You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The reason: bufferTiles will only buffer if data is available. We now solve this using 'makeSquareTiles', but that inserts nodata.
Option 1
What we can do to solve this is to make buffer sizes a multiple of tile sizes. Effect should be that we load an extra full tile rather than only the border.
Option 2
Create 'buffered' tiles at load time immediately instead of as an extra step. This would also speed up the whole process?? Caveat is that subsequent processes would need to properly preserve the buffering when transforming tiles, up until the apply_neighbourhood??
Also the 'buffertiles' call returns an RDD that is no longer the of type <SpaceTimekey,MultibandTile>, so it won't be compatible with anything downstream. Also if we would convert back to multi band tile, it will still be of incorrect type...
Workaround option
As a workaround, users can simply increase the extent of the load_collection that needs to be buffered, and then clip away bad data after the apply_neighborhood.
The text was updated successfully, but these errors were encountered:
In option 1, does the load an extra full tile refer to an internal tile (set with the tileSize feature flag) or an actual S2 Tile? (assuming the first but wanted to be sure)
In that case the first option might be the fastest solution with limited performance given that the tileSize is set to a small extent
apply_neighborhood adds nodata at cube borders rather than properly buffering
https://github.com/Open-EO/openeo-geotrellis-extensions/blob/222ff52c8d2146029f3ef6a0847660b3be3bbd54/openeo-geotrellis/src/main/scala/org/openeo/geotrellis/OpenEOProcesses.scala#L1071
The reason: bufferTiles will only buffer if data is available. We now solve this using 'makeSquareTiles', but that inserts nodata.
Option 1
What we can do to solve this is to make buffer sizes a multiple of tile sizes. Effect should be that we load an extra full tile rather than only the border.
Option 2
Create 'buffered' tiles at load time immediately instead of as an extra step. This would also speed up the whole process?? Caveat is that subsequent processes would need to properly preserve the buffering when transforming tiles, up until the apply_neighbourhood??
Also the 'buffertiles' call returns an RDD that is no longer the of type <SpaceTimekey,MultibandTile>, so it won't be compatible with anything downstream. Also if we would convert back to multi band tile, it will still be of incorrect type...
Workaround option
As a workaround, users can simply increase the extent of the load_collection that needs to be buffered, and then clip away bad data after the
apply_neighborhood
.The text was updated successfully, but these errors were encountered: