Application package patterns based on data stage-in and stage-out behaviors commonly used in EO workflows

Application packages expose several patterns in terms of data access and result publishing.

This repository contains CWL descriptions implementing these patterns to verification and validation activities

The patterns are:

1. one input/one output

The CWL includes:

• one input parameter of type Directory
• one output parameter of type Directory

This scenario typically takes one input, applies an algorithm and produces a result

Implementation: delineate water bodies using NDWI and Otsu automatic threshold taking as input a Landsat-9 acquisition

2. two inputs/one output

The CWL includes:

• two input parameters of type Directory
• one output parameter of type Directory

This scenario typically takes as input one pre-event acquisition, one post-even acquisition, applies an algorithm and produces a result

Implementation: delineate water bodies using NDWI and Otsu automatic threshold taking as input Landsat-9 acquisitions

3. scatter on inputs/one output

The CWL includes:

• scatter on an input parameter of type Directory[]
• one output parameter of type Directory

This scenario typically takes as input a stack of acquisitions, applies an aggregation algorithm and produces a result

Implementation: process the NDVI taking as input a stack of Landsat-9 acquisitions producing a STAC Catalog with n STAC Items

{
    "water_bodies": [
        {
            "value": "s3://results/pattern-3-9451lj/catalog.json",
            "type": "https://raw.githubusercontent.com/eoap/schemas/main/string_format.yaml#URI"
        },
        {
            "value": "s3://results/pattern-3-ezt7al/catalog.json",
            "type": "https://raw.githubusercontent.com/eoap/schemas/main/string_format.yaml#URI"
        }
    ]
}

4. one input/two outputs

The CWL includes:

• one input parameter of type Directory
• two output parameters of type Directory

This scenario takes as input an acquisition, applies two algorithms and generates two outputs.

Implementation: process the NDVI and NDWI taking as input a Landsat-9 acquisition. The output include two STAC Catalogs, each with a single STAC Item

5. one input/scatter on outputs

The CWL includes:

• one input parameter of type Directory
• scatter on an output parameter of type Directory[]

This scenario takes as input an acquisition, applies an algorithm and generates several outputs

Implementation: process the NDVI and NDWI taking as input a Landsat-9 acquisition and generating a stack of STAC Catalogs

6. one input, no output

The CWL includes:

• one input parameter of type Directory
• there are no output parameters of type Directory

This corner-case scenario takes as input an acquisition, applies an algorithm and generates an output that is not a STAC Catalog

Implementation: derive the NDVI mean taking as input a Landsat-9 acquisition

Output:

{
    "mean": 0.07260978
}

7. optional inputs, one output

The CWL includes:

• one optional input parameter of type Directory?
• one output parameter of type Directory

This scenario may take as input an acquisition, an optional input, applies an algorithm and generates an output

Implementation: process the NDVI taking as input Landsat-9 acquisitions

8. one input, optional output

The CWL includes:

• one input parameter of type Directory
• one output parameter of type Directory?

This scenario takes as input an acquisition, applies an algorithm and may or may not generate and output

Implementation: process the NDVI taking as input a Landsat-9 acquisition with a parameter to create or not the output

9. one input, optional outputs

The CWL includes:

• one input parameter of type Directory
• output parameter of type Directory[]?

This scenario takes as input an acquisition, applies an algorithm and may or may not generate outputs

10. multiple inputs, multiple outputs

The CWL includes:

• input parameter of type Directory[]
• output parameter of type Directory[]

This scenario takes as input an array of acquisition, applies an algorithm to each of them.

Typed CWL development

The feature checkout uses CWL v1.2 with embedded URI, bounding box, CRS, spectral band and vegetation index definitions. Existing shared OGC and GeoJSON schemas remain explicit imports. Metadata and port documentation are validated by python scripts/cwl-docs.py --validate --output /tmp/cwl-docs.

The lessons are Markdown pages backed by twelve executable CWL examples. eoap-cwlwrap composes each example with the mastering stage-in/out commands; wrapper-introduced URI definitions are embedded in the resulting standalone document. No CWL packing step is used.

Install and validate the release environment:

uv venv --python 3.12 .venv
uv pip sync --python .venv/bin/python --require-hashes requirements-notebooks.txt
source .venv/bin/activate
task test:contracts
task wrap

Run all twelve patterns and four optional-input/output cases against real Landsat and DEM data using the application-package-patterns module in dev-platform-eoap. Its task labs uses Calrissian on Kubernetes and the deployed SeaweedFS S3 endpoint. For a local Podman runner, configure PATTERNS_S3_ENDPOINT, prepare the bucket with python scripts/prepare-s3.py, then run python scripts/run-acceptance.py --runner cwltool.

The acceptance runner verifies published STAC links, readable raster assets, coordinate reference systems and intersection with the requested area. Pattern 6 returns a scalar normalized-index mean; its acceptance check verifies a finite value within [-1, 1]. Evidence includes execution logs, outputs and scientific checks for each case.

The staging provenance and immutable GHCR image references are recorded in mastering-staging.lock.json. Deployment uses GitHub resources and published containers; it requires no sibling host checkout.