Application of PLS-modelling within the environmental field
Öberg, T.
Lecture at the 4^th Unscrambler User Conference, October 2-4, 1991, Kolbotn, Norway.

Abstract
In this presentation I will describe three different applications of PLS-modelling related to environmental protection and air quality.

Classical response surface designs and multiple linear regression has proved to very useful tools in optimizing chemical processes (1). However there are certain conditions that must be at hand to use these methods, that are not always fulfilled in real life. I will with a practical example, emissions from combustion processes, illustrate some obstacles that can come up and how PLS can help in solving these (e.g. variable constraints, random variations in influential process variables and partly distorted experiments)

Many well known applications of PLS-regression can be found in different spectroscopy techniques. I will give another example of a useful application in analytical/environmental chemistry, namely indirect measurement of dioxins and other highly toxic ultra-trace components in flue gases from waste heating plants and electric arc furnaces (2-3).

PLS-discriminant analysis has successfully been applied to source apportionment for inorganic constituents in aerosol samples (4). I will discuss a similar problem with organic air pollutants, from an air-field, where gaussian dispersion models proved to be insufficient. The contribution to immissions of organics in a residential area was calculated based on GC-patterns in ambient air samples and calibration data from several possible sources.

References:

1. Box, G E P; Draper, N R: Empirical Model-Building and Response Surfaces. Wiley, 1987.
2. Öberg, T; Bergström, J: Chemosphere 16, 1221-1230 (1987).
3. Öberg, T; Bergström, J: Chemosphere 19, 337-344 (1989).
4. Vong, R; Geladi, P; Wold, S; Esbensen, K: J. Chemometrics 2, 281-296 (1988).