Chemosphere, Vol. 16, No. 6, pp 1221-1230, 1987

The emission and chlorination pattern of polychlorinated dioxins and dibenzofurans from waste combustion shows a close covariation with that of chlorinated benzenes and phenols. This covariation can be utilized to predict the emission levels of specific isomers as well as the pattern of congeners.

Organic micro-pollutants and especially chlorinated organics have given rise to concern in the last few years. Combustion of waste and other fuels containing chlorine are well-known sources of emission. Sampling and analysis of micro-pollutants in flue gases increase in complexity and cost as the detection limit is lowered; in particular this applies to ultra-trace components like polychlorinated dioxins (PCDD) and dibenzofurans (PCDF). With suitable indicator parameters it may be feasible both to enhance the precision in the measurements as well as to reduce the costs. The correlation between chlorinated benzenes and PCDD/PCDF, as reported by us and others (1,2,3), can form the basis for indirect measurements.

Here we report data from combustion studies with different types of waste demonstrating the possibility not only to predict emission levels, but also the chlorination pattern from one group of chlorinated aromatics to another.

The sampling and analytical methods used have been described thoroughly elsewhere (4,5).

Emission data from four different plants (6, 7, 8, 9) were evaluated together with production data (concentrations upstream from the flue gas cleaning system) from a fifth plant (10). The reason for using production data in this latter case is the proven high separation efficiency of PCDD/PCDF in this particular plant.

Measurement data were analyzed and evaluated using principal component analysis (PCA) and partial least square modelling with latent variables (PLS), using the data program SIMCA-3B from Sepanova AB, Sweden (license agreement no 861210-SR1). The purpose of PCA is to reduce the multidimensional (multivariate) data set to a few significant principal components, describing as completely as possible the variation in the original data. The significance of these components is determined with cross-validation.

With PLS (a method related to principal component regression), relations between the blocks of independent and dependent variables are constructed to give the best possible correlation. In the prediction phase new objects, classified as similar to the original class (training set), are introduced to give predicted values of the dependent variables.

The analysis and modelling of multivariate data with SIMCA has been reviewed by Wold et al (11).

The flue gas samples show considerable discrepancies both in concentrations of chlorinated aromatics, as well as chlorination pattern. The variation in flue gas concentrations of chlorinated aromatics can primarily be attributed to variable combustion conditions, while variations in chlorination pattern result from different chlorine input (5, 12, 13, 14). In table 1 we summarize flue gas concentrations and percentage of the highest chlorinated isomer in each substance group.

Table 1

Plant	D (ref 7)	G (ref 8)	K (ref 9)	N (ref 6)	S(ref 10)
Fuel	Municipal waste	Municipal waste	Municipal waste	Hazardous waste	Hazardous waste
Sum chlorinated benzenes*	66/150	11/190/6.3	17/7.8	16/18	38/29/19
Hexachlorobenzene %	70/67	24/58/13	31/3	75/81	68/66/63
Sum chlorinated phenols*	38/42	11/140/4.7	24/7.7	0.23/0.20	8.0/11/9.5
Pentachlorophenol %	47/50	9.4/7.1/19	5.4/9.2	26/31	49/40/33
PCDD**	670/2400	37/3800/37	180/92	45/200	330/310/150
OCDD %	60/75	32/38/46	36/36	91/90	73/70/65
PCDF**	910/1800	130/3000/98	220/90	260/630	540/470/380
OCDF %	18/14	2.4/18/1.6	8.7/10	67/72	25/33/10

* µg/m³ standard dry gas at 10 % CO₂ ** ng/m³ standard dry gas at 10 % CO₂

Flue gas concentrations (both particles and gas phase) of 20 different isomers of chlorinated benzenes and phenols, 5 different isomers and 10 different groups of congeners of PCDD/PCDF were used as variables for PCA and PLS-analysis. Concentrations below detection limits were treated as missing values.

The significance of the PCA was checked with cross-validation. The first two principal components were found to be significant. In figure 1 we show a score plot of these principal components, each point represents one object/ flue gas sample.

Figure 1
PC-plot

Similar samples are grouped together and we see four different subgroups. Moving from the lower left-hand corner of the plot to the upper right-hand we notice a decreasing degree of chlorination, with the two N samples representing the most extreme condition. If we move along the other diagonal a change of flue gas concentrations is evident.

Since the measurement results for different groups and isomers of chloroaromatics obviously show a considerable amount of covariation, we have attempted to construct a multivariate calibration model with PLS. Chlorinated benzenes and phenols were defined as the block of independent and PCDD/PCDF as the block of dependent variables. Three significant components were found as checked by cross-validation. Only 27 % of the standard deviation in the block of dependent variables was left unexplained.

Modelling power is a measure of how much of the variable that contributes to the model. After the third principal component has been added five independent variables show a modelling power above 0.75, and four of these are chlorinated benzenes. One likely reason for the higher predictive value of chlorinated benzenes than phenols is that sampling and analysis can be executed with high precision and repeatability for this group of chlorinated aromatics.

In figures 2 - 6 we illustrate the results from the PLS-modelling for some isomers and groups of congeners of PCDD/PCDF. The predicted logarithmic flue gas concentrations are plotted against actual measurement values (also logarithmic).

Figure 2

Figure 3

 

Figure 4

 

Figure 5

 

Figure 6

The concentrations for different isomers and groups of congeners of PCDD/PCDF vary by a factor of between 20 and 1 000 in these 12 samples. As can be seen, from the figures shown, the correlation between predicted and measured flue gas concentrations is high (r >0.9).

It was stated above that chlorinated benzenes have a higher predictive value, for PCDD/PCDF, than chlorinated phenols. This observation can also he illustrated by comparing PLS-models for PCDD/PCDF with either chlorinated benzenes or phenols as independent variables. These two models calculated with chlorinated benzenes or phenols as the block of independent variables explain 72 % respectively 55 % of the standard deviation in the block of dependent variables (PCDD/PCDF).

The strength of the PLS-model described above was finally tested with six additional samples from plant G and 3 + 2 samples from two other plants. These other plants were a MSW-incinerator respectively a batch operated incinerator for special wastes. These additional samples were not included whilst building the model. In figure 7 we show the result for one variable.

Figure 7
Test samples

In table 2 we present numerically measured and predicted flue gas concentrations of some PCDD/PCDF for the new samples (test objects).

Table 2
Measured and predicted flue gas concentrations of PCDD/PCDF in ng/m³ standard dry gas at 10 % CO₂. Data presented as measured/predicted.

Sample/variable:	1	2	3	4	5	6
Sum TCDD	10/4.5	31/13	2.6/1.6	0.46/0.89	3.4/3.6	2.2/2.4
Sum PeCDD	39/21	160/73	26/8.6	4.6/4.3	16/22	12/12
Sum HxCDD	17/31	59/93	8.6/12	1.7/8.2	18/28	8.5/17
Sum TCDF	130/87	440/220	79/40	50/26	57/82	51/53
Sum PeCDF	170/99	590/260	93/50	49/36	73/110	99/64
Sum HxCDF	76/33	290/170	35/30	15/24	38/67	29/39
Sample/variable:	7	8	9	10	11
Sum TCDD	31/13	4.2/1.8	81/29	15/2.8	35/10
Sum PeCDD	61/77	9.5/7.3	260/190	35/16	91/50
Sum HxCDD	94/68	15/10	480/240	68/26	150/62
Sum TCDF	68/200	28/40	460/460	67/69	120/160
Sum PeCDF	140/200	49/44	1000/570	140/100	230/170
Sum HxCDF	140/110	40/24	1100/400	140/69	210/110

The data presented in table 2 show that the agreement is good between actual measurements and predictions of flue gas concentrations of PCDD/PCDF. The deviation is seldom more than a factor of two.

Our conclusions from the results presented are that:

• both the emission levels and the isomer or chlorination pattern of PCDD/PCDF can be predicted, from chlorinated benzenes and phenols, with high accuracy over a concentration range of 2 - 3 orders of magnitude.

• chlorinated benzenes show a higher predictive value than chlorinated phenols for PCDD/PCDF.

• the indirect "measurement" of PCDD/PCDF with chlorinated benzenes as indicator parameters can be performed not only to a lower cost but also with higher precision than conventional sampling and analysis. This latter fact may prove important when establishing emission limits for chlorinated aromatics.

We gratefully acknowledge the use of data from measurements carried out for Dala Miljö AB, Göteborgsregionens Avfallsaktiebolag, Västra Mälardalens Renhållnings AB, Norsk Hydro Plast AB and Svensk Avfallskonvertering AB.

1. OLIE, K, LUSTENHOUWER, JWA, HUTZINGER, 0
   Polychlorinated dibenzo-p-dioxins and related compounds. Impact on the environment. Editors HUTZINGER O, Frei RW, Merian E, Pocchiari F, Pergamon Press, Oxford, 1982.
2. ÖBERG, T, BERGSTRÖM, JGT
   Chemosphere 14, 1081-1086 (1985).
3. BENESTAD Ch, OEHME, M
   Presented at the ISWA/WHO/DAKOFA Specialized Seminar: Emission of trace organics from municipal solid waste incinerators, 20 - 22 January 1987 in Copenhagen, Denmark.
4. BERGSTRÖM, JGT, EKLUND, G, WARMAN, K
   Presented at the Sixth International Symposium on Polynuclear Aromatic Hydrocarbons, 27 - 29 October 1981 in Colombus, Ohio, USA.
5. BERGSTRÖM, JGT, WARMAN, K
   Presented at the ISWA/WHO/DAKOFA Specialized Seminar: Emission of trace organics from municipal solid waste incinerators, 20 - 22 January 1987 in Copenhagen, Denmark.
6. AITTOLA, J-P, BERGSTRÖM, JGT
   MKS Teknisk Rapport 85/68 (In Swedish)
   Environmental Consultants at Studsvik, 1985.
7. ERIKSSON, K
   MKS Teknisk Rapport 86/5 (In Swedish)
   Environmental Consultants at Studsvik, 1986.
8. LARSSON, G
   MKS Teknisk Rapport 85/105 (In Swedish)
   Environmental Consultants at Studsvik, 1985.
9. ÖHRSTRÖM, T
   MKS Teknisk Rapport 85/91 (In Swedish)
   Environmental Consultants at Studsvik, 1985.
10. BJÖRNER, B
    MKS Teknisk Rapport 86/75 (In Swedish)
    Environmental Consultants at Studsvik, 1986.
11. WOLD, S et al
    Pattern recognition: Finding and using regularities in multivariate data. In Food Research and Data Analysis. Editors Martens H, Russum Jr H Applied Science Publishers, London, 1983.
12. ÖBERG, T, AITTOLA, J-P, BERGSTRÖM, JGT
    Chemosphere 14, 215 - 221 (1985).
13. ÖBERG, T, BERGSTRÖM, JGT
    Chemosphere 15, 2041 - 2044 (1986).
14. ENGKVIST, T, ÖBERG, T
    MKS Teknisk Rapport 85/90 (In Swedish)
    Environmental Consultants at Studsvik, 1985.

(Received in Germany 16 February 1987; accepted 15 March 1987)

Reprinted from Chemosphere, Volume 16, Öberg, T, Bergström, JGT, Emission and chlorination pattern of PCDD/PCDF predicted from indicator parameters, Pages No. 1221-1230, Copyright (1987), with permission from Elsevier Science. Single copies of the article can be downloaded and printed for the reader's personal research and study.

DOI: 10.1016/0045-6535(87)90059-2