Hello all,
I have a question. When measured and estimated data do not have the same unit, what are the best error criteria to assess model efficiency or estimation accuracy?
For example, suppose that we want to compare the estimated soil water index (dimensionless and range from 0 to 1) with the measured soil moisture content (g/g).
Thanks in advance,
Reza
The first suggestion would be to have them in the same units. If your model/method is trained for estimating a parameter why are the reference values in a different unit? How did you train it?
However, if you are trying to compare two parameters of a sample that have different sources what you are trying to do is assess their correlation, for which you can use R or R2.
if they are both on a continuous scale can you do a correlation cofficient
Hello,
I suggest to first convert both cases to one dimension and then use a statistical index such as RMSE, RE%, SI, WI, NS, KGE or etc. to evaluate the accuracy.
Hope it be benefit for your target.
It depends on your goal. If both time series are in the same dimension, you can use many criterias, but if they are not the same, CC can show their correlation and can be useful.
From my point of view KGE is an excellent criteria to compare two time series. In this metric we have CC, Bias and Variability which makes it a complete criteria.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts