Australian Coastline Imagery (tech.marksblogg.com)

Digital Earth Australia has published a “Tidal Composites” archive: cloud‑free, low‑ and high‑tide mosaics of Australia’s coasts, estuaries and reefs derived from Sentinel‑2 imagery (2016–2023). The collection (~2.6 TB) is hosted on AWS S3 (dea-public-data) as Cloud‑Optimized GeoTIFFs (COGs) and includes individual single‑band tiles across Sentinel‑2’s 13 spectral bands at ~10 m resolution. That makes a large, ready‑to‑use trove for coastal monitoring, change detection, habitat mapping, spectral‑index research and ML model training—especially for problems that need multispectral inputs or time‑aware labels (high vs low tide). Technically, each COG is a tiled, multi‑resolution TIFF (largest pyramid level 3200×3200, with overviews down to 100×100) using DEFLATE compression and predictor=2, so web apps can efficiently request small byte ranges rather than whole files. The dataset is organized by x/y tiles and year/tide in S3; a 325k‑line JSONL manifest was used to build DuckDB tables for fast metadata queries. Per‑band yearly volumes are roughly 26–35 GB (except 2016), and common tools used to explore it include GDAL, QGIS and DuckDB (with spatial extensions). For the AI/ML community this means immediate access to analysis‑ready, multispectral coastal imagery suited to model training, spectral‑indices extraction, and scalable serverless or cloud‑native workflows.