[Classification layer accuracy and area report (for simple random sampling)] wrong confidence intervals for area estimates

EnMAP-Box / enmap-box

EnMAP-Box source code repository. See https://enmap-box.readthedocs.io for documentation

GNU General Public License v3.0

34 stars 16 forks source link

[Classification layer accuracy and area report (for simple random sampling)] wrong confidence intervals for area estimates #909

Closed janzandr closed 4 weeks ago

janzandr commented 1 month ago

Discussed with @d-pflugmacher.

Currently, we are using the Stehman (2014) formulars (https://doi.org/10.1080/01431161.2014.930207) also for the simple random sampling case, by assuming all samples to be in the same stratum. This is producing 0-width confidence intervalls, which is wrong.

It is proposed to skip the area estimates for the simple random sampling case.

janzandr commented 4 weeks ago

FYI @d-pflugmacher, area estimation is now removed from the report:

Also, I don't use the Stehman2014 formulars in this case anymore, but the standard formulars. For SE calculations I use: