Powered by FOSS

Reflections: Life experience and calibration model robustness

In light of Senior Chemometrician Mark Westerhaus’ recent award for outstanding achievements in NIR spectroscopy, Senior Manager Lars Nørgaard reflects upon a life time of achievements within calibration development, NIR application and research.


Mark Westerhaus receiving the award for outstanding achievements in NIR

November 2014 I had the pleasure of attending the Eastern Analytical Symposium in Somerset, New Jersey, US. A multitude of vendors present the latest developments in analytical solutions and the program offers typically more than 500 papers on all aspects of analytical science.
Each year EAS honors scientists that have contributed significantly within different important research areas and this year Mark Westerhaus won the award for outstanding achievements in NIR spectroscopy.

Mark’s contributions cover many algorithmic developments, and they are all based on a lifelong experience with applications of NIR spectroscopy to real life complex problems in the food and feed industry. Mark presented his many developments in a very nice lecture and it was clear to me that the only way to be able to come up with methods used by many, many developers around the world is by being able to combine the mathematical development with a thorough knowledge of NIR spectroscopy and the application in question – a true chemometric approach! Not many people are able to do this but it is a fact that Mark can.

Introducing the repeatability file concept

One of the methods Mark has introduced is the repeatability file concept. This concept introduces a very elegant way to include variations in a calibration model to make it robust to these variations. If you want to make sure that the calibration covers a temperature range from 5 to 30 degrees Celsius you simply analyze a few samples in this temperature range and include them in an intelligent way in the calibration model – and you don’t even have to know the reference values of the samples. The repeatability file concept has shown to be an extremely efficient way to make calibration models robust and this is recently supported, again, by a nice comparative study by U.K. Acharyain et al. from Central Queensland University in Journal of Near Infrared Spectroscopy (JNIRS 22, 279–286, 2014).

If you are interested in more information on repeatability files I recommend that you have a look at this video interview with Chemometrician Robin Malm who explains the use of repeatability files for stable calibration development, or read the paper Stabilizing predicted values from NIR using REP files by Chemometrician Martin Andersson.

Lars Nørgaard