STATISTICAL ANALYSIS AND EVALUATION OF PM CONCENTRATIONS DURING THE DUST STORMS AT MAY 2020 FOR SELÇUKLU DISTRICT OF KONYA CITY, TURKEY

: Dust storms are widespread events that occur several times a year and spread over many countries of the world relating the wind direction and speed. Especially particulate matter (PM) is the main pollutant spread over by these storms. Because of the dust storms, PM concentrations increase rapidly in the areas found on the way of dust storm passes In this study, statistical evaluation was made accordingly the PM data measured with personal measurement device in Selçuklu District of Konya and the meteorological and the air pollution data provided from air quality monitoring station, which is affiliated by the Ministry of Environment and Urbanization, located nearby. Pearson correlation test has been applied to both data sets and a significant relationship has been detected between the measured and provided data. Moreover, multiple linear regression was applied to the data for PM 2.5 and PM 10 separately. Adjusted R 2 of the analysis has been found as 0.573 and 0.559 respectively for PM 2.5 and PM 10 which explains almost half of the relationship between PM and meteorological variables. The highest positive effect on PM pollution was determined as air temperature. Finally, principal component analysis (PCA) was applied to both data and 4 different principal components were detected. Measured PM 2.5 and PM 10 , air temperature, and relative humidity were clustered at the same component group.


INTRODUCTION
Air pollution and its effects have become global issues since middle of 19s. Air pollution is transported to long distances with air movements and has global effects. The greenhouse effect major cause of global warming and depletion of ozone layer with effect of many primary and secondary pollutants are major effects of air pollutants. (Zannetti 1990). Main primary air pollutants which are contaminants causing some adverse effects on environment are particulate matter (PM), sulphur compounds (e.g., SO2, H2S), nitrogen compounds (e.g., NO, NH3), carbon compounds (e.g., HCs, CO, CO2) and halogen compounds (e.g., fluorides, bromides, chlorides).
Sources of air pollution are separated as natural and anthropogenic. Anthropogenic air pollution is the waste remaining from the production, transportation, and energy generation ways of humans (Vesilind et al., 2010). Anthropogenic sources of air pollution are classified as the industrial sources, utilities, and personal sources. The main source of industrial air pollution originates from raw materials in manufacturing processes. For example, during mining activity dust and SO2 emissions, during briquetting & coking of coal dust, gases and impurities of coal, during metal smelting SO2 and various volatile metals such as Hg, As, Pb, Cd, and from chemical industry HCl, HF, H2S, NOx, NH3, HCs, VOC are emitted. Utilities are the important source of anthropogenic air pollution because most of them produce electricity by converting energy. This procedure emits huge amount of carbon dioxide, nitrous and sulphur oxides to the atmosphere. Moreover, personal sources such as mobile vehicles, furnaces and stoves in homes, barbeque grills, and burning of leaves in open area contribute to anthropogenic air pollution (Boubel et.al 1994). The amount of gaseous and particulate matter in the air which harms the living organisms increases because majorly of the combustion of fossil fuels. Impurities in the fuel, poor air-to-fuel ratio, too high, and low combustion temperatures lead to the pollutants (Boubel et al., 1994 Besides, air pollution is produced by natural sources such as volcanic eruptions, oceans, forest fires, dust storms, etc. Particulate matter (PM) which is also known as particulate pollution naturally occurs in the dust from the earth's surface, sea salt, and biological material (Morand and Maesano, 2004). Volcanic eruptions produce huge amount of particulate matter and other pollutant gases like SO2, H2S, and methane. These gases and particulate matter stay in the air very long time. Forest fires are another main source of natural air pollution (Boubel et.al 1994).
PM is composed of very small particles and droplets of liquid. Many constitutions such as acids of sulfide and nitrate, organic compounds, metals, and dust particles produce particulate pollution. Moreover, particle size is an important factor affecting the health of all living creatures. Especially particles equal to or smaller than 10 microns are very important for investigators and they must be removed since these particles can pass through the nose and throat and affect the lungs and heart. This can lead to very serious health problems (EPA, 2020). Particulate matters or pollutants are classified into two categories according to the Environmental Protection Agency of the United States (EPA). These are named inhalable coarse particles and fine particles. The diameter of inhalable course particles (PM10) is bigger than 2.5 microns and smaller than 10 microns. Fine particles (PM2.5) have a diameter smaller than 2.5 microns. These particles are found in the atmosphere as a form of smoke and haze that are emitted directly from forest fires and power plants (EPA, 2020).
Turkey is one of the most influenced countries from dust storms coming from middle east countries because of the location. With the help of strict regulations and precautions, PM pollution levels have been lowered last years in most of the country. Dust storms are known as natural processes resulting in high concentrations of PM and they mainly originate from desert areas (Jaafari et al., 2018;Wang et al., 2005). PM levels of Turkey are also affected by these events negatively. Generally, these storms are produced by strong turbulent winds and convective haboobs and fronts are effective forces (Jaafari et al., 2018;Miller et al., 2008). Dust coming from desert areas sometimes may reach up to 6,000 μg/m 3 and create serious effects on daily life such as reducing visibility and respiratory problems on humans (Jaafari et al., 2018;Song et al., 2007). With many epidemiological studies, a relationship has been found between the level of air pollution caused by PM and death cases due to respiratory diseases, lung, cardio and respiratory problems (Karakaş, 2015). Particulate matter exposure may create serious health effects concerning the results of several studies. In these studies, short-term exposure to PM has been found correlated with different health problems (Goudarzi et al., 2013;Linares et al., 2010;Malig and Ostro, 2009;Brunekreef and Forsberg, 2005;Graff et al., 2009;Host et al., 2008;Qiu et al., 2012).
In this study, PM10 pollution measured during a dust storm event in Konya city was evaluated by comparison with previous years' data. Moreover, statistical analysis was conducted for understanding the relation of meteorological parameters and other pollutants concentrations along with PM2.5 and PM10 concentrations.

Data
Data including PM10 (µg/m 3 ), wind speed (m/s), air temperature ( o C), relative humidity (%), atmospheric pressure (mbar), wind direction (degree), SO2 (µg/m 3 ), and CO (µg/m 3 ) was provided from Ministry of Environment and Urbanization Air Quality Monitoring Station (AQMS). Also, the data including PM2.5 (µg/m 3 ) and PM10 (µg/m 3 ) was measured with Particle Counter Dust Measuring Device PCE-PCO 1. The Dust Meter PCE-PCO 1 is a universal measuring instrument developed for the measurement of the density of particles in air. The Dust Meter can measure particles of 6 different sizes. In addition, the Dust Meter can also measure temperature and air thanks to its sensor, thus making the Dust Meter a versatile device. The built-in camera allows us to connect with video and photographic measurement data. The Dust Measuring Device is a device developed to determine the pollution level precisely. All data were measured and provided hourly for 10 days' period between 16 th and 26 th May 2020. This period was determined according to the dust storm satellite images of NASA. After 26 th May 2020, dust storm above Konya region was moved to the north-west direction and the measurements were stopped.

Study Area
Konya city in the Central Anatolia Region of Turkey with the biggest surface area. It is also located on the passageway of the dust storm according to the satellite map images from NASA (NASA, 2020). In Figure 1, the PM2.5 pollution map of Turkey and neighboring countries was shown for 18 May 2020. Figure 1 shows that the PM2.5 values during that day were seen slightly high almost 100 μg/m 3 around the middle part of Turkey. These high PM concentrations are mainly originating from the dust storm events. Day and night dust score and aerosol optic depth value comparison of the same area were given in Figure 2 for 18 May 2020. Day and night dust scores represent the dust intensity on the area with a value between 400-500 (AIRS Level 1B). On 18 May 2020, some parts of Turkey had a high dust intensity with a score of around 500 (AIRS Level 1B). Also, the aerosol optic depth of the area indicating the measure of the extinction of the solar beam by dust and haze was seen close to the highest amount 5 in some parts of Turkey. That is dust particles in the atmosphere were blocked the sunlight by absorbing or scattering light. In Figure 3, a dust surface mass map of Turkey and neighbor countries for 18 May 2020 was given. This map points that there was a huge dust storm formed in the Middle East and Northern Africa. Turkey and other closed countries are affected by the impacts of the dust storm. Thus, the Selçuklu district of Konya which is one of the most affected cities of Turkey from this dust storm was selected as a study area. There are two different measurement location presents in this study. To understand the correction of the personal measurements in one location (personal measurement point), the data taken from nearby measurement station (ministry measurement point) affiliated by Ministry of Environment and Urbanization was used. Figure 4 shows the map of Turkey, the map of Konya and Selçuklu District, and the location of the measurement point and monitoring station. According to Figure 4 (c) locations of the personal measurement point and ministry measurement point are found in the same latitude and they are very close to each other.

Pearson Correlation Test
Pearson correlation test was applied to all data including both measured and taken from the monitoring station. This test mainly explains the linear relationship between pairs of continuous variables. The coefficient called "r" represents the strength of this relationship. The Pearson correlation coefficient for two different parameters (a and b) is computed as equation 1. This coefficient may only take values between -1 and +1. The sign of this coefficient indicates the relationship's direction. The strength of this relation is indicated with the magnitude of the r value that is the closeness to -1 and +1.

Multiple Linear Regression
Two different multiple linear regression analysis was conducted for measured values of PM2.5 and PM10 together with the meteorological and pollution data taken from the monitoring station. In Table 2 and Table 3, variables entered into the regression models were given. Measured PM2.5 and PM10 were used as dependent variables and atmospheric pressure, air temperature, wind direction, CO concentration, SO2 concentration, wind speed, and relative humidity taken from AQMS were used as independent variables for each model. The results of these models show how many percent of the variance of the dependent variable was explained by the independent variables.

Principal Component Analysis
The data with several variables can be classified with the help of principal component analysis (PCA) which is known as the most useful and common method for revealing the potential structure of the data set. PCA method mainly transforms the variables in a huge data set to smaller independent data sets called principal components. The linear combinations of the original variables from huge data are used to create uncorrelated and orthogonal principal components (Azid et al., 2014).
In this study, PCA was applied on 9 components (measured PM10, measured PM2.5, monitored PM10, SO2, CO, wind speed, air temperature, relative humidity, atmospheric pressure, and wind direction) together. Before the PCA analysis, z scores were calculated by SPSS to normalize the data. Kaiser-Meyer-Olkin (KMO) and Bartlett's Test were applied to the data to understand the usability of the data for PCA. KMO measure of sampling adequacy represents the proportion caused by underlying factors and bigger than 0.5 represents that data is useful to be used in factor analysis. Besides, Bartlett's test of sphericity indicates the relation of the variables, and significance which is smaller than 0.05 represents that data is appropriate for the factor analysis (IBM Knowledge Center, 2020) During PCA analysis, the rotation method was selected as Kaiser-Varimax rotation. This method increases the factor loading by maximizing the variance squared loadings. This method highlights the small number of variables and the results may be easily interpreted (Anonymous-4, 2020).

Graphical Comparison
All data used in this study were compared with each other on the graph during 10 days of monitoring. Moreover, PM10 data measured during 10 days between 16-26 th May 2020 was compared with the data of previous years (2015, 2016, 2017, 2018) from the same monitoring station used in the analysis. These graphical comparisons explain the increase at PM10 levels during this period concerning other parameters and previous years.

Pearson Correlation Test Results
Pearson correlation test result of all variables including measured and station data is given in Table 4. It explains that there was a medium positive relationship between measured ((PM10)M) and station (PM_10) PM10 pollution data. Moreover, strong positive relationships were detected between measured PM10 and measured PM2. These results indicate that the measurements made by personal measurement device and measurements of ministry's monitoring station were compatible with each other. Therefore, in regression models and PCA analysis personal measurement data and meteorological data from monitoring station were used together. Moreover, the correlation between the parameters were also supports the results of multiple linear regression and PCA.

Multiple Linear Regression Results
Model summary of multiple linear regression for measured PM2.5 pollution is given in Table  5. Because of the units of independent variables, the adjusted R square value explains the regression results. This result indicates that the independent variables explain 57.3% of the variance of the dependent variable (PM2.5). The coefficients and significance of the independent variables on PM2.5 pollution were explained in Table 6. Wind speed, air temperature, relative humidity, and CO concentrations were found significant, and they explain the change in dependent variable differently. Wind speed affects the dependent variable negatively with a standardized beta coefficient of -0.225. Air temperature, relative humidity, and CO concentrations affect PM2.5 pollution positively with 0.890, 0.276, and 0.154 standardized beta coefficients, respectively. Model summary of multiple linear regression for measured PM10 pollution is given in Table  7. According to Table 7, the independent variables explain 57.3% of the variance of the dependent variable (PM10) In Table 8 coefficients and significance of independent variable for pollution are given. Wind speed, air temperature, relative humidity, SO2, and CO concentrations were found significant. Wind speed explains the change in PM10 pollution negatively with a standardized beta coefficient of -0.194. Air temperature, relative humidity, SO2, and CO concentrations explain the change in PM10 positively with a standardized beta coefficient of 0.851, 0.250, 0.153 and 0.143, respectively. The study made by Fong et al. PM10 were forecasted by multiple linear regression with independent variables such as CO, NO2, SO2, ambient temperature, relative, humidity, wind speed, Mean Sea Level Pressure and rainfall amount. R 2 value was found as 0.715 and major positive contributors was found as CO, SO2, temperature, pressure and negative contributors was found as relative humidity and rainfall (Fong et al. 2018). In another study conducted by Çelik and Kadı, relation between PM pollution and meteorological factors was determined. The results show that temperature, wind speed and humidity affected PM pollution negatively with R 2 of 0.661 (Çelik and Kadı, 2007). Study of Sritong-aon et al. applied multiple linear regression on PM pollution and meteorological factors and found that PM pollution was affected negatively form temperature, relative humidity, wind speed and rainfall and positively from pressure and fire hotspots (Sritong-aon et al., 2021).
The multiple linear regression models in this study had given similar results with previous works. Meteorological factors have significant effect on PM pollution. With respect to the climatic features of the study area, the effects of the parameters change.

Principal Component Analysis Results
To evaluate the suitability of the data for factor analysis, KMO and Bartlett's test was applied. The results of KMO and Bartlett's test are given in Table 9. KMO measure of sampling adequacy value was found as 0.636 which is greater than 0.5 indicating the suitability of data for PCA. Also, the significance of Bartlett's test of sphericity was found as 0.000. .000 Table 10 shows that the total variance explained by PCA and cumulative percent of loading. According to this table, 4 principal components (PC) were detected, and cumulatively 78.108% percent of the total variance was explained by these components. 32.716% of the total variance was explained by PC1, 19.335% was explained by PC2, 13.275% was explained by PC3 and 12.782% was explained by PC4. The rotated component matrix showing the clusters of components is given in Table 11. PC1 includes PM2.5, PM10, air temperature, and relative humidity which are evaluated in the same group. PC2 includes wind speed and SO2 pollution, and PC3 includes wind direction and pressure and PC4 includes CO concentration. The results of this analysis support the results of the Pearson correlation. Also, PM10 and PM2.5 pollution in that area is mainly affected by air temperature and relative humidity. Relative humidity has a negative effect on particulate matter while air temperature has a positive effect on PM concentrations. The component plot in the rotated space is given in Figure 5. In Figure 5, the appearance of the variables in 3D space was shown.

Figure 5: Component plot in rotated space
Several studies were conducted about PCA application on understanding the effect of meteorological factors on particulate matter concentrations and source identification. In Table 12 the results of similar studies were compared with the results of this study on the application of PCA. According to this table, PM concentrations were generally grouped with other pollutants and some meteorological factors such as temperature, wind speed and relative humidity. PCA study conducted by Abdullah at al. resulted that PM10 pollution was grouped with other pollutants like CO, NO2 which were represented as traffic originated emissions . Another study conducted by Zu´ska et al. found that average, maximum and minimum temperature had the greatest effect on PM10 pollution (Zu´ska et al., 2019). Moreover, other study concluded that PM10 pollution is influenced positively from pressure (Fong et al., 2018). Comprehensive study conducted by Khan et al. aimed to identify potential sources of PM10 in the residential area. PCA study results showed that PM10 was mainly contributed by traffic emissions such as O3, CO, NOx, NO, NO2, SO2, CH4 and nonmethane hydrocarbon (NMHC). The other PM10 source was identified as meteorological factors. Temperature and wind speed affected PM10 concentrations positively and relative humidity affected PM10 concentrations negatively. Last component of the study was detected as wind direction which has a negative impact on PM10 concentration levels (Khan et al., 2015). The study of Hashim also applied PCA on air pollution and meteorological data and found that PM10, CO, SO2 were grouped in same component, and they had strongly positive relation with each other as a traffic emission (Hashim et al., 2018). The study conducted by Rahman et al. for Malaysia concluded that PM10 concentrations were affected positively by temperature and negatively by relative humidity (Rahman et al., 2015). When the results of this study were compared with the results of previous studies, it is concluded that PM pollutant concentrations were affected from meteorological factors such as temperature and relative humidity and similar results were found with other studies. Moreover, SO2 and CO were not found contributing factors like some previous studies because the source of PM in this study is dust storm not traffic.

Graphical Comparison Results
The daily mean value of PM10 and PM2.5 measurements and monitoring station data during the measurement period (16-26 th May 2020) was compared in Figure 6. According to this graph, Especially PM2.5 concentrations were detected very high until the 22 nd of May 2020. After that day, the PM2.5 concentrations decreased very sharply with the effect of rain and the effects of dust storm were passed away. Similarly, PM10 concentrations also decreased after the 22 nd of May. Air temperature showed a similar trend with PM concentrations, the decrease in the air temperature was detected after the 22 nd of May. Besides, relative humidity values increased after that day. However, the other parameters such as wind speed, wind direction, pressure, SO2, and CO concentrations showed different trends than PM concentrations, air temperature, and relative humidity. In previous years PM10 values were used to make comparisons about PM10 pollution during the measurement period. In Figures 7 and 8, daily mean PM10 concentrations between 16 and 25 th May of previous years were compared with a daily mean value of 2020 measurements. The PM10 concentrations measured at 16-26 th May 2020 were detected significantly higher than the previous four years (2015,2016,2017,2018) and the average of these years. These graphs prove that the extreme PM concentration increase was observed around the sampling area on 16-22 th May 2020.

CONCLUSION
Dust storms are natural processes affecting PM pollution levels for the short term. Several health issues have been related to the short-term high PM pollution levels. Therefore, it is important to monitor PM levels during dust storm events. In this investigation, PM pollution originating from dust storms was measured 10 days period and the measurement results were compared statistically with the data taken from the closest air quality monitoring station. As a result of statistical analysis, it has been investigated that there is a significant relationship between some meteorological parameters and PM measurements. Air temperature and relative humidity are the major factors affecting PM pollution levels. Besides, change in PM2.5 and PM10 concentrations were explained by meteorological parameters and other pollution concentrations as 57.3% and 55.9 %, respectively. PCA analysis showed that 4 principal components were determined for the data set. PM2.5, PM 10 , air temperature, and relative humidity were clustered in the same component. The statistical analysis supports that PM concentrations are mainly affected by relative humidity and air temperature. Air temperature is the major contributing factor affecting PM levels positively during dust storms oppositely relative humidity has a negative contribution on PM levels. Similar results were found by the PCA study conducted for three different regions such as urban, suburban, and rural areas. PCA results of the rural area indicate that the major contributing factors to PM10 pollution were found as the air temperature with .754 positive effect and relative humidity with -.774 negative effect .
PM levels measured in this study showed that the limit values determined by EU legislation and WHO were exceeded during the dust storm period. Sensitive people who may be affected by high levels of PM pollution should be careful these days. Moreover, this study about monitoring PM during dust storm events is helpful to understand fluctuations in annual measurement and the effects of meteorological factors on particulate matter. It may be also possible to assume the potential PM levels during dust storms concerning meteorological predictions. Future studies may be conducted to produce prediction models for the dust storm cases following the results of this study.