Validated uncertainty calibration (PIC). Each PHASE deposit now ships PIC_-.csv and surfaces picp/mpiw as DQV measurements under DQ_UsabilityElement, tied to the BSE COG via dqv:computedOn. PICP/MPIW come from a dedicated 5-fold cross-validation, separate from the published all-data fit, so the rasters are never degraded to obtain the calibration. Coverage DWD Plant ID 202 (Triticum aestivum L., "Winterweizen" in NamesPlantPhase.txt) 7 phenological phases: 10 (Bestellung Beginn = sowing start); 12 (Auflaufen Beginn = emergence start); 15 (Schossen Beginn = shooting / stem-extension start); 18 (Ährenschieben Beginn = ear / panicle emergence start); 19 (Milchreife Beginn = milk ripeness start); 21 (Gelbreife Beginn = yellow ripeness start); 24 (Ernte = harvest) 33 years (1993–2025) 1 km × 1 km grid in EPSG:25832, Federal Republic of Germany (GeoNames 2921044) Each phase ships as a multi-band Cloud-Optimised GeoTIFF, one band per year (band names "1993"…"2025"), so terra::rast(file, lyrs = "2020") returns the 2020 surface directly. Two layers per phase: DOY (entry day-of-year, posterior mean) and BSE (per-pixel basis-spline standard error). The DWD catalogue (NamesPlantPhase.txt) lists additional phase code(s) for this crop that this deposit does not model: 22 (Vollreife Beginn / full ripeness start); 23 (Ernte von Hand / manual harvest). Phase 23 duplicates phase 24 (Ernte / harvest) and is omitted everywhere in the deposit family; phase 22 is omitted across the cereal subset because the calibration data is sparser than the modelled subset. The published phase set is the modelled subset of the DWD catalogue rather than its full set. Uncertainty. The BSE layer is the posterior standard error of the fitted mean — a model-internal statistical uncertainty (anchored to AGROVOC "statistical uncertainty", c_28975), not a validated predictive uncertainty. Its calibration is reported separately as PICP (prediction-interval coverage probability) and MPIW (mean interval width) at nominal 90%, computed by k-fold cross-validation (5 folds) independent of the published all-data fit: each station is held out exactly once, and the interval combines each fold's se.fit with the production residual variance. These are surfaced as DQV quality measurements under DQ_UsabilityElement, tied to the BSE COG via dqv:computedOn — so a consumer can distinguish what the layer is (model-internal uncertainty) from how well-calibrated it turned out (validated coverage). Status as canonical deposit This is the reference deposit of the PHASE family on Zenodo; every other crop's record points at concept DOI 10.5281/zenodo.19571847 under Is part of → Dataset. Provenance Software: PhenoPhaseR v1.7.2 — 10.5281/zenodo.18743008 Upstream dataset (filter variants, Hook A): concept DOI 10.5281/zenodo.19483111 Input data (DWD): 10.5281/zenodo.18594963

This dataset provides interpolated phenological phase entry dates (day-of-year) with a per-pixel uncertainty layer, from the PhenoPhaseR processing pipeline (https://doi.org/10.5281/zenodo.18743008) for winter wheat (Triticum aestivum L., DWD Plant ID 202), covering the 7 DWD-reported phenological phases modelled by the pipeline for the entire territory of Germany for the period 1993–2025 at 1 km resolution. PhenoPhaseR is an open-source R workflow that couples the PHASE model — a growing-degree-day (GDD) approach with adaptive outlier filtering — to derive spatially consistent, quality-controlled phase-start dates from the DWD volunteer observer network, then interpolates them to a 1 km grid with a spatial generalized additive model. The component provided here corresponds to processing Step 7 (spatial interpolation) of the seven-step pipeline.