search

digitalearthpacific / dep-coastlines

GNU General Public License v3.0
3 stars 1 forks source link

Certainty masking #27

Open jessjaco opened 12 months ago

jessjaco commented 12 months ago

Should be a drop in to use deafrica-coastlines.vector.certainty_masking, etc. but we need stdev for mndwi (or ndwi) to use their cutoff for sd

jessjaco commented 12 months ago

Just to note

  1. I don't think the standard deviation accurately quantifies the uncertainty of the median
  2. I think using n (number of clear readings) consistently indicates overall quality, since many readings still have clouds
jessjaco commented 12 months ago

Also, none of this quantifies uncertainty in the actual coastline location due to

  1. Bad georegistration of the image
  2. The water index not accurately capturing the location of the coastline
  3. Variation due to tides
jessjaco commented 5 months ago

After months of reflection, I've softened my stance a bit on this issue. However, what I would like to do, perhaps in addition to these things, is get some data for bad vs good areas and see how the values differ. In doing so I would need to consider

  1. Whether the "bad" areas are the lines themselves of the land they contain (or the ocean they don't!). If used for certainty masking we want to include the line, but that may not be the actual issue.
  2. It's not always easy to determine a "good" area for a specific year. That is, even if the line is still close to the previous / next years it could still be bad