Closed BaptisteVandecrux closed 10 months ago
During the bufr processing, a linear regression is fitted to the available transmitted coordinates. Unfortunately, this regression and the OLS minimization is to sensitive to outliers. E.g. these outliers:
... lead to this linear regression:
Switching to a RANSAC regressor leads to more robust estimation:
More testing on the way before I make this update.
Examples where RANSAC don't work:
another problem at KAN_M where only coordinates from SBD messages are available:
For the moment I am switching off the gap-filling with SBD coordinates and using SBD average when no coordinates are available.
This is now fixed:
red dots are monthly averaged transmitted positions