[en] The paper discusses the ability of recognition models to classify malodours in the environment three years after their calibration, with the same sources releasing odours. Two methodologies applicable either in the lab or in the field are used. The effect of the long term drift of the sensors is highlighted on the results of principal component analysis and of discriminant function analysis.
The paper examines also the applicability of some drift counteraction methods proposed in the literature. Handled with some care, a correction based on the drift direction in the principal components subspace should be applicable to the classification of real odorous sources in the field.