geoai-datacubes: AI-ready multi-mission satellite data cubes for Earth-observation machine learning

geoai-datacubes is an open-source Python pipeline that turns raw multi-mission Earthobservation imagery into AI-ready data cubes for machine-learning and deep-learning workflows.From a single configuration block -a region of interest, a time window, a set of bands -it fetches, co-registers, fuses, and tiles imagery from 26 missions.The direct sensor observations are Sentinel-2 L1C and L2A (Drusch et al., 2012), Sentinel-1 RTC (Torres et al., 2012), Landsat 8/9 C2 L2 (Roy et al., 2014), NAIP, PlanetScope 4-band and 8-band imagery, MODIS reflectance and land-surface temperature, HLS, ALOS PALSAR, Copernicus DEM at 30 m and 90 m, and USGS 3DEP.The derived products are ESA WorldCover (Zanaga et al., 2022), ALOS Forest / Non-Forest, USDA CDL, USGS LCMAP, Impact Observatory annual LULC, JRC Global Surface Water, Hansen Global Forest Change, and Chloris biomass.Per-mission details are in docs/data_layers.md.Four data providers are unified behind one dispatcher -Element 84 Earth Search, Microsoft Planetary Computer, the commercial Planet Orders API, and a direct_http path for non-STAC datasets such as Hansen GFC on Google Cloud Storage.The output is a multi-band GeoTIFF or Zarr cube on a common Universal Transverse Mercator (UTM) grid at a user-specified resolution, ready to be consumed by a PyTorch Dataset that streams training tiles on the fly without ever materialising them to disk.Three self-bootstrapping Colab notebooks ship with the repository as reviewed examples: a grand tour of every mission, an LULC classification notebook that trains and compares logistic regression, random forest (Breiman, 2001), XGBoost (Chen & Guestrin, 2016) and a lightweight U-Net (Ronneberger et al., 2015) against a conditional NDWI (McFeeters, 1996) spectral-index baseline, and an integration notebook that composes geoai-datacubes with the opengeos/geoai modelling library (Wu, 2026) across a multi-AOI Cleveland / Cincinnati / Columbus experiment.The smoke-tests/ folder includes Slurm and Bash scripts for every mission.

Authors

Institutions

Publication Details

Journal
The Journal of Open Source Software
Published
2026-11-13
DOI
https://doi.org/10.21105/joss.11034
Primary Topic
Remote-Sensing Image Classification
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

geoai-datacubes: AI-ready multi-mission satellite data cubes for Earth-observation machine learning

H. J. Hsu, Aswathnarayan Radhakrishnan, Joachim Moortgat, Satyaki Roy Chowdhury et al.
The Journal of Open Source Software
Remote-Sensing Image Classification
article

geoai-datacubes: AI-ready multi-mission satellite data cubes for Earth-observation machine learning

H. J. Hsu, Aswathnarayan Radhakrishnan, Joachim Moortgat, Satyaki Roy Chowdhury, bhavika1512
article en

Abstract

geoai-datacubes is an open-source Python pipeline that turns raw multi-mission Earthobservation imagery into AI-ready data cubes for machine-learning and deep-learning workflows.From a single configuration block -a region of interest, a time window, a set of bands -it fetches, co-registers, fuses, and tiles imagery from 26 missions.The direct sensor observations are Sentinel-2 L1C and L2A (Drusch et al., 2012), Sentinel-1 RTC (Torres et al., 2012), Landsat 8/9 C2 L2 (Roy et al., 2014), NAIP, PlanetScope 4-band and 8-band imagery, MODIS reflectance and land-surface temperature, HLS, ALOS PALSAR, Copernicus DEM at 30 m and 90 m, and USGS 3DEP.The derived products are ESA WorldCover (Zanaga et al., 2022), ALOS Forest / Non-Forest, USDA CDL, USGS LCMAP, Impact Observatory annual LULC, JRC Global Surface Water, Hansen Global Forest Change, and Chloris biomass.Per-mission details are in docs/data_layers.md.Four data providers are unified behind one dispatcher -Element 84 Earth Search, Microsoft Planetary Computer, the commercial Planet Orders API, and a direct_http path for non-STAC datasets such as Hansen GFC on Google Cloud Storage.The output is a multi-band GeoTIFF or Zarr cube on a common Universal Transverse Mercator (UTM) grid at a user-specified resolution, ready to be consumed by a PyTorch Dataset that streams training tiles on the fly without ever materialising them to disk.Three self-bootstrapping Colab notebooks ship with the repository as reviewed examples: a grand tour of every mission, an LULC classification notebook that trains and compares logistic regression, random forest (Breiman, 2001), XGBoost (Chen & Guestrin, 2016) and a lightweight U-Net (Ronneberger et al., 2015) against a conditional NDWI (McFeeters, 1996) spectral-index baseline, and an integration notebook that composes geoai-datacubes with the opengeos/geoai modelling library (Wu, 2026) across a multi-AOI Cleveland / Cincinnati / Columbus experiment.The smoke-tests/ folder includes Slurm and Bash scripts for every mission.

The Journal of Open Source SoftwareVol. 11(127)
The Ohio State University (US)
Ohio State University, U.S. Geological Survey
Remote-Sensing Image Classification
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon