Indicator parameters for PCDD/PCDF - plant specific models

Multivariate calibration using PLS-regression has been proved to be very useful in many chemometric applications. Since 1987 we have ourselves used this method for multivariate data analysis to develop indirect measurement methods for polychlorinated dioxins (PCDD) and dibenzofurans (PCDF) emitted from different thermal processes. Previously presented results have shown that the emissions of PCDD/PCDF can be modeled and predicted from less costly and more precise isomerspecific analyses of different chlorinated benzenes and phenols^1-4. The new results we report here comes from a study aimed to further utilize these possibilities and develop specific models for each plant to accomplish maximum precision and accuracy in the calibration process.

Multivariate calibration using PLS have been extensively described in the literature^5-6.

All flue gas sampling was done according to the recommended Nordic method⁷. Chlorinated benzenes and phenols as well as PCDD/PCDF were analyzed using HRGC-MS/SIM as described in an earlier paper⁸. The amounts in each sample have been calculated from the recoveries of labeled surrogates added prior to extraction and clean-up.

Plant specific models have been developed for two municipal waste combustion (MWC) facilities and one metallurgic industry. In figure 1-3 we show the predictions in each cross-validation segment as a function of measurement results for TCDD-equivalents. By using cross-validation we also get a conservative estimate of the likely spread in future samples.

Figure 1
Predicted vs. analytical results for TCDD-equivalents (Eadon), log-scale 3.2-1000 ng in sample. Plant-specific model for a metallurgical industry.

Figure 2
Predicted vs. analytical results for TCDD-equivalents (Nordic), log-scale 6.3-250 ng in sample. Plant-specific model for MWC plant A.

Figure 3
Predicted vs. analytical results for TCDD-equivalents (Nordic), log-scale 6.3-250 ng in sample. Plant-specific model for MWC plant B.

The samples with the narrowest variation range, figure 3, are particularly interesting to study. In these samples we can predict 2378-substituted isomers with much higher precision than for the groups of congeners, figure 4-7. The specific 2378-substituted isomers are of course analyzed with higher accuracy and precision due to the availability of 13-C-labelled surrogates.

Figure 4
Predicted vs. analytical results for 12378-P5CDD, log-scale 1.3-10 ng in sample. Plant-specific model for MWC plant B.

Figure 5
Predicted vs. analytical results for sum P5CDD, log-scale 13-100 ng in sample. Plant-specific model for MWC plant B.

Figure 6
Predicted vs. analytical results for 23478-P5CDF, log-scale 6.3-32 ng in sample. Plant-specific model for MWC plant B.

Figure 7
Predicted vs. analytical results for sum P5CDF, log-scale 79-320 ng in sample. Plant-specific model for MWC plant B.

Our interpretation of these results are that the overall measurement precision and accuracy for PCDD/PCDF sets the limits how far we can go in improving the calibration models.

Plant specific multivariate calibration models can simplify flue gas sampling procedures since the levels of chlorinated benzenes and phenols are much higher than the levels of PCDD/PCDF. Furthermore, the cost of the analysis is reduced and therefore a greater number of samples can be used as a basis for calculation of the annual emissions.

The main source of error for calculation of annual "dioxin" emissions is not uncertainties in the performed analyses, but the lack of representativity. Predicting PCDD/PCDF from plant specific models, with an increased number of test points, will give much better estimates than single-shot conventional measurements. Simplified sampling and analysis also facilitate optimization of both combustion processes and flue gas cleaning equipments.

1. Öberg T, Bergström J. Emission and chlorination pattern of PCDD/PCDF predicted from indicator parameters. Chemosphere 1987;16:1221-1230
2. Öberg T, Bergström J. Indicator parameters for PCDD/PCDF. Chemosphere 1989;19:337-344.
3. Öberg T, Bergström J. Organic micro-pollutants from steel plants. Report D621. Jernkontoret, 1988. (In Swedish).
4. Öberg T. Multivariate calibration models for organic micropollutants from industrial processes. Presentation at the First Scandianavian Symposium on Chemometrics, October 6-8, 1988, Lappenranta, Finland.
5. Martens H, Næs T. Multivariate calibration. John Wiley & Sons, 1989.
6. Wold S et al. Pattern recognition: Finding and using regularities in multivariate data. In: Martens H, Russwurm H, eds. Food research and data analysis. Applied Science Publishers, London, 1983.
7. Jansson B, Bergvall G. Recommended methodology for measurements of PCDD and PCDF in the Nordic countries. Waste Management & Research 1987;5:251-255.
8. Öberg T, Warman K, Bergström J. Brominated aromatics from combustion. Chemosphere 1987;16:2451-2465.

Preprint of short paper published in Organohalogen Compounds 8, 197-200 (1992).