I have a question regarding the batch API and its flexibility:
I am currently testing your excellent eo-learn API to look into roughly 100k scattered field parcles over a larger region (each parcel has a different time period of interest). The final aim with this data is to perform a crop type classification. So currently my eopatches are single field parcels and each eopatch has a different start- and enddate as a time interval (the length of the interval is however the same).
I am interested in implementing/downlading this with the batch API to speed things up and also potentially lower costs, BUT: Is this use case possible with the batch API? As far as I could see, one can only provide a general bounding box, a general time interval, and a predefined tiling grid. Subsequently all data for this bounding box and dates are downloaded as tiles based on this grid. Or is it possible to download custom tiles (e.g. field parcels?) and provide a different timeinterval for each of them?
gridding system is a necesseity for Batch processing, so it does not make sense in your case. Working with parcel-size patches it is probably better to use process API for generation of patches, as described in this Jupyter Notebook.
That being said, if you are planning to perform object based crop classification, we would recommend to take a look at the Statistical API. We are working, at this moment, on the next version of this API, which will as well be available in “Batch mode”. It will still take some time to have it available for the public, but we are happy to provide you an early access, when available, to give it a try. Again, this only makes sense if you are planning to do ML modelling on statistical data, rather than the original raster data.
Dear @gmilcinski, thank you for you swift and detailed response!
I will keep playing around with the process API for my use case then and also explore FIS along the way.
I will be glad to come back to your offer regarding the early access FIS batch version if needed too!