PhenoPhaseR: Reproducible processing workflow for interpolating phenological DWD observations

PhenoPhaseR is a reproducible R workflow for downloading, filtering, modelling, and spatially interpolating phenological observations provided by the German Weather Service (DWD). It implements the PHASE approach (Gerstmann et al. 2016), which combines growing degree day models with geostatistical interpolation to produce area-wide phenological predictions across Germany at 1 km spatial resolution.

• Since v1.2.0, two in-pipeline publish hooks package the intermediate filter variant results and the final PHASE entry-date COGs as self-contained RO-Crate 1.2 deposits with W3C-anchored provenance (PROV-O), dataset descriptors (DCAT 3 / Dublin Core Terms), and quality metadata (DQV with SKOS bridge to ISO 19157-1).
• As of v1.3.0, Hook B aggregates the per-(phase, year)  intermediate outputs into per-phase multi-band Cloud-Optimised GeoTIFFs (one band per year, named by year) and wide-format per-phase VAM CSVs, reducing the published artefact count from ~896 to 21 files while preserving the per-year ISO 19157-1 quality elements through DQV temporal tagging.
• v1.4.0 adds two new published quality artefacts to Hook B (CAL, in-sample BAM model-fit diagnostics; GEM, spatial quantiles of the BSE uncertainty raster), auto-writes a README.md into each crate documenting its validation stance, and generates a self-contained ro-crate-preview.html for browser-based inspection
    without external tooling.
• v1.5.0 restructures the DQV quality metadata: each per-year quality measurement is now a first-class top-level entity in the JSON-LD @graph with its own @id, and the per-phase Datasets reference them by @id only — satisfying JSON-LD's node-reference rule and resolving 33 REQUIRED-severity violations previously  reported by `roc-validator`.
• v1.6.0 turns the publish hooks into a multi-crop blueprint: a  single shared `_crop_specs.R` file holds per-crop metadata (DWD Plant ID, binomial, AGROVOC concept URI, Wikidata QID) for seven crops (winter wheat, winter rye, winter barley, winter rapeseed, spring barley, oats, maize), a family-wide creators list with explicit DataCite roles, and a layered keyword scheme; both  builders are parametric on this configuration. Domain semantics are added via AGROVOC subject terms (schema:about / dct:subject) with DefinedTerm and DefinedTermSet entities in the @graph, alongside the existing W3C / ISO 19157 vocabulary stack.
• v1.6.1 is a patch release fixing a `promise already under evaluation: recursive default argument reference` crash in the v1.6.0 publish hooks (the parameter `crop_spec` is renamed to  `crop` to break a self-shadowing default); switching the Hook B (PHASE) deposit to a flat layout by default so downstream pipelines can stream individual multi-band COGs from Zenodo via GDAL's `/vsicurl/` without downloading the full deposit; and documenting the intentional Schema.org `http://` namespace choice in both builders. No numerical changes from v1.6.0; no changes to the multi-crop blueprint or the AGROVOC integration.
• v1.6.2 adds resilient temporal-gap handling so a (phase, year) whose interpolation fails because too few stations survived filtering no longer aborts the run: the filter-variant selector logs the dropped cell and reason to `opt_scores/GAPS_<plant>.csv`, the interpolation writes a full-extent NA surface instead of calling `stop()`, and the Hook B publisher detects every all-NA band as an ISO 19157-1 `DQ_CompletenessOmission` measurement.  Every COG remains a complete 32-band cube where band *i* maps to  year *i*. v1.6.2 also adds sugar beet (DWD plant ID 253, *Beta vulgaris* ssp. *vulgaris* var. *saccharifera*) to the blueprint, bringing the deposit family to eight crops; introduces two optional visualization helpers (`plot_phenology_raster_maps.R`  for a DOY/BSE raster pair over a named AOI, and  `plot_phenology_window_timeseries.R` for an AOI-aggregated phase-window time-series across 1993–2024); corrects the RO-Crate profile claim from Workflow Run Crate to Process Run Crate (the WRROC base profile that the existing CreateAction already satisfies); and suppresses GDAL PAM sidecars (`.tif.aux.json`, `.tif.aux.xml`) from the Hook B deposit since their content is already in the COG's TIFF tags. No numerical changes to any output  for the existing seven crops.
• v1.6.3 removes the AGROVOC controlled-vocabulary subject anchors introduced in v1.6.0 after the hand-curated URIs in `_crop_specs.R` were found to point at unrelated concepts when resolved against `agrovoc.fao.org` — e.g. the URI labelled "winter rye" resolved to "sawlogs", the URI labelled "phenology" resolved
    to "local authorities", the URI labelled "Germany" was a different geographic concept. Only the winter-wheat URI verified correctly. Rather than swap to another vocabulary (Wikidata, GEMET) and  inherit the same verification burden — controlled-vocabulary subject anchoring is supposed to *prevent* exactly the typo-points-at-wrong-concept bug we found — the entire subject-anchor layer is removed. The v1.6.3 deposits use only free-text `keywords` (indexed by Zenodo, OpenAIRE, BonaRes, and Google Dataset Search) and the GeoNames `spatialCoverage` URI for subject anchoring; the deposits remain valid RO-Crate 1.2 since `schema:about` is optional in the profile. v1.6.3 also makes the auto-written PHASE README's gap-handling section build-aware, separating the *mechanism* description (unchanged across deposits) from a conditional "What this deposit reports" subsection that states whether the gap chain actually fired for the build at hand. Already-published v1.6.1 and v1.6.2 crop deposits retain their original `schema:about` entries on Zenodo as historical record; the cleanup lands on their next coordinated re-release. No
  numerical changes to any output, no changes to the publish-hook machinery, the PROV-O lineage, or the DQV quality measurements.