Instrumental Odour Monitoring Systems (IOMS) represent the only tool available for environmental monitoring capable to perform real-time characterization of ambient air. They have been commonly used to assess odour impact at receptors thanks to their capability to detect odours and identify their provenance. An emerging application of IOMS concerns the real-time monitoring of emissions at plant fencelines. To do this, IOMS must provide a fast and accurate measurement of the odour concentration.
The most common approach, currently applied for odour quantification models, involves simplified regression algorithms, neglecting the classification of detected odours before quantification. This results in poorly accurate estimations of the odour concentration since IOMS responses to samples having the same odour concentration, but representative of different sources, may differ significantly.
Carmen Bax*, Beatrice Lotesoriere, Laura Capelli
Politecnico di Milano, Department of Chemistry, Materials and Chemical Engineering “Giulio Natta”, Piazza Leonardo da Vinci 32, 20133 Milano, Italy. *Carmen Bax
Competing interests: The author has declared that no competing interests exist.
Academic editor: Carlos N. Díaz.
Content quality: This paper has been peer reviewed by at least two reviewers. See scientific committee here.
Citation: Carmen Bax, Beatrice Lotesoriere, Laura Capelli. 2021. IOMS for the real-time monitoring of odour concentration at a msw landfill, 9th IWA Odours & VOC/Air Emission Conference, Bilbao, Spain, www.olores.org.
Copyright: 2021 Olores.org. Open Content Creative Commons license. It is allowed to download, reuse, reprint, modify, distribute, and/or copy articles in olores.org website, as long as the original authors and source are cited. No permission is required from the authors or the publishers.
Keywords: e-nose, quantification, regression, environmental monitoring, performance verification, uncertainty.
Instrumental Odour Monitoring Systems (IOMS) represent the only tool available for environmental monitoring capable to perform real-time characterization of ambient air. They have been commonly used to assess odour impact at receptors thanks to their capability to detect odours and identify their provenance. An emerging application of IOMS concerns the real-time monitoring of emissions at plant fencelines. To do this, IOMS must provide a fast and accurate measurement of the odour concentration. The most common approach, currently applied for odour quantification models, involves simplified regression algorithms, neglecting the classification of detected odours before quantification. This results in poorly accurate estimations of the odour concentration since IOMS responses to samples having the same odour concentration, but representative of different sources, may differ significantly. This paper proposes a new approach for estimating the odour concentration by IOMS, which is based on multi-phase model involving specific regression models for each odour source of the plant under examination. The paper describes the monitoring of odours from a MSW landfill by one IOMS (Rubix WT1) installed at plant fenceline. The work focuses on describing the experimental protocol used for IOMS training and performance verification in the field, and comparing the proposed regression model with a most commonly applied model that doesn’t involve qualitative classification of analysed air. With our model, the odour concentration estimated for almost all landfill samples fell within the confidence interval of the olfactometric measurements, whereas the simplified model generally resulted in overestimation of the odour concentration, especially for Fresh Waste samples. Thus, sample classification should be used as input for implementing effective quantification models.
In recent years, Instrumental Odour Monitoring Systems (IOMS) are increasingly applied as air quality monitoring tools (Sohn et al., 2008; Pan and Yang, 2009; Laor et al., 2014; Deshmukh et al., 2015; Licen et al., 2018; Bax et al., 2020a; b). They are commonly involved for a direct assessment of the odour impact at receptors, i.e., where the odour nuisance is lamented. The odour impact is calculated in terms of the frequency of odour events attributable to the plant under examination occurring during the monitoring period (Bax et al., 2021).
An emerging application of IOMS in the environmental field concerns the real-time monitoring of odours at plant fenceline, which, besides providing information about odour nuisance, offers many advantages for plant management. Indeed, the setting of “warning” thresholds for the odour concentration at fenceline and the real-time signalling of threshold exceedance enables the instantaneous identification of plant malfunctions, thereby allowing to prevent odour episodes at receptors (Bax et al., 2021). In Italy, this approach is gaining acceptability and the installation of IOMS at plant fenceline is being prescribed more and more frequently (Bax et al., 2021; Cangialosi et al, 2018).
For this purpose, IOMS must provide a continuous, fast and accurate measurement of the odour concentration. Some scientific papers have already proposed the adoption of IOMS for odour quantification (Deshmukh et al., 2017; Hu et al., 2014; Laor et al., 2014). However, the most common approach used for building quantification models based on gas sensor signals involves simplified regression algorithms, neglecting the classification of the detected odours previous to quantification. Since IOMS responses to samples having the same odour concentration, but belonging to different odour classes, may differ significantly, those models are often ineffective and poorly accurate.
This paper proposes a new approach for odour quantification by IOMS, and describes a case study related to the monitoring of odours from a landfill for non-hazardous waste, carried out by an IOMS, i.e., the RubiX WT1, installed at the Southern fenceline of the landfill along the prevalent wind direction. The work focuses on describing the experimental protocol used for IOMS training and building the quantification model. The proposed approach is based on a double-step model. As first step, the model classifies detected odours. Then, the concentration of the recognized odours is estimated based on the specific regression model. To investigate the importance of classification before quantification, the proposed multi-phase regression model (model “A”) was compared with a most commonly applied model (model “B”), which does not involve the qualitative classification of the analysed air. Since IOMS data are gaining legal value, this paper proposes an experimental protocol for IOMS performance verification in the field aimed at the verification of the reliability of IOMS outcomes. The proposed performance testing procedure involves the execution of verification tests after IOMS installation at the monitoring site to test its capability of detecting, recognizing and quantifying odours, by using samples representative of odour sources of the plant under examination independent from data used for IOMS training.
2. Materials and methods
2.1. IOMS used for the study
The IOMS used for the study is an outdoor e-nose commercialized by RubiX S&I SAS. It is equipped with 6 commercial sensors: 4 MOS sensors for odours and 2 electrochemical cells specific for H2S and NH3, respectively. This instrument is capable to supply real-time alerts, based on the combination of the sensors outputs. For this specific case study, the WT1 was installed at the plant fenceline, close to the landfill entrance, along the prevalent wind direction (i.e., from North to South). The instrument performed a continuous characterization (i.e., acquisition frequency 1Hz) of the ambient air to detect, classify and quantify odours from the landfill that might generate odour events at the receptor, located at about 2 km south the landfill, where the presence of odour attributable to waste disposal is lamented.
2.2. IOMS training
The IOMS training consists in the creation of a reference dataset, i.e., the Training Set (TS), to be used during the monitoring phase to characterize the analysed air. The first step of the training phase involved the identification and the olfactometric characterization of odour emission sources of the landfill under investigation. Fresh waste disposal and pre-treatment sections, landfill gas collection wells and leachate collection tanks were considered. Two olfactometric campaigns were carried out under different meteorological conditions (i.e., sunny and foggy), to include in the training set the intrinsic variability of landfill emission sources. To build the TS, based on their odour concentration measured by dynamic olfactometry (EN13725:2003), the odour samples were presented to the WT1 pure or diluted with odourless ambient air. The odour concentration range of training samples should be representative of the concentration level that the WT1 will be exposed in the field. Based on this principle, samples with an odour concentration ranging from 80 to 1000 ouE /m3 were used for training. Also ambient air samples, collected at the monitoring site when no odour could be perceivable by operators, were analysed, to create an olfactory class “Air” as a reference (Bax et al., 2021). The PCA score plot relevant to the WT1 TS (Figure 1)Error: Reference source not found pointed out the potentialities of the WT1 to differentiate the landfill odour sources: “Air”, “Fresh Waste”, “Landfill Gas” and “Leachate”. Samples representative of different classes cluster in different regions of the plot. However, some samples belonging to the “Landfill Gas” class, which were sampled at wells collecting the landfill gas produced in an area of the landfill still in cultivation, fall very close to the “Fresh Waste” cluster. Probably, because the process of waste biological degradation was not over, their chemical composition and, thus, their odour fingerprint resulted somewhere in between the fresh waste and the landfill gas classes.
Figure 1. PCA score plot relevant to IOMS training for the landfill monitoring.
A two-step decisional model was built on training data. The proposed model involves as first step a 5-NN (Nearest Neighbour) classifier to provide a classification of the odours detected at the landfill fenceline, and 3 Partial Least Squares (PLS) regression models (i.e., one for each landfill odour class) to estimate the odour concentration. The model was implemented considering the responses of both 4 MOS sensors and specific H2S and NH3 sensors relevant to the analysis of the training samples. The resistance of MOS sensors and the concentration assessed by electrochemical sensors were used as features.
2.3. Performance testing
According to the experimental protocol proposed in our previous works (Bax et al., 2020a, Bax et al. 2021), new samples independent form the TS were sampled at landfill emissions sources, analysed by dynamic olfactometry and presented to the IOMS after its installation at the monitoring site pure or diluted with odourless ambient air based on their odour concentration. Odour samples were alternated to odourless ambient air samples, to simulate the odour events that might occur during the monitoring.
The detection and classification performances were expressed in terms of accuracy indexes, respectively AIdetection and AI classification (Bax et al., 2020a). Conversely, the IOMS quantification capability was assessed by means of Bland and Altman method, which allows comparing two uncertain methods by studying the mean difference (i.e., bias) and constructing limits of agreement (LoA) (Giavarina, 2015). It is expected that the 95% limits include 95% of differences between the two measurement methods (Bland and Altman, 1995; Bland and Altman, 1999). The definition of LoA is reported in our previous publication (Bax et al., 2021).
3. Results and discussion
3.1. IOMS performance testing
During the monitoring phase, 22 odourless air samples and 37 landfill samples at different concentrations (i.e, from 80 ouE/m 3 to 1000 ouE/m3), independent from the TS, were presented to the WT1 to verify its detection, classification and quantification capability. The WT1 proved capable to detect and correctly classify landfill odours with accuracy indexes above 90% for both the detection and the classification of landfill odours (Table 1).
Table 1. Test characteristics relevant to IOMS detection and classification capability
Accuracy Index detection
Accuracy Index classification
Figure 2. Bland&Altman plot relevant to field verification tests.
In order to evaluate the IOMS performance in terms of odour quantification capabilities, the odour concentration values measured by the WT1, expressed as logarithms, were compared with the values measured by olfactometry by applying the B&A analysis. The B&A plot (Figure2) pointed out that the odour concentration estimated by the IOMS for almost all landfill samples fell within the confidence interval of the olfactometric measurement. In fact, only for one “Landfill Gas” sample, misclassified as Fresh Waste, the estimated odour concentration resulted slightly overestimated, even though it was within the confidence interval of the upper LoA. This result highlighted the need of using sample classification as input for the construction of quantification models.
3.2. Comparison of the multi-phase quantification model with the most common approach involved for odour quantification
To investigate the importance of classification before quantification, the model “A” was compared with a simplified regression model neglecting the qualitative characterization of detected odours (model “B”). More in detail, the IOMS estimations by mean of model A and model B for samples involved for field verification tests were compared with the dynamic olfactometry, i.e., the reference method for odour quantification, to assess their reliability (Figure 3).
Figure 3. Comparison of IOMS estimations by model “A” and model “B” with dynamic olfactometry.
Model B seemed to overestimate the odour concentration of samples, especially in the case of Fresh Waste samples: model B estimations resulted to be significantly higher than the upper limit of the confidence interval of the real odour concentration, confirming that quantification models built without first considering the sample class are imprecise. Conversely, the odour concentrations estimated by model A were very similar to ones assessed by dynamic olfactometry. The LoA (Table 2) highlighted that the IOMS uncertainty (i.e., about 3) was comparable with the one of the olfactometry (i.e., about 2). Based on these results, it is possible to state that the multi-phase model proved effective to provide reliable real-time measurements at the landfill fenceline.
Table 2. Limits of agreement assessed by Bland&Altman analysis.
This paper proved the possibility to use IOMS to provide an accurate real-time estimation of the odour concentration at landfill fenceline. This estimation could be used to identify “warning” concentration thresholds that may allow preventing odour problems in the surroundings of the plant, thereby offering many advantages from a management point of view. The comparison of the quantification performance achieved by model A with the one achieved by model B pointed out the importance of a proper IOMS training accounting for all odour sources of the plant under investigation to ensure reliable IOMS estimations during the monitoring phase.
Bax, C., Sironi, S., Capelli, L. 2019. Application and performance verification of electronic noses for landfill odour monitoring. In: CISA, Sardinia2019 17th International Waste Management and Landfill Symposium.
Bax, C., Sironi, S., Capelli, L. 2020. Definition and Application of a Protocol for Electronic Nose Field Performance Testing: Example of Odor Monitoring from a Tire Storage Area. Atmosphere, 11, 4, 426 (a).
Bax, C., Sironi, S., Capelli, L. 2020. How Can Odors Be Measured? An Overview of Methods and Their Applications. Atmosphere, 11, 1, 92 (b).
Bax, C., Lotesoriere, B.J., Capelli, L. 2021. Real-time monitoring of odour concentration at a landfill fenceline: performance verification in the field. CEt, 85, 19-24.
Bland, J.M., Altman, D.G. 1999. Measuring agreement in method comparison studies. Stat Methods Med Res, 8, 2, 135-60.
Cangialosi, F., Intini, G., Colucci, D. 2018. On line monitoring of odour nuisance at a sanitary landfill for non-hazardous waste. Chemical Engineering Transactions, 68, 127-132.
Deshmukh, S. Bandyopadhyay, R., Bhattacharyya, N., Pandey, R. A., Jana, A. 2015. Application of electronic nose for industrial odors and gaseous emissions measurement and monitoring – An overview. Talanta, 144, 329-340.
EN13725: Air Quality-Determination of Odour Concentration by Dynamic Olfactometry. Brussels, Belgium, 2003.
Giavarina, D. 2015. Understanding Bland Altman analysis. Biochemia medica, 25, 2, 141-151.
Laor, Y., Parker, D., Page, T. 2014. Measurement, prediction, and monitoring of odors in the environment: A critical review. Reviews in Chemical Engineering, 30, 139-166.
UNI11761: Emissioni e qualità dell’aria – Determinazione degli odori tramite IOMS (Instrumental Odour Monitoring Systems). Milano. 2019.