Double-counting consecutive overlaps?

[jessjaco]

Consider that now since we are pulling all data within a tile footprint we may have consecutive overlaps which are biasing median (and other) calcs.

[alexgleith]

What do you mean by consecutive overlaps?

The same scene twice?

[jessjaco]

Adjacent scenes on a path which are taken consecutively on the same day, like these

[alexgleith]

Got it.

This is the kind of the the Open Data Cube is really good at.

I know it's late in the process, but it might be worth exploring the use of ODC STAC instead of STAC Stack: https://odc-stac.readthedocs.io/en/latest/

This loads data based on STAC items, but it'll do it in a way where the adjacent scenes are carefully overlaid.

Another little thing that might make a big difference is in picking a CRS. I saw were using an Equal Earth projection, but Robbi picks the most common UTM projection. I know he found better results when not using regridded data over Australia. The effects will likely be very minor, though.

[jessjaco]

I don't think using groupby=solar_day in the odc.stac.load call is going to work, as we need the exact times for tidal filtering. Will think about whether it's possible to stage the steps in another way but for now I think just doing .groupby("1D").first() or some similar on the all-times dataset may work better

[alexgleith]

Check in with Robbi on that one.

I think groupby solar day is good though, as the images are captured as a swath so are basically at the same time.

It means you get a single slice for each day, rather than it being broken up by scenes, I think.