openwatcherhub

This is a proof of concept for a locally hosted version of the featureset SentinelHub has, the product that is being pushed as the easiest way to work with Sentinel data.

The project attempts to reimplement certain interfaces of SentinelHub's APIs, and provide a somewhat equivalent feature-set.

The project currently has two main components:

• copernicus-dataspace-s3-cache
• openwatcherhub-api

copernicus-dataspace-s3-cache

A modified nginx-s3-gateway docker image configured to download and cache Sentinel satellite bands from the Copernicus S3 repository. The local caching helps keep the bandwidth usage down, and enables fast reprocessing of the same bands.

You will need S3 credentials from Copernicus Dataspace to configure the container properly.

Copy example.env to prod.env and fill in the values from their website.

openwatcherhub-api

This is currently an all-in-one API service that can respond to certain SentinelHub requests. It is designed without ANY kind of security, and is only suitable for local installations.

It is also terribly inefficient (although now less inefficient). It will be rewritten into a proper rendering pipeline and only then will performance be taken into account.

Currently supported

• /api/v1/process endpoint
• basic evalscript, only older ecmascript support (no arrow functions etc)
  • supports numpy vectorized functions
• single input, band selection
• single output to tiff only
• only nearest resampling
• sentinel2 1c and 2a products, no cloud coverage masks (they show up in code but not computed)
• no mosaicking, just order by cloud coverage and pray

evalscript is implemented by transpiling JavaScript to Python (partially) and executing it through python exec

evalscript with numpy

To use numpy in evalscript, the directive //VECTORIZE must be present at the top of the file after version. After that, each band exposed in a sample will represent ALL pixels for that band, and vectorized math can be used to quickly process the bands. Additionally, a new module is exposed called Vec, which contains all the functions supported by numpy for vectorized computations.

Running this locally

docker compose is the easiest way for now. Make sure to change (and NOT commit) the s3 access key and secret.

Pointing SentinelHub code to this instance

Start by pointing the SHConfig base url to the instance:

config = SHConfig()
config.sh_client_id = "DENADA"
config.sh_client_secret = "NOTHING"
config.sh_token_url = ""
config.sh_base_url = "http://127.0.0.1:8081"

Then set the URL for the specific data collection for a request:

request_true_color = SentinelHubRequest(
    evalscript=evalscript_true_color,
    input_data=[
        SentinelHubRequest.input_data(
            data_collection=DataCollection.SENTINEL2_L2A.define_from(
                name="s2", service_url="http://127.0.0.1:8081"
            ),
            time_interval=("2023-07-01", "2023-07-20"),
            other_args={"dataFilter": {"mosaickingOrder": "leastCC"}},
        )
    ],
    responses=[SentinelHubRequest.output_response("default", MimeType.TIFF)],
    bbox=aoi_bbox,
    size=aoi_size,
    config=config,
)

Because this system doesn't use OAuth2 at all, you must trick SentinelHub python library to work without it.

Add this snippet somewhere at the top of your file/notebook, which patches the SentinelHubSession to not attempt to fetch an auth token.

from sentinelhub.download.session import SentinelHubSession
SentinelHubSession._collect_new_token = lambda self: {'access_token': "herpderp", "expires_at": 0}

Full example available in openwatcherhub_test.ipynb.