Probabilistic and Technology-Specific Modeling of Emissions from

ARTICLE pubs.acs.org/est

Probabilistic and Technology-Specific Modeling of Emissions from Municipal Solid-Waste Incineration Annette Koehler,* Fabio Peyer, Christoph Salzmann, and Dominik Saner Group for Ecological Systems Design, Institute of Environmental Engineering, ETH Zurich, Wolfgang-Pauli-Strasse 15, 8093 Zurich, Switzerland

bS Supporting Information ABSTRACT: The European legislation increasingly directs waste streams which cannot be recycled toward thermal treatment. Models are therefore needed that help to quantify emissions of waste incineration and thus reveal potential risks and mitigation needs. This study presents a probabilistic model which computes emissions as a function of waste composition and technological layout of grate incineration plants and their pollution-control equipment. In contrast to previous waste-incineration models, this tool is based on a broader empirical database and allows uncertainties in emission loads to be quantified. Comparison to monitoring data of 83 actual European plants showed no significant difference between modeled emissions and measured data. An inventory of all European grate incineration plants including technical characteristics and plant capacities was established, and waste material mixtures were determined for different European countries, including generic elemental waste-material compositions. The model thus allows for calculation of country-specific and material-dependent emission factors and enables identification and tracking of emission sources. It thereby helps to develop strategies to decrease plant emissions by reducing or redirecting problematic waste fractions to other treatment options or adapting the technological equipment of waste incinerators.

’ INTRODUCTION Municipal solid-waste incineration plays an important role in the waste management of many European countries. Around 20% of the municipal solid waste (MSW) generated in Europe is treated by incineration, while percentages of MSW combusted in individual states range from 0% (e.g., Greece) to 54% (Denmark, 2005).1 The total number of MSW-incineration installations already exceeds 400 plants, and, due to European legislation,2 a rapid expansion of the MSW incineration sector is expected over the next 10-15 years.3 In order to provide reliable information on emission releases and the impacts of technical and legislative reduction measures, adequate models are required for proper estimation of emission loads from the changing waste-management sector. Several technology and waste-specific mass-flow models exist for municipal solid-waste incineration to simulate contaminant releases to air and water (e.g. refs 4-8). With the exception of the ecoinvent waste-incineration model,6 these tools are limited to deterministic model parameters expressed as point estimates. Although the model parameters are known to be uncertain, the tools disregard the inherent variability in the type, size, and operational modes of incineration plants, the considerable fluctuation in waste composition, and the uncertainties associated with imprecision in waste-composition and emission measurements. Some other studies quantify uncertainties of single specific issues which are restricted to, for instance, incineration-emission measurements,9 modeling stack concentrations for health risk analysis,10,11 and analyzing changes in chemical-waste compositions and bottom ash.12,13 Only very few surveys investigate the interrelation of element behavior in MSW incinerators (MSWI) and the variability in determining factors as outlined above (e.g., refs 14 and 15). Regulatory guidance documents, in addition, supply confidence r 2011 American Chemical Society

intervals for generic emission factors16 and emission concentrations,3 yet these are mainly based on expert judgment. Hence, a comprehensive MSW incineration-emission model which accounts for relevant stochastic and epistemic uncertainties in the model parameters and combines them consistently throughout the model is still lacking. The objective of this work is to develop a probabilistic and technology-specific emission model for municipal solid-waste incineration in Europe for application in environmental assessments. The model facilitates the computation of stochastic emission estimates as function of the waste composition and technologies applied. It allows for the prioritization of waste components significantly contributing to both overall emission loads and their associated uncertainties. The stochastic estimates enhance the interpretation of modeled emissions and give insight in relative relevance of influencing factors.

’ METHODS Model Scope and System Boundaries. The emission model describes municipal solid waste incineration in Europe focusing on grate incineration which by far is the most employed combustion technology (>90% of all plants). The variety of existing flue-gas treatment techniques was grouped into three main purification systems (for further details see Supporting Information SI, Section 1.1): (i) systems with fly ash separation only employing cyclones, electrostatic precipitators, or fabric filters where additional cleaning stages are missing (CYC/ESP/FF); (ii) Received: June 28, 2010 Accepted: February 1, 2011 Revised: December 6, 2010 Published: March 16, 2011 3487

dx.doi.org/10.1021/es1021763 | Environ. Sci. Technol. 2011, 45, 3487–3495

Environmental Science & Technology

ARTICLE

Figure 1. Overview of model scope and structure.

dry and semidry systems with a scrubbing stage removing acid gases with solid neutralizing absorbents or neutralizing-agent solutions and filters eliminating the solid residues and dust (DRY); and (iii) wet systems with fly ash separation and a two-stage acidic pollutant removal in aqueous solution and an on-site wastewater treatment of wet-scrubber effluents (WET). Removal of nitrogen compounds by selective (non)catalytic reduction was additionally simulated (see SI, Section 1.1), while extra treatment by

activated carbon for elimination of dioxins and trace elements (e.g., mercury) was disregarded because data which are compatible with the chosen model design were missing. The core emission model can be connected to detailed information on the incineration-waste input. The model’s system boundaries thus comprise the waste amounts combusted, the waste’s material and elemental compositions, the element transfer in the incineration process, and the resulting emission loads to air and water (Figure 1). 3488

dx.doi.org/10.1021/es1021763 |Environ. Sci. Technol. 2011, 45, 3487–3495


ARTICLE

Figure 2. Schematic segmentation of flue-gas cleaning systems into subprocesses and probabilistic modeling procedure of element transfers throughout the incineration system.

Being restricted by data availability for all required model parameters, we determined the year 2005 as reference year for the analysis. Probabilistic and Technology-Specific Model Design for Element Transfers Throughout the Incineration Process. The incineration process was simulated as waste-input dependent and technology-specific substance-flow model. The combination of furnace and three different types of flue-gas cleaning with the additional option of denitrification generates six distinct MSWI configurations. Emissions that could be causally related to the waste input were simulated using linear technology-specific transfer coefficients (tc). Transfer coefficients quantify the partitioning of contaminants into slag, fly ash, exhaust gas, slurry, and scrubber effluent. Using Student’s t-distribution, uncertainty ranges for transfer coefficients were computed from plant-specific measurements of various studies, which apply different tc determination procedures. If reference data were unavailable, generic uncertainty margins were derived (for detailed methods see SI, Section 1.3). Note that such statistical analyses of transfer coefficients show their individual variability between plants and thus do not add to 100% for a certain element (see SI, Tables 10, 11, and 12). Data analysis indicated the choice of normal distributions for most transfer coefficients, while merely for small transfer

coefficients below a threshold of 0.015 log-normal distributions are advocated in order to avoid negative values. Particularly for heavy metals with extremely small transfers e.g. into exhaust gas, tc values are generally not precisely reported but stated to fall below detection limits. In such cases, the detection limits were applied as median estimates with generic uncertainty margins. Elemental waste mixtures entering combustion were computed by matrix calculation using matrices for waste types and material and elemental compositions (for mathematical model see SI, Section 1.4). Uncertainties for all model parameters (i.e., waste composition, transfer coefficients) were propagated throughout the model utilizing Monte Carlo simulation (@Risk software17), Latin Hypercube sampling, 1000 iterations, thus generating probability distributions for substance flows into environmental media (via exhaust air and wastewater effluent). The calculation routine assumes independent parameters and covariance terms to be negligible. In order to obey the rule of mass conservation, for each technology-specific subprocess first the transfer coefficient determined by measurements was stochastically simulated and its complementary tc calculated as difference to 100%. A scheme describing the modeling procedure for each technology and its segmentation into generic subprocesses is given Figure 2. 3489


Environmental Science & Technology Due to the probabilistic estimates for waste compositions, waste type, material, and elemental compositions were customized to a 100% mass balance applying different modeling procedures (SI, Section 1.4). Because water, oxygen, and hydrogen represent environmentally uncritical chemical constituents, they acted as balance against 100% of the elemental composition. For waste types and material compositions, the largest fraction in each category was selected to complete the mass balance as only these fractions were sufficiently large to accomplish the adjustment. This choice also accounts for the fact that the relative error of the difference between the adjusted value and a value randomly selected from the fraction’s probability distribution is smaller for large fractions compared to small fractions. Emissions of chemical compounds into air were either calculated stoichiometrically from the emitted element mass (for SO2, CO2), computed on the basis of the element transfer and constant species shares estimated from literature (for NOx, NH3, N2O), or derived by statistical fitting of monitoring data related to 1 m3 of exhaust gas, independent of the elemental waste composition (for NMVOC, CO, PCDD/F, PM10). Exhaust-air volumes were calculated from the emission mass-flows of the waste constituents C, H, S, and N and the input combustion-air volume. For an overview of data sources applied, transfer coefficients quantified, emission factors calculated, and calculation procedures see SI (Section 1). Thermal Waste-Treatment Plants under Study. An inventory of all grate incinerators operated in Europe in 2005 was established. All plants were characterized concerning their geographic location, flue-gas purification technologies, and actual waste quantities combusted. The inventory covers 16 European Union member states as well as Iceland, Norway, and Switzerland. The individual plant’s flue-gas cleaning installations were assigned to the three pollution-control systems included in the model. Waste amounts burned per incineration plant in 2005 were retrieved from various bibliographic sources or approximated with a tiered estimation procedure. A generic uncertainty factor for waste amounts was determined. For plant inventory, waste amount estimation and uncertainty information see SI (Section 1.1). Compositions of Wastes Modeled. The model considers four waste types, including household waste, MSW-like commercial and medical waste, and sewage sludge. As the data availability did not allow for quantifying the exact shares of waste types for each incinerator, generic shares were derived for all plants from literature and standard deviations computed employing the Student’s t-distribution. Household, commercial, and medical wastes were subdivided into various material fractions. Material fractions considered include paper, cardboard, polyethylene terephthalate (PET), polyethylene (PE), polypropylene (PP), polystyrene (PS), mixed plastics, composite packaging, diapers, glass, textiles, inert minerals, kitchen/garden waste, wood, other organic material, ferrous metal, nonferrous metals, electronic/electrical equipments, accumulators/batteries, and other household hazardous waste (e.g., printer cartridges). The material composition of household waste, medical waste and MSW-like commercial waste was derived from various literature sources (see SI, Section 1.1). For medical and commercial waste generic material compositions as derived from literature were applied for all MSWI plants and countries. Standard deviations for material compositions of municipal solid waste were calculated from refs 18 and 19 and applied as generic uncertainty ranges for the respective materials in household, commercial, and medical waste. Rather large material fractions are reported to follow a normal

ARTICLE

probability distribution, whereas small fractions seem to be lognormally distributed.1 Using various surveys, all materials and sewage sludge were further differentiated by their chemical-element composition and split into one compound (H2O) and 34 chemical elements (Al, As, Ba, Br, C, Ca, Cd, Cl, Co, Cr, Cu, F, Fe, H, Hg, K, Li, Mg, Mn, Mo, N, Na, Ni, O, P, Pb, S, Sb, Se, Si, Sn, Ti, V, and Zn). Uncertainty distributions for elemental fractions could generally only be derived directly from literature for selected chemical elements (C, S, N, P, Cl, and common heavy metals Cr, Cu, Hg, Ni, Pb, Zn) of paper, cardboard, plastics, kitchen/garden waste, accumulators/batteries, electronic/electrical equipment, and sewage sludge. For other elements and elements of further waste materials generic uncertainties were derived. Element fractions in different materials are sometimes very small, and therefore lognormal distributions were assumed for median elemental concentrations below 0.02 kg element/kg material in order to avoid negative results in model simulations. Waste-type fractions, material and elemental compositions, their uncertainty ranges, data sources, and procedures used for estimating generic uncertainties are given in the SI (Section 1.2). Model Evaluation with Monitoring Data and Reported Emission Factors. To assess the model performance, modeled emission concentrations for a selection of air pollutants were directly evaluated against monitoring data comprising major stack pollutants regularly measured (e.g., NOx, SO2) and compared to regulatory emission limits.20 Monitoring data were collected from 83 different European MSWI plants representing different fluegas purification systems and geographic regions and considering data availability for 2005 or adjacent years (e.g., Austria, Italy) (see SI, Section 1.5). Probability distributions were fitted to monitored emission concentrations in the flue gas with @Risk software.17 Their uncertainty ranges were compared with the model results on European average level and for individual countries with sufficient plant coverage regarding monitoring data (Austria, Switzerland, and Germany). Furthermore, simulated waste-mass dependent emission factors were compared to reference values reported in European guidance documents3,16 to assess the model accuracy. Sensitivity Analysis of Influencing Parameters. In order to identify the most influencing parameters, the contribution-tovariance (CTV) of each uncertain input parameter i to an emission output j was quantified by the following equation ni

∑

CTV i, j ¼ r2i, j 3 ð r2i, j Þ-1 i¼1

ð1Þ

where ri,j is the rank-order correlation coefficient between the parameter i and the score of emission j, and ni is the number of input parameters contributing to the variance. This analysis displays the sensitivity information as percentage of emission-parameter variance due to the uncertainty of each input and model parameter. CTV values were computed for the European mix of flue-gas technologies considering the actual waste amounts incinerated in 2005. Additionally, the relative contribution of waste materials to the overall emission load on European level was assessed to identify priority materials that should be avoided or might better be directed to alternative treatment options instead of incineration.

’ RESULTS Modeled Emission Factors. To demonstrate the capabilities of the model, European average, country-specific, and 3490



ARTICLE

Figure 3. (a) Technology-specific contaminant emission factors per unit mass of waste incinerated and (b) comparison of modeled emission concentrations with monitoring data of selected European MSWI plants and regulatory limit values.

material-dependent emission factors per unit mass of waste burned were calculated for all three incineration systems modeled (Figure 3a, SI Sections 2.3 and 2.4). 95% confidence intervals (CI) of the computed technology-specific air emission factors generally span over 1-3 orders of magnitude showing for most air emissions higher variability for the CYC/ESP/FF and DRY technology compared to the WET technology (Figure 3a). Reference values from the EMEP/EEA emission inventory guidebook,16 which are estimated by expert judgments, indicate uncertainty ranges of up to 3-4 orders of magnitude, but these confidence intervals display the full range of all three off-gas purification systems studied here. Simulated air emission factors

for NOx, SO2, As, and Cd fall rather well into the range of the reference data. For most heavy metals (e.g., Pb, Cr) the model returns CYC/ESP/FF and DRY emission factors slightly higher than referenced by the guidance document, while WET emission factors cover the lower percentiles of the reported reference values (Figure 3a). Similar overlaps with the lower reference-data range result for compound emissions such as NMVOC, CO, and PM10, whereas modeled dioxins/furans fall below. Modeled European average water-emission factors indicate a large spread in uncertainty ranges (SI, Section 2.5), and for some metals (e.g., Cr, Cu, Pb) they reflect the data reported in the IPPC guidance document,3 which however characterizes water emissions from one single plant only. 3491



ARTICLE

Figure 4. Relative contribution of waste materials to emission load.

Overall, the results show correspondence between computed and reported emission factors as 95% confidence intervals overlap, and thus the distributions are not significantly different. In comparison to the modeled European average values, minimummaximum ranges of country-specific emission factors pinpoint high variability in pollutant releases among the different European countries (for detailed values see SI Section 2.3). Model Evaluation. On European average level, for both the dry and wet flue-gas purification systems, overall there is no significant difference between the simulated and measured concentrations as indicated by the overlapping 95% confidence intervals (Figure 3b). A particularly good agreement is observed for nitrogen emissions. The uncertainty ranges for monitoring data are generally larger than for simulated pollutant concentrations and reveal 97.5% percentiles for WET emission levels substantially higher than for simulated values. Evidently, monitoring data display no significant difference between emission levels of dry and wet abatement techniques, while modeled concentrations for dry and wet treatment clearly diverge for SO2, HCl, HF, Cd, and Hg releases into air. Computed 97.5% percentiles for the WET system stay substantially below the regulatory emission threshold, whereas measured concentrations and simulated concentrations from DRY pollution control partly hit these limits (e.g., for HF and Cd). Country-level comparisons of air emissions from MSWI plant with WET purification systems in Austria and Switzerland show rather good correspondence of computed and monitoring concentrations. Larger discrepancies are apparent for the DRY system of plants in Germany, yet also for this case monitoring and simulated emission levels do not show significant differences (for detailed results see SI, Section 2.6). Relative Importance of Waste Materials Regarding Overall Emission Loads. For most pollutant emissions, varying sets of up to three different waste fractions were accountable for the majority of total release (51-97%) (Figure 4). With a reliability

of 97.5% kitchen/garden waste and paper are among the three major producers of nitrogen (N2O, NOx, NH3), sulfur (SO2), and carbon (CO2) emissions into air. The CO2 fraction of biogenic origin amounts in average to 61% (see SI, Section 2.7). Mixed plastics particularly contribute to CO2, HCl, Cd, and Sb offgas releases. Apart from some special contributors (e.g., wood for Zn and composite packaging for V emissions), heavy metal releases are primarily caused by electronics, accumulators/batteries, ferrous metals, and household hazardous waste. These results largely correlate with the distribution of heavy metals among waste materials found in MSW by Burnley.21 Waste materials substantially contributing to the emission of a certain pollutant influence the country-specific emission factors according to their weight fraction. Yet, also materials may significantly add to pollutant releases which contain rather medium-size contaminant contents but are dominant in terms of mass contribution in the feed-waste mixture (e.g., relatively high Pb and Cr emission factors for Germany are attributable to quite large paper amounts in incineration-waste input). Sensitive Parameters. The results of the contribution-tovariance analysis show that both waste-input specific parameters and transfer coefficients highly affect the variance in the model outcomes (Table 1). For (heavy) metal and halogen emissions the uncertain transfer coefficients to slag and fly ash are among the top three contributors to overall variance (3-62%). Further, the uncertain (heavy) metal compositions particularly of accumulators/batteries and electronic/electrical equipment have a significant contribution-to-variance (4-66%) for (heavy) metal releases to air and water. For N, S, and F emissions into air the elemental fraction in kitchen/garden waste plays a prominent role with CTV values of 26-64%. Other relevant waste-input parameters such as the share of the pollutant containing material fractions (e.g., accumulators/batteries fraction) and the amount of household and commercial wastes in the waste feed represent 3492



ARTICLE

Table 1. Results of the Contribution-to-Variance Analysis for Selected Pollutants Indicating the Three Parameters Contributing Most to the Uncertainty of Simulated Emissions pollutant emissions into air

contribution-to-variance [%]

NOx, N2O, NH3

N fraction in kitchen/garden waste 64

fraction of household waste 6.3

fraction of commercial waste 5.7

CO

emission factor of CO 94

exhaust gas volume 5.0

amount of incinerated waste 0.5

Zn

transfer coefficient of Zn to fly ash 62

transfer coefficient of Zn to slag 16

fraction of Zn in accumulators/batteries 15

Cu

transfer coefficient of Cu to slag 43

fraction of Cu in electrical/electronic

transfer coefficient of Cu to fly ash 13

equipment 33

additional important contributors to variance (

Probabilistic and Technology-Specific Modeling of Emissions from

Recommend Documents