capellaInterface

Interface for downloading and performing transformations on Capella SAR datasets.

capella is a CLI + Python library that:

1. Downloads Capella Space SLC (the single Capella product that is slant-plane, complex, single-look) — processing level L1.
2. Derives all higher processing levels locally from that SLC — L2, L2LCR, L2ML.

Capella's other stock products (GEO/GEC/SIDD detected-multilooked, SICD/CPHD) are not used; everything above L1 is built locally from SLC.

Processing levels

Level	Meaning	Built from
L1	slant-plane image (complex, single-look)	Capella SLC — download
L2	ground-plane image (complex, single-look)	L1 — coherent geocoding
L2LCR	ground plane, registered across looks	L1 — sub-aperture looks → geocode → co-register
L2ML	ground-plane, multi-looked + detected	L2LCR — average co-registered looks + detect

Pipeline order: L1 → L2 → L2LCR → L2ML.

Status: the SLC downloader (L1) and search/download CLI are implemented. The SAR processing backend and the L2/L2LCR/L2ML transforms are not yet built — their dependencies (rasterio, sarsen/sarpy, numpy/scipy, geopandas) and a Copernicus DEM input will be added when that work lands.

Setup

Environment is managed with uv. From the repo root:

uv sync --extra dev        # creates .venv, installs editable package + dev deps
uv run capella --help

Quickstart

Discover SLC scenes covering an area (no files written):

uv run capella search --point -121.87085436,37.318451845 --limit 10
uv run capella search --bbox -121.9,37.2,-121.7,37.4 --datetime 2025-08 --json

Guided download (writes the SLC bundle — complex COG + extended-metadata JSON — to <dest>/<item_id>/). Downloads are resumable: a file already complete is skipped; a partial file is continued via HTTP Range; and a dropped connection mid-stream is retried by resuming from the bytes on disk (up to 5 attempts). Re-running download after an interrupted run picks up where it left off.

# non-interactive
uv run capella download --point -121.87085436,37.318451845 --range all --yes --dest ./data

# interactive (prompts for area, destination, range, and confirmation when flags are omitted)
uv run capella download

IEEE Data Contest 2026

SLC search walks the IEEE Data Contest 2026 collection by default alongside the capella-open-data-slc product-type collection. The contest collection ships paired SLC + GEO Spotlight items over a Big Island of Hawaiʻi AOI (~lon −155.3, lat 19.4); only its SLC items are surfaced (GEO pairs are skipped), and acquisitions appearing in both collections are deduped by item.id.

Note (verified against the live catalog): the contest's 791 SLC items are a subset of the 2274 SLC items already in capella-open-data-slc. So contest SLC scenes appear in search results regardless of the contest collection — walking it guarantees coverage if Capella ever publishes contest-only SLC. --no-ieee-contest skips walking the contest collection; it does not remove SLC items, since they are already present in the slc collection. The contest collection's distinct contribution is its GEO pairs, which are skipped for SLC-only download. The contest collection's 791 GEO item links are now skipped without a fetch (id pre-filter), so --no-ieee-contest is mainly a small further opt-out that avoids resolving the contest's 791 SLC candidates.

uv run capella search --bbox -155.4,19.3,-155.1,19.5 --limit 5   # contest AOI SLC scenes
uv run capella search --bbox -155.4,19.3,-155.1,19.5 --limit 5 --no-ieee-contest

CLI reference

capella search [--bbox MINLON,MINLAT,MAXLON,MAXLAT | --point LON,LAT]
               [--datetime 2025-08 | 2025-08-01/2025-08-31]
               [--mode spotlight|sliding_spotlight|stripmap]
               [--limit N] [--json] [--no-ieee-contest]

capella download [--bbox ... | --point LON,LAT]
                 [--datetime ...] [--mode ...]
                 [--range 3-8 | all | 5]
                 [--dest DIR] [--yes] [--no-ieee-contest]
                 [--level L1|L2|L2LCR|L2ML]

download only ever fetches SLC (L1). Calling it with --level L2|L2LCR|L2ML prints the matching local transform command instead of downloading.

Catalog & data facts

• Static STAC catalog (not a STAC API): https://capella-open-data.s3.us-west-2.amazonaws.com/stac/catalog.json
• SLC product-type collection: …/stac/capella-open-data-slc/collection.json
• IEEE Data Contest 2026 collection: …/stac/capella-open-data-ieee-data-contest/collection.json
• Anonymous access; assets live in s3://capella-open-data/data/ (region us-west-2).
• Spatial filtering is client-side (shapely-free bounding-box/point intersection) over both collections. Each collection lists items as flat item links with no bbox/datetime in the link, so the item JSON must be fetched to filter spatially. Search resolves the candidate items concurrently (64 worker threads over a shared keep-alive session) and pre-filters by the item id — which encodes product type (SLC/GEO), instrument mode (SP/SS/SM), and the acquisition start/end timestamps — so contest GEO pairs, wrong-mode items, and out-of-range datetimes are skipped without a fetch. A filtered search (bbox + datetime and/or mode) typically returns in a few seconds; an unfiltered bbox search fetches more candidates and takes longer. --limit N stops early once N matches are found.

Development

uv run pytest -q
uv run ruff check .