Mono- to tri-chlorinated dibenzodioxin (CDD) and dibenzofuran (CDF) congeners/homologues as indicators of CDD and CDF emissions from municipal waste and waste/coal combustion

Article abstract of DOI:10.1016/S0045-6535(99)00347-1

Total homologue concentrations and select congener concentrations from amongst the mono- to tri-chlorinated dibenzodioxins (CDDs) and dibenzofurans (CDFs) are used to model both Total (mono- to octa-) CDD + CDF emissions and the toxicity equivalent (TEQ) of the 2,3,7,8-chlorine-substituted emissions. Analysis of emission data from two facilities indicates that use of total homologue concentrations shows limited, facility-specific correlations with Total CDDs/CDFs and TEQ. Concentrations of select mono- to tri-CDD/CDF congeners show promising correlation with CDD/CDF TEQ across facilities, suggesting that these compounds can act as TEQ indicators. (C) 2000 Elsevier Science Ltd.

Full text of DOI:10.1016/S0045-6535(99)00347-1

Chemosphere 40 (2000) 1015±1019
Mono- to tri-chlorinated dibenzodioxin (CDD) and
dibenzofuran (CDF) congeners/homologues as indicators of
CDD and CDF emissions from municipal waste and waste/
coal combustion
a,
Brian K. Gullett , Evalena Wikstrom
b
*

a
National Risk Management Research Laboratory, Air Pollution Prevention & Control Division, US Environmental Protection Agency
(MD-65), Research Triangle Park, North Carolina 27711, USA
b
Institute of Environmental Chemistry, Umeꢀa University, S-901 82 Umeꢀa, Sweden
Abstract
Total homologue concentrations and select congener concentrations from amongst the mono- to tri-chlorinated
dibenzodioxins (CDDs) and dibenzofurans (CDFs) are used to model both Total (mono- to octa-) CDD + CDF
emissions and the toxicity equivalent (TEQ) of the 2,3,7,8-chlorine-substituted emissions. Analysis of emission data
from two facilities indicates that use of total homologue concentrations shows limited, facility-speci®c correlations with
Total CDDs/CDFs and TEQ. Concentrations of select mono- to tri-CDD/CDF congeners show promising correlation
with CDD/CDF TEQ across facilities, suggesting that these compounds can act as TEQ indicators. Ó 2000 Elsevier
Science Ltd. All rights reserved.
1. Introduction
binomial reactivity functions (Funcke and Hemming-
haus, 1993). The homologue pro®les appear much more
dynamic than the congener patterns and have been
linked through Principal Component Analyses (PCAs)
to ash characteristics and operating parameters (Pitea
et al., 1990; Manninen et al., 1996) and through a
combination of Poisson process/structural equations
modelling (Gullett and Dunn, 1995) to gas concentra-
tions and operating parameters. Recent e€orts (Gullett
et al., 1998) have modelled homologue pro®les and
concentrations using generalized additive models to de-
velop a multivariate model with input of six operating
and ¯ue gas parameters.
Knowledge of the congener patterns and homologue
pro®les of polychlorinated dibenzodioxins and poly-
chlorinated dibenzofurans (PCDD/Fs) is important be-
cause toxicity is linked to the concentration distribution
of the 17 congeners that are fully chlorinated in the
2,3,7,8 positions. A mechanistic understanding of how
chlorine (Cl) partitions among and within the homo-
logues will enable approaches toward reducing overall
formation of PCDD/Fs and/or reducing the concentra-
tion of the 17 congeners that comprise the toxic equiv-
alency (TEQ) value. E€orts have been made to model
equilibrium distributions of congeners using computa-
tional molecular modelling (Unsworth and Dorans,
1993). Modelling of homologue distributions has used
The e€ort reported here uses homologue concentra-
tions and a subset of the mono- to tri-chlorinated
PCDD/F congeners to predict both PCDD/F TOTAL
(sum of mono- to octa- CDD/F concentrations) and
TEQ values. The ability to establish correlations be-
tween PCDD/F measures (particularly TEQ values) and
concentrations of a limited subset of the 74 mono- to tri-
CDD and CDF congeners is particularly valuable since
* Corresponding author. Tel.: +1-919-541-1534; fax: +1-919-
541-0554.
E-mail address: gullett.brian@epa.gov (B.K. Gullett).
0045-6535/00/$ - see front matter Ó 2000 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 4 5 - 6 5 3 5 ( 9 9 ) 0 0 3 4 7 - 1

1016
B.K. Gullett, E. Wikstrom / Chemosphere 40 (2000) 1015±1019
current instrumentation development shows promise for
measuring these lower chlorinated congeners in an on-
line, real-time mode (Oser et al., 1998). Sampling results
from two facilities and a wide range of operating con-
ditions are used to establish the robustness of the pre-
dictions for application to other facilities and fuels.
Strong relationships between lower chlorinated cong-
eners and PCDD/F measures (such as TEQ) coupled
with further development of methods to monitor the
concentrations of these congeners in ¯ue gas should
provide a valuable tool for understanding the mecha-
nism of PCDD/F formation and ®nding process control
methods to reduce or prevent their formation.
dominance of PCDFs over PCDDs. As is commonly
found, the intra-group PCDD/F TOTAL and TEQ
values appear linearly correlated (Fig. 1). Variation in
the slopes is consistent with di€erences found in inter-
source comparisons (Cleverly et al., 1997). Each groupÕs
ꢀ
homologue pro®les (Norfolk ± Fig. 2, Umea-1 ± Fig. 3,
and Umea-2, Fig. 4) exhibit consistent, intra-group
ꢀ
pro®les, although the magnitude and peak homologue
group vary considerably between the three test groups.
The stability of the homologue pattern is not surprising,
ꢀ
although Umea-1, with planned operation beyond the
limits of desirable combustion conditions, might have
been expected to exhibit a larger pro®le variation based
on work by Gullett et al. (1999). Their work showed
shifts or magnitude changes in the homologue pro®le
that were associated with changes in operational pa-
rameters of sulfur dioxide (SO2) and hydrogen chloride
(HCl) concentrations, residence time, and quench rate.
SASâ procedure REG (SAS/STAT User's Guide,
1990) was used to obtain the best group-speci®c one-,
two-, and three-predictor models (Table 1) of run-
2. Experimental
A municipal waste combustor near Norfolk, Virgin-
ia, burning processed ¯u€ refuse derived fuel (RDF),
was the site for 13 pre-spray-dryer sampling tests
(``Norfolk''). This facility ®red RDF-only or co-®red
RDF with two types of coal during this testing. One coal
was low sulfur (S, 0.7 wt%) and one coal from Illinois
was high S (3.5 wt%). Co-®ring up to 5 wt% high S coal
with RDF reduced the pre-spray-dryer PCDD/F con-
centration by up to 70% from the initial baseline. Sam-
pling and analytical methods were expanded to quantify
mono-, di-, and tri-CDD/CDF congener totals as well as
select isomers using the isotope dilution method. La-
belled di- and tri-CDF and di- and tri-CDD surrogates
were also added to XAD-2 prior to sampling to assess
overall measurement performance.
ꢀ
The pilot scale Umea reactor is a solid fuel comb-
ustor that burns arti®cial pellets on an under®re/over®re
air supplied grate. The nominal burn rate is 18 MJ/h
(1 kg/h). A triple looped convection section, 5 m per
section, allows for simulating quench rates in ®eld units.

Further details can be found in Wikstrom et al. (1998),

Wikstrom and Marklund (1998).
Fig. 1. Comparison of Total PCDD + PCDF [mono- to octa-
PCDD/F (pmol/dscm)] and TEQ (ng/dscm) for Norfolk (n),
ꢀ
Two test groups were run in this reactor. ``Umea-1''
run parameters varied total air ¯ow from 90 to 150
L/min, ¯ow in the secondary air varied from 20 to 70%,
and the temperature of the secondary air varied from 50
ꢀ
Umea-1 (r), and Umea-2 (m) tests.
ꢀ

to 350°C (the test matrix is shown in Wikstrom and
ꢀ
Marklund, 1997). The ``Umea-2'' tests were run under
non-varied combustion conditions. The purpose of these
tests was to simulate full scale combustion with an ar-
ti®cial municipal solid waste to study formation of
PCDD/Fs.
3. Results, analyses, and discussion
Thirteen tests at the Norfolk site and 16 tests at the
ꢀ
Umea pilot plant resulted in a fairly wide range of
PCDD/F TOTAL and TEQ values, with the common
Fig. 2. Run-speci®c homologue pro®les from Norfolk tests.

B.K. Gullett, E. Wikstrom / Chemosphere 40 (2000) 1015±1019
1017
ꢀ
the Norfolk and Umea-2 data result in single-predictor
models which show good prediction of TEQ (R2 ˆ 0.767
and 0.948, respectively). The low (<0.05) P values show
that each selected predictor is not likely selected by
ꢀ
chance. However, the Umea-1 data were not successfully
modelled and none of the three data sets resulted in se-
lection of a common, single-homologue predictor.
Extension of the analyses to include up to three
predictors improves the modelsÕ R2. This is particularly
2
ꢀ
true for both Norfolk and Umea-2 where R ˆ 0.868 and
0.996, respectively, are obtained. These models also re-
sult in selection of the same predictor homologues.
However, the same predictors for each model have dif-
ferent coecient signs (+/)) and the Norfolk predictors
have high (>0.05) P values, lending doubt to the modelsÕ
ꢀ
Fig. 3. Run-speci®c homologue pro®les from Umea-1 tests.
universal mechanistic signi®cance. The three-predictor
2
model for the Umea-2 group data has a high R value
ꢀ
with low P values (<0.03) for all three terms, suggesting
that homologue correlations with TEQ may be appro-
priate under some circumstances for facility-speci®c
application.
ꢀ
Prediction of TEQ with the Umea-1 homologues is
poor, even when extended to a three-predictor model.
ꢀ
This is particularly noteworthy in that the Umea-1
pro®le is a result of combustor operation at extreme
conditions of air ¯ow and temperature, unrepresentative
of the normal operating mode. The homologues are
dominated by mono-CDF (MCDF) and di-CDF
(DiCDF) homologues, yet the concentrations of these
lower chlorinated compounds do not readily re¯ect the
higher chlorinated compounds that comprise the TEQ
measure. These results suggest that homologue totals
may be useful in predicting TEQ only on a facility-spe-
ci®c basis.
ꢀ
Fig. 4. Run-speci®c homologue pro®les from Umea-2 tests.
speci®c ln (TEQ, ng/dscm). For each test group, predic-
tors were selected from among the six total concentra-
tions [pmol/m3, 7% oxygen (O2)] of the mono- to tri-
CDD/F homologues for each run. Logarithms of the
dependent variable were evaluated to ensure a normal
distribution of the data. Model selection criterion is
based on maximization of R2 and minimization of the
predictor p values. The p values for each predictor indi-
cate the probability of rejecting the null hypothesis (no
linear relationship between the predictor and the depen-
dent variable) in error. Use of the mono- to tri-CDD/F
homologues as potential predictors of TEQ shows that
Switching now from a regulatory focus (TEQ) to a
mechanistic focus (TOTAL), Table 2 shows the best
one-, two-, and three-predictor models for PCDD/F
TOTAL from among the mono- to tri-CDD/Fs. Single
homologue totals do well at predicting TOTAL, al-
though (as in Table 1) the selected predictors are not
consistent between the data sets. Only modest im-
provements in R2 are noted with selection of additional
Table 1
Regression models for TEQ using mono- to tri-CDD/F homologue totals
ꢀ
Umea-1 (10 runs)
ꢀ
Umea-2 (6 runs)
Norfolk (13 runs)
Predictor(s)
TrCDF
P value
R2
Predictor(s)
MCDF
P value
R2
Predictor(s)
TrCDD
P value
R2
0.000
0.767
0.141
0.250
0.001
0.948
DiCDF,
TrCDF
0.199,
0.003
0.813
0.868
MCDF,
TrCDF
0.262,
0.619
0.278
0.390
TrCDD,
DiCDF
0.003,
0.243
0.969
0.996
DiCDD,
DiCDF,
TrCDF
0.129,
0.053,
0.008
DiCDD,
TrCDD,
MCDF
0.296,
0.286,
0.412
DiCDD,
DiCDF,
TrCDF
0.013,
0.021,
0.003

1018
B.K. Gullett, E. Wikstrom / Chemosphere 40 (2000) 1015±1019
Table 2
Regression models for TOTAL using mono- to tri-CDD/F homologue totals
ꢀ
Umea-1
ꢀ
Umea-2
Norfolk
Predictor(s)
DiCDD
P value
R2
Predictor(s)
DiCDF
P value
R2
Predictor(s)
TrCDD
P value
R2
0.001
0.761
0.002
0.731
0.010
0.838
MCDD,
DiCDD
0.410,
0.001
0.782
0.787
DiCDF,
TrCDF
0.000,
0.048
0.852
0.855
MCDD,
TrCDD
0.281,
0.014
0.897
0.917
MCDD,
DiCDD,
TrCDD
0.505,
0.199,
0.676
MCDF,
DiCDF,
TrCDF
0.744,
0.020,
0.074
TrCDD,
MCDF,
TrCDF
0.176,
0.331,
0.394
Table 3
Regression models for TEQ selecting from 12 mono- to tri-CDD/F congenersa
ꢀ
Umea-1
ꢀ
Umea-2
Norfolk
Predictor(s)
1,2,3-TrCDF
P Value
R2
Predictor(s)
P Value
R2
Predictor(s)
P Value
R2
0.001
0.679
2,4,6-TrCDF
0.037
0.438
2,4,6-TrCDF
0.442
0.442
1,2,3-TrCDF,
1,6-DiCDD
0.009,
0.000
0.843
1,2,3-TrCDF,
2,4,6-TrCDF
0.005,
0.006
0.832
1,2,3-TrCDF,
2,4,6-TrCDF
0.060,
0.025
0.856
a 1-MCDD; 2-MCDD; 1,6-DiCDD; 2,3-, 2,7-, 2,8-,DiCDD; 1,2,3-, 1,7,8-TrCDD; 2-MCDF; 4-MCDF; 2,4-DiCDF; 2,8-DiCDF; 1,2,3-
TrCDF; 2,4,6-TrCDF, 2,4,8-TrCDF.
2
ꢀ
predictors. Unlike Table 1, the Umea-1 R value is high
of the concentration of a single congener, 1,2,3-TrCDF,
allows prediction of over 67% of the variation in the
TEQ values for the Norfolk data. Signi®cant improve-
ment in R2 (>80% for all three data sets) is obtained with
addition of a second predictor. It should be noted that
selection of the speci®c congeners cited in Table 3 is not
necessarily unique; other statistically selected congeners
(not shown) may be substituted without large loss of R2.
The actual prediction of ln (TEQ) in units of ng/dscm
can be written as a weighted linear combination of the
selected predictors using the regression coecients (not
shown) as weights. These results suggest that the gas-
phase concentrations of a limited number of mono- to
tri-chlorinated congeners can serve as indicators of ¯ue
gas CDD/CDF concentrations. Recent instrumental
developments (Oser et al., 1998) in the ability to monitor
these congeners in real time and at observed ¯ue gas
concentrations suggest a possible method for predicting
TEQ emissions and providing feedback to control
combustor performance.
for even the single-predictor model, likely re¯ecting the
ability of the homologue to predict itself (the pro®le is
dominated by MCDFs and DiCDFs). This di€erence in
ꢀ
the ability of the Umea-1 homologue data to predict
TOTAL versus TEQ may be indicative solely of the high
concentration of the lower chlorinated compounds in
this sample. For all of the data sets there is surprising
frequency of PCDD homologue (rather than PCDF
homologue) selection to re¯ect TOTAL, since TOTAL is
dominated by PCDF compounds. When models are
extended beyond a single predictor, the P values increase
signi®cantly, lending doubt to the signi®cance of their
inclusion within the models.
Finally, we evaluated the ability of select mono- to
tri-CDD/F congener concentrations (pmol/m3, 7% O2)
to predict the logarithm of TEQ either singly or in
combination with another congener (Table 3). Twelve
potential predictors were evaluated from among the
concentrations of 15 congeners (some predictors con-
sisted of the sum of co-eluting peaks) for which we
(EPA) had chromatography standards. These same 12
predictors were used as candidates to model the Nor-
Acknowledgements
ꢀ
ꢀ
folk, Umea-1, and Umea-2 datasets for TEQ. The joint
model selection criteria are maximization of R2 and
preference for non-co-eluting congeners. Table 3 shows
that the concentrations of select TrCDF compounds,
particularly 1,2,3-TrCDF and 2,4,6-TrCDF, show rea-
sonable correlations with TEQ. For example, knowledge
B. GullettÕs work was partly sponsored by the Illinois
Clean Coal Institute (ICCI). Any opinions, ®ndings,
conclusions, or recommendations expressed herein do
not necessarily re¯ect the view of the ICCI.

B.K. Gullett, E. Wikstrom / Chemosphere 40 (2000) 1015±1019
1019
Oser, H., Thanner, R., Grotheer, H.-H., Gullett, B.K.,
Natschke, D., Raghunathan, K., 1998. Lowly chlorinated
dibenzodioxins as TEQ indicators. A combined approach
using spectroscopic measurements with DLR Jet-REMPI
and statistical correlations with waste combustor emissions.
Comb. Sci. and Tech. 134, 201±220.
References
Cleverly, D., Schaum, J., Schweer, G., Becker, J., Winters, D.,
1997. The congener pro®les of anthropogenic sources of
chlorinated dibenzo-p-dioxins and chlorinated dibenzofur-
ans in the United States. Organohalogen Compounds 32,
430±435.
Pitea, D., Lasagni, M., Moro, G., Todeschini, R., Clementi, S.,
Cruciani, G., Chiesa, G., 1990. Response surface models for
the formation of PCDD and PCDF in a pilot plant
combustion of MSW. Chemosphere 20 (10±12), 1973±1979.
SAS/STAT UserÕs Guide, 1990. Vol. 2, version 6, Fourth ed.,
SAS Institute, Incorporation, Cary, NC.
Funcke, W., Hemminghaus, H.-J., 1993. Evaluation of PCDF/
D, PBDF/D and PBCDF/D pro®les in ¯ue gas of combus-
tion facilities using
Organohalogen Compounds 11, 345±350.
a statistical distribution function.
Gullett, B.K., Dunn, J.E., 1995. Modelling of polychlorinated-
dioxin and -furan congener pro®les from municipal waste
combustion. In: Proceedings of the International Specialty
Conference on Solid Waste Management: Thermal Treat-
ment and Waste-to-Energy Technologies, EPA/AEERL,
Research Triangle Park, NC, April 18±21.
Unsworth, J.F., Dorans, H., 1993. Thermodynamic data for
dioxins from molecular modeling computations: prediction
of equilibrium isomer composition. Chemosphere 27 (1±3),
351±358.

Wikstrom, E., Andersson, P., Marklund, S., 1998. Design of a
laboratory scale ¯uidized bed reactor. Rev. Sci. Instruments
69 (4), 1850±1859.
Gullett, B.K., Dunn, J.E., Bae, S.-K., Raghunathan, K., 1998.
E€ects of combustion parameters on polychlorinated
dibenzodioxin and dibenzofuran homologue pro®les from
municipal waste and coal co-combustion. Waste Manage-
ment 18, 473±483.

Wikstrom, E., Marklund, S., 1997. A two step procedure to
investigate the importance of combustion parameters in the
formation of products of incomplete combustion. Organo-
halogen Compounds 31, 532±537.

Manninen, H., Perkio, A., Vartiainen, T., Russkanen, J., 1996.

Wikstrom, E., Marklund, S., 1998. Combustion of an arti®cial
Formation of PCDD/PCDF ± e€ect of fuel and ¯y ash
composition on the formation of PCDD/PCDF of refuse-
derived and packaging-derived fuels. Environmental Science
and Pollution Research 3 (3), 129±134.
municipal solid waste in a laboratory ¯uidised bed reactor.
Waste Manag. & Res. 16, 342±350.