Missing / incomplete tiles where rasters' valid data borders each other

[zstatmanweil]

I am working with rasters that correspond to watersheds and so have valid data within these boundaries and nodata without. When Rasters are float32, 5m resolution. I view them with titiler-pgstac at zoom 11, I have no issues:

When I zoom in, I start seeing oddities like here you can see the tile is only filled in by one of the watersheds values:

I zoom in more and some tiles are completely missing data:

I tried to play around with pixel_selection, resampling and scale but the same issue occurred in the same places.

I am a little unclear on which DEBUG env variable name I should use so I tried all three of these but none seemed to actually show any logs:
- DEBUG=TRUE
- TITILER_DEBUG=TRUE
- TITILER_PGSTAC_API_DEBUG=TRUE

Any pointers on troubleshooting would be much appreciated!

[drnextgis]

If you use titiler-pgstac which comes with eoAPI then to enable DEBUG you should use EOAPI_RASTER_DEBUG.

[zstatmanweil]

Thanks @drnextgis, I am not running via eoAPI though

[zstatmanweil]

When I set both exitwhenfull and skipcovered to False, I no longer have this issue. I think I was assuming it would factor in the nodata value but since those parameters are used on the pgstac side, it makes sense it is just looking at the geometry, not assessing the raster or its nodata value.

No issues here!

[zstatmanweil]

Yes, exactly. Our valid data matches watershed boundaries but our STAC item geometries are just the entire raster's footprint. So a somewhat niche problem

[vincentsarago]

maybe we should add settings to configure

titiler-pgstac/titiler/pgstac/mosaic.py

Lines 276 to 280 in d353b9d

	scan_limit = scan_limit or 10000
	items_limit = items_limit or 100
	time_limit = time_limit or 5
	exitwhenfull = True if exitwhenfull is None else exitwhenfull
	skipcovered = True if skipcovered is None else skipcovered

[zstatmanweil]

Do you mean so they don't have to passed as query params every time? In my use case that would definitely be useful

[vincentsarago]

yes

[zstatmanweil]

Thanks for opening an issue. I may have time to work on it.