Landsat-based cloud-free mosaics covering small island developing states across the world. Coverage is annual (using up to three years of data) and for all years from 2020-2025.
Annual cloud-free 30 m Landsat mosaics for Small Island Developing States (SIDS) and Pacific Community (SPC)'s countries/territories for 2000-2025, generated using robust pixel compositing methods for long-term environmental monitoring, land cover mapping, coastal change detection, and machine learning applications.
Created by: Auspatious
Code: https://github.com/auspatious/ldn-lulc
License: Creative Commons BY
Visualisation app: https://mmufb4pjqf.execute-api.us-west-2.amazonaws.com/
Coverage includes:
Many countries/territories belong to both of these groups. In that case, they are placed in the Pacific region.
ASMCOKFJIPYFGUMKIRMHLFSMNRUNCLNIUMNPPLWPNGPCNWSMSLBTKLTONTUVVUTWLFAIAATGABWBHSBRBBLZBMUVGBCPVCYMCOMCUBCUWDMADOMGRDGLPGNBGUYHTIJAMMDVMTQMUSMSRPRIKNALCAVCTSYCSGPSXMSURSTPTLSTTOTCAVIRTiling is done using ODC Gridspec to partition the data into spatial chunks for processing. Two gridspecs are used because the Pacific region straddles the antimeridian, which would cause a single grid in EPSG:6933 to break at that boundary.
Output files are Cloud-Optimized GeoTIFFs structured by grid index and year (x/y/year). Example:
https://source.coop/auspatious/geomad-sids/ci_ls_geomad/0-2-1/118/125/2000/ci_ls_geomad_118_125_2000_bcmad.tif
All outputs for a tile/year are included in a STAC item JSON e.g.
https://data.source.coop/auspatious/geomad-sids/ci_ls_geomad/0-2-1/118/125/2000/ci_ls_geomad_118_125_2000.stac-item.json
A cloud-free annual surface reflectance composite generated using the geometric median of all valid observations. Unlike independent per-band median compositing, the geometric median preserves spectral relationships between bands, producing more physically realistic reflectance values.
See more technical details here.
A robust measure of temporal variability useful for quantifying uncertainty, identifying unstable surfaces, detecting environmental change, and supporting machine learning workflows.
See more technical details here.
A count band is also included, recording the number of valid observations used per pixel.
These products are designed for change detection and land surface analysis. Auspatious uses them to classify land use/land cover across SIDS and Pacific countries/territories, with annual outputs feeding into Land Degradation Neutrality calculations for UN SDG Indicator 15.3.1 (proportion of land that is degraded over total land area).
Landsat Collection 2 Tier 1 data is preferred. Where insufficient timesteps are available, Tier 2 data is included, and if still insufficient, data from ±1 year is incorporated. A small number of tiles remain incalculable even after these relaxations.
Cloud masking is aggressive - missing pixels are preferred over cloudy outputs. Masked layers include cloud, dilated cloud, cirrus, cloud shadow, and snow (flagged from qa_pixel). Additional custom whiteness and blueness indices catch residual unmasked cloud. Morphological filters clean up the mask and remove small isolated pixel groups surrounded by cloud. Saturated pixels are masked using qa_radsat (although saturation is much less impactful than cloud).
Snow is included as a mask because of its unlikelihood to be present in this AOI, and its common presence in the input data.
Some areas remain partially incomplete. Cloud-persistent regions such as Fiji can lack sufficient cloud-free observations in certain years even after broadening the input data window.
Data is distributed through Source Cooperative: https://source.coop/auspatious/geomad-sids. The 2 regions are subfolders.
A combined index for both regions is available at:
https://data.source.coop/auspatious/geomad-sids/ls_geomad/0-2-1/ls_geomad.parquet
Queryable directly - no need to download the full catalog. Within seconds you can search (spatially and temporally), load, and visualise the data using the following Python code.