Waste Management 27 (2007) 389–397 www.elsevier.com/locate/wasman Continuous monitoring of odours from a composting plant using electronic noses Selena Sironi a, Laura Capelli a,*, Paolo Céntola a, Renato Del Rosso a, Massimiliano Il Grande b a Olfactometric Laboratory, Department of Chemistry, Materials and Chemical Engineering ‘‘Giulio Natta’’, Politecnico di Milano. P.za Leonardo da Vinci 32, 20124 Milan, Italy b Progress S.r.l., Via di Rocca Cencia 95, 00132 Rome, Italy Accepted 31 January 2006 Available online 29 March 2006 Abstract The odour impact of a composting plant situated in an urbanized area was evaluated by continuously monitoring the ambient air close to the plant during a period of about 4 days using two electronic noses. One electronic nose was installed in a nearby house, and the other one inside the perimeter of the composting plant in order to compare the response of both instruments. The results of the monitoring are represented by tables that report the olfactory class and the odour concentration value attributed to the analyzed air for each of the 370 measurements carried out during the monitoring period. The electronic nose installed at the house detected the presence of odours coming from the composting plant for about 7.8% of the monitoring total duration. Of the odour detections, 86% (25 of 29 measurements) were classified as belonging to the olfactory class corresponding to the open air storage of the waste screening overflows heaps, which was therefore identified to be the major odour source of the monitored composting plant. In correspondence of the measurements during which the electronic nose inside the house detected the presence of odours from the composting plant, the olfactory classes recognized by both instruments coincide. Moreover, the electronic nose at the house detected the presence of odours from the composting plant at issue in correspondence of each odour perception of the house occupants. The results of the study show the possibility of using an electronic nose for environmental odours monitoring, which enables the classification of the quality of the air and to quantify the olfactory nuisance from an industrial source in terms of duration and odour concentration. Ó 2006 Elsevier Ltd. All rights reserved. 1. Introduction In the last 30 years, odour nuisances from different industrial sources have become a serious environmental concern, especially in the case of odour emissions from waste treatment plants. One reason for the increasing number of odour complaints is that industrial plants are situated nearer to urbanized and residential areas. Furthermore, even though odorous substances are not * Corresponding author. Tel.: +39 02 23993206; fax: +39 02 23993291. E-mail address: laura.capelli@chem.polimi.it (L. Capelli). 0956-053X/$ - see front matter Ó 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.wasman.2006.01.029 directly toxic or harmful to human health, and their concentration in air is normally far below the threshold limit value (TLV) fixed by health authorities (American Conference of Governmental and Industrial Hygienists, 1999), the population has become more sensitive to air quality issues and is rarely tolerant of odours that originate from industrial activities. For these reasons, the regulatory bodies require a reliable method to evaluate odour impact originating from industrial installations. In many cases, it would be advisable to analyze air continuously, in order to detect the odours from an industrial source and to determine exactly when such odours are perceived. Since it is not always possible to correlate the 390 S. Sironi et al. / Waste Management 27 (2007) 389–397 chemical composition of a gaseous mixture with its odour properties (Defoer et al., 2002; Gostelow et al., 2001), analytical measurements are not suitable for odour determination. A method for continuous odour monitoring could be based on the use of electronic noses (Guy et al., 2004). The applications of electronic noses to the food industry have been proven (Ampuero and Bosset, 2003; Pardo and Sberveglieri, 2002; Schaller et al., 1998), whereas the possibility of applying these instruments for monitoring odour emissions of environmental interest (Abbas et al., 2001; Helli et al., 2004; Negri and Reich, 2001; Stuetz and Bourgeois, 2004), such as odours originating from waste treatment works and composting facilities (Baby et al., 2005; Rajamäki et al., 2005), is still under study. One difficulty of using electronic noses for environmental applications is that odours often are complex mixtures containing dozens of different compounds (Davoli et al., 2003; Smet et al., 1999). The greatest problem associated with the continuous monitoring of environmental odours is given by the multitude of different odours that can be present and complicate the recognition of emissions originating from the monitored source (Nicolas and Claude, 2004). Furthermore, in order to use electronic noses for environmental uses, the instruments should be able not only to qualify odours, but also to quantify the odour concentration as well (Guy et al., 2004). The aim of this work was to evaluate the possibility of using an electronic nose for the analysis of odours emitted from a composting plant and assessing the presence of odours in a nearby house whose owners often complained about the unpleasant odours originating from the composting plant. The electronic nose response was correlated with the odour concentration values measured by dynamic olfactometry (EN 13725, 2003), in order to use the instrument for the continuous odour concentration measurement of the odour source. In order to optimize the electronic nose and to maximize its capacity of qualitatively and quantitatively recognizing odours, it was necessary to make some operational choices. These choices concern the sensors (number, typology and composition) (Nagle et al., 1998), the sensor operating conditions (temperature, catalysts, etc.) (Kalinina et al., 2003; Yamazoe et al., 2003), the analysis planning (number of measurements, duration, etc.) and the data elaboration methods (algorithms for the odour recognition and quantification) (Lavine, 1999). In order to choose the parameters that are necessary for the electronic nose optimization, some test analyses were carried out with odorous air samples collected in the composting plant at issue. Some difficulties, which could affect the instrument accuracy and reliability, were encountered while using electronic noses. The principal problems that will have to be faced in the next studies are the sensor sensitivity to humidity and the instability of the sensor baseline, even though this difficulty has been partially solved by introducing a silica gel and an activated carbon filter for the deodorization and dehumidification of the reference air. 2. Materials and methods 2.1. Olfactometric analysis The olfactometric analysis was conducted in conformity with the European Norm EN 13725 (2003) in the Olfactometric Laboratory of the Politecnico di Milano (Italy). Dynamic olfactometry is a sensorial technique that determines the odour concentration of an odorous air sample using selected panellists. The odour concentration is expressed in European odour units per cubic metre (ouE/ m3), and it represents the number of dilutions with neutral air that are necessary to bring the concentration of the sample to its odour perception threshold concentration. The analysis is carried out by presenting the sample to the panel at increasing concentrations by decreasing serial dilutions, until the panel members perceive an odour that is different from the reference neutral air. The odour concentration is then calculated as the geometric mean of the odour threshold values of each panellist, multiplied by a factor that depends on the olfactometer dilution step factor. An olfactometer ECOMA Mannebeck model TO7, based on the ‘‘yes/no’’ method, was used as a dilution device. This instrument with aluminium casing has 4 panellist places in separate open boxes. Each box is equipped with a stainless steel sniffing port and a push-button for ‘‘yes’’ (odour threshold). The measuring range of the TO7 olfactometer starts from a maximum dilution factor of 1:64,000, with a dilution step factor 2. All of the measurements were conducted within 30 h after sampling, relying on a panel composed by 8 panellists (4 + 4). The odour concentration was calculated as geometric mean of p the ffiffi odour threshold values of each panellist, multiplied by 2. 2.2. Electronic noses 2.2.1. Instrument description The electronic noses that were employed for this experimentation have been developed in collaboration with Sacmi Group, Imola, Italy (Pardo and Sberveglieri, 2004). The electronic noses that were employed for this monitoring worked with six thin film metal oxide semiconductor (MOS) gas sensors, because of their properties of long term stability and reproducibility (Sberveglieri, 1995). The reducing volatile compounds in the sampled air are adsorbed on the sensor active layer producing a variation in the sensor electric resistance (Yamazoe et al., 2003). The sensors used in both electronic noses have a high sensitivity, especially towards sulphur compounds and alcohols. The selected combinations of sensors and temperatures are shown in Table 1. 2.2.2. Training The first step of the work consisted in the electronic nose training, which is fundamental in order to calibrate the instrument and to create a complete database for the odour recognition. For this reason, it was necessary to identify the 391 S. Sironi et al. / Waste Management 27 (2007) 389–397 Table 1 Sensors and operating temperatures of the electronic noses EOS 3 and EOS 9 installed inside the house and inside the composting plant Electronic nose Sensor Material T (°C) EOS 3 (House) 1 2 3 4 5 6 Tin oxide catalyzed with gold (SnO2 + Au) Tin oxide catalyzed with silver (SnO2 + Ag) Tungsten oxide (WO3) Tin oxide catalyzed with molybdenum (SnO2 + Mo) Tin oxide + indium oxide (SnO2 + In2O3) Indium oxide (In2O3) 450 450 400 425 425 400 EOS 9 (Composting plant) 1 2 3 4 5 6 Tin oxide catalyzed with gold (SnO2 + Au) Indium oxide (In2O3) Tungsten oxide (WO3) Tungsten oxide (WO3) Tin oxide catalyzed with silver (SnO2 + Ag) Tin oxide catalyzed with molybdenum (SnO2 + Mo) 400 475 350 375 425 425 olfactory classes corresponding to the principal potential odour sources in the monitored composting plant. The considered odour sources were the open air stocked heaps and the biofilter. Therefore, air samples were collected directly on the heaps of ground green waste, curing compost and waste screening overflows, and over the biofilter surface. Furthermore, some odorous ambient air samples were collected near the stocking area of the heaps. For the sample collection on the heaps, a flux chamber system was employed (Eklund, 1992; Reinhart et al., 1992; Sironi et al., 2002). This system consists of a hood, which is positioned on the heaps for a sufficient time to enable the establishing of equilibrium conditions between gas and solid phase. Subsequently the air sample was collected by sucking the air by means of a depression pump inside a Nalophane bag with a Teflone inlet tube inserted into the hood. The sampling of the air coming out from the biofilter was made by means of a static hood, which has the function of isolating the sampling point from the external conditions, and to channel the air stream in a stack from where the sample was finally collected with a depression pump (Bockreis and Oxbol, 2004). Subsequently, some air samples corresponding to odours that are not attributable to the composting plant but that are present in proximity of the house where one electronic nose was installed were collected, because these odours might have been detected by the instrument during the following monitoring phase. For this purpose, air samplings were carried out along the street that passes in front of the house, and above some grass heaps recently cut in the adjoining meadows. Finally, it was necessary to analyze some non odorous ambient air in order to add a reference olfactory class corresponding to neutral air to the electronic nose database. The odour concentration of all samples collected for the electronic nose training was measured by dynamic olfactometry (Table 2). The quantification of odour concentration by means of an electronic nose requires a particular training: air samples corresponding to each olfactory class are diluted at different odour concentration values and analyzed in order to create a database that can be used for the estimation of the Table 2 Odour concentration of the samples collected for the electronic nose training Sample no. Sample description Odour concentration (ouE/m3) 1 2 3 4 5 6 Biofilter outcoming air Plant ambient air Ground green waste heap Curing compost heap Waste screening overflows heap Cut grass 53 72 5000 2100 4200 2000 odour concentration of unknown air samples by interpolation of the training points. Once the odour concentration of the collected samples was measured, it was possible to decide which samples to dilute and which dilution ratios to apply in order to obtain samples with odour concentration values included in the typical odour concentration range of odorous ambient air (30–300 ouE/m3). Among the air samples collected inside the plant, only the ones taken directly on the heaps were diluted, because the other ones presented too low of odour concentration values. The dilution device that was used is the same olfactometer employed for the olfactometric analyses (Table 3). Fig. 1 shows the response curves of sensor 1 and sensor 3 of the electronic nose EOS 3 relevant to the analysis of the same air sample (collected on the curing compost heaps) at different odour concentration values (the odour Table 3 Dilution factors and odour concentrations of the samples for the electronic nose training Sample ouE/m3 Dilution factor ouE/m3 Ground green waste heap 5000 113 56.6 14.1 44 88 355 Curing compost heap 2100 56.6 28.3 7.1 37 74 296 Waste screening overflows heap 4200 113 56.6 14.1 37 74 298 392 S. Sironi et al. / Waste Management 27 (2007) 389–397 Fig. 1. Response curves of sensor 1 and sensor 3 of the electronic nose EOS 3 at different odour concentration values. concentration value is indicated under each curve). It is possible to observe that there is a correlation between the sensor resistance variation and the odour concentration of the analyzed air. 2.2.3. Monitoring The electronic nose EOS 3 was installed inside a house located at about 300 m from the composting plant at issue. The instrument was positioned in the attic, with the Teflone inlet tube for the gas suction coming out from a window and directed towards the composting plant. The electronic nose EOS 9 was installed inside the composting plant, inside an office in a building positioned between the biofilter and the heaps stocking area. The stabilization of the sensor response baseline was obtained by passing the reference air through a silica gel and an activated carbon filter, in order to have it humidity free. The electronic noses ran an air analysis every 12 min during a period of about 4 days, starting the 19th April 2004 at 10.00 a.m. until the 23rd April 2004 at 11.00 a.m. During each time slot, the instruments sucked the ambient air and analyzed it for 3 min, whereas for the following 9 min the reference air was fluxed inside the sensor chamber in order to clean the sensors and to bring their resistance value back to the baseline for the next measurement. The silica gel for the reference air dehumidification loses adsorption capacity after a certain time depending on the air humidity content. For this reason, it was necessary to substitute the filters of both electronic noses every 12 h during the whole monitoring period. 2.2.4. Data processing The first step of data processing is the extraction of some significant features from the sensor response curves in order to produce a dataset that can be elaborated by the electronic nose pattern recognition system. For the feature selection, consideration was given to the fact that one of the principal problems associated with the odour monitoring by means of an electronic nose is represented by the baseline instability, which is influenced by different factors (e.g., humidity or temperature variations). The features considered were the minimum sensor resistance value (corresponding to the lowest peak of the response curve) and the first 10 coefficients of the Fourier transform, which approximates the sensor response curve. As far as concerns the pattern recognition algorithms, the KNN (K-nearest-neighbours) algorithm (González Martı́n et al., 2001; Roggo et al., 2003) in the electronic nose software was used to classify the odour quality. This algorithm calculates the Euclidean distance between the point relevant to the sample that has to be classified and the points of the training dataset. The sample is then attributed to the olfactory class to which the k nearest points belong, where k is a parameter that must be set by the operator. In this case, the parameter k was set equal to 1, i.e., the unknown sample quality was attributed to the olfactory class corresponding to its nearest neighbour. Even through its apparent simplicity, this algorithm gives reliable results and it is very effective in the qualitative classification of unknown samples (Sironi et al., 2004). The odour concentration quantification was obtained by linear interpolation of the training dataset. The variables that are used for the linear regression correspond to the features that are extracted from the sensor response (i.e., minimum sensor resistance value and 10 coefficients of the Fourier transform which approximates the sensor response curve). The feature extraction from the signal produced by the analysis of an unknown sample produces a feature set, which can be used in order to estimate its odour concentration. This feature set is composed by parameters that have a different and not a priori definable dependence from the stimulus that originates the sensor signal. Given the variability of the laws that regulate the dependence of the feature set from the sample concentration, a linear interpolation was used. The choice of considering a narrow odour concentration range as training dataset further justifies the use of a linear interpolation for the odour concentration estimation of unknown samples, even though the relation between sensor signal and sample concentration is not linear. For this reason, in order to improve the accuracy of the estimation, only the data relevant to the diluted 393 S. Sironi et al. / Waste Management 27 (2007) 389–397 Fig. 2. Linear interpolation of the training datasets used for the odour concentration estimation. samples were used as training dataset, i.e., the samples with typical odour concentration values of an ambient air (between 30 ouE/m3 and 100 ouE/m3). The results of the linear interpolation of the training datasets for both electronic noses are shown in Fig. 2. 3. Results and discussion The results are represented by large tables that report the olfactory class and the odour concentration value attributed to the analyzed air for each measurement carried out during the monitoring period. A synthetic representation of the qualitative air classification results is given in Figs. 3 and 4. The abscissa reports the measurement date and hour, while the ordinate reports the olfactory class attributed to the analyzed air. Fig. 3 also indicates the periods during which the owners of the house, where one electronic nose was installed, perceived the presence of odours arising from the near composting plant. Figs. 5 and 6 illustrate the results of the odour concentration estimation for each measurement. Among the 370 useful measurements executed by the electronic nose EOS 3 in the house during the monitoring period (corresponding to a total time of 74 h), 29 were attributed to the composting plant. This result indicates that the instrument detected the presence of odours coming from the plant at issue for about 7.8% of the monitoring total duration. The qualitative recognition of the analyzed air by the electronic nose EOS 3 shows that 86% of the odour detections (25 of 29 measurements) were classified as belonging to the olfactory class of the ‘‘waste screening overflows’’ heaps. As shown in Table 5, this is confirmed by the electronic nose EOS 9 at the same time. Based on this consideration, it may be affirmed that these open air stocked heaps represent the major odour source of the composting plant. citizens odour complaints waste screening overflows heap curing compost heap plant ambient air neutral air 19/04/2004 0.00 19/04/2004 12.00 20/04/2004 0.00 20/04/2004 12.00 21/04/2004 0.00 21/04/2004 12.00 22/04/2004 0.00 22/04/2004 12.00 23/04/2004 0.00 23/04/2004 12.00 24/04/2004 0.00 Fig. 3. Quality of the ambient air analyzed by the electronic nose EOS 3 inside the house and citizens odour complaints. 394 S. Sironi et al. / Waste Management 27 (2007) 389–397 waste screening overflows heap curing compost heap ground green waste heap biofilter outcoming air plant ambient air neutral air 19/04/2004 19/04/2004 20/04/2004 20/04/2004 21/04/2004 21/04/2004 22/04/2004 22/04/2004 23/04/2004 23/04/2004 24/04/2004 0.00 12.00 0.00 12.00 0.00 12.00 0.00 12.00 0.00 12.00 0.00 Fig. 4. Quality of the ambient air analyzed by the electronic nose EOS 9 inside the composting plant. 120 100 80 60 40 20 0 19/04/2004 0.00 19/04/2004 12.00 20/04/2004 0.00 20/04/2004 12.00 21/04/2004 0.00 21/04/2004 12.00 22/04/2004 0.00 22/04/2004 12.00 23/04/2004 0.00 23/04/2004 12.00 24/04/2004 0.00 Fig. 5. Odour concentration values estimated by the electronic nose EOS 3 inside the house. The detection of odours referable to the composting plant in the house where the electronic nose was installed was concentrated in some specific time bands, particularly the evening (from 7.00 p.m. to 10.00 p.m.) and the night (from 3.40 a.m. to 7.40 a.m.). These time bands are not associated with specific operational practices (i.e., waste or compost processing, moving or turning). For this reason it is reasonable to suppose that the odour perception outside of the plant is strongly influenced by the nightly meteorological conditions. A comparison of the sensations of the house owners with the electronic nose classification results demonstrated that in correspondence of each odour perception of these people the electronic nose EOS 3 detected the presence of odours from the composting plant at issue. The quality of this correspondence can be evaluated with a kind of ‘‘confusion matrix’’ (Kohavi and Provost, 1998) showing the number of measurements for which the electronic nose odour detections and the house owners perceptions match (Table 4). In general, the accuracy of a classification system can be evaluated as the proportion of the total number of predictions that were correct, i.e., in this case, the proportion of the total number of observations that coincide: 395 S. Sironi et al. / Waste Management 27 (2007) 389–397 160 140 120 100 80 60 40 20 0 19/04/2004 0.00 19/04/2004 12.00 20/04/2004 0.00 20/04/2004 12.00 21/04/2004 0.00 21/04/2004 12.00 22/04/2004 0.00 22/04/2004 12.00 23/04/2004 0.00 23/04/2004 12.00 24/04/2004 0.00 Fig. 6. Odour concentration values estimated by the electronic nose EOS 9 inside the composting plant. Table 4 Confusion matrix relevant to house owners odour perceptions and electronic nose odour detections EOS 3 qualitative classification House owners perceptions Odour complaint No odour complaint AC ¼ Odours from composting plant Neutral air 12 132 0 327 12 þ 327 ¼ 72% 12 þ 327 þ 132 þ 0 The large number of measurements during which the electronic nose EOS 3 recognized the presence of odours from the composting plant while the house owners did not, does not necessarily indicate a failure of the classification system. The first reason is that the electronic nose, which analyzes the air continuously, has a higher and more regular sampling frequency than the frequency of the observations of the house occupants. Moreover, the instrumental detection limit may be lower than the odour detection threshold of the house owners. It is possible to make a similar discussion in order to compare the results of the qualitative classification of the electronic nose installed inside the house with those of the electronic nose installed inside the composting plant. In this case, the ‘‘confusion matrix’’ shows the number of measurements for which the olfactory classes attributed by both instruments to the analyzed air match (Table 5) and allows the existence of a correspondence between these classes to be highlighted. An important observation is that for 26 of 28 measurements during which the electronic nose EOS 3 installed inside the house detected the presence of odours from the composting plant, the olfactory classes recognized by both instruments coincide. The accuracy in this case can be calculated as the proportion of the total number of measurements for which the olfactory classes match: 24 þ 1 þ 1 þ 67 24 þ 1 þ 1 þ 67 þ 9 þ 19 þ 133 þ 6 þ 25 þ 1 þ 1 þ 1 ¼ 7% AC ¼ The low accuracy is due to the high number of measurements in which the electronic nose EOS 9 recognized the Table 5 Confusion matrix relevant to the olfactory classes attributed by both electronic noses to the analyzed air EOS 9 qualitative classification Waste screening overflows heaps EOS 3 qualitative classification Waste screening overflows heaps Curing compost heaps Plant ambient air Ground green waste heap Biofilter outcoming air Neutral air Curing compost heaps Plant ambient air Ground green waste heap Biofilter outcoming air Neutral air 24 0 1 0 0 0 0 0 0 0 91 1 0 0 0 19 0 1 0 0 133 0 0 0 0 6 0 1 0 0 25 1 0 0 0 67 396 S. Sironi et al. / Waste Management 27 (2007) 389–397 presence of odours from the composting plant, while the electronic nose EOS 3 installed inside the house classified the air as ‘‘Neutral air’’. These mismatches do not indicate a failure of the classification system, but they are due to the fact that the perception of odours inside the plant does not necessarily implicate that these odours reach the house where the monitoring took place. This situation is relatively frequent, and it depends prevalently on the meteorological conditions (e.g., principally wind direction and wind speed, but also solar radiation). This consideration is useful for understanding why there is a correspondence between the odour detections with the electronic nose EOS 9 installed inside the plant and the waste and compost processing, or the material turning or moving operations (the plant director was asked to make a note of all the operations on waste and compost that were carried out during the monitoring period), whereas the detection of odours attributed to the composting plant by the electronic nose EOS 3 installed inside the house is not directly connected to these operations. A further confirmation of the results was given by the fact that the measurements during which the electronic nose inside the house detected the presence of odours imputable to the composting plant were contemporaneous to periods in which relatively high odour concentration values were measured inside the plant (70–100 ouE/m3) as compared to values of the remaining part of the day (30–60 ouE/m3). 4. Conclusions Odours imputable to the composting plant at issue were detected by the electronic nose EOS 3 installed inside the house for about 7.8% of the monitoring total duration, and the major odour source of the plant was identified to be the open air area for the storage of the waste screening overflows heaps. The possibility of identifying the principal odour source represents an useful project tool: in this case, the results highlight that a significant odour impact reduction could be achieved by closing this area. The results show that once properly trained, electronic noses are able to qualitatively and quantitatively recognize odours. Electronic noses can therefore be successfully used for the continuous monitoring of composting odours. Future studies should focus on the solution of the problems due to the sensor sensitivity to humidity and to the baseline instability. References Abbas, M.N., Moustafa, G.L., Gopel, W., 2001. Multicomponent analysis of some environmentally important gases using semiconductor tin oxide sensors. Analytica Chimica Acta 431, 181–194. American Conference of Governmental and Industrial Hygienists (ACGIH), 1999. 1999 TLVs and BEIs: Threshold Limit Values for Chemical Substances and Physical Agents, Biological Exposure Indices. Cincinnati, OH. Ampuero, S., Bosset, J.O., 2003. The electronic nose applied to dairy products: a review. Sensors and Actuators B 94, 1–12. Baby, R.E., Cabezas, M.D., Labud, V., Marqui, F.J., Walsöe de Reca, N.E., 2005. Evolution of thermophilic period in biosolids composting analyzed with an electronic nose. Sensors and Actuators B 106, 44–51. Bockreis, A., Oxbol, A., 2004. Sampling techniques of odour in Denmark and Germany. In: Environmental Odour Management, VDI-Berichte 1850. VDI Verlag GmbH, Düsseldorf, Germany, pp. 247–260. Davoli, E., Gangai, M.L., Morselli, L., Tonelli, D., 2003. Characterisation of odorants emissions from landfills by SPME and GC/MS. Chemosphere 51, 357–368. Defoer, N., De Bo, I., Van Langenhove, H., Dewulf, J., Van Elst, T., 2002. Gas chromatography–mass spectrometry as a tool for estimating odour concentrations of biofilter effluents at aerobic composting and rendering plants. Journal of Chromatography A 970, 259–273. Eklund, B., 1992. Practical guidance for flux chamber measurements of fugitive volatile organic emission rates. Journal of Air & Waste Management Association 42, 1583–1591. EN 13725, 2003. Air Quality – Determination of Odour Concentration by Dynamic Olfactometry. Comité Européen de Normalisation, Brussels, Belgium, pp. 1–70. González Martı́n, Y., Cerrato Oliveros, M.C., Pérez Pavón, J.L., Pinto, C.G., Cordero, B.M., 2001. Electronic nose based on metal semiconductor sensors and pattern recognition techniques: characterisation of vegetable oils. Analytica Chimica Acta 449, 69–80. Gostelow, P., Parsons, S.A., Stuetz, R.M., 2001. Odour measurements for sewage treatment works. Water Research 35 (3), 579–597. Guy, C., Giasson, F., Purenne, P., 2004. Automated electronic nose network for measuring industrial odours. In: Environmental Odour Management, VDI-Berichte 1850. VDI Verlag GmbH, Düsseldorf, Germany, pp. 431–439. Helli, O., Siadat, M., Lumbreras, M., 2004. Qualitative and quantitative identification of H2S/NO2 gaseous components in different reference atmospheres using a metal oxide sensor array. Sensors and Actuators B 103, 403–408. Kalinina, M.V., Moshnikov, V.A., Tikhonov, P.A., Tomaev, V.V., Mikhailichenko, S.V., 2003. Temperature dependence of the resistivity for metal-oxide semiconductors based on tin oxide. Glass Physics and Chemistry 29 (4), 422–427. Kohavi, R., Provost, F., 1998. Glossary of terms. Editorial for the special issue on application of machine learning and the knowledge of discovery process. Machine Learning 30, 271–274. Lavine, B.K., 1999. Clustering and classification of analytical data. In: Robert, A. Meyers (Ed.), Encyclopedia of Analytical Chemistry, Chemometrics. John Wiley & Sons Ltd, Chichester, UK. Nagle, H.T., Gutierrez-Osuna, R., Schiffman, S.S., 1998. The how and why of electronic noses. IEEE Spectrum 35 (9), 22–31. Negri, R.M., Reich, S., 2001. Identification of pollutant gases and its concentrations with a multisensor array. Sensors and Actuators B 75, 172–178. Nicolas, J., Claude, A.C., 2004. Establishing the limit of detection and the resolution limits of odorous sources in the environment for an array of metal oxide gas sensors. Sensors and Actuators B 99, 384–392. Pardo, M., Sberveglieri, G., 2002. Coffee analysis with an electronic nose. IEEE Transactions on Instrumentation and Measurement 51 (6), 1334–1339. Pardo, M., Sberveglieri, G., 2004. Electronic olfactory systems based on metal oxide semiconductor sensor arrays. MRS Bulletin 29 (10), 1–6. Rajamäki, T., Arnold, M., Venelampi, O., Vikman, M., Räsänen, J., Itävaara, M., 2005. An electronic nose and indicator volatiles for monitoring of the composting process. Water, Air, and Soil Pollution 162, 71–87. Reinhart, D.R., Cooper, D.C., Walker, B.L., 1992. Flux chamber design and operation for the measurement of municipal solid waste landfill gas emission rates. Journal of Air & Waste Management Association 42, 1067–1070. S. Sironi et al. / Waste Management 27 (2007) 389–397 Roggo, Y., Duponchel, L., Huvenne, J.P., 2003. Comparison of supervised pattern recognition with McNemar’s statistical test: application to qualitative analysis of sugar beet by near-infrared spectroscopy. Analytica Chimica Acta 477, 187–200. Sberveglieri, G., 1995. Recent developments in semiconductor thin-film gas sensors. Sensors and Actuators B 23, 103–109. Schaller, E., Bosset, J.O., Escher, F., 1998. ‘Electronic noses’ and their application to food. Lebensmittel-Wissenschaft und -Technologie 31, 305–316. Sironi, S., Il Grande, M., Céntola, P., Del Rosso, R., Gronchi, P., 2002. Odour control in composting plants. In: Pierucci, Sauro (Ed.), AIDIC Conference Series, vol. 5. Editoriale Elsevier, Milano, Italy, pp. 293– 300. 397 Sironi, S., Capelli, L., Rossi, A.N., Il Grande, M., Giorgi, M. 2004. Monitoring and controlling the release of landfill gas. In: Proceedings of the ISWA World Congress 2004. 17–21 October, Rome, Italy. Smet, E., Van Langenhove, H., De Bo, I., 1999. The emission of volatile compounds during the aerobic and the combined anaerobic/ aerobic composting of biowaste. Atmospheric Environment 33, 1295–1303. Stuetz, R.M., Bourgeois, W., 2004. Application of non-specific gas sensors for detection of odour abatement process failure. In: Environmental Odour Management, VDI-Berichte 1850. VDI Verlag GmbH, Düsseldorf, Germany, pp. 463–471. Yamazoe, N., Sakai, G., Shimanoe, K., 2003. Oxide semiconductor gas sensor. Catalysis Surveys from Asia 7 (1), 63–75.