aggregate_temporal_period after load_stac gives unexpected results

[VincentVerelst]

In job j-240902ff18204cc38e0256e8872702b9 on the Terrascope backend, I load in a STAC collection. The STAC collection can be found in the attachment. The temporal extent ranges from 2016 to 2018. I then apply median compositing using the aggregate_temporal_period process. Finally, I use aggregate_spatial to extract some points to geoparquet format.

The resulting geoparquet file unexpectedly contains a timeseries ranging from 1970-2062 in monthly timesteps. Due to this, the batch job took a long time to run and used many credits.

Without applying aggregate_temporal_period, the timeseries in the resulting geoparquet varies from 2016 to 2018, as expected. This leads me to believe that the problem lies within the aggregate_temporal_period process being used after load_stac.

collection.json

[VincentVerelst]

Explicitly setting a temporal_extent in load_stac fixes this issue.

[jdries]

Proposed fix: fall back to temporal extent of stac collection when no extent is provided in load_stac.

[bossie]

@VincentVerelst the fix for #852 should also cover this; please retry on openeo-dev.

[VincentVerelst]

@bossie, sorry forgot to comment on this one: confirmed that this one is fixed as well!