Calibration Transfer for Predicting Lake-Water pH from near Infrared Spectra of Lake Sediments

Abstract

With near infrared spectra of lake sediment samples, it is possible to predict lake-water pH using a partial least squares (PLS) calibration model. The calibration models have large model and prediction residuals and it is important to understand the residuals. All the sediment samples were measured on five different instruments. This allows the use of each of the five instruments as master with the other four as field instruments. When using more than one near infrared instrument, prediction becomes unreliable unless calibration transfer is used. A number of techniques including Savitzky–Golay Transform (SGT), finite impulse response (FIR), piecewise direct standardisation (PDS), orthogonal scatter correction (OSC) and wavelet transform (WT) were compared. The quality of the predictions was expressed as root mean square error of prediction ( RMSEP), but the calibration transfer methods were also compared on practical usefulness. In these data, the OSC filtering worked best and gave adequate calibration transfer results.