[general]
name=02EarthEmbeddings - Earth Embeddings Lab
qgisMinimumVersion=3.28
qgisMaximumVersion=4.99
description=Map with Google AlphaEarth satellite embeddings in QGIS: read the public 64-band annual embeddings without an account, then run similarity search, label-free clustering, few-label classification with spatial block cross-validation and an area of applicability, design-based accuracy and area estimates, multi-year change and polygon embeddings, with an HTML report.
version=0.1.1
author=Geophilo
email=yusuf.eminoglu@deu.edu.tr
about=02EarthEmbeddings goes from the AlphaEarth Foundations Satellite Embedding dataset (annual, 2017 to 2025, 10 m, 64 channels, CC BY 4.0) to an answer you can defend, in one dock panel and in Processing, with no extra installs.
    Data: the public Cloud-Optimised GeoTIFFs are read anonymously from Google Cloud or the Source Cooperative mirror (or from a local folder). Bottom-up tiles are turned north-up, de-quantised and mosaicked across UTM zones into one working CRS in metres; an estimate of pixels, download size and tiles is shown before anything is fetched.
    Similarity: example-based search from points, polygons or map clicks (mean prototype, maximum or contrast with negative examples), Otsu, top-percent or fixed thresholds, a minimum mapping unit and ranked region polygons.
    Clusters: spherical k-means with k-means++ seeding, the number of clusters chosen from silhouette, Calinski-Harabasz and Davies-Bouldin curves, cluster cards with share, area, coherence and the nearest cluster.
    Classification: k-nearest neighbours, nearest prototype and logistic regression in pure numpy (random forest when scikit-learn is installed), spatial block cross-validation beside random cross-validation so the optimism gap is visible, quantity and allocation disagreement, and an area-of-applicability mask.
    Accuracy and area: stratified random sample design sized for a target precision, a labelling helper and Olofsson et al. (2014) area estimates with 95 percent confidence intervals.
    Change: distances between years, calibrated thresholds, change year and change regions. Zones: polygon embeddings with coherence and similarity to a reference zone for planning units.
    Report: one self-contained HTML file with provenance, attribution, every formula and reference.
    The AlphaEarth Foundations Satellite Embedding dataset is produced by Google and Google DeepMind. Only qgis.core, GDAL, numpy and PyQt are required.
    Manual: https://geophilo.com/zero2earthembeddings/
repository=https://gitlab.com/geophilo1/zero2earthembeddings
tracker=https://gitlab.com/geophilo1/zero2earthembeddings/-/work_items
homepage=https://geophilo.com/zero2earthembeddings/
category=Raster
tags=alphaearth,satellite embeddings,embeddings,foundation model,google deepmind,remote sensing,land cover,classification,clustering,similarity search,change detection,accuracy assessment,area estimation,olofsson,spatial cross-validation,area of applicability,cloud optimized geotiff,geoai,raster,geophilo
icon=icons/icon.png
experimental=False
deprecated=False
hasProcessingProvider=yes
license=GPL-2.0-or-later
changelog=
    0.1.1 - Charts and themes: report line-chart legends sit in their own band above the plot and wrap, x labels no longer collide; under the QGIS Night Mapping and Blend of Gray themes the rail buttons lose the grey boxes, table rows and combo box text are readable, and empty chart placeholders wrap instead of being clipped in a narrow dock. Enum names in their scoped form only, so the Hub Qt6 Check passes.
    0.1.0 - First release. Anonymous access to the AlphaEarth annual embeddings (Google Cloud, Source Cooperative mirror or a local folder) with a download estimate, north-up mosaics across UTM zones and a PCA colour view; similarity search, spherical k-means clustering with model-selection curves, few-label classification with spatial block cross-validation and an area of applicability, design-based accuracy and area estimates, multi-year change, polygon embeddings, a synthetic demo cube and a self-contained HTML report.
