Sensors | Free Full-Text | Improving the Quality of Measurements Made by Alphasense NO2 Non-Reference Sensors Using the Mathematical Methods

The main goal of this study was to investigate the measurement data quality of NOAlphasense (NO2-B43F) sensors. The research focused on interfering factors and lead to the calculation of expected errors in comparative measurements. Results showed that without improving correction methods, the range of measured air pollution concentrations may be greater than their actual values in ambient air. Therefore, based on the conducted comparative measurements with professional devices, the article proposes an innovative algorithm for converting raw measurements, taking into account air temperature and relative humidity. The proposed formula does not require complex analytical tools and a history of measurements, so it can be used in simple microcontrollers. Its effectiveness has been checked on measurements carried out in a location other than the original measurements. The research flowchart is presented in Figure 1

Many authors (e.g., mentioned below) suggest that to improve the quality of measurements, a mathematical correction is necessary; however, determining the correction functions remains relatively poorly presented [ 22 ]. Lin et al. examined O, NO(Aeroqual), and particulate matter (RTI) sensors and determined simple, linear correction functions [ 23 ]. Additionally, the mutual influence of particular pollutants was analyzed. The study concluded that better results are obtained by the correction and the use of a data set from many sensors, instead of relying only on data from the sensor for which the correction function is determined and which were subjected to comparative measurements. In different research scenarios, after carrying out the measurements with the Alphasense NOelectrochemical sensor and reference device and determining the correction function, the coefficients of determination for the tested set of sensors increased from the range of 0.3–0.7 to 0.6–0.9 [ 24 ]. The authors concluded that measurement campaigns using low-cost sensors based on modern generations of electrochemical NOsensors can provide useful complementary data on local air quality in an urban environment. In the study of Wei et al., electrochemical sensors (Alphasense B4 series) for carbon monoxide (CO), nitric oxide (NO), nitrogen dioxide (NO), and oxidants (O) were evaluated under controlled laboratory conditions to identify the influencing factors with sensor outputs [ 10 ]. Based on the laboratory tests, the authors developed different correction methods to compensate for the impact of ambient conditions.

In other research, the authors calibrated samples of three NO2-B43F sensors and three OX-B431 sensors with NOand Oexclusively and conducted mixture experiments over a range of 0–1.0 ppm NOand 0–125 ppb Oto evaluate the ability of the paired sensors to quantify NOand Oconcentrations in the mixture [ 21 ]. Although the slopes of the response among samples of three sensors of each type varied by as much as 37%, the individual response of the NO2-B43F sensors to NOand OX-B431 sensors to NOand Owere highly linear over the concentrations studied (R> 0.99).

One of the biggest studies in comparative measurement focused on validating electrochemical (EC) sensor measurements of CO, NO, NO, and Oat an urban neighborhood site with pollutant concentration ranges: [NO] = 11.7 ± 8.3 ppb and [O] = 23.2 ± 12.5 ppb [ 20 ]. Through the use of high-dimensional model representation, the authors showed that interference effects derived from the variable ambient gas concentration mix and changing environmental conditions over three seasons can be effectively modeled for the Alphasense sensors.

Bauerová et al. evaluated low-cost electrochemical sensors in field testing measurements [ 19 ]. The data from different sensors (for SO, NO, O, and CO) were compared with co-located reference monitors used within the Czech National Air Quality Monitoring Network. The results showed that in addition to the given reduced measurement accuracy of the sensors, the data quality depends on the early detection of defective units and changes caused by the effect of meteorological conditions (effect of air temperature and humidity on gas sensors). This author concluded that comparative measurement is necessary before each sensor’s field application.

Enhancing the spatial and temporal resolution of air pollution monitoring is nowadays one of the emerging challenges [ 6 ]. With the development of low-cost air quality sensors lot of research has been focused on achieving accurate, robust, and reliable air quality data [ 7 8 ]. The equipment was evaluated according to its performance in different environments, seasons, and meteorological conditions [ 9 10 ]. Critical issues faced by researchers are mostly correction methods [ 11 12 ] and the long-term stability of the sensors [ 13 ]. To characterize interferences and improve correction functions, both laboratory and ambient tests were conducted [ 14 18 ].

Life quality and human health are affected by air pollution, especially in urban areas, where most of the population lives [ 2 3 ]. Europe’s most problematic pollutants in terms of health are PM, NO, and ground-level O 4 ]. Nitrogen dioxide (NO) is one of the major pollutant gases, and its emanation is mainly caused by traffic [ 5 ].

Measurement instruments incorporating these sensors recorded the values of the appropriate voltages every few seconds, and then, every minute relayed these data to the server. The data were aggregated, as a result of which, 1 h mean values were determined. The study includes the comparative analysis of the data obtained in this way and 1 h measurement data from SEM’s air quality monitoring station. The following statistical measures were used to compare the measurements from Alphasense sensors to measurements from the SEM station: Pearson’s correlation coefficient, mean error, mean percentage error, mean absolute error, mean absolute percentage error, and mean square error. All these statistical measures were determined for particular months of the measurement periods.

In the comparative measurements carried out at the monitoring station in Nowy Sącz, two NO 2 -B43F sensors were used. For each of these electrochemical sensors, called later in the text as NO2_1 and NO2_2, the following set of parameters was available: WE E , AE E , WE 0 , and AE 0 .

For the NO-B43F sensors the manufacturer proposes the use of one of the following two equations to determine the corrected voltage (we will refer to them also as method (1) and method (2), respectively):

To convert the measured value to the pollutant concentration, depending on the sensor used, first, the corrected voltage value (WE c ) is calculated and then the appropriate conversion factor (mV/ppb), also, given individually for each sensor, is applied. The last step is to convert the concentration to the appropriate unit (e.g., from (ppb) to (µg/m 3 )).

The values of the above-mentioned factors from the manufacturer’s documentation for the tested type of sensors are shown in Table 1

AE e —the value of the electronic offset for the used ISB plate for the auxiliary electrode (mV);

WE e —the value of the electronic offset for the used ISB (Alphasense Individual Sensor Board) for the working electrode (mV);

At the same time, the following values are provided individually for each sensor (as a result of calibration performed by the manufacturer):

Alphasense sensors record the voltage related to the measured quantity at the two outputs of each sensor:

From the research perspective, the city of Nowy Sącz is interesting due to its frequently occurring high air pollutant concentrations and varied weather conditions [ 28 ]. The town is situated in a diversified area—the lowest point of the city is located at an altitude of 272 m above sea level, and the highest point—475 m above sea level. There are typically urban areas with tenement houses, parks, and green areas as well as single-family houses. Some areas are supplied by the district heating network, while others, especially single-family houses, are equipped with individual boilers or fireplaces (mainly for solid fuels), being a key source of particulate matter emissions. On the other hand, Warsaw, which is the capital of Poland, has a slightly different climate but is also quite diverse in terms of the variability of pollutants and main sources of emission [ 29 ].

The measurement performance analysis of Alphasense (NO2-B43F) sensors was conducted in two research fields. Sensors were placed at the air quality monitoring stations, carrying out measurements based on reference methods. The stations are part of the State Environmental Monitoring (SEM)— Figure 2 . The locations and timing of comparative measurements were as follows ( Figure 3 ):

According to the manufacturer’s description, together with a dedicated electronic board, they enable the measurement of even small concentrations of nitrogen dioxide (ppb level). To test the sensors measuring devices were built. The main element was a small measuring chamber in which two Alphasense sensors were placed. To take the air directly from the surroundings, a small fan was installed at the inlet to the measuring chamber. The measuring chamber (with direct air intake from the environment) was placed in a larger housing, which contained electronic components necessary to power the sensors and microcontroller to acquire the results (voltages from electrodes acquired from the sensor’s transmitter board) and transfer them to the server. The inlet to the measuring chamber was directed downwards (analogous to the air outlet). As a result, the measuring chamber was not susceptible to wind force and rain/snow.

Alphasense NO(NO2-B43F) sensors are popular, low-cost sensors for measuring the concentration of nitrogen dioxide in the ambient air using the electrochemical method (4-electrodes). The working principle of the NOelectrochemical sensor is based on electrochemical reactions [ 5 ]. When the air passes through, it creates a reaction in the electrochemical cell. The surface of the working electrode is the site for the first half-reaction (oxidation), generating an electronic charge, balanced by the second half-reaction (reduction) that occurs at the counter electrode [ 25 ]. This type of sensor provides high selectivity, low limit of detection, low power consumption, and linear response to the target gas [ 26 27 ].

3. Results

In order to determine the quality of the obtained measurements, the measured voltage at both outputs of individual sensors was converted into concentrations with the use of both methods presented above. For the obtained results, the basic statistics for each month were calculated. The results are shown in Table 2 . These statistical parameters were calculated by comparing the data from a given Alphasense sensor with the values obtained from the SEM measurement station.

Table 3 shows the average monthly temperature measured in the chamber of the device. It should be emphasized that the instrument was intentionally left unsheltered and therefore the maximum temperatures inside the device during the summer period often exceeded +40 °C.

The data presented in Table 2 show relatively different errors; in particular, months of the measurement period. For both tested sensors and both methods for converting voltages into concentrations suggested by the manufacturer, the lowest values of this factor were recorded in July. The best parameters were recorded in October. To some extent, this may have been related to the variability of actual concentrations.

The highest absolute percentage error occurred for the measurements conducted in warm months. The decrease in the average temperature lowered the error. While in July, the absolute percentage error exceeded 130%; in November, it oscillated around 50%.

A similar tendency is also visible in case of the absolute errors. The absolute errors decreased from 11–15 µg/m3 in July to 8–10 µg/m3 in November. It is also worth mentioning the high correlation of low-cost sensors’ indications—for the entire period, it was r = 0.952 after applying the method (1) and r = 0.949 after applying the method (2). This means that the readings of the sensors were repeatable and that they both reacted similarly to the parameters and changes in the atmospheric air, and the actual concentration values (and possibly other pollutants). This repeatability also implies the assumption that a single, universal method of mathematical correction of sensor indications can be created.

2 and the temperature during one of the study days are shown in

During warm and hot days, when the temperature was approximately or exceeding 30 °C, both sensors regularly generated voltage that corresponded to negative concentrations of nitrogen dioxide after applying methods (1) or (2). The variability of 1 h mean concentrations of NOand the temperature during one of the study days are shown in Figure 4 and Figure 5

Individually for both methods (1) and (2), deviations due to high temperature were corrected. After calculations, negative WEC values and the relevant minimum TWEC temperature were determined in this set. For method (1), it was TWEC = 23.1 °C, while for method (2), TWEC = 23.8 °C. Then, for all hours for which the average temperature was higher or equal to the determined minimum, the average temperature and the determined voltages WEU and AEU were collected. For each hour, the desired WEC° voltage was determined, which would correspond to the NO2 concentration measured in the SEM station. It means that the WEC° value denotes the result of the relationship (1) or (2) where the result from the Alphasense sensors would correspond exactly to the value determined by the SEM station. In the next step, the difference WEC′ = WEC° − WEC (offset) was calculated, determined by how much the current indication from the low-cost sensor should be corrected to obtain the value corresponding to the measurement from the SEM.

Multiple regression was applied to the set of 1 h mean values (T, WEU, AEU, and WEC′), where the independent variables were: 1 h mean temperature (T), 1 h mean voltage WEU, and 1 h mean voltage AEU. The dependent variable was the WEC′ value.

WEC′ = −3.489(WEU − WEE) + 2.448(AEU − AEE) + 0.103T + 3.39,

(3)

WEC′ = −2.815(WEU − WEE) + 2.118(AEU − AEE) − 0.238T + 12.357,

(4)

As a result, for (1) and (2) the following relationships were obtained, respectively:

C′ values were added to the WEC for measurements where the TWEC was higher than the predetermined cutoff for both methods. In the case of the previously presented data from 20 July 2019, the new volatility patterns are presented in

These additional WE′ values were added to the WEfor measurements where the Twas higher than the predetermined cutoff for both methods. In the case of the previously presented data from 20 July 2019, the new volatility patterns are presented in Figure 6

2 concentrations were recalculated for both Alphasense sensors. The results for July and August 2019 (being the months with the highest temperatures) are presented in

Equations (3) and (4) were added to (1) and (2), respectively, and again NOconcentrations were recalculated for both Alphasense sensors. The results for July and August 2019 (being the months with the highest temperatures) are presented in Table 4

The use of an offset (methods (3) or (4)) in most cases improved the quality of the results. First of all, the “negative” concentrations were removed, and the correlation with the measurements from the SEM stations was significantly improved. The highest improvement appeared in August, where correlation coefficients in relation to the SEM stations were r = 0.8. Correlation between NO2_1 and NO2_2 sensors in July were: r = 0.934 (for (3)) and r = 0.923 (for (4)), while in August: r = 0.959 (for (3)) and r = 0.946 (for (4)). The mean values of the absolute percentage errors and thus the absolute errors also improved significantly. In most cases, they were reduced by half.

The next step was to analyze the set of measurement data and determine the new correction functions. First, the relationship between the Alphasense sensor indications and the measurements from the SEM stations was examined.

Figure 7 presents that the best fit function was a 2-degree polynomial regression.

NA′ = 0.0085NA2 + 0.4215NA + 5.7901,

(5)

The obtained relationship is as follows:where

• NA—NO2 concentration measured by the Alphasense sensor, determined from the equation:

NA = (WEC + WEC′) sA nA,

(6)

where

• sA—conversion factor (mV) to (ppb) given by Alphasense individually for each sensor;

• nA—factor for converting the NO2 concentration from (ppb) to (µg/m3);

• WEC—described by Equation (1);

• WEC′—described by Equation (3).

In the last step, multiple regression was used to bind the sensor indications with the meteorological parameters. The independent variables were: 1 h mean temperature (T), 1 h mean relative humidity (H), 1 h mean NO2 concentration expressed by the Equation (5), 1 h mean measured voltage (WEU − WEE), and the 1 h mean measured voltage (AEU − AEE). The dependent variable was the concentration of NO2: NA″.

Although the Equation (5) is based on the measured WEU and AEU voltages, it was decided to include them as additional, independent variables (or in fact, as differences in relation to the WEE and AEE values) to obtain a relationship with the interval, to which these measured voltages belong (as opposed to Equation (5), where information about the measured voltage values is lost). Secondly, the lack of additional consideration of WEU and AEU voltages led to worse results than those presented below.

NA″ = 1.059NA′ − 0.244(WEU − WEE) + 0.463(AEU − AEE) − 0.304T − 0.023H + 5.54,

(7)

The obtained, final relationship for the original method (1) is as follows:

NA′ = 0.0124NA2 + 0.0973NA + 8.4452,

(8)

The analysis for the second method (described by the original Equation (2)) was carried out similarly. The best fit (among regression: linear, exponential, polynomial—see Table 5 ) between the indications of Alphasense sensors and measurements from the SEM stations was obtained by a polynomial regression of the second degree ( Figure 8 ). The obtained relationship is as follows:

NA″ = 1.05NA′ − 0.087(WEU − WEE) + 1.091(AEU − AEE) − 0.116T − 0.019H − 3.32,

(9)

After applying multiple regression, the final form of the relationship for the original method (2) is as follows:

Statistical parameters after the conversion of the measurement results for NO2_1 and NO2_2 sensors using Equations (7) and (9) are presented in Table 6

2-B43F (indicated later in the text as NO2_3) sensor during the measurement campaign conducted at the Warsaw-Chrościckiego SEM station from December 2019 to May 2020. The results were calculated using Equation (7) and the basic equation recommended by the manufacturer. The results are presented in

The effectiveness of the determined relationship was checked with the use of a new NO-B43F (indicated later in the text as NO2_3) sensor during the measurement campaign conducted at the Warsaw-Chrościckiego SEM station from December 2019 to May 2020. The results were calculated using Equation (7) and the basic equation recommended by the manufacturer. The results are presented in Table 7

Table 3,

During the analyzed period, the statistical parameters describing the measurements calculated using method (1) were closer to the results received from the SEM air quality monitoring station than those presented in Table 2 Table 4 and Table 5 . Therefore, the application of the correction method did not bring such a spectacular improvement as in the measurements in Nowy Sącz. The reason for this could be the fact that in the analyzed period, there were no very high temperatures, as was the case during the measurements carried out between July and September 2019.

Nevertheless, the application of the correction function resulted in the improvement of the measurement data quality. The value of the correlation coefficient improved each month—the highest increase was recorded in May and the lowest in March. In all months, the mean absolute percentage error improved, with the best values obtained for the colder months, and slightly worse for the warmer months. Similarly, to the measurements carried out in Nowy Sącz, and also in Warsaw-Chrościckiego, it can be observed that the sensors during the colder months tend to underestimate the measurements, and during the warmer months—to overestimate (even after applying the correction method).