Closed ctroupin closed 2 years ago
There are some strong gradients in the data and the low values just might be due to these gradients reflected in the analysis (undershooting where you have a strong gradient).
Also note that if you are doing 3D analyses, values from deeper or uppers layers might have some influence.
Are there no data at all below 250 in your data set ?
I forgot to send the 2nd email:
I did notice that if I increase the correlation length, the situation will get slightly better. My guess is that the bullseyes are not caused by some actual low values in that spot. Instead, they are caused by: (a) a lack of data in the spot, and (b) a nearby trend of decreasing value towards that spot. Under this situation, the system would automatically extrapolate to this data sparse spot and give it a low value.
So yes, probably due to strong gradients + lack of data. We can suggest to change the analysis parameters, but this will only dilute the problem...
Or create a specific background field?
Or if it is clear that there are never values below a certain value apply a change of variables which ensures that.
I close that isssue because it seems related to the data distribution itself rather then issues with DIVAnd
Opening the issue from an email received on December 4: