freezebase

Lightweight geospatial helpers for reproducible MGRS grids and S3-backed raster workflows.

freezebase grew out of large-scale glacier mapping, where scattered areas of interest require several local projections and large raster collections live in object storage.

The centrepiece of freezebase is MGRSGrid: a 10 km grid for the UTM-covered world that gives every MGRS code a dataset-independent target backed by an odc.geo.GeoBox. Rasters resampled to the same code and resolution share their CRS, transform, and shape. See The MGRS grid for the reasoning.

Installation

Requires Python 3.12 or newer, and uv or pip.

freezebase is on PyPI:

# uv
uv add freezebase

# pip
pip install freezebase

S3 support is optional; see Raster and S3 helpers below.

MGRS quick start

Build the 10 km grid covering an area of interest:

from shapely import box

from freezebase.mgrs import MGRSGrid

grid = MGRSGrid(box(8.4, 47.3, 8.6, 47.5))  # Zürich, WGS84
gdf = grid.to_geodataframe()  # mgrs_code, zone, epsg, geometry, ...
square = grid[0]  # an odc.geo GeoBox per grid square

Each square is an MGRSGeoBox — a subclass of odc.geo.GeoBox carrying the CRS, affine transform, and shape needed to resample any raster onto it. Reconstruct one from its code alone, without building a grid:

from freezebase.mgrs import MGRSGeoBox

square = MGRSGeoBox.from_mgrs("32TMS35", resolution=10.0)  # 1000 x 1000 px, EPSG:32632

Raster and S3 helpers

The raster helpers use Rasterio for local I/O. For object storage, freezebase bridges Rasterio/GDAL with Universal Pathlib, so filesystem operations and raster I/O use the same per-path credentials and endpoint configuration.

Rewrite a local GeoTIFF as a Cloud-Optimized GeoTIFF:

from freezebase.raster import COG_PROFILE, rewrite_tiff

rewrite_tiff("in.tif", "out.tif", profile=COG_PROFILE)  # copy
rewrite_tiff("in.tif", "out.tif", profile=COG_PROFILE, move=True)  # move

The s3 extra

S3 support needs the optional s3fs, fsspec, and boto3 dependencies:

uv add "freezebase[s3]"

Credentials and endpoint configuration travel with each path. Named profiles make it possible to select different stores without attaching raw credential values:

from freezebase.s3 import make_s3_upath

aws_path = make_s3_upath("s3://aws-bucket/data", profile="research")
ceph_path = make_s3_upath(
    "s3://ceph-bucket/data",
    profile="ceph-research",
    endpoint_url="https://objects.example.org",
)

An explicitly selected profile also prevents boto from falling back to ambient AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY values from the shell.

The same configured path works for pathlib-style discovery and Rasterio reads:

from freezebase.raster import rasterio_open

scene = aws_path / "scene.tif"
if scene.exists():
    with rasterio_open(scene) as src:
        data = src.read(1)

Public buckets use anon=True. See Rasters on S3 for authentication patterns, custom endpoints, and S3-to-S3 copies.

Other helpers

Download a file with retries and a progress bar:

from freezebase.download import HTTPDownloader

with HTTPDownloader(auth=("user", "pass")) as download:
    path = download("https://example.org/data.zip", "cache/")  # -> Path

Documentation

Full documentation, including the MGRS grid guide and the S3 raster guide, is at https://lqgentner.github.io/freezebase/.

Support

This is a small internal core library maintained on a best-effort basis. Please file bugs and questions on the issue tracker.

Contributing

git clone https://github.com/lqgentner/freezebase.git
cd freezebase
uv sync --all-extras
uv run pytest

See the contributing guide for the full checks, the S3 integration tests, and how to build these docs.