Uncertainty in rate constants estimated from spectral data with baseline drift

This paper concerns the impact of baseline drift and noise on some methods for rate constant estimation from spectroscopic reaction monitoring data, a nd means to assess the uncertainty in the estimates. The methods studied ar e a target testing procedure and non-linear least squares fitting. Both sim ulated and experimental data are used. The experimental data are from in si tu UV/VIS and in situ ATR mid-IR monitoring of eater hydrolysis reactions. For non-linear least squares fitting, biased estimates are obtained when ba seline drift is present, since the drift violates the model assumptions. Th e target testing procedure, on the other hand, can be used to reliably dete rmine rate constants from spectroscopic data with baseline drift. Owing to the use of a projection onto a stochastic subspace, bias can occur in the e stimated rate constants also for this method. However, the results indicate that in most cases this bias is not significant in practice. Bias and prec ision in the estimated rate constants are dependent on noise level and over lap in the spectral and time domains of the data. Bootstrapping can be used to calculate confidence intervals that have good coverage properties in mo st cases. Copyright (C) 2000 John Wiley & Sons, Ltd.