[general]
name=GLUB
qgisMinimumVersion=3.28
qgisMaximumVersion=4.99
supportsQt6=True
description=Seagrass and shallow seabed mapping: Sentinel-2 or Landsat, water-column correction and supervised classification with a depth mask.
version=1.0.2
author=Daniel Ibarra-Marinas, Alejandro Fenollar-Rueda, Ana Mónica de Jhesú García-García, Ángela Bellido-Solano, Dulce Mata-Chacón, Marta Serrano-Vicente, Arturo Mora-Olivo
email=daniel.ibarra@uat.edu.mx
about=GLUB (GIS Looking Under the Blue) maps shallow benthic habitats (seagrass, sand, rock...) from satellite imagery (Sentinel-2, Landsat and others). It downloads Sentinel-2 imagery from the Copernicus Data Space Ecosystem, prepares masked surface reflectance from Sentinel-2 L2A or Landsat 4-9 Collection 2 Level-2 (or imports ACOLITE), or any multiband reflectance raster (PlanetScope, drones), removes sunglint (Hedley), builds median composites of several dates (with a water-clarity ranking of the scenes), computes the Lyzenga depth-invariant bottom index and classifies the seabed with your training points or polygons (Random Forest or maximum likelihood) or, without samples, into k-means groups that you name afterwards, optionally with bottom texture (local standard deviation) as extra features. A three-state depth mask (bottom visible, bottom not visible, too deep for seagrass) uses a bathymetry such as StarShoal's for the optical limit and an independent bathymetry for the ecological limit. Validation with spatial blocks by class or common to all classes; HTML report with confusion matrix, producer's and user's accuracy, F1, accuracy by depth and Olofsson et al. (2014) error-corrected areas. An optional minimum probability leaves doubtful pixels unclassified. A validation tab draws stratified random validation points by map class and assesses the map with them once labelled (area-weighted accuracy and corrected areas with 95-percent intervals). A change tab compares two class maps of the same place (transition map and matrix, gains and losses, only where both dates see the bottom). Class and change maps can be exported to polygons with class, area and perimeter. With Sentinel-2 the realistic target is seagrass / not seagrass in clear, shallow water. One window with tabs, in Spanish and English; every tool is also in the Processing Toolbox. Dependencies: GDAL and numpy (shipped with QGIS); Random Forest also needs scikit-learn (on Windows, from the OSGeo4W Shell: python -m pip install scikit-learn); without it, maximum likelihood is used. Downloads need a free Copernicus Data Space Ecosystem account.
tracker=https://github.com/adanielibarra/glub/issues
repository=https://github.com/adanielibarra/glub
hasProcessingProvider=yes
changelog=1.0.2: unsupervised classification (k-means groups, numpy only) in tab 6 and in Processing, with the same features and depth mask as the supervised one; a 'none' choice for the green, red and NIR bands and for the optional layers of tab 6; one switch to use or not the reflectance bands instead of a tick box per band; 'training samples' instead of 'reference data'; updated quick guide. 1.0.1: passes the plugins.qgis.org security scan (no silently ignored errors), new icon. 1.0.0: first release, built on StarShoal 1.0.0 (download, preparation, ACOLITE import, Hedley sunglint); median composite; Lyzenga depth-invariant index with a floor at deep-water noise (no holes, floored pixels flagged); supervised classification (Random Forest, maximum likelihood) with a three-state depth mask and a per-pixel bottom-signal test, spatial-block validation by class, accuracy by polygon and by pixel, class balance off by default, strip-by-strip processing (a whole tile fits in memory), date checks in composites, polygon shrinking, optional majority filter and minimum mapping unit, common spatial blocks, log and report in Spanish or English, QGIS 4 enum names, an accuracy report with error-corrected areas, an optional minimum probability (low-confidence pixels left unclassified), stratified random validation points and map assessment with them, a water-clarity ranking of the scenes, optional bottom texture (local sd) as features, Landsat 4-9 Collection 2 Level-2 import, any reflectance raster (PlanetScope, drones), export of class and change maps to polygons, simple and advanced mode, help buttons that open the manual at the right section, save and load of the settings of every tab, change between two class maps, citation files (CITATION.cff, .zenodo.json), and automatic depth sign from the values for rasters without StarShoal metadata
tags=landsat,change detection,seagrass,posidonia,benthic habitat,seabed,classification,sentinel-2,remote sensing,coastal,ocean,water column,accuracy assessment,validation
homepage=https://github.com/adanielibarra/glub
category=Raster
icon=icon.png
experimental=False
deprecated=False
