High-performance Python STAC tooling, backed by Rust


🐍 🤝 🦀

Pete Gadomski is a geospatial engineer at Development Seed

STAC

Usage

STAC's sweet spot is L1 to L2

— me

Issues

Original libraries (e.g. pystac) must move slowly

Building for

data providers 👩‍🍳 developers 💁‍♂️ data users 😋

requires different strategies and methods

Sometimes (but not always) performance matters

            
$ curl -s https://landsatlook.usgs.gov/stac-server/search | jq .numberMatched
45955446

"Strategy"

  1. Build lots of small, bespoke libraries
  2. Stitch them together
  3. ...
  4. Don't profit?! (it's open source, after all)

stac-geoparquet

Format Size
json 21 MB
json.gz 614 kB
parquet 488 kB
parquet (compressed) 179 kB

1000 sentinel-2 items

stacrs

            
              $ python -m pip install stacrs
              

items = stacrs.search(
    "https://landsatlook.usgs.gov/stac-server",
    collections="landsat-c2l2-sr",
    intersects={"type": "Point", "coordinates": [-105.119, 40.173]},
    sortby="-properties.datetime",
    max_items=1000,
)
stacrs.write("items.parquet", items)

gadomski/stacrs

stacrs-cli

            
              $ python -m pip install stacrs-cli
              $ stacrs search https://landsatlook.usgs.gov/stac-server/search --max-items 1000 \
                | stacrs translate - items.parquet

stac-utils/stac-rs

stac-fastapi-geoparquet

👷‍♂️

Coming March 2025...

cql2

we need more CQL2 STAC examples in the wild, feels very sparse - and not even clear on easy way for a user to know if said API supports CQL2

— Alex (@wildintellect.com) January 8, 2025 at 5:19 PM

cql2-rs

CLI


$ echo "(\"landsat:scene_id\" = 'LC82030282019133LGN00')" | cql2 -o json
{"op":"=","args":[{"property":"landsat:scene_id"},"LC82030282019133LGN00"]}

Python


from cql2 import Expr

expr = Expr("landsat:scene_id = 'LC82030282019133LGN00'")
print(expr.to_json())
print(expr.to_sql())

Implementations

Backend Supported?
Postgres (pgstac) ✅
Elasticsearch ✅
DuckDB 👷

DuckDB

🦆

stacrs (again)

CLI

            
              $ stacrs search items.parquet --max-items 1000

Python

            
from stacrs import DuckdbClient

client = DuckdbClient()
item_collection = client.search("items.parquet", ...)

STAC API search

"Lingua franca"

  • Datetime(s)
  • Geometry (or bbox)
  • cql2
  • sort
  • fields

Putting it together

Repo Description
developmentseed/cql2-rs Rust library and Python bindings for parsing the OGC CQL2 filter language
geoarrow/geoarrow-rs GeoArrow in Rust, Python, and JavaScript (WebAssembly) with vectorized geometry operations
object_store A focused, easy to use, idiomatic, high performance, async object store library for interacting with object stores.
stac-utils/stac-rs Tools and libraries for the SpatioTemporal Asset Catalog (STAC) specification, written in Rust
gadomski/stacrs A small no-dependency Python package for STAC, using Rust under the hood
stac-utils/pgstacrs Python async API for pgstac, backed by Rust
stac-utils/pystac Python library for working with any SpatioTemporal Asset Catalog (STAC)
stac-utils/pystac-client Python client for searching STAC APIs
developmentseed/obstore The simplest, highest-throughput interface to Amazon S3, Google Cloud Storage, Azure Storage, and S3-compliant APIs like Cloudflare R2
stac-utils/stac-fastapi-pgstac PostgreSQL backend for stac-fastapi using pgstac (https://github.com/stac-utils/pgstac)

Rust and Python

🐍 🤝 🦀

Closing thoughts

Fin

Thank you for your time.

Github | Bluesky | These slides