Statistical API: Exceeding width and length limits with goal of calculating tile statistics

[guyon.duifhuizen]

Hi team,

I want to calculate the percentage of clouds in a tile.

However, I’ve encountered that the tile’s width exceeds the allowed 2,500 range.
Would the most effective approach be to split the polygons into smaller chunks and process iteratively, or is there a more efficient workaround anyone can suggest on the tile level?

I’m keen to maintain accuracy but also want to avoid excessive overhead if possible.

Thanks in advance for any insights!

[william.ray]

Hi Guyon,

I assume that are using one of the Sentinel-2 data collections. If so, you do not need to calculate the cloud cover percentage yourself. When using Catalog API this is already available as a filter. Below is an example of how to do this using SH PY:

search_iterator = catalog.search(
  DataCollection.SENTINEL2_L2A,  
  bbox=bbox,
  time=('2023-08-13', '2023-09-13'),  
  filter={
    "eo:cloud_cover": {
        "<": 50
    }
},  
)

Let us know if you need some more guidance.

[guyon.duifhuizen]

Super cool!
I’m, however, sourcing from a byoc-collection containing fusion tiles.

I want to count the amount of pixels where a quality layer==0 (clear-pixel).
So perhaps this is not pre-calculated, or do you have additional methods?

[gmilcinski (Grega Milcinski)]

For Statistical API, one simple workaround for “too large areas” is to simply reduce the resolution. I.e. if you change resolution from 3m to 10m, you will be able to process 10-times more data. As the clouds are typically much larger than 3x3 meters (or 10x10m),and noise at the border of the clouds is significant, there should be very little (or none) impact to the results (average of the average is still average).
You might even test with some significant reductions, i.e. 30x30m, so that you get results much faster (and consume less processing units)

[guyon.duifhuizen]

Wow. I love the versatility. Thanks!