1. Introduction

With the development of society and the economy, the pollution of particulate matter (PM) in the air around the world has become more and more serious. A large number of studies (Mukherjee and Agrawal, 2018) have shown that no matter in developed or developing countries, PM in the air poses a huge threat to people's health and life. The WHO (WHO, 2021) has continuously updated its air quality guidelines since it first published them in 1987. There is now stronger evidence than the previous understanding of how air pollution affects different aspects of health at much lower concentrations. Thus in the latest 2021 update, the guidelines have been revised downward by 5 ug/m³ from 2005 (WHO, 2006;WHO, 2021).

It is estimated that exposure to air pollution still causes millions of deaths and loss of healthy life years each year. The burden of disease from air pollution is now estimated to be comparable to other major global health risks such as unhealthy diet and smoking. Therefore, in order to improve the quality of life of residents and increase life expectancy, real-time monitoring of the concentration of PM in the atmosphere is particularly important (Snyder et al., 2013). At the same time, indoor PM concentrations in elementary schools near PM pollution hotspots (Lee et al., 2022), as well as indoor PM concentrations in public transportation vehicles (An and Park, 2022;Yun et al., 2022), have recently received high attention. These increasingly widespread PM monitoring needs are stimulating the emergence of new and innovative measurement technologies and methods.

The U.S. EPA has designated three types of methods for atmospheric particulate matter measurement and monitoring: the Federal Reference Method (FRM); the Federal Equivalent Method (FEM); and other non-FRM/ FEM methods (Gilliam and Hall, 2016) . The FRM is a formal EPA reference method that collects the mass of particles from a 24-hour sample (12:00 a.m. to 11:59 p.m.) on a weighing filter in a low-humidity environment. FEM can be any measurement method that has been shown to be equivalent to the FRM method according to EPA regulations, including the Tapered Element Oscillating Microbalance (TEOM), Beta Attenuation Monitor (MetOne BAM), and Light Scattering Method. Unlike filter-based FRM measurements, most FEMs generate data semi-continuously and in real-time. TEOM uses a microbalance to measure the mass gain on the filter. BAM measures the absorption of β radiation by particles collected on the filter strip to calculate the mass of the particulate matter. Light scattering methods, on the other hand, derive particle concentrations from the intensity of light scattered by the particles.

β ray measurement is the most common method of monitoring the atmospheric particulate matter concentration. It continuously measures the mass concentration of atmospheric particles by collecting dust on filter papers of different cut-off sizes and measuring the attenuation of beta-ray energy. It has the advantages of simplicity, convenience, and low maintenance, but due to its high price and large equipment, single measurement type, and long measurement cycle (mostly 1-hour average), it is difficult to measure multiple particles at the same time with high frequency.

The light scattering method determines the concentration of particles in the air by measuring the change in the intensity of scattered light and converting it into the concentration of particles (Alfano et al., 2020). Under certain conditions, the intensity of scattered light has a good measurement relationship with the concentration of particles. Its advantages are low price, real-time measurement, and ease to carry. Therefore, the device can measure particles of various sizes, such as PM_2.5, PM₁₀, TSP, etc. Its disadvantage is that, due to the low price of the equipment, its measurement accuracy may not be as good as the β-ray absorption method (Giordano et al., 2021). Another factor is that in a high-humidity atmospheric environment, the inherent hygroscopicity of dust particles will cause the diameter of the particles to increase significantly after absorbing water vapor, thereby affecting the accuracy of the measurement results. Generally speaking, if the relative humidity is greater than 35%, atmospheric particles will absorb water vapor and expand, and if the relative humidity is greater than 60%, the water vapor adsorption capacity will be greater (Jeong et al., 2008). However, in another study, it was found that the optical method measurements only gradually deviated from the reference value at humidity levels greater than 75% (Jayaratne et al., 2018). The mixing state of different hygroscopic particles also had an effect on the optical sensor measurements (Wang et al., 2021). The effect of humidity can be corrected by the κ- Köhler theory (Crilley et al., 2018), and a correction algorithm based on the particle size distribution can significantly improve the accuracy of the sensor (Di Antonio et al., 2018). And in the winter when the humidity is less than 60%, it is found that the measurement results of the light scattering sensor seem to be related to the temperature (An et al., 2021).

Some studies believe that although water vapor does not directly affect light scattering, water vapor can cause changes in the properties of particles, which in turn affects the results of light scattering measurements (Han et al., 2020). Related research in North America found that most atmospheric particulate matter is composed of inorganic substances, such as sulfate, nitrate, chloride, sodium, and ammonium ions. The inorganic salt formed by these ions will deliquesce and absorb water in a high humidity environment. Therefore, as the relative humidity of the environment changes, the absorption and loss of water by aerosols will significantly change the lightscattering characteristics of the aerosol itself (Day and Malm, 2001). Under conditions of high relative humidity, the measurement results of atmospheric particulate matter concentration by the light scattering method will bring certain errors. For example, when the relative humidity is greater than 80%, two or more light scattering enhancement factors are observed (Day et al., 2000;McInnes et al., 1998;Rood et al., 1987). Aerosol types show strong but relatively different humidity dependence under low and high relative humidity conditions (Sun et al., 2013). Atmospheric aerosol particles are composed of a mixture of different substances. These substances adsorb or absorb water vapor under high humidity, but have weak hygroscopicity under low humidity (Winkler, 1973).

The above studies show that the relative humidity of the atmosphere is one of the important factors that affect the accuracy of the optical method to determine the concentration of particulate matter. Therefore, it is necessary to further study the influence of the relative humidity of the atmosphere on the concentration of atmospheric particulate matter measured by the light scattering method. In this study, the standard salt solution was used to evaluate the performance of the humidity sensor, and then the optical particle concentration sensor and the beta-ray particle concentration measuring instrument were placed outdoors for field experiments. After a period of time, the data was collected and analyzed. The effect of concentration on sensor measurements was first analyzed. Then the effect of humidity was examined in the range where the effect of concentration is low. The degree of deviation of the optical method measurement results under different humidity conditions is studied, and the model is built and evaluated to correct optical measurement results.

2. Experimental materials and methods

2.1 Selection and calibration of humidity sensor

The humidity sensor determines the environmental relative humidity according to the change in resistance or capacitance of the humidity sensitive material when it absorbs water vapor in the air. There are many types of humidity sensors on the market, and their performance varies greatly. In order to accurately determine the humidity, we compared the characteristics of four kinds of humidity sensors (Table 1), and finally selected the DHT22 sensor. The humidity measurement range of the DHT22 sensor is 0 to 100%, and the resolution is 0.1%. It can accurately measure the temperature and humidity changes of the environment, and output a stable digital signal at a sampling rate of 2Hz.

Preparation of saturated aqueous solutions for humidity standards: Before the field experimental measurement, we used a set of saturated aqueous solutions for humidity standards in the laboratory to evaluate the accuracy of the humidity sensor. These solutions are saturated solutions of MgCl₂, NaI, KI, KCl, NaCl, and K₂SO₄. At 20°C, the theoretical humidity values corresponding to the headspace of the closed container are 33.07%, 39.65%, 69.90%, 85.11%, 75.47%, and 97.59%, respectively (Table 2, Carotenuto and Dell’Isola, 1996).

Humidity sensor calibration: Transferring the prepared saturated solution into a 1L jar, passing the three wires through the pre-drilled holes in the bottle lid, fixing the wires with hot melt glue and sealing the hole, and then connecting the DHT22 temperature and humidity sensor. Covering the bottle and screw tightly to seal. The DHT22 sensor is connected to the Arduino microprocessor board outside the sealed container. After connecting the Arduino (with the uploaded sketch) to the computer with a USB type A-B cable, the humidity data can be obtained in real-time (Fig. 1).

2.2 Measurement of outdoor PM_2.5, PM₁₀ concentration

In this experiment, we used both the light scattering method and the β-ray absorption method to measure the concentration of dust particles in the atmosphere. The β- ray absorption method uses the E-BAM-9800 device (Met One Instruments, Inc, USA). The instrument is equipped with BX-802 PM₁₀ inlet and BX-807 cyclone which are the inlet heads of PM₁₀ and PM_2.5 respectively, which are used to detect the concentration of PM₁₀ and PM_2.5 (designated by the Environmental Protection Agency). The 9250 AT sensor is used to determine the ambient temperature. EBAM-9800 uses the relationship between the attenuation of beta particles and the particle deposition on the glass filter belt to continuously measure the particle concentration every hour (Macias and Husar, 1976;Schweizer et al., 2016). E-BAM is suitable for portable applications that require rapid deployment and shortinterval real-time measurement.

The PMS 7003M particle concentration sensor is used for the determination of PM₁₀ or PM_2.5 particle concentration level by the light scattering method. The internal and external design of the sensor is shown in Fig. 2. When dust particles in the air are captured by the light beam, light scattering occurs. There is a quantitative relationship between light scattering intensity and dust concentration and dust particle diameter. By measuring the intensity of scattered light perpendicular to the direction of the incident light, the concentration of dust particles can be inferred (Li and Wu, 2014).

The relevant information on the light scattering method measuring instrument is shown in Table 3. The minimum resolution particle size of PMS 7003M is 0.3 microns, which makes it easier for us to observe changes in low concentrations. Taking PM_2.5 as the standard, the effective range of PMS 7003M is 0~500 μg/m³, and the maximum range is greater than 1,000 μg/m³. The sensor has a built-in microprocessor, which can directly output the measurement signal in digital form after internal processing. The working temperature range of PMS 7003M is -10°C ~ 60°C, and the working humidity range is 0 to 99%, which can meet the requirements of the general environment.

2.3 Data processing and analysis

We measured PM concentration for about a month, 2020.5.21 ~ 2020.6.20 at Daejeon University campus. After reading data from the data logger of the optical sensor, unreasonable data in the data set were removed, such as values of PM concentration exceeding the measurement range of the sensor, incorrect temperature and humidity values, and other records. After data cleaning, hourly averages were calculated to obtain the optical sensor measurement data set with n=469.

For the BAM instrument measurement results, the records with Alarm=0 were first removed, and then the data less than 4.0 (whose minimum detection limit < 4.0 μg/m³ typical) were deleted according to the operation manual. Thus, the final data set of PM₁₀ measurements with n=493, and PM_2.5 measurements with n=421 were obtained.

The three cleaned data sets were combined in a timestamped alignment to obtain a data set with n=312. The filtering strategy is to judge outliers based on three times the standard deviation of PM concentration, concentration difference, and normalized concentration. Data with normalized concentrations over 3.5 is also deleted. In addition, the data with a PM_2.5 concentration of less than 10 clearly distinguish one group from other data, so it is not included in the entire data set.

To evaluate the performance of the optical sensor, the concentration difference and normalized concentration of the two methods were calculated and added to the dataset. Python 3.8 software package was used for statistical and regression analysis in this study, including numpy, pandas, scipy and matplotlib.

3. Results and discussions

3.1 Basic Statistical Analysis

Fig. 3 shows the comparison of the sensor data with the BAM data after calibration. It is clear that all PM_2.5 data points are located at the vicinity of the 1:1 line after calibration. The difference between the sensor calibration and the BAM data as a function of the measured BAM concentration is shown in Fig. 4. The concentration difference decreases with increasing PM concentration for both PM_2.5 and PM₁₀, thus showing a clear concentration effect, which will be discussed in detail in the next section.

As can be seen from Table 4, the means of the calibrated sensor data are the same as the BAM. The standard deviations of both methods for the coarse particle concentration measurements are larger than the corresponding fine particle measurements. It can also be seen that the standard deviations of the measurements for the optical sensors are both smaller than those of the corresponding BAM.

The distribution of the data pairs for these two particle concentration measurements is shown in Fig. 5, while the humidity color bar is listed on the right side of the figure. As can be seen from the graph, the data pairs with high particulate matter concentrations correspond to high humidity. There is a clear inflection point for PM_2.5 data points, while PM₁₀ data are more dispersed. The figure also clearly shows that the low-concentration area is accompanied by low humidity, while the high-concentration area also has high humidity. Therefore, the effects of PM concentration and humidity on light scattering sensor measurements will be discussed in detail below.

Table 5 shows that there is a very good correlation (>0.97) between PM_2.5 and PM₁₀ from the optical sensor. In contrast, the correlation between the two particle concentrations measured by BAM is only 0.73. Regarding the effect of meteorological conditions, there is a positive correlation between the concentrations measured by both methods and the ambient humidity, while there is a weak negative correlation with the ambient temperature.

The errors in the light scattering sensor measurements are expressed as root mean square error (RMSE) and normalized root mean square error (NRMSE). The former is the square root of the squared mean of the difference between the light scattering sensor concentration and the paired BAM reference measurements. The latter is the RMSE divided by the mean of the BAM reference measurements, which is 0.28 and 0.35 for PM_2.5 and PM₁₀, respectively.

3.2 Effect of PM concentration

The effect of PM on the measurement results of the light scattering sensor is shown in Fig. 6. The dispersion of the PM concentration difference for PM₁₀ does not differ significantly over the entire measurement concentration range, but the dispersion of the PM concentration difference for PM_2.5 shows a decreasing trend as the concentration increases. The dispersion of PM_2.5 concentration difference showed convergence with increasing concentration. For normalized concentrations, this high dispersion at low concentrations and rapid convergence at high concentrations is more evident. This highlights the concentration effect in light scattering sensor measurements. The effect of PM concentration on light scattering sensor measurements suggests that minimizing or eliminating concentration effects is an important consideration in improving sensor performance when analyzing light scattering data.

To investigate the effect of concentration on the measurement results of the light scattering sensor in different concentration ranges, all the measurement data were divided into five concentration groups according to the BAM measurement results, namely 'A[0,15]', 'B[15,25]', 'C[25,40]', 'D[40,55]', and 'E[55,100]'. The concentration differences and concentration ratios for each concentration group were calculated separately and plotted against the BAM measurements as shown in Fig. 7. It is clearly seen that both the concentration difference and concentration ratio of PM_2.5 decrease with increasing concentration, indicating that the deviation of light scattering measurements from BAM is large at low concentrations, while the deviation is reduced at high concentrations. the variation of the concentration ratio of PM₁₀ is basically similar to that of PM_2.5, but the variation of the concentration difference is very dissimilar. The maximum concentration difference of PM₁₀ appears in group B[15,25], while that of PM_2.5 appears in the maximum concentration group E[55,100], the concentration difference reverses to a negative value. Such results indicate that the PM concentration has a significant effect on the light scattering sensor measurements.

It can be seen that the concentration has a greater effect on the light scattering sensor measurement of PM₁₀ than PM_2.5. In the low-concentration region, the light scattering sensor measures higher concentrations than the BAM method, and conversely in the high-concentration region the light scattering sensor measures lower concentrations. This indicates that the β-ray method is less sensitive in the low-concentration part, and conversely, the light scattering sensor measures lower concentrations in the high-concentration part due to interference effects, etc. It indicates a concentration effect between the two methods. This implies that when examining the effect of humidity on sensor measurement results, the additional effect from concentration should be eliminated, or that it is appropriate to determine the humidity effect using only data from the medium concentration region excluding the low and high concentration regions.

3.3 Effect of Humidity

As discussed in the previous section, in order to explore the effect of humidity on PM light scattering sensor measurements, data for intermediate concentrations (15–55 μg/m³) were retained for humidity effect analysis. To examine the effect of humidity on light scattering sensor measurements, the graphical analysis of the normalized concentration (the ratio of a sensor to BAM concentration) and the variation of the concentration difference with humidity is used here. From Fig. 8 and Fig. 9, it can be seen that both the concentration difference and the normalized concentration do not show a clear trend in the direction of increasing humidity, but a random distribution.

This indicates that the increase in humidity has almost no effect on the PM measurements of the light scattering sensor. Although many papers have reported the existence of a non-negligible effect of humidity on PM optical sensor measurements, a few recent studies have also shown that humidity has almost no effect on such sensor measurements. The results of this study at least illustrate that the effect of humidity on PM optical sensor measurements presents different results with different regions. Since the effect of humidity requires further correction, the presence or absence of humidity effects should be determined before deploying such sensors.

The shapes of each group of the violin plot of Fig. 9 are different because the data size and distribution of each humidity group vary. For PM_2.5, the differences and ratios among the groups are small, indicating that humidity has almost no effect on the sensor measurements of PM_2.5. As for PM₁₀, the difference between the groups is not significant except for the small difference of group B, and thus the effect of humidity is almost negligible.

4. Conclusion

In this paper, we investigated the performance of the light-scattering PM sensor deployed with the BAM reference PM monitor in measuring ambient atmospheric PM concentrations in the field. The measured data were cleaned and then used for data analysis such as statistical and graphic analysis. The results show that although meteorological conditions such as ambient humidity have some correlation with PM concentration, it has almost no effect on the measurement results of the light scattering sensor. In contrast, the effect of PM concentration on the measurement results cannot be ignored. PM concentration has a greater impact on PM₁₀ measurements than on PM_2.5 measurements. PM concentration seems to have a greater impact on sensor measurements than humidity. The reasons may be related to particle size and interference effect through the light path. The results of bias, linearity, and error indicate that the performance of this light scattering sensor is somewhat different from the recommended method, and how to improve the performance of such sensors will be the focus of future research work.