Aunir Glossary - RMSEP

RMSEP stands for Root Mean Square Error of Prediction

SEP stands for Standard Error of Prediction

RMSEP or SEP is the simplest and most efficient measure of the uncertainty in NIR predictions. This value is a measure of the average uncertainty that can be expected when predicting new samples. RMSEP is expressed in the same units as the predicted parameter. The results of future predictions can then be presented as predicted values ± 2 x RMSEP. This measure is valid provided that the new samples are similar to the ones used for the calibration otherwise the prediction error might be much higher.

There is another form of SEP that is commonly used and is called SEPc (or SEP-b) which is the bias corrected version. It can be shown that the following relationship exists between RMSEP and SEPc

RMSEP= SEPc+ bias2

If the bias is small then SEPc will be similar to RMSEP (SEPc ~ RMSEP)

A word of caution: It is discouraged to use the values of SEPc on its own as it is usually over-optimistic. So you need to either use RMSEP or bias and SEPc.



Davies, A.M.C and Fearn, T. 2006. Back to basics: calibration statistics. Spectroscopy Europe. 18: 31-32.

The Unscrambler methods (Camo software AS), 2013.