taco

Read and write TACO datasets in Python.

pip install taco-eo
python examples/minimal.py

Published wheels include the native TACO reader. Building from the source distribution requires a C++23 compiler, CMake, Ninja, pkg-config, libcurl 7.83 or newer, and OpenSSL 3 or newer.

import taco

samples = taco.read("dataset.zip")
parts = taco.read(["part-0.zip", "part-1.zip"])

dataset = taco.open_dataset("dataset.zip")
targets = dataset.read(files="target.tif")
train = dataset.sql("SELECT * FROM dataset WHERE \"ml:split\" = 'train'")

export() writes the complete samples selected by a SQL query. The query may use dataset, sample, or any declared metadata level. A match at a lower level still copies the whole sample. Collection fields are inherited unless they are replaced. For a remote source, only payloads from matching samples are downloaded. Pass overwrite=True to replace an existing TACO output.

source = "https://data.source.coop/major-tom/core-dem/"
taco.export(
    source,
    "core-dem-sample.zip",
    sql='SELECT * FROM sample ORDER BY id LIMIT 10',
)

Remote reads and exports show download progress in interactive terminals. Writers show their build progress when opened with progress=True.

GeoEnrich

GeoEnrich uses the public 10 km MajorTOM index on Source Cooperative by default, so it needs no Earth Engine account. Place MajorTOM(dist_km=10) in the same metadata level before using it:

taco.Level(
    "sample",
    stac=taco.extensions.STAC(),
    majortom=taco.extensions.MajorTOM(dist_km=10),
    geoenrich=taco.extensions.GeoEnrich(
        ["elevation", "temperature", "admin_countries"],
    ),
)

Set backend="earthengine" explicitly to retain centroid-based Earth Engine sampling. That backend requires taco-eo[geoenrich] and an authenticated Earth Engine installation.

Examples

Every example is self-contained, uses synthetic data, and writes its output in the current directory.

Spatial and temporal metadata use separate profiles: Spatial for regular spatial grids, ISpatial for irregular footprints, and Temporal for time alone. STAC combines regular spatial + temporal metadata; ISTAC combines irregular spatial + temporal metadata.