How to filter Open Data Cube satellite data by cloud coverage similar to the Sentinel API?

Question

I'm planning to use Sentinel API for a small region and have a time-space dataset filtered out by a cloud coverage percentage.

Does someone know if I can do the same with Open Data Cube? i.e. when indexing a product like s2_l2a

Answer 1

Yes, once you have Sentinel-2 data indexed into an Open Data Cube instance you can use some existing tools to filter by cloud percentages: for example, the load_ard function from dea-tools allows you to specifiy a min_gooddata percentage which will return only satellite observations with at least X% of non-cloudy pixels in your region. There's some nice worked examples here! https://github.com/GeoscienceAustralia/dea-notebooks/blob/develop/Frequently_used_code/Using_load_ard.ipynb

For example, to filter to only images with less than 1% cloud:

!pip install --extra-index-url="https://packages.dea.ga.gov.au" git+https://github.com/GeoscienceAustralia/dea-notebooks.git#subdirectory=Tools

import datacube
from dea_tools.datahandling import load_ard

# Connect to datacube
dc = datacube.Datacube(app='Using_load_ard')

# Create a spatiotemporal query
query = {
    'x': (153.38, 153.47),
    'y': (-28.83, -28.92),
    'time': ('2019-01', '2019-05'),
    'measurements': ['nbart_green'],
    'output_crs': 'EPSG:3577',
    'resolution': (-30, 30),
    'group_by': 'solar_day'
}

# Load available data filtered to 99% clear observations
ds_noclouds = load_ard(dc=dc,
                       products=['s2a_ard_granule'],
                       min_gooddata=0.99,
                       **query)

All of those examples were written for the Digital Earth Australia datacube, so you might need to edit the source code a little depending on the structure of your indexed Sentinel-2 data - the key parts are here where the function calculates the percentage of cloudy pixels in each image: https://github.com/GeoscienceAustralia/dea-notebooks/blob/develop/Tools/dea_tools/datahandling.py#L351-L365

Another thing to try is checking out the same function on the Digital Earth Africa repo, which I think already supports s2_l2a data: https://github.com/digitalearthafrica/deafrica-sandbox-notebooks/blob/253ef52d37e63c98c77f04d8727ba395c496ba55/Tools/deafrica_tools/datahandling.py#L102-L602