Source apportionment using PLS-discriminant analysis
Öberg, T.
T. Öberg Konsult AB, 1992.

Abstract
Partial least squares regression (PLSR) have been extensively reviewed in a recent text-book¹. Vong et al suggested some years ago discriminant PLSR as a method to solve source apportionment problems². This technique was applied to measurement data for different elements in aerosols.

Here principal component analysis (PCA) and discriminant PLSR is applied to identify and estimate the influence of different sources of organics in ambient air within a residential area. A simple measurement technique for C6-C18-hydrocarbons was used, Tenax-adsorption followed by thermal desorption and HRGC-FID-analysis. It is shown that the GC-pattern can be used directly to differentiate between emission sources, and also calculate their relative contributions to environmental samples.

The experimental data was obtained from measurements on different activities and sources within an air field, in a closely situated residential area and in background samples (influenced to some extent by other urban activities). 31 samples were collected. The ability of the calibration model to correctly differentiate between different sources in the design matrix is presented in table 1.

References:

1. Martens H, Næs T. Multivariate Calibration. Wiely, 1989.
2. Vong R, Geladi P, Wold S, Esbensen K. Source Contributions to Ambient Aerosol calculated by Discriminat Partial Least Squares Regression (PLS). Journal of Chemometrics, Vol. 2, 281-296 (1988).

Table 1 Design matrix and prediction results for calibration objects in %.

Sample	Combustion gases and fuels:		Background:
	Design	Prediction	Design	Prediction
A	100	101	0	-0.7
B	100	99.8	0	0.2
C	100	101	0	-1.5
D	100	99.6	0	0.4
E	100	99.6	0	0.1
F	100	98.4	0	1.6
G	0	-0.9	100	101
H	0	3.1	100	96.9
I	0	-2.5	100	103
J	0	0.6	100	99.4