Each low-cost sensing unit was comprised of two sensors (one oxidative gas sensor and one CO gas sensor), two adapters (one for each sensor, made by the Alphasense Ltd.), one microcontroller for reading and storing the voltages obtained from each sensor, and a custom-built circuit to connect and integrate all constituents built in-house. The microcontroller was developed to record detected signals by the sensors as a voltage (ranging between 0 and 5 V) on a memory card every 2.5 sec. The integrated circuit board was responsible for connecting the microcontroller to the sensor adaptors and implementing the electric voltage gains. To obtain proper resolution at the concentration range of interest, the electric voltage gain, the mean ratio of the voltage signal at the output port of the circuit to that of at the input of the circuit, was 5 for the oxidative gas sensor and 1 for the CO sensor. The integrating custom-built circuit (see Figure 1b) was an Arduino Yun board (https://www.arduino.cc/en/Main/ArduinoBoardYun). Arduino is an open-source and inexpensive electronics platform with an easy programming language for fast prototyping and creation of interactive communication between the sensor and the memory card. The Yun model of Arduino board used in this study provided the advantage of direct posting the recorded voltages over an embedded webserver software. The Arduino board was powered by connection to a 5-V electric supply and had the capability for wireless connection to a computer via Wi-Fi. The integrating board included circuitry that stored and transmitted voltages detected by the sensors. The recorded voltages were stored and appended in a text file after each 2.5 sec. Then, a Visual Basic for Application (VBA) code was developed to sort and average the data over 5 min periods.

The main purpose of evaluating the oxidative gas sensor was to correlate the recorded voltage outputs by the Alphasense oxidative gas sensors to the concentration values of either O₃ or NO₂ measured by reference monitors. The oxidative gas sensors were exposed separately to different targeted concentrations of either NO₂ or O₃ and their responsiveness to alterations in the gas concentrations was examined. All measurements were conducted inside an acrylic chamber (0.5 × 0.5 × 0.5 m), tinted to minimize interactions of O₃ with light. For statistical reliability, three units of each sensor type were tested at a time. The linearity of the response was investigated and equations describing the relationship between the sensor response voltage and concentration were developed. The choice of concentration range for each of O₃ (25–150 ppb) and NO₂ (0.2–1.5 ppm) gases was based on typical concentration of these gases in the atmosphere and in occupational environments and are all below regulated occupational exposure limits.^[24]

Ozone was introduced to the chamber using a bench-top ozone generator (Model 1008-PC, Dasibi Environmental Co., Glendale, CA, USA) that irradiates ultraviolet (UV) light from a cold cathode Hg lamp on the zero air. The zero air was produced by filtering compressed house air using a combination of a HEPA filter, silica gel, purafilm, and charcoal columns. Generation of O₃ was controlled through a cycle consisting of three phases of: (1) ascending, wherein the concentration was increasing from the baseline concentration to the target concentration; (2) steady state, wherein O₃ concentration was maintained at the targeted concentration (25, 50, 75, or 100 ppb); and (3) descending, wherein the concentration was decreasing from the target concentration to the baseline concentration (Figure 3). The flow rate of the O₃ into the chamber for all experiments was adjusted to 5.5 Lpm, as suggested by the manufacturer of the O₃ generator. Before the start of each experiment, zero air was blown into the chamber at a flow rate of10 Lpm to create the lowest possible O₃ concentration for the baseline level. The O₃ concentration was measured continuously using a bench-top O₃ analyzer (Model 1008-PC, Dasibi Environmental Co., Glendale, CA, USA), two higher-cost personal O₃ monitors (Model POM, 2B Technologies Inc., Boulder, CO, USA), and two lower-cost personal O₃ monitors (Series 500, Aeroqual LLC., Auckland, New Zealand) (see Figure 2a). All O₃ monitoring instruments were adjusted to report the O₃ concentration at 5 min intervals. Depending on the target concentration value, the time to reach the target concentration in the chamber from the baseline and the time to return to the baseline concentration from the target concentration was 30–60 min.

Specification of the O₃ and NO₂ monitoring instruments are summarized in Table 1. The O₃ analyzers manufactured by Dasibi Environmental Co. and 2B Technologies Inc. both operate on the principle of ultraviolet (UV) absorption. Designated as a reference method by the USEPA, the UV absorption mechanism is an accurate and well characterized method employing a mercury lamp that emits light primarily of wavelength 254 nm, which corresponds to the maximally absorbed wavelength of O₃.^[25] The reference ozone-free air flow for O₃ detection is generated using an ozone-specific scrubber on a substrate inside the instruments. The personal O₃ monitor manufactured by Aeroqual LLC. utilizes a semi-conductor oxide O₃ sensor (OZU 0–0.15 ppm), and has been used commonly for O₃ detection in air quality monitoring networks.^[26–28] The operating principle of the Aeroqual S500 includes drawing sample air into an enclosure wherein the oxidizing capacity of the ozone molecules modifies the electrical state of a heated thin film of tungsten oxide as the sensing material.^[26] Since the electrical conductivity of the material is a strong function of the temperature and RH, devices operating based on this semi-conduction oxidizing mechanism are not a reference method for the detection of O₃.

Exposure to NO₂ was performed separately by introduction of compressed NO₂ into the chamber, after dilution with zero air, using a dynamic gas calibrator (Model 146i, Thermo Fisher-Scientific Inc., Franklin, MA, USA). A bench-top liquid sorbent NO₂ analyzer (RM Series, Interscan Co., Simi Valley, CA, USA) was used for measurement of the NO₂ inside the chamber (see Figure 2a). Although analysis by chemiluminescence is considered the most reliable and preferred method for NO₂ detection, the liquid sorbent-based NO₂ analyzers (e.g., Interscan NO₂ analyzer, used in this study) have demonstrated a high NO₂ measurement accuracy when used in the laboratory (less than 5% bias compared to the chemiluminescence method for NO₂ concentrations below 5 ppm).^[29] The NO₂ monitor was factory-calibrated before the first experiment, and was zeroed and re-calibrated with zero air before each experiment. Tested target concentrations of NO₂ included 0.2, 0.5, 1, and 1.5 ppm, which are common NO₂ concentrations in urban areas.^[30]

Performance of sensors was evaluated by estimating accuracy and linearity of the sensors compared to the reference instruments. As formulated in Equation (1), the bias (B) as a measure for accuracy of the sensors was quantified as the average ratio of the difference between concentration values estimated by the Alphasense sensor (C_Alphasense, using derived regression lines) and that by the reference instruments (C_Actual) to the C_Actual for each measurement during the steady-state period of the experiment. As a measure of intra-unit variability of the sensors, the coefficient of variation (CV) of the sensors was calculated using Equation (2): (1)B(%)=1n∑i=1nCAlphasense,i−CActual,iCActual,i×100 (2)CV(%)=1n∑i=1nσiμi×100, where σ_i is the standard deviation at concentration i and μ_i is the mean of the measurements at the i^th concentration from three replicate sensors for each test. Linearity of the sensors was evaluated by estimating the r-squared as a measure of closeness of the recorded voltages to the fitted regression line for different reference concentrations.

Specifications for the CO monitoring instruments are summarized in Table 2. The experimental procedure for evaluation of the CO sensors was similar to that of the oxidative gas sensors. The same cycle of ascending, steady state, and descending periods for target concentrations of (2, 5, 10, 12, 15, 25, 30, and 50 ppm) was followed for CO as for O₃ and NO₂. The CO gas was introduced at a controlled and constant concentration into the chamber from a compressed tank after dilution with zero air and adjustment by the gas calibrator. The flow rate provided by the gas calibrator to the chamber varies depending on the selected concentration of the target gas. In this study, the reference measurement of CO concentrations was estimated using an electro-chemical portable probe developed by (GrayWolf Advanced Pro, GrayWolf Sensing Solutions LLC., Shelton, CT, USA). Reliability verification of this reference electrochemical monitor along with detailed discussion on comparison to EPA reference methods have been reported in literature.^[31,32] Prior to measurements, the instrument was factory-calibrated to measure CO concentrations between 0–100 ppm with a precision of ±0.2 ppm. Each Alphasense sensor was co-located with a commercially available CO sensor (EL-USB-CO, Lascar Electronics Inc., Essex, UK) to compare the recorded voltages by the Alphasense sensor with respect to another electrochemical sensor (see Figure 2b). The same performance metrics as for the oxidative gas sensors were estimated for the CO sensors.

Response time is the time required for a sensor to reach a certain percentage of the steady-state value. We evaluated the response time of the sensors to identify their ability to capture transient peaks in exposure and to determine the needed time when bump-checking these sensors in the field with calibration gases. The response time of three sensors of each type was reported as t₅₀ and t₉₀ values, defined as the time required for the sensor voltage to reach 50% and 90% of the steady-state reading for the span tests (15 ppm CO for CO sensors or 2 ppm NO₂ for oxidative gas sensors). The reason for using NO₂-only to evaluate the response time of the oxidative gas sensors was availability of compressed NO₂ cylinders on the market. For this purpose, a factory-fabricated shroud was used. The shroud had a single entry and supplied the gas directly from the tank to each sensor, one at a time. The zero and span gas tests were performed on each of the CO and oxidative gas sensors, separately. Two exposure flow rates of 0.25 Lpm (Test A) and the manufacturer recommended 0.5 Lpm (Test B) were regulated through the shroud. The response time estimated in this study is based on direct injection of zero air or span gases into the sensor head. This response time is required for obtaining insight into the field calibration protocol.

Recorded voltages corresponding to the target concentrations indicated a highly linear relationship with ozone (see Figure 3a). The change in the relationship after nine months of continuous operation is also displayed in Figure 3b. Although the electro-chemical sensors maintained the linear association with O₃ concentrations measured inside the chamber (with almost the same R² of 0.98), there was an increase in the line slope from 0.330 to 0.395 (about 20%) and a reduction in the intercept as an aging influence. The mean change in the recorded voltages agreed with the drift reported by the manufacturer (2% drift in recorded values per month). We also observed an increase in intra-sensor variability after the 9-month period. The average size of the error bars at different concentrations was ± 22 mV, which was about 2–7 times greater than the variability observed when the sensors were new. This suggests the sensors not only drift, but also become less precise over time or the sensors may not drift equally over time. The manufacturer suggested lifetime for Alphasense oxidative gas sensors is 24 months, when the sensors experience a 50% change in their reading as compared to the first day of operation.

The time series of O₃ concentrations measured by high- and mid-cost instruments listed in Table 1 (reference instrument and two types of personal O₃ monitors) were compared to the low-cost sensors, as displayed in Figure 4. The concentrations associated with the oxidative gas sensors were estimated using the equation obtained from the regression analysis when relating the steady-state target concentration to the recorded voltages (Figure 3a). Although the three Alphasense sensors overestimated the concentration in the beginning of the test when O₃ concentrations were close to the background (< 10 ppb), they were more consistent with the reference instrument for concentrations above 25 ppb. The fact that the POM monitor and the reference instrument operate on the same detection principle is probably the reason for obtaining similar O₃ concentrations by them throughout the measurement periods. However, the POM monitors overestimated the concentration when the targeted concentration was relatively high (100 or 150 ppb, as seen in Figures 4c and 4d). In contrast, the Aeroqual monitors consistently underestimated the O₃ concentration, which was more obvious at higher concentrations. O₃ monitors using semi-conduction oxidizing mechanism are sensitive to environmental conditions, particularly temperature and RH.^[33] The temperature inside our chamber increased by 5°C (starting at ~25°C and ending at ~30°C) and the RH decreased by 20% (starting at about 65% and ending at 45%) over the course of the experiment. These changes in temperature and relative humidity might contribute to underestimation of the real concentrations by the Aeroqual monitor and Alphasense sensors.

The voltages recorded by the Alphasense oxidative gas sensor were strongly linear (R² = 0.996) at steady-state NO₂ concentrations (Figure 5). Interestingly, the slope of the regression line for NO₂ (0.37) was similar to that for O₃ (0.33). This finding indicates that the oxidative gas sensors are almost equally sensitive to O₃ and NO₂, when exposed separately (see Figure 5). Since both selected ranges of concentrations for O₃-only and NO₂-only included the range commonly measured in the ambient air, the oxidative gas sensors are recommended for use in situations where only O₃ or NO₂ are present, or when knowledge of the combined concentration is acceptable, as it is not possible to distinguish between O₃ and NO₂.

Each oxidative gas sensor is provided with a slope and an intercept specified by the manufacturer. According to the manufacturer, the mean slope and intercept for O₃ of the three oxidative gas sensors tested in this study were 0.318 mV/ppb and 208 mV, respectively. The equivalent values for the sensors when exposed to O₃ in our test conditions were 0.33 mV/ppb and 8.25 mV (Figure 5). Thus, the estimated slopes for O₃ from this study was fairly similar compared to those provided by the manufacturer. However, due to the considerable difference between the intercept reported by the manufacturer and that determined in this study, direct application of the manufacturer’s calibration line for estimation of the environmental O₃ concentration using Alphasense oxidative gas sensor is cautioned.

The bias and CV of the Alphasense oxidative gas sensors relative to reference analyzer (by Dasibi Environmental Co.) for different target concentrations were summarized in Table 3. All sensors/monitors showed a lower accuracy at the lowest target concentration (25 ppb of O₃ or 0.2 ppm of NO₂). The Alphasense sensors were less accurate in environments with low (25 ppb) or high (150 ppb) O₃ concentrations. Both the bias and CV values for cases at 50 and 100 ppb concentrations were below 10%. The Alphasense sensors showed better accuracy for NO₂ exposures. The Aeroqual monitors exhibited the least accuracy among the O₃ sensors (bias values ranging from −27 to −40.6%) and the highest intra-sensor variability (CV values ranging from 6.9–24.6%). The Aeroqual monitors underestimated the target concentration in all cases. As expected, the mid-cost O₃ monitors manufactured by 2B Technologies Inc. demonstrated the best performance among all monitors (bias values ranging from −3.9 to 7.7%). The intra-unit variability of the 2B Tech monitors was higher when the O₃ concentration was the lowest (25 ppb) or the highest (150 ppb) and lower for intermediate concentrations. These monitors overestimated the concentration for target O₃ concentrations below 100 ppb. In general, bias and CV values obtained from Alphasense oxidative gas sensors were more accurate than the Aeroqual monitors but less accurate than the one by 2B Technologies (except for target concentration of 50 ppb). Except for NO₂ concentration of 0.2 ppm, the CV values for tested NO₂ concentrations were within the 10% acceptance CV value by the EPA for test instruments.

We observed a strong linear relationship (R² > 0.998) between voltages from the Alphasense CO sensors and the reference instrument for CO concentrations below 12 ppm (Figure 6). There were negligible increases in the voltage (from 4.5 to 4.8 mV) recorded by the Alphasense CO sensors when the target concentration increased over a wide CO concentration ranging from 12–50 ppm. Thus, these sensors are inappropriate for use in environments with CO concentrations above 12 ppm. Using the linear relationship between the Alphasense sensors and the reference instrument, the CO concentration values for voltages recorded by the Alphasense sensors at different target concentrations were calculated and plotted against Lascar CO monitors in Figure 7. For all target concentrations less than 12 ppm (Figures 7a–7d), both the Lascar electro-chemical sensors and the Alphasense CO sensors demonstrated a very similar response. However, the Lascar CO sensors overestimated the CO concentrations (up to 10%) when the targeted concentration was 12 ppm. This overestimation by the Lascar sensors was also observed at target concentration above 12 ppm. The Lascar CO sensors showed a wider range of proper CO measurement for target CO concentrations up to 50 ppm. The intra-sensor variability of the measurements by the Lascar sensors for all targeted CO concentrations was greater than the Alphasense CO sensors. Considering the lower cost of the Alphasense sensors (about half of the Lascar sensors cost) yielding similar set of the results, the Alphasense sensors may be appropriate for use in an air quality monitoring network where concentrations generally below 12 ppm are expected.

Compared to the oxidative gas sensors, the Alphasense CO sensors revealed much lower bias and CV values over its measurement range of 2–12 ppm. In fact, all biases were below 10% in all tested target CO concentrations (2.2–5.9%, as listed in Table 4). CV as a measure for the mean intra-unit variability was consistently below 5%. The Lascar CO sensors showed a poor accuracy for CO concentration below 5 ppm (biases above 21%) but they showed a good accuracy over a wider range of CO concentration values (5–25 ppm). These sensors overestimated CO concentrations considerably (bias > +10%) when the target CO concentration exceeded 25 ppm.

In summary, the use of the Alphasense CO sensor, as set up we observed, is not recommended for situations where concentrations above 12 ppm are expected (see Figure 6). However, the maximum daily concentrations of the CO measured in ambient air of seven major cities of the US have been below 5 ppm.^[34] In occupational settings, CO concentrations can be considerably higher than 12 ppm. Thus, the use of Alphasense CO sensors are not recommended for occupational monitoring where high concentrations are anticipated.

The response times, t₅₀ and t₉₀, of the oxidative gas and CO sensors for different tests are summarized in Table 5. Except for Test A for span NO₂ concentration of 2 ppm, wherein the response times of t₉₀ was not achieved in 600 s, the response time of the sensors decreased with an increase in the flow rate. The voltages from the oxidative gas sensors were all negative during the entire zero test period. Since there is no physical meaning for the negative voltage values the electro-chemical oxidative gas sensors at near-zero NO₂ concentrations may not be reliable. Although the behavior of the oxidative gas sensors at different zero air flow rates of 0.25 and 0.5 Lpm were similar (see Figure 8a), the response time of the higher flow rate was shorter (e.g., t₉₀ = 198 at the flow rate of 0.25 Lpm vs 147 s at the flow rate of 0.5 Lpm). When first exposed to zero air, the voltage dropped rapidly and then slowly approached zero. However, none of the tested cases reached voltages higher than about −15 mV after 300 s. This result presents a challenge when conducting a zero air test on these oxidative gas sensors for field calibration. We speculate that the negative voltages are associated with very low (less than 10%) relative humidity inside the bumping shroud, which rarely occurs in ambient air. In fact, the decrease in relative humidity due to direct exposure to zero air lead to further oxidation of the sensors than during passive exposure to environmental O₃.^[35] Thus, the recorded voltages showed a drift toward negative values. For the span test at the flow rate of 0.5 Lpm, saturation of the oxidative gas sensors with the span gas (2 ppm NO₂) was achieved faster (t₅₀ = 20 and t₉₀ = 48 s) than the span test at the flow rate of 0.25 Lpm, which could not reach the steady state during the entire test time (150 sec), as seen in Figure 8b.

Similar to the oxidative gas sensors, all recorded voltages were negative after 80 s of zero air exposure, likely due to the decrease in RH. The CO sensors required a longer time than the oxidative gas sensors to report a steady voltage (about −60 mV, as displayed in Figure 8c). In contrast to the oxidative gas sensors, there were shorter response times for the span test of the CO sensors at both tested flow rates compared to the oxidative gas sensors. The flow rate had a negligible effect on the response time (t₉₀ of 45 vs. t₉₀ of 44 s) at the span concentration of 15 ppm of CO (Figure 8d). In conclusion, the higher flow rate, 0.5 Lpm recommended by the manufacturer, is suggested only for field calibration of the oxidative gas sensors. A lower flow rate (e.g., 0.25 Lpm) is recommended for field calibration of the CO sensors as it will require less gas. No flow rates higher than 0.5 Lpm were tested in this study, as it was not recommended by the manufacturer.