[general]
name=Terranova
qgisMinimumVersion=4.0
qgisMaximumVersion=4.99
description=Earth-observation toolkit for QGIS: STAC search, classical + deep-learning classification, time-series change detection.
about=Terranova is an open-source QGIS plugin for modern Earth observation: STAC catalogue search and Cloud-Optimised-GeoTIFF download from Planetary Computer / Earth Search / Copernicus Data Space, an embedded interactive map for picking areas of interest and inspecting scene footprints, classical and deep-learning classification, accuracy assessment with PDF and Excel reports, and per-pixel time-series change detection. It offers native Qt dialogs plus a modern React dock panel via QtWebEngine. Requires QGIS 4.0 or newer (built on PyQt6); it is not supported on the QGIS 3.x / PyQt5 series.
    External dependencies (NOT bundled with QGIS): the core features need the Python packages pystac-client, odc-stac, rasterio, rioxarray, scikit-learn and reportlab; the LightGBM and XGBoost classifiers additionally need lightgbm and xgboost; Excel export needs openpyxl; the optional Beta deep-learning features (Prithvi / Clay fine-tuning, SAM segmentation) further require torch, terratorch and segment-geospatial. Install them into the QGIS Python environment — on Windows use the OSGeo4W Shell, e.g. "pip install pystac-client odc-stac rasterio rioxarray scikit-learn reportlab openpyxl". Full per-platform instructions are in the repository README. Any feature whose dependency is missing reports a clear message instead of crashing. Licensed under GPL-3.0-or-later.
version=1.0.3
author=Cole Battell, Arne de Klerk
email=colebattell@gmail.com
homepage=https://github.com/TerranovaEO/terranova
repository=https://github.com/TerranovaEO/terranova
tracker=https://github.com/TerranovaEO/terranova/issues
icon=resources/icon.svg
experimental=False
deprecated=False
hasProcessingProvider=yes
tags=remote sensing, classification, machine learning, deep learning, sentinel, landsat, stac, change detection, time series, segment anything, prithvi
category=Raster
license=GPL-3.0-or-later
plugin_dependencies=
changelog=
    1.0.3 (2026-06)
    - Officially target QGIS 4.0+ only (qgisMinimumVersion raised to
      4.0). The plugin is built on PyQt6 and uses QtWebEngine for its
      map dock; the QGIS 3.x / PyQt5 LTR series is not supported.
    1.0.2 (2026-06)
    - Declare the plugin's external Python dependencies and how to
      install them in the About metadata (per plugins.qgis.org
      guidelines).  No functional change.
    1.0.1 (2026-06)
    - Beta-features toggle: time-series, SAM, foundation models and CDSE
      are hidden by default behind a header switch.
    - Catalogue composite is now discoverable via a checkbox under the
      date range, with a max-images cap.
    - Accuracy: validation-vector vs random-points modes; interactive
      point-by-point labelling pad with class-name persistence;
      metrics shown inline (OA as a percentage) and exported to Excel.
    - Faster, lighter dropdowns on lower-spec / HiDPI machines.
    - Fixes: AOI-rectangle persistence, percentage formatting in PDF
      and Excel reports, plus packaging cleanups.
    1.0.0 (2026-05)
    First public release.
    - STAC catalogue search (Planetary Computer / Earth Search / CDSE) with
      embedded interactive Leaflet map for AOI picking and overlap-aware
      scene-footprint inspection.  Multi-select batch download as COG;
      optional mean/median composite from up to N images.
    - Supervised classification: Random Forest, Extra Trees, Gradient
      Boosting, LightGBM, XGBoost, KNN, Logistic Regression, MLP — each
      with a 2-4 sentence pros/cons hint in the panel.
    - Unsupervised classification: K-Means + ISODATA, fit on a random
      subsample with output as a labelled COG.
    - Accuracy assessment: validation-vector mode (sample raster at every
      pixel covered by a vector) AND random-points mode (generate
      random / stratified / equalized-stratified points; step through
      each one in a labelling pad with auto-pan + zoom-to-pixel +
      class-name persistence; compute confusion matrix straight from
      the labelled points file).  PDF + Excel report with OA / kappa /
      per-class UA / PA / F1; OA shown as a percentage.
    - Beta tabs (off by default): time-series + per-pixel change detection
      (CuSum + BFAST/LandTrendr); SAM segmentation; Prithvi / Clay
      foundation-model fine-tune scaffolding; CDSE OAuth sign-in.
    - Dual UI: native Qt dialogs for compatibility, React-in-QWebEngine
      dock with embedded OpenStreetMap / Esri satellite basemap for
      the primary surface.
    - NDVI Processing algorithm; CI on Linux/Windows/macOS.
