Model-based signal tracking in the quantitative analysis of time series of NMR spectra

Hard modeling of NMR spectra by Gauss-Lorentz peak models is an effective way for dimensionality reduction. In this manner high-dimensional measured data are reduced to low-dimensional information as peak centers, amplitudes or peak widths. For time series of spectra these parameters can be assumed to be smooth functions in time. We suggest to model these time-dependent parameter functions by cubic spline functions, which makes a stable quantitative analysis of NMR series possible even for crossing, highly overlapping peaks. Applications are presented for the batch distillation of methanol and diethylamine, and the reaction of acetic anhydride with 2-propanol.