{"name": "GEE Time Series Analyzer & Classifier", "package_name": "gee_timeseries_analyzer", "version": "1.13.1", "experimental": true, "qgis_min": "3.22.0", "qgis_max": "3.99.0", "downloads": 30, "uploaded_by": "franz-geoeco", "upload_datetime": "2026-11-04T07:11:10.243373", "changelog": "1.13.1 Security scan fixes: Sentinel-2 metadata parsed with DTDs and entities refused (no XML entity attacks); no silent exception handlers; unused imports removed.\n1.13.0 Spectra: local Sentinel-2 L2A scenes (.SAFE.zip / .SAFE, offline, reflectance offset and SCL mask applied, TCI on the map); samples can be renamed; names of layer features from a chosen column.\n1.12.0 New first section 'Spectra': surface-reflectance spectra of points / polygons from the Sentinel-2 and Landsat 8/9 image closest to a chosen date (worldwide, cloud-aware), charts per sensor or per sample, closest scenes on the map, CSV export.\n1.11.0 Performance settings: CPU cores for the offline engine (Random Forest training / prediction, cross-validation, model application, tree export; remembered; also in the Processing algorithm) and parallel Earth Engine requests (tile downloads, online cross-validation folds).\n1.10.0 Hover info buttons on every group and on the less obvious settings; HTML reports contain no local folder paths; GitHub README with screenshots; build script.\n1.9.0 Renamed to GEE Time Series Analyzer & Classifier; the dock is organised in three sections (Time series, Classification, Export imagery) with a flatter, calmer look; new icon.\n1.8.0 Field-level results (majority class, share, runner-up per polygon); class profiles (mean \u00b1 1\u03c3 per class) in a tab and the report; training check (suspicious / likely mislabelled training polygons, exclude & retrain); spatial k-fold cross-validation; multi-year training by pooling samples of other models; Rotation tab (areas per year, transition matrices, flow diagram, change map, CSV); Processing algorithms: apply model (Earth Engine year / local stack), aggregate to fields, transition matrix.\n1.7.0 Export of all Random Forest / CART trees (Earth Engine or offline) as an interactive HTML viewer: forest overview with variable usage, zoomable tree view with click-to-expand, depth control, variable highlight and class-coloured leaves.\n1.6.0 HTML accuracy report (summary cards incl. CI / kappa scale / balanced accuracy / quantity & allocation disagreement, sortable per-class table with sample-size check, colour-coded confusion matrices, most frequent confusions, charts, settings). Classification plots scale with the number of classes.\n1.5.0 Export tab: composites / time-series stacks with the full recipe (years, season, dekads to whole season, any reducers, gap filling, S2 indices/bands with Cloud Score+ or SCL mask and scene filter, S1), stack / per-period / per-variable files, Float32 or scaled Int16, WorldCover band, RGB quicklooks, live summary of periods and sizes, band metadata and JSON sidecars. The Classify tab shares the same recipe panel; offline classification reads scaled stacks.\n1.4.0 Offline classification: download a feature stack (GeoTIFF with feature names, WorldCover band, recipe) or use any local raster stack; train, assess and classify locally with scikit-learn; models are interchangeable between Earth Engine and offline. Composite export as one multi-band stack.\n1.3.0 Monthly composites: only the first layer is switched on; export of the composites as GeoTIFF (download or Drive) for a polygon layer / shapefile, a drawn rectangle or the map view. 0 line in the index charts.\n1.2.0 Classify tab: supervised crop classification (Random Forest / GTB / CART) from Sentinel-2 + Sentinel-1 time-series composites. Training polygons and classification area from QGIS layers or GEE assets; selectable season, aggregation period, reducers, gap filling, indices / bands, cloud mask, WorldCover mask, sampling and train/test split. Accuracy (OA, kappa, confusion matrices, PA/UA/F1, importance), CSV export, direct GeoTIFF download (class codes, colour table) and Drive / asset export of the map. Models are saved as JSON and can be applied to other years.\n1.1.0 Compare tab (many polygons on shared axes, mean +/- 1 sigma band, outlier ranking, click a line to highlight its polygon on the map and vice versa); clicked fields are kept and cached so only new polygons are queried; previous/next and map-selection switching of the charts\n1.0.0 initial release", "external_deps": null, "download_url": "https://plugins.qgis.org/plugins/gee_timeseries_analyzer/version/1.13.1/download/"}