PHASE: Interpolated phenological phases of Winter Wheat (Triticum aestivum) in Germany, 1993–2024

Crop-specific PHASE dataset for Winter Wheat (Triticum aestivum, JKI/DWD crop code 202). Contents (35 data files + 1 metadata file): DOY COGs (7 files): Interpolated Day-of-Year per phenological phase. Multi-band Cloud Optimized GeoTIFF; 32 bands (1993–2024); Int16; 1 km; EPSG:25832. BSE COGs (7 files): BAM posterior standard error (uncertainty). Same multi-band structure; Float32 rounded to 2 decimal places. VAM CSVs (7 files): Holdout-validation accuracy metrics. BAM fitted on 75% training stations, evaluated at 25% withheld test stations. Wide-format CSV; one row per year (1993–2024, 32 rows). Columns: PLANT, PHASE, YEAR, TN, ON, VN, METHOD, BAM_K, RMSE, MAE, MSE, R2. CAL CSVs (7 files): In-sample calibration / model-fit diagnostics computed from the fitted BAM object (mgcv::bam). Useful for assessing model complexity and goodness-of-fit on the training data, complementing the out-of-sample VAM metrics. Wide-format CSV; one row per year (32 rows). Columns: PLANT, PHASE, YEAR, BAM_K, AIC, BIC, EDF, DEV_EXPLAINED. GEM CSVs (7 files): Spatial quantiles of the per-pixel BSE raster — the global distribution of prediction uncertainty across Germany for each year. Useful for uncertainty propagation in downstream applications. Long-format CSV; five rows per year (1993–2024, 160 rows). Columns: YEAR, Quantile, Value. Quantile levels: 0%, 25%, 50%, 75%, 100%. ro-crate-metadata.json: RO-Crate 1.2 provenance document linking all files to the filtered phenological observation shapefiles and to the PhenoPhaseR software. Embeds W3C-anchored metadata (DCAT 3, DQV with SKOS bridge to ISO 19157-1, PROV-O, Dublin Core Terms). File naming: DOY_202-{PHASE}.tif, BSE_202-{PHASE}.tif, VAM_202-{PHASE}.csv, CAL_202-{PHASE}.csv, GEM_202-{PHASE}.csv. Phases covered: 10, 12, 15, 18, 19, 21, 24. Band names in the multi-band COGs encode the year — subset semantically in R with terra::rast(file, lyrs = as.character(year)) or via gdal_translate -b N for the corresponding band index. Derived from filtered and optimized phenological observation shapefiles (doi:10.5281/zenodo.19483112) using PhenoPhaseR v1.3.0. What's new in v1.2.0: CAL and GEM diagnostic tables added. 14 new CSV files documenting model-fit diagnostics (CAL: AIC, BIC, EDF, deviance explained) and spatial uncertainty distribution (GEM: 5-point quantile summary of the per-pixel BSE raster) for every phase × year combination. Closes the gap between the published validation metrics (VAM) and the underlying model behaviour. RO-Crate metadata file renamed and restructured. The previous PHASE_202_metadata.json is now ro-crate-metadata.json (the canonical RO-Crate 1.2 filename), so the dataset is discoverable by RO-Crate consumers (Describo, ro-crate-py, FAIR validators). Concept DOI used as the dataset identifier. The root identifier now points to the all-versions DOI 10.5281/zenodo.19571847 rather than a version-specific DOI, so citations remain stable across releases. Validator-clean metadata. Resolves the remaining DCAT 3 type-range warnings flagged by NovaCrate (dct:creator → dct:Agent, dct:publisher → dct:Agent, dct:license → dct:LicenseDocument, dct:spatial → dct:Location), removes the prov:Activity type label whose schema is not resolvable (provenance properties continue to work via the prov: prefix in @context), and adds explicit context mappings for wasGeneratedBy and spatialResolution. Produced by PhenoPhaseR v1.3.0. First dataset release built from the publish-hook output of PhenoPhaseR's build_phase_cog_ro_crate().面向冬小麦（普通小麦，*Triticum aestivum*，JKI/DWD作物编码202）的专属物候期（PHASE）数据集。 数据集内容：共包含35个数据文件与1个元数据文件： DOY COGs（7个文件）：各物候期的插值日序值。采用多波段云优化GeoTIFF（Cloud Optimized GeoTIFF，简称COG）格式，包含32个波段（对应1993–2024年），数据类型为Int16，空间分辨率1 km，投影坐标系为EPSG:25832。 BSE COGs（7个文件）：BAM后验标准误差（不确定性度量）。采用与DOY COGs一致的多波段结构，数据类型为Float32，数值保留两位小数。 VAM CSVs（7个文件）：留出法验证精度指标。模型基于75%的观测站点训练集拟合BAM，在剩余25%的测试站点上进行评估。采用宽格式CSV存储，每年对应一行（1993–2024年共32行），字段包括：PLANT、PHASE、YEAR、TN、ON、VN、METHOD、BAM_K、RMSE、MAE、MSE、R2。 CAL CSVs（7个文件）：基于拟合完成的BAM对象（mgcv::bam）计算得到的样本内校准/模型拟合诊断指标，用于评估训练集上的模型复杂度与拟合优度，作为留出法验证指标（VAM）的补充。采用宽格式CSV存储，每年对应一行（共32行），字段包括：PLANT、PHASE、YEAR、BAM_K、AIC、BIC、EDF、DEV_EXPLAINED。 GEM CSVs（7个文件）：逐像素BSE栅格的空间分位数数据，即德国全境各年份预测不确定性的全局分布情况，可用于下游应用中的不确定性传递分析。采用长格式CSV存储，每年对应5行（1993–2024年共160行），字段包括：YEAR、Quantile、Value，分位数层级为0%、25%、50%、75%、100%。 ro-crate-metadata.json：采用RO-Crate 1.2标准的溯源文档，将所有文件与经过筛选与优化的物候观测形状文件（shapefile）以及PhenoPhaseR软件相关联。内嵌基于W3C标准的元数据（DCAT 3、通过SKOS桥接至ISO 19157-1的DQV、PROV-O、都柏林核心元数据术语（Dublin Core Terms））。 文件命名规则：DOY_202-{PHASE}.tif、BSE_202-{PHASE}.tif、VAM_202-{PHASE}.csv、CAL_202-{PHASE}.csv、GEM_202-{PHASE}.csv。涵盖的物候期编号为10、12、15、18、19、21、24。 多波段COGs的波段名称以年份编码，可在R语言中通过`terra::rast(file, lyrs = as.character(year))`按语义选取波段，或通过`gdal_translate -b N`根据对应波段索引选取。 本数据集基于经过筛选与优化的物候观测形状文件（doi:10.5281/zenodo.19483112），通过PhenoPhaseR v1.3.0生成。 v1.2.0版本更新内容： 1. 新增CAL与GEM诊断表：为每个物候期×年份组合新增14个CSV文件，包含模型拟合诊断指标（CAL：AIC、BIC、EDF、偏差解释度）与空间不确定性分布（GEM：逐像素BSE栅格的5分位数汇总），填补了已发表的验证指标（VAM）与底层模型行为之间的信息缺口。 2. RO-Crate元数据文件重命名并重构：原PHASE_202_metadata.json现已更名为ro-crate-metadata.json（RO-Crate 1.2标准的规范文件名），使数据集可被RO-Crate工具（如Describo、ro-crate-py、FAIR验证器）识别。 3. 使用概念DOI作为数据集标识符：根标识符现已指向全版本DOI 10.5281/zenodo.19571847，而非特定版本的DOI，确保不同版本引用的稳定性。 4. 元数据通过验证器校验：修复了NovaCrate标记的DCAT 3类型范围警告（如dct:creator → dct:Agent、dct:publisher → dct:Agent、dct:license → dct:LicenseDocument、dct:spatial → dct:Location），移除了模式无法解析的prov:Activity类型标签（溯源属性仍可通过@context中的prov:前缀正常使用），并为wasGeneratedBy与spatialResolution添加了显式上下文映射。 本数据集由PhenoPhaseR v1.3.0生成，为首次发布的数据集，基于PhenoPhaseR的build_phase_cog_ro_crate()函数的发布钩子输出构建。