A brief STAC history

SpatioTemporal Asset Catalog (STAC)

---
config:
    theme: neutral
    look: handdrawn
---
timeline
    2017 : Initial commit to what became the stac-spec repository
            : Sprint 1 (Boulder)
            : v0.1.0
    2018 : Name changed to STAC
            : Sprints 2 (Ft. Collins) and 3 (Menlo Park)
            : v0.4.0 to v0.6.0
    2019 : Sprint 4 and 5 (Arlington)
            : v0.7.0 to v0.8.0
    2020 : Sprint 6 (Virtual)
            : v0.9.0
    2021 : v1.0.0
    2023 : Sprint 7 (Philadelphia)
            : API v1.0.0
    2024 : v1.1.0
    2025 : Sprint 8 (Italy)
            : OGC Community Standard
                        

Cribbed heavily from https://element84.com/geospatial/stac-a-retrospective-part-1/

What the STAC?

STAC entities

Uses of STAC

Based on commonly-used, open source projects such as stac-fastapi

Uses of STAC: search

I need all Landsat images with cloud cover less than 20% over Summit County, Colorado in 2025 for a snow distribution research project

Uses of STAC: full scan

I need every NAIP image to train a model.

Uses of STAC: ingest

I need to update my data store each day with a new analytic product

Uses of STAC: near real-time

I need to be notified as soon as a new weather forecast is available, so I can update my market model

Modes and uses

Cloud-native geospatial metadata

stac-geoparquet specification

Initial commit May 2022 by Tom Augspurger

https://github.com/radiantearth/stac-geoparquet-spec

Uses of STAC: full scan

https://radiantearth.github.io/stac-browser/#/external/planetarycomputer.microsoft.com/api/stac/v1/collections/io-lulc-annual-v02

Data are useful at rest

                            
                                $ rustac search \
                                    https://raw.githubusercontent.com/developmentseed/labs-375-stac-geoparquet-backend/refs/heads/main/data/naip.parquet \
                                    --intersects='{"type":"Point","coordinates":[-105.1019,40.1672]}'
                            
                        

No geospatial moats

                                
                                    D select max(datetime), min(datetime) from 'naip.parquet';
                                    ┌──────────────────────────┬──────────────────────────┐
                                    │      max(datetime)       │      min(datetime)       │
                                    │ timestamp with time zone │ timestamp with time zone │
                                    ├──────────────────────────┼──────────────────────────┤
                                    │ 2022-08-27 10:00:00-06   │ 2019-09-18 18:00:00-06   │
                                    └──────────────────────────┴──────────────────────────┘
                                
                            

Data are smaller

1000 sentinel-2 items

Major downside: inflexible schema

Workarounds:

• DuckDB's union_by_name
• The new variant type
• Just re-write it

Experiments

🧑‍🔬

1. stac-fastapi-geoparquet
2. ITS_LIVE

stac-fastapi-geoparquet

stac-fastapi-geoparquet full-scan

https://github.com/developmentseed/labs-375-stac-geoparquet-backend/

stac-fastapi-geoparquet search

https://github.com/developmentseed/labs-375-stac-geoparquet-backend/

ITS_LIVE

• 10M items as 29GB of JSON in AWS s3
• Heterogeneous spatial distribution
• Full-scan (or at least bulk) requests

ITS_LIVE

Searching by id against ten million items was reduced from 187s to 3s by partitioning and sorting

For more on spatial partitioning, see Enabling Cryo Science at Scale: How ITS_LIVE is Providing a Serverless Analysis Ready Archive of the Future by Joe Kennedy, Luis López, et al

https://www.gadom.ski/posts/stac-geoparquet-organization/
https://stac-utils.github.io/rustac-py/latest/notebooks/its-live/

Large-scale on source.coop

The rise of the machines

Large, statically hosted datasets that rich text descriptions can be quite useful when fed to a robot.

By https://eleven.com.au/futurama-characters.html, Fair use, Link

Tooling

stac-map

Visualize stac-geoparquet in your browser, uses Deck.gl and DuckDB.

https://developmentseed.org/stac-map

stac-map

Download search results as stac-geoparquet, and upload stac-geoparquet from your local filesystem into your browser.

https://developmentseed.org/stac-map

Software ecosystem

Next steps

Extract the spec

https://radiantearth.github.io/stac-geoparquet-spec
https://github.com/radiantearth/stac-geoparquet-spec

More examples and documentation!

TODO @gadomski patterns and recipes for sorting and partitioning

Push to v1

⛵︎

Open questions around a "catch-all" properties column, how we handle assets, and more.

Head on over to the repo's issues to weigh in!