Field application of cost-effective sensors for the monitoring of NH₃, H₂S, and TVOC in environmental treatment facilities and the estimation of odor intensity

ABSTRACT

Odor is usually a complex mixture of various compounds. In many countries, odor complaints have been addressed using the air dilution olfactory method (ADOM) to reduce their malodor complaint. In this study, continuous monitoring of ammonia, hydrogen sulfide, and total volatile organic compounds (TVOC) using sensors was conducted in facilities for municipal and livestock wastewater treatment (LWT), and for food waste composting (FWC). Odor intensity was modeled by multivariate linear regression using sensor monitoring data with air dilution measured by the ADOM. In testing the performance of sensors in the lab, all three sensors showed acceptable values for linearity, accuracy, repeatability, lowest detection limit, and response time, so the sensors were acceptable for application in the field. In on-site real-time monitoring, the three sensors functioned well in the three environmental facilities during the testing period. Average ammonia and hydrogen sulfide concentrations were high in the LWT facility, while TVOC showed the highest concentration in the FWC facility. A longer sampling time is necessary for ammonia monitoring. Odor intensity from individual sensor data correlated well to complex odor measured by the ADOM. Finally, we suggest a protocol for field application of sensor monitoring and odor data reproduction.

Implications: We suggest a protocol for the field application of sensor monitoring and odor data estimation in this study. This study can be useful to a policy maker and field operator to reduce odor emission through the determination of a more effective treatment technology and removal pathway for individual odorants.

Introduction

The importance of air quality monitoring has increased to meet demands for mitigation and management of air pollution as awareness of the need for environmental protection has grown. Many methods have been widely applied to measure pollution levels of air pollutants in various fields. However, current air quality monitoring systems are limited in application mainly because of costs (Penza and EuNetAir Consortium 2014; Szulczyński and Gebicki 2017). Cost-effective monitoring using sensors is one alternative that can be applied in real-time for small grids (Rock, Barsan, and Weimar 2008; Stetter et al. 2008; Gebicki 2016). Wide application of sensor monitoring in this manner is the expected trend of the “fourth” industrial revolution.

Odorants consist not only of human sense (olfaction) constituents of smell but also chemical air pollutants (Peng et al. 2009; Gebicki 2016). Besides being a nuisance, odorants may cause symptoms of illness (Van, 2001; Peng et al. 2009; Nicell 2009; Belgiorno et al. 2012; Lee et al. 2013; Kang et al. 2020). Odor is a social issue that is recognized as a major cause of civil complaints in Korea (Lee et al. 2013; Park 2020). Environmental treatment facilities such as municipal wastewater treatment, livestock manure treatment, food waste treatment, and others, are particularly an issue because they are often located adjacent to residential areas.

Major odorants produced from these facilities are ammonia, hydrogen sulfide, and volatile organic compounds (VOCs). Ammonia (NH₃) is a typical odorous compound produced from the decomposition of organic nitrogen; ammonia is also a precursor of secondary inorganic aerosols (Walker et al. 2003; Peng et al. 2009; Joo et al. 2015a). Hydrogen sulfide (H₂S), a strong odorous and toxic compound that affects animal and human health, is mainly produced from anaerobic digestion of organic sulfur (Ni and Heber 2009; Peng et al. 2009). Concentrations greater than 100 ppm of H₂S are dangerous to life and health (OSHA 2005) low concentrations, less than 300 ppb, are responsible for the foul odor of rotten eggs (Pandey et al. 2012; Joo et al. 2015b). Volatile organic compounds (VOCs) contribute to particulate air pollution, health problems and many have a malodorous character. Hydrocarbons such as benzene, toluene, ethylbenzene, and xylene, and esters, organic acids, aldehydes, and phenols in VOCs are significant concerns in many industries (Szulczyński and Gebicki 2017; Wang et al. 2017).

Various types of sensors such as electrochemical (EC), metal oxide semiconductor (MOS), photo ionization detector (PID), ultraviolet spectrometers, portable gas chromatographic system, non-dispersive infra-red, and electronic noses can be used for monitoring various odorants and air pollutants (Yuwono and Lammers 2004; Penza and EuNetAir Consortium 2014). Gas sensors to measure individual odorous compounds and complex odor mixtures have been developed and tested in various fields (Fusaya et al. 1999; Hudon et al. 2000; Kampa and Castanas, 2008; Hossein Babaei and Amini, 2014; Wang et al. 2017). Researchers have sought to improve sensitivity, selectivity and response velocity. Nowadays, many sensors are highly sensitive and selective and are capable of measuring even to the ppb level of target compounds (Pandey et al. 2012; Spinelle et al. 2015; Brown, Hall, and Schoenfisch 2019). Nevertheless, field application of sensor monitoring is still limited because of some uncertainties, such as their stability against changes in humidity and temperature, especially for long-term operation, the lack of maintenance technology and other reasons (Romain et al. 2005; Capelli et al. 2012; Zeng et al. 2016).

Frequent or real-time monitoring of odor intensity is important in resolving complaints. Characterization of the spatiotemporal pattern of odor emission and dominant odorous species must be clarified prior to control and regulation. Air dynamic olfactory methods (ADOM) have been recognized as a standard to assess odor intensity of complex odors in most counties (Ishikawa & Nishida 2000; Nagata and Takeuchi 2003; National Institute of Environmental Research (NIER), 2005; Jiang et al. 2006; National Institute of Environmental Research (NIER), 2006; Jacobson et al. 2008; Sun et al. 2008; Kim and Jeong 2009; Guillot et al. 2012). The Korean government allows only ADOM (fold unit), in which five panels (odor inspection specialist) manually measure air dilution factors in the laboratory within 45 hours from the field sampling (NIER 2005). However, ADOM has to be updated; several issues, such as the representative sampling for sampling location and time, minimizing the artifact of each panel characteristics, etc. (Sorokowska et al. 2015; Kim et al. 2019). Concentration data for individual odorous compounds such as ammonia, hydrogen sulfide, and VOCs from odor emissions play important roles in improving treatment efficiency and designing odor treatment methods (Ko et al. 2013; Jo et al. 2015). Therefore, real-time monitoring to measure individual odorants and complex odors using sensors can be a useful alternative solution in odor treatment and malodor complaints.

Up to now, studies of sensor monitoring in real-time in various environmental management plants such as piggery effluent ponds, composting facilities, anaerobic digesters, and biofilters, have demonstrated the possibility of sensor application in those fields (Vandecruys et al. 1998; Romain et al. 2005; Sohn et al. 2006; Capelli et al. 2012; Adam et al. 2013; Di et al. 2013). However, the application of sensor monitoring in actual fields still has some limitations owing to large gap from laboratory tests and sensor performances such as linearity, precision, reproducibility, lower detection limit (LDL), and so on in various environmental conditions (EPA 2014; Rai et al., 2017). Previous studies demonstrated that most of tested MOS and EC sensors showed significantly high coefficient of determination (R²) in laboratory studies with their reference measurements, while only a few of them showed significant values of R² in the field studies (Rai et al., 2017). It means that laboratory sensor performance tests should be carried out under various levels of environmental conditions such as concentration, temperature and humidity, in addition sensor monitoring data collected from fields must be scientifically verified by reference methods or their equivalent methods or. One of the scientific approach to verify field monitoring data, a few studies using correlation analysis between complex odor or odor intensity and individual odorants have been tried (Fusaya et al. 1999; Lee et al. 2013; Park 2020). However, more practical studies of sensor monitoring for both individual odorant and odor intensity in actual fields should be conducted. Environmental treatment facilities such as wastewater, livestock manure and food waste treatment facilities can be an optimum test-bed to verify the feasibility of sensor monitoring, because those facilities are operated by public agencies.

In this study, we applied an environmental monitoring system equipped with ammonia, hydrogen sulfide, and total VOCs (TVOC) sensors in facilities for treatment of municipal wastewater, livestock manure and food waste composting. Prior to field application, fundamental performance of those sensors was tested under laboratory conditions. Concentrations of ammonia, hydrogen sulfide, and total volatile organic compounds (TVOC) were continuously monitored in real-time and the feasibility of the monitoring system using sensors for long-term periods was evaluated. In addition, odor intensity was produced using individual sensor monitoring data and then compared to complex odor (by dilution factors, fold) measured by ADOM. Finally, we suggest a protocol to apply sensor monitoring in environmental monitoring fields.

Materials and methods

Sensors used in this study

Table 1 shows the specifications of three gas sensors, which are electrochemical sensors for monitoring ammonia and hydrogen sulfide, and photoionization detection sensors for monitoring TVOC.

Table 1. Specification of odor sensors used in this study.

	H2S sensor	NH3 sensor	TVOC sensor
Principal	Electrochemical (EC)	Electrochemical (EC)	Photoionization detection (PID)
Model	Model #SO1198	Model #SO1N8	MinPID
Manufacture	Senko LTD. (Korea)	Senko LTD. (Korea)	Base-Mocon LTD. (USA)
Temperature	−10∼50℃	−10∼50℃	−20∼60℃
Humidity	15∼90%)	15∼90%)	0∼90%)
Measurement range	5∼5000(ppb)	100∼5000(ppb)	5∼5000(ppb)
T90 response time	<90 seconds to 90% RH	<30 seconds to 90% RH	<30 seconds to 90% RH
Weakness	Long initial running time	Low Selectivity	high cost, Short lifetime

Monitoring sites

The three monitoring sites were as follows; i) a municipal wastewater treatment facility (O-MWT) located in Gyeonggi-do, Korea, ii) a livestock manure treatment facility (Y-LWT) located in Chungcheong-do, Korea, and iii) a food waste composting facility (S-FWC) located in Gyeonggi-do, Korea. These three facilities are entirely closed systems with an odor treatment plant (OTP) at the outlet of the system. Table 2 shows the treatment status of those facilities and odor treatment plants.

Table 2. Information on three environmental treatment facilities and their odor treatment plants.

	O-MWT	Y-LWT	S-FWC
Treatment media	Municipal wastewater	Piggery wastewater	Food waste
Treatment method	DeNiPho	Natural purification + HCR	Aerobic composting
Treatment capacity	5,180 m^3/d (wastewater)	141 m^3/d (wastewater)	37 ton/d (food waste)
Data periods of sensor monitoring	Oct. 18 to Dec. 12, 2019	Oct. 23 to Dec. 25, 2019	Nov. 22 to Dec. 26, 2019
Odor treatment process	Scrubber + Biofilter	Biofilter + Activated carbon	Scrubber + Biofilter
Odor treatment capacity	80 m^3/min	140 m^3/min	450 m^3/min

Field monitoring system and maintenance

The environmental monitoring system used in the three monitoring sites was composed of four main parts as shown in Figure 1: first) sensor part which contained the sensor, filter, flow controller, and suction tube, second) display part which contained a minimal computer, communication modem, and touch screen, third) electric part which contained an electricity controller, a circuit breaker, noise filter, fan/temperature controller, heater, and ground circuit, fourth) optional part which might contain an airflow meter, and relative humidity/temperature sensors. Two sets of the monitoring system were placed between the In-OTP and Out-OTP at the MWT and LWT facilities while one monitoring system was placed to measure both the In-OTP and the Out-OTP at the FWC facility with a rotation time of 15 minutes. A pretreatment device (desolvator-1800, Nara control, Korea) was additionally installed at the FWC facility to minimize any adverse effects of humidity.

Figure 1. Customized environmental monitoring system including odor sensors to measure ammonia, hydrogen sulfide and TVOC.

Performance tests of odor sensors

Fundamental performance tests for ammonia, hydrogen sulfide and TVOC sensors were carried out using a lab-scale test chamber. The lab-scale test chamber was composed of three main parts: a gas dilution system, a control system for humidity and temperature, and a measurement system. The lower detection limit, reproducibility and precision, linearity, and response time were tested. Ten second average values for analog-to-digital conversion were used in the data analysis. More details of experimental methods for sensor performance tests were described in our previous report (Han et al. 2019).

Data acquisition and processing

Measured data from a field monitoring system were automatically collected and transferred to the data acquisition system by the Wi-Fi and LTE network. Hourly average data converted from raw data (10 s data) were used in this paper. Unusual data such as mechanical-electrical spikes or data that was unresponsive for a long time were eliminated in the data processing step. As shown in Table 2, data periods were October 18 to December 12, 2019 (56 days), October 23 to December 25, 2019 (64 days) and November 22 to December 26, 2019 (35 days) for the MWT, LWT and FWC facilities, respectively.

Results and discussion

Performance tests of odor sensors using a laboratory testing chamber

Table 3 shows the results of the fundamental performance tests using a chamber. The linearity of the three sensors was considerably high, ranging from 0.985 to 0.999; the hydrogen sulfide sensor showed the highest linearity. Lower detection limits were observed at 22.3 ppb for hydrogen sulfide, 18.6 ppb for ammonia and 26.7 ppb for the TVOC sensor. The lower detection limit for ammonia was better than that in previous reports (Han et al. 2012), while the lower detection limit of the TVOC sensor was slightly worse than that in the report by Choi (Choi et al. 2014). All three sensors showed high values for accuracy and repeatability. The hydrogen sulfide sensor showed the highest accuracy and the TVOC sensor showed the highest repeatability. The response time for the ammonia sensor was 9 to 12 minutes according to concentration levels; the hydrogen sulfide and TVOC sensors showed less than 2 minutes. Gong et al. reported that tested odor sensors showed good linearity with >0.97 of R² in the conditions less than 500 ppb (H₂S, NH₃ and VOC) and they suggested standard performance criteria, i.e., less than 5% of repeatability, 20% of linearity and 10% of accuracy, respectively (Gong et al., 2015). Another previous study presented that the H₂S sensor showed less than 1.7% of repeatability, 2.7% of accuracy and 400 seconds of response time, respectively (Zeng et al., 2016). Eventually, we decided that all three sensors were suitable for field application.

Table 3. Summary of performance tests for the three sensors used in this study (Han et al. 2019).

		H2S sensor	NH3 sensor	TVOC sensor
Linearity	R^2	0.999	0.986	0.985
	Range	0–1000 ppb	0–1300 ppb	0–300 ppb
Lower detection limit		22.3 ppb	18.6 ppb	26.7 ppb
Accuracy		0.09%	5.2%	5.7%
Repeatability		0.3%	0.1%	0.03%
Response time		120 s	540 720 s	60 s

Field monitoring data from environmental treatment facilities

Municipal wastewater treatment facility

Table 4 and Figure 2 show sensor monitoring results at the MWT facility. The data expression ratios of ammonia against the total number of data were 91% at the In-OTP and 82% at the Out-OTP. The data expression ratios for hydrogen sulfide were 91% at the In-OTP, 83% at the Out-OTP, and the ratios for TVOC were 91% at the In-OTP and 70% at the Out-OTP. The data expression ratio for TVOC at the Out-OTP was worse than for ammonia and hydrogen sulfide. The hourly average for ammonia showed 200–600 ppb (average 383 ppb) at the In-OTP and 0–150 ppb (average 24 ppb) at the Out-OTP. Concentrations of hydrogen sulfide showed 300–2,000 ppb (average 828 ppb) at the In-OTP and 0–200 ppb (average 35 ppb) at the Out-OTP. The TVOC pattern at the In-OTP was unreasonable, and the TVOC concentration showed 100–400 ppb (average 210 ppb) at the Out-OTP. Temperature and relative humidity at the Out-OTP fluctuated widely because the monitoring system was located outside of the building. The temperature ranged from 10°C to 33°C, and the relative humidity ranged from 20% to 70%. In the previous report by Jun et al., ammonia concentrations from MWT facilities were distributed from 110 ppb to 420 ppb, the hydrogen sulfide concentration varied from 200 ppb to 5,000 ppb depending on treatment processes, and the toluene concentrations were 100 ppb to 700 ppb (Jeon et al. 2010). When those data were compared with sensor monitoring data in this study, the sensor monitoring data from the O-MWT facility appears to be at a reasonable concentration.

Figure 2. Sensor measurement data at a municipal wastewater treatment facility (black solid line: In-OTP, and brown dotted line: Out-OTP).

Table 4. Sensor monitoring data from a municipal wastewater treatment facility.

	In-OTP					Out-OTP
	NH3 (ppb)	H2S (ppb)	TVOC (ppb)	Temp. (°C)	RH (%)	NH3 (ppb)	H2S (ppb)	TVOC (ppb)	Temp. (°C)	RH (%)
Mean ± SD	383 ± 90	828 ± 363	29 ± 8	26 ± 4	39 ± 7	24 ± 27	35 ± 20	210 ± 54	17 ± 5	46 ± 13
Max./Min.	960 / 221	1,904 / 253	43 / 10	33 / 17	57 / 27	148 / 0	201 / 0	395 / 37	30 / 6	70 / 18
Expressed data (n)	1,217	1,217	1,217	1,217	1,217	1,089	1,100	932	1,100	1,100
Data expression ratio (%)	91	91	91	91	91	82	83	70	83	83

Livestock manure treatment facility

Table 5 and Figure 3 show sensor monitoring data at the LWT facility. Data expression ratios of ammonia against the total data were 62% at the In-OTP and 97% at the Out-OTP. The ratios for hydrogen sulfide were 64% at the In-OTP, 97% at the Out-OTP, and the ratios for TVOC were 29% at the In-OTP and 97% at the Out-OTP. The data expression ratio for TVOC at the In-OTP was lower than for ammonia and hydrogen sulfide. The hourly concentration average for ammonia showed 400–25,000 ppb (average 10,059 ppb) at the In-OTP and 0–2,000 ppb (average 314 ppb) at the Out-OTP. Concentrations of hydrogen sulfide showed 300–25,000 ppb (average 2,258 ppb) at the In-OTP and 0–5,000 ppb (average 484 ppb) at the Out-OTP. Ammonia and hydrogen sulfide concentrations from the LWT facility were significantly higher than those at the MWT facility. The TVOC concentration at the In-OTP was only successfully monitored from Nov. 19 to Dec. 10 (about 20 days), and the TVOC concentration showed 0–15,000 ppb (average 1,889 ppb) at the In-OTP and 0–5,000 ppb (average 823 ppb) at the Out-OTP. Hydrogen sulfide and TVOC concentrations frequently peaked during the dumping time for livestock manure transported from the manure collection in individual livestock farms, and hydrogen sulfide and TVOC concentrations were highly correlated with each other (R = 0.82, data not shown). The temperature ranged from 0°C to 30°C, and relative humidity ranged from 20% to 90%. In the previous report, ammonia concentration from LMT facilities were distributed from 6,800 ppb to 7,500 ppb, hydrogen sulfide concentration showed 2,200 ppb to 4,400 ppb, and toluene concentration were 0.5 ppb to 6.7 ppb at the inlet hopper and liquid tank, respectively (Jeon et al. 2010). In the study presented by Choi and Lee, ammonia, hydrogen sulfide and toluene concentrations were 8,900 to 15,900 ppb, 81000 to 114,000 ppb, 20 to 240 ppb, respectively (Choi and Lee 2011). Sensor monitoring data for ammonia and hydrogen sulfide were reasonable, while TVOC concentrations were high because other VOC groups (acetic acid, propionic acid, acetaldehyde, cresol, phenol, ketones, for example), in LWT facilities might be emitted more than toluene and styrene (Choi and Lee 2011; Liu et al. 2011; Parker et al. 2012; Blazy et al. 2015).

Figure 3. Sensor measurement data at a livestock wastewater treatment facility (black solid line: In-OTP, and brown dotted line: Out-OTP).

Table 5. Sensor monitoring data from a livestock wastewater treatment facility.

	In-OTP					Out-OTP
	NH3 (ppb)	H2S (ppb)	TVOC (ppb)	Temp. (°C)	RH (%)	NH3 (ppb)	H2S (ppb)	TVOC (ppb)	Temp. (°C)	RH (%)
Mean ± SD	10,059 ± 3,237	2,258 ± 3,159	1,889 ± 2,719	16 ± 6	56 ± 14	314 ± 204	484 ± 686	823 ± 445	13 ± 6	62 ± 13
Max./Min.	24,472 / 3,089	27,271 / 116	14,222 / 42	31 / 0	91 / 19	2,334 / 120	5,135 / 1	4,371 / 166	34 / 0	96 / 31
Expressed data (n)	945	972	442	932	932	1,468	1,468	1,468	1,468	1,468
Data expression ratio (%)	62	64	29	61	61	97	97	97	97	97

Food waste composting facility

Table 6 and Figure 4 show sensor monitoring data at the FWC facility. Data expression ratios for ammonia, hydrogen sulfide and TVOC all showed 100% at both the In-OTP and the Out-OTP. Hourly average concentrations of ammonia were 200–4,000 ppb (average 930 ppb) at the In-OTP and 200–4,000 ppb (average 954 ppb) at the Out-OTP. The ammonia concentration pattern was nearly identical for the In-OTP and the Out-OTP owing to the time delay of ammonia with sequencing measurement times of 15 minutes. Many studies have demonstrated that more than 15 minutes of sampling time should be necessary to reach target concentrations of ammonia (Rom and Zhang 2010; Santos et al. 2017). This fact demonstrated that the rotation time of the sampling location should be longer than 15 minutes. In addition, the ammonia concentration was highly correlated with the temperature condition (R = 0.89) and ammonia emission was influenced by oxidation activity in the biological composting process. Concentrations of hydrogen sulfide were 100–200 ppb (average 151 ppb) at the In-OTP and 80–150 ppb (average 119 ppb) at the Out-OTP. As in the LWT facility, ammonia concentrations were higher than hydrogen sulfide. Nitrogen removal technology should be important in livestock and food waste treatment facilities. TVOC was successfully monitored at both the In-OTP and the Out-OTP. TVOC concentrations showed 1,000–12,000 ppb (average 3,669 ppb) at the In-OTP and 500–2,000 ppb (average 1,004 ppb) at the Out-OTP. Humidity might strongly influence the TVOC sensor; reducing humidity with a desolvator was effective in the successful monitoring of the TVOC sensor. The TVOC concentration level at the FWC facility was highest among the three facilities. Temperature and relative humidity at both the In-OTP and the Out-OTP fluctuated widely and showed the same pattern because both monitoring systems were located outside of the building compared to those at the Out-OTP. Humidity at the Out-OTP was slightly lower than that at the In-OTP. In the previous report, ammonia concentrations from FWC facilities were distributed from 360 ppb to 1,190 ppb, hydrogen sulfide concentrations showed 5 ppb to 34 ppb, and acetaldehyde concentrations were 43 ppb to 550 ppb. The sensor monitoring data from the O-FWC facility showed higher concentration levels in the previous report (Bae et al. 2009; Yoo et al. 2012).

Figure 4. Sensor measurement data at a food waste composting facility (black solid line: In-OTP, and brown dotted line: Out-OTP).

Table 6. Sensor monitoring data from a food waste composting facility.

	In-OTP					Out-OTP
	NH3 (ppb)	H2S (ppb)	TVOC (ppb)	Temp. (°C)	RH (%)	NH3 (ppb)	H2S (ppb)	TVOC (ppb)	Temp. (°C)	RH (%)
Mean ± SD	930 ± 657	151 ± 16	3,669 ± 2,069	17 ± 6	57 ± 10	954 ± 652	119 ± 10	1,004 ± 286	17 ± 6	49 ± 9
Max. / Min.	2,833 / 222	210 / 117	11,954 / 1,190	29 / 6	71 / 35	2,853 / 269	142 / 93	1,911 / 603	29 / 6	68 / 31
Expressed data (n)	758	758	758	743	748	758	758	758	742	747
Data expression ratio (%)	100	100	100	98	99	100	100	100	98	99

Correlation analysis between individual sensor and complex odor data

Figure 5 shows the results of the correlation analysis between complex odor (air dilution factor) measured by ADOM and each sensor measurement concentration for NH₃, H₂S, and TVOC in three environmental treatment facilities. Individual odorant measurement data from the LWT facility were highly correlated to air dilution factor (Figure 5b). Concentration data for hydrogen sulfide, ammonia and TVOC at the LWT facility showed high correlation coefficients of 0.82, 0.75 and 0.74 related to complex odor, respectively. While, only hydrogen sulfide at the MWT facility and TVOC at the FWC facility were correlated with complex odor correlation coefficients of 0.57 and 0.80, respectively. These results show that the correlation between ADOM and odor sensor monitoring can be significant at the conditions of high odor concentrations.

Figure 5. Correlation between odor by ADOM and each sensor measurement concentrations of NH₃ (black), H₂S (brown), TVOC (blue) in MWT (a), LWT (b), FWC (c), respectively.

To verify the relationship between complex odor and sensor data, a multivariate linear regression (Equation (1)(1) $y\left(odor\right)=A\cdot x1\left(N{H}_3\right)+B\cdot x2\left(H_{2}S\right)+C\cdot x3\left(TVOC\right)+Intercept$(1) ) between both measurement data presented in Figure 5 was performed with IBM SPSS Statistics 20 as shown in Table 7.

Table 7. Multivariate linear regression between complex odor (dependent variable) measured by ADOM and sensor measurement data.

Regression	Independent variables	B (coefficients for independent variables)	T (coefficient /standard error)
R = 0.821 p value = 0.000	Intercept	228.444	0.373
	NH3	0.153	1.094
	H2S	1.339	2.051
	TVOC	0.789	2.994

(1) $y\left(odor\right)=A\cdot x1\left(N{H}_3\right)+B\cdot x2\left(H_{2}S\right)+C\cdot x3\left(TVOC\right)+Intercept$(1)

As shown in Equation (1)(1) $y\left(odor\right)=A\cdot x1\left(N{H}_3\right)+B\cdot x2\left(H_{2}S\right)+C\cdot x3\left(TVOC\right)+Intercept$(1) , complex odor can be expressed as a function of ammonia, hydrogen sulfide and TVOC concentration measured by sensors. A, B and C are coefficients (beta values) for ammonia, hydrogen sulfide and TVOC, respectively (i.e., A = 0.153, B = 1.339, C = 0.789, and Intercept = 228.444). A multivariate linear regression was significant with a p value <0.001. Odor estimated by a multivariate linear regression was highly correlated with complex odor (R = 0.89) as shown in Figure 6a.

Figure 6. Correlation comparison between odor by odor estimated by sensor data (a) and ADOM and odor divided by the odor threshold values (b) and (black: MWT, brown: LWT, blue: FWC).

Odor intensity was regarded as the sum of individual values divided by a threshold for each individual odor compound (Bae et al. 2009; Han et al. 2012). Odor thresholds were applied as 2,930 ppb for ammonia, 0.43 ppb for hydrogen sulfide, and 3.756 ppb for TVOC. The average value of five representative VOC compounds (acetaldehyde, butyraldehyde, propionaldehyde, n-valeraldehyde, i-valeraldehyde) was used for the TVOC threshold (Han et al. 2012). Odor intensity divided by the odor threshold of individual odor compounds correlated with complex odor (R = 0.80) as shown in Figure 6b.

Figure 7 shows the relationship between odor intensity estimated by the regression using Equation (1)(1) $y\left(odor\right)=A\cdot x1\left(N{H}_3\right)+B\cdot x2\left(H_{2}S\right)+C\cdot x3\left(TVOC\right)+Intercept$(1) with sensor measurement data and odor intensity divided by each odor threshold manually measured data at three treatment facilities. Both odor intensities in Figure 7 were highly correlated to each other (correlation coefficient more than 0.97). The slope for the LWT facility was 1.06, implying that both odor intensities were similar. On the other hand, slopes for the MWT and FWC facilities were considerably larger or smaller than 1, meaning that odor estimated by sensor data can be overestimated at the FWC facility and underestimated at the MWT facility.

Figure 7. Correlation between odor estimated by correlation equation (eq.1) and odor divided by threshold (black: MWT, brown: LWT, blue: FWC).

Estimation of odor intensity from individual sensor monitoring data

Figure 8 and Table 8 show odor intensity estimated from sensor measurement data using a regression Equation (1)(1) $y\left(odor\right)=A\cdot x1\left(N{H}_3\right)+B\cdot x2\left(H_{2}S\right)+C\cdot x3\left(TVOC\right)+Intercept$(1) and removal efficiencies at the MWT, LWT and FWC facilities. Odor intensity in the MWT facility was 1,000 to 4,000 O.U. at the In-OTP (average 2,039) and 300 to 1,000 O.U. at the Out-OTP (average 419), with odor removal efficiencies ranging from 50 to 95% (average 78%). Odor intensity at the LWT facility was widely distributed from 2,000 to 40,000 O.U. at the In-OTP (average 5,557) and 1,000 to 10,000 O.U. at the Out-OTP (average 1,573), with odor removal efficiencies ranging from 10 to 95% (average 60%). Odor intensity at the FWC facility was 2,000 to 10,000 O.U. at the In-OTP (average 3,4688) and 1,000 to 3,000 O.U. at the Out-OTP (average 1,326), with odor removal efficiencies ranging from 10 to 80% (average 58%). The removal efficiencies at the LWT and FWC facilities were not high enough; odor treatment technologies such as microbial media or a scrubber will need to be developed or replaced. This estimated odor intensity can be useful to determine an odor mitigation strategy in odor emission industries. In previous reports, odor intensity from environmental treatment facilities were consistent with data from this study: distributed from 368 to 3,000-fold at MWT facilities, 2,080 to 30,000-fold at LWT facilities and 1,442 to 2,080-fold at FWC facilities (Jeon et al. 2010; Choi and Lee 2011; Yoo et al. 2012; Yi et al. 2020).

Figure 8. Estimated odor unit from sensor measurement data and removal efficiency at a MWT (a), LWT (b), FWC (c), respectively (black solid line: In-OTP, and brown dotted line: Out-OTP, blue solid line: removal efficiency).

Table 8. Odor intensity generated by the linear regression using sensor measurement data.

	MWT		LWT		FWC
	In-OTP	Out-OTP	In-OTP	Out-OTP	In-OTP	Out-OTP
Mean ± SD (O.U)	2,039 (±518)	419 (±93)	5,557 (±5,392)	1,573 (±1,220)	3,468 (±1,664)	1,326 (±281)
Removal ratio ± SD (%)	78 (± 8)		60 (± 25)		58 (± 12)
Max./min. (O.U.)	4,280 / 1,025	686 / 0	45,358 / 455	10,380 / 0	10,077 / 1,444	2,167 / 970
Expressed data (n)	1,217	1,088	972	1,469	748	748
Data expression ratio (%)	91	82	64	97	99	99

A protocol to estimate odor intensity using individual sensor monitoring data

Finally, we suggest a protocol for the field application of sensor monitoring and odor data estimation as shown in Figure 9. The first step is the selection of a reliable and stable sensor, comparative study of various information such as the original specification, a case study of application, applicable environmental conditions, and so on should be carried out in this step. The performance of the selected sensors must be tested precisely under controlled conditions in the laboratory. The monitoring system must also be composed of pre-tested sensors in the laboratory, because the actual performances in field tests are often different from the original specifications in terms of precision and reproducibility (EPA 2014; Rai et al., 2017). The next step is installation and operation of the monitoring system in the field. Once field monitoring is started, complex odors must be measured periodically and simultaneously by an ADOM to correlate sensor monitoring data (odors must be sampled from the same location and at the same time). Periodical maintenance such as biweekly or monthly zero and span check and multi-point calibration (if needed) is very important in the collection of reliable monitoring data. Correlation analysis between ADOM data and sensor monitoring data is conducted (Figure 6) and odor intensity (dependent variable) can be estimated from the correlation regression using sensor monitoring data as individual independent variables (Figures 6 and 8). The estimated odor intensity can be useful to a policy maker and field operator to reduce odor emission through the determination of a more effective treatment technology and removal pathway for individual odorants. When sensors are applied in the fields, the highlighted steps 1, 2, 3, and 4 in Figure 9 have not been considered, because these steps are not mandatory up to now. However these scientific steps play an important role in the collection of reliable measurement data from a sensor monitoring method. The protocol suggested in this study can be expected to provide a useful guideline for the field application of environmental monitoring sensor.

Figure 9. Suggested protocol for field application of sensor monitoring and odor data estimation.

Highlights

• Continuous monitoring of odorants using sensors was conducted in environmental treatment facilities.

• Odor sensors for NH₃, H₂S and TVOC functioned well in the three environmental facilities.

• Odor intensity was modeled by sensor monitoring data.

• Modeled odor intensity correlated well to complex odor measured by the air dilution olfactory method.

• This study suggests a protocol for field application of sensor monitoring and odor intensity reproduction.

Data availability statement

The data that support the findings of this study are available from the first author,[HSJ], corresponding author, [JSH], upon reasonable request.

Additional information

Funding

This research was supported by the Korea Ministry of Environment as Advanced Industrial Technology Project (No. 2017000120004) and R&D Project for Management of Atmosphere environment Project (No. 2021003410002), FRIEND (Fine Particle Research Initiative in East Asia Considering National Differences) Project through the NRF and Ministry of Science and ICT (No. 2020M3G1A1114999), and the support of Cooperative Research Program for Agriculture Science & Technology Development (Project No. PJ017075) Rural Development Administration of Korea.

Notes on contributors

HungSoo Joo

HungSoo Joo is an assistant professor of Department of Environmental Engineering in Anayang University, Anyang-si, Gyeonggi-do, Korea.

Sang-Woo Han

Sang-Woo Han is a Ph.D. student of Department of Environmental Engineering in Anayang University, Anyang-si, Gyeonggi-do, Korea.

Chun-Sang Lee

Chun-Sang Lee is a Ph.D. student of Department of Environmental Engineering in Anayang University, Anyang-si, Gyeonggi-do, Korea.

Hyun-Seop Jang

Hyun-Seop Jang is a research professor and director of Zero Emission Center of Sungkyunkwan University, Suwan-si, Gyeonggi-do, Korea.

Sung-Tae Kim

Sung-Tae Kim is a leader of E2M3 Inc., Anyang-si, Gyeonggi-do, Korea.

Jin-Seok Han

Jin-Seok Han is an associate professor of Department of Environmental Engineering in, Anayang University, Anyang-si, Gyeonggi-do, Korea.