Role of Meteorological Parameters in the Diurnal and Seasonal Variation of NO2 in a Romanian Urban Environment

The main purpose of this study was to investigate whether meteorological parameters (temperature, relative humidity, direct radiation) play an important role in modifying the NO2 concentration in an urban environment. The diurnal and seasonal variation recorded at a NO2 traffic station was analyzed, based on data collected in situ in a Romanian city, Braila (45.26° N, 27.95° E), during 2009–2014. The NO2 atmospheric content close to the ground had, in general, a summer minimum and a late autumn/winter maximum for most years. Two diurnal peaks were observed, regardless of the season, which were more evident during cold months. Traffic is an important contributor to the NO2 atmospheric pollution during daytime hours. The variability of in situ measurements of NO2 concentration compared relatively well with space-based observations of the NO2 vertical column by the Ozone Monitoring Instrument (OMI) satellite for most of the period under scrutiny. Data for daytime and nighttime (when the traffic is reduced) were analyzed separately, in the attempt to isolate meteorological effects. Meteorological parameters are not fully independent and we used partial correlation analysis to check whether the relationships with one parameter may be induced by another. The correlation between NO2 and temperature was not coherent. Relative humidity and solar radiation seemed to play a role in shaping the NO2 concentration, regardless of the time of day, and these relationships were only partially interconnected.


Introduction
Monitoring atmospheric pollution and the possibility to predict its evolution are of high interest. One major atmospheric pollutant is nitrogen dioxide (NO 2 ), which may cause many health problems [1]. The NO 2 gas has a reddish-brown color, is nonflammable, and has a detectable, pungent odor, perceptible from concentrations of approximately 190 µg/m 3 [2]. Nitrogen dioxide reacts with the hydroxyl radical (-OH) in the atmosphere, forming the highly-corrosive nitric acid, but it can also form toxic organic nitrates. Nitrogen oxides known as NO x (i.e., NO + NO 2 ) are involved in the formation of tropospheric ozone and smog, mediated by light through photolysis. Due to the significant environmental impact, the NO x compounds are strictly monitored and legislative limits were set at EU and national levels [2]. Part of the NO 2 molecules in the atmosphere are of primary nature (i.e., directly emitted), while most NO 2 results from nitrogen monoxide (NO) [1,3]. The latter is produced via natural and anthropogenic processes. About one-fifth of NO x is released in the atmosphere during The number of vehicles in Braila has increased during the last few years, which led to a rather increased concentration of atmospheric pollutants. BR1 is a traffic station; it is located on Calea Galaţi No. 53 (The Agency of Environmental Protection Braila: http://apmbr.anpm.ro/) and monitors on a continuous basis the pollution levels generated mainly by traffic emissions, with medium and high flows, from the neighboring streets. Calea Galati Street is one of the busiest traffic streets in Braila. The width of the street is 16 m and it is covered with an asphalt carpet. The area around the air quality monitoring station has apartment blocks with four floors and some green spaces. The air quality monitoring station is located approximately 20 m from Calea Galati Street and 15 m from the apartment buildings. The surface is flat, characteristic of a plain area. The sources of pollution in the area consist mostly of road traffic and domestic heating.
The hourly atmospheric concentrations of NO 2 are determined in situ, by using the chemiluminescence technique [8,21]. This method implies the reduction of NO 2 to NO in the presence of a Molybdenum surface, at a temperature of roughly 310 • C. The resulted NO reacts with ozone (O 3 ), leading to the formation of fluorescent NO 2 , whose emission is detected by a sensor [4,21]. Data between 2009 and 2014 were used in this study. Besides, concentrations of NO 2, meteorological parameters were also recorded.
Satellite measurements were provided by the Ozone Monitoring Instrument (OMI) onboard the Earth Observing System Aura satellite [22,23]. The OMI measures several important pollutants, such as O 3 , NO 2 , SO 2 , and aerosols, with a spatial resolution of 13 km × 24 km, i.e., at a near urban scale resolution, https://aura.gsfc.nasa.gov/omi.html [24]. OMI is a remote sensing space instrument onboard the Aura satellite that provides a raw NO 2 product called DSCD (Differential Slant Column Density), which is retrieved using Differential Optical Absorption Spectroscopy (DOAS) in the 405-465 nm range. Satellite measurements provide valuable data about atmospheric pollutants, including NO 2 [25][26][27][28][29][30][31][32] at a large scale. The most refined product of OMI is the tropospheric Vertical Column Density (VCD), which is the result of a near real-time retrieval algorithm that gives a 0.7 × 10 15 molec./cm 2 uncertainty for each individual pixel [26]. In our study, we have used level 3 data, which are a conversion of OMI 13 × 24 km 2 data to a 0.25 • × 0.25 • resolution [33]. Such data collected between 2005 and 2015 by OMI confirmed a reduction in NO 2 concentration over Northern China, Eastern Europe, and USA, while an increase in NO 2 concentration was detected over the Persian Gulf and India [30].

Methods
Meteorological parameters recorded at an hourly rate were used, in order to verify to what extent they are relevant for modulating the NO x variability by means of correlation analysis. Correlations/anticorrelations (i.e., positive/negative correlation coefficients) between two variables are not the decisive proof for direct cause-effect relationships; however, they suggest that a link may exist if a physical mechanism supports it. When two variables are not correlated, the first thought is that there is no link between the two variables, which, however, may not always be true. Moreover, meteorological parameters (temperature, humidity, solar radiation) are not independent (e.g., solar radiation at the ground is a measure of the cloud cover, which is linked to relative humidity but also to temperature). Consequently, the correlation analysis was refined and a partial correlation analysis was used [34,35], complementary to the bivariate correlation. Such an analysis is useful when the relationship between two variables, e.g., X and Z, measured by the bivariate correlation coefficient, C XZ , may be induced or suppressed by a third (intervening) variable, e.g., Y, which affects both variables X and Z. The partial correlation P X(Y)Z corresponds to the link between X and Z when Y is constant. To be more specific, in this particular case the main variables, X and Z, were the NO 2 concentration and one meteorological factor (e.g., temperature) and the intervening variable, Y, was another meteorological parameter (e.g., radiation). The radiation (Y) may affect both the NO 2 concentration (X) and the temperature (Z). The difference D X(Y)Z = P X(Y)Z − C XZ between the partial correlation coefficient, P X(Y)Z , and the direct correlation coefficient, C XZ , measures the degree of intervention for Y. If P X(Y)Z is smaller than C XZ , i.e., the difference and the direct coefficient have opposite signs and C XZ ×D X(Y)Z < 0, then Y is responsible for part of the correlation, or it is said that Y "intervenes". If the two coefficients, C and D, have the same sign, i.e., C XZ × D X(Y)Z > 0, then the correlation is real and Y even suppresses, partially, the correlation. If the partial and direct coefficients are equal, i.e., D = 0, then the Y variable does not intervene. Table 1 summarizes possible results and interpretations for various combinations of direct and partial correlation coefficients. The correlation analysis has been applied to standardized data (by their standard deviation) in order to identify possible links between NO 2 concentration in the atmosphere and the variations of several meteorological parameters. . These values are smaller than those measured by traffic stations in Bucharest [28].

Diurnal and Seasonal Variation
The NO 2 content varied with year; it was low in 2009 and in the first part of 2010, reached a maximum in the winter between 2010 and 2011, and then slightly decreased towards the end of the interval. The period is too short for assessing whether a trend does exist.
The seasonal variation of NO 2 was evident; NO 2 concentrations were clearly higher during winter than during summer. The seasonal variation may be explained by: (1) the strong variation of the anthropogenic sources contributing to NO 2 emissions, i.e., traffic and combustion, but also by (2) a longer lifetime of the NO 2 during winter, taking into account that the NO 2 lifetime and its concentration in the atmosphere are affected by the seasonal variation of the photochemical activity [7,8,26,31], which is reduced during winter. Additionally, during winter, the soil is cold, and thus the nearby air is heavy so that the emitted pollutants may stay close to the ground longer [2]. Another explanation for the seasonal NO x variation relates to the variation of the boundary layer height with the season, which influences the wind pattern that, in turn, has a very important role in pollutant dispersion [7].
Two diurnal peaks of the NO 2 concentration could be observed, centered around 09:00 Local Time (LT) and 20:00 LT, which were more evident during cold months. The diurnal peaks are in agreement with previous findings [8] and are associated with peaks in traffic [7]. Morning peaks, observed around 9:00, were smaller than the evening peaks. The NO 2 concentration increased during the afternoon-evening period, between 17:00 and 24:00, with maxima progressing towards earlier time during winter months. This was seen in all seasons except summer. In December, these diurnal maxima were smaller, for all years, which can be linked to the reduced traffic and industrial activity due to holidays. earlier time during winter months. This was seen in all seasons except summer. In December, these diurnal maxima were smaller, for all years, which can be linked to the reduced traffic and industrial activity due to holidays.  Figure 2 shows the variation of the tropospheric NO2 VCD (Vertical Column Density) over Braila, derived from OMI, together with in situ measurements at 11:00 LT, which is the approximate hour when the satellite overpasses the city. The tropospheric NO2 VCD varied between 0.8 × 10 15 molec./cm 2 for February 2011 and 4.7 × 10 15 molec./cm 2 for December 2012. Note that the comparison in Figure 2 is strictly qualitative, since OMI measures the number of NO2 molecules in a column over a grid of 0.25° [33], while in situ instruments measure the volumetric concentration near the ground; thus a direct quantitative comparison makes no sense. The annual NO2 VCD ranges between 1.73 and 2.23 × 10 15 molec./cm 2 . Previous studies have shown that OMI measurements give reliable results about NO2 emissions of anthropogenic sources at a large scale in cities [26,32,36], above large industrial platforms (located away from cities) [25,37] and given by vehicle emissions and also by residential emissions [38]. However, other studies have shown that satellite observations do not correctly evaluate the NO2 pollution caused by traffic or by other point sources [27,28]. Figure 2 shows that the two time series agreed relatively well except for a period between the winter of 2009-2010 and the summer of 2011. The correlation coefficient between the two series was close to 0.40 (after values in December 2010 were removed) and did not change when NO2 concentrations measured earlier, between 08:00 and 11:00 LT, were considered for the mean.
The seasonal variation of the tropospheric VCD was more regular: it was small during warm months and high during cold months. In 2010, the two time series did not coincide, which is mainly due to a peculiar behavior of the in situ measurements that showed an unexpected wide maximum during summer and autumn, and a minimum in December. The OMI instrument "sees" a large surface area (312 km 2 ), and thus integrates the emissions of many sources. Consequently, it cannot accurately measure the local variability, captured by the in situ measurements. This opposing variation suggests that a punctual, temporary, source of NO2 was added in the second part of 2010 close to the measuring station.  Figure 2 shows the variation of the tropospheric NO 2 VCD (Vertical Column Density) over Braila, derived from OMI, together with in situ measurements at 11:00 LT, which is the approximate hour when the satellite overpasses the city. The tropospheric NO 2 VCD varied between 0.8 × 10 15 molec./cm 2 for February 2011 and 4.7 × 10 15 molec./cm 2 for December 2012. Note that the comparison in Figure 2 is strictly qualitative, since OMI measures the number of NO 2 molecules in a column over a grid of 0.25 • [33], while in situ instruments measure the volumetric concentration near the ground; thus a direct quantitative comparison makes no sense. The annual NO 2 VCD ranges between 1.73 and 2.23 × 10 15 molec./cm 2 . Previous studies have shown that OMI measurements give reliable results about NO 2 emissions of anthropogenic sources at a large scale in cities [26,32,36], above large industrial platforms (located away from cities) [25,37] and given by vehicle emissions and also by residential emissions [38]. However, other studies have shown that satellite observations do not correctly evaluate the NO 2 pollution caused by traffic or by other point sources [27,28]. Figure 2 shows that the two time series agreed relatively well except for a period between the winter of 2009-2010 and the summer of 2011. The correlation coefficient between the two series was close to 0.40 (after values in December 2010 were removed) and did not change when NO 2 concentrations measured earlier, between 08:00 and 11:00 LT, were considered for the mean.
The seasonal variation of the tropospheric VCD was more regular: it was small during warm months and high during cold months. In 2010, the two time series did not coincide, which is mainly due to a peculiar behavior of the in situ measurements that showed an unexpected wide maximum during summer and autumn, and a minimum in December. The OMI instrument "sees" a large surface area (312 km 2 ), and thus integrates the emissions of many sources. Consequently, it cannot accurately measure the local variability, captured by the in situ measurements. This opposing variation suggests that a punctual, temporary, source of NO 2 was added in the second part of 2010 close to the measuring station.  In general, the diurnal variation was relatively regular for fall, when a relatively small data spread was seen. A higher variability was observed in winter, spring and summer, when the data spread was larger. There were no outliers in the middle part of the day, when the traffic is slightly reduced compared to the morning and afternoon. The outliers during morning and afternoon may be associated with peaks or drop-offs of the traffic. The least regular season was summer, when the number of outliers was high regardless of the hour and their large majority lied in the upper part of the plot. This suggests that the NO2 loading was significantly higher than the average for a short period of time. This is confirmed by the next analysis ( Figure 4). The explanation might relate to road construction works that resulted in deviation of the traffic. Starting with 2010, landslides occurred on streets near the measuring station. In 2011, the asphalt was removed and some excavations were done, in order to check the underground water and sewerage pipes. Subsequently, utility pipes were installed, the foundation was compacted and the traffic flow returned to normal values.
Apparently, the NO2 concentration and the temperature were anticorrelated, especially during daytime: high NO2 concentrations during morning corresponded to the lowest temperatures, while the afternoon reduction in NO2 coincided with the highest temperatures. However, this is just coincidental, since the daytime variation of NO2 is governed by traffic [8], whose peaks happen to occur at times when the temperature is lowest. Moreover, during nighttime, the NO2 concentration and the temperature seemed to be correlated.

Monthly Deviation of NO2 Concentration from the Seasonal Mean
Obviously, the diurnal variation of NO2 was not the same for each year. Figure 4 shows the difference between the hourly NO2 concentrations at ground level and the hourly seasonal mean during each month of each year. Hourly seasonal means are averages of hourly NO2 values measured during 2009 and 2014 over those months that define a season, i.e., November-February for winter, March-April for spring, May-August for summer and September-October for fall.  In general, the diurnal variation was relatively regular for fall, when a relatively small data spread was seen. A higher variability was observed in winter, spring and summer, when the data spread was larger. There were no outliers in the middle part of the day, when the traffic is slightly reduced compared to the morning and afternoon. The outliers during morning and afternoon may be associated with peaks or drop-offs of the traffic. The least regular season was summer, when the number of outliers was high regardless of the hour and their large majority lied in the upper part of the plot. This suggests that the NO 2 loading was significantly higher than the average for a short period of time. This is confirmed by the next analysis ( Figure 4). The explanation might relate to road construction works that resulted in deviation of the traffic. Starting with 2010, landslides occurred on streets near the measuring station. In 2011, the asphalt was removed and some excavations were done, in order to check the underground water and sewerage pipes. Subsequently, utility pipes were installed, the foundation was compacted and the traffic flow returned to normal values.
Apparently, the NO 2 concentration and the temperature were anticorrelated, especially during daytime: high NO 2 concentrations during morning corresponded to the lowest temperatures, while the afternoon reduction in NO 2 coincided with the highest temperatures. However, this is just coincidental, since the daytime variation of NO 2 is governed by traffic [8], whose peaks happen to occur at times when the temperature is lowest. Moreover, during nighttime, the NO 2 concentration and the temperature seemed to be correlated. Int. J. Environ. Res. Public Health 2020, 17, x 7 of 15 The analysis of the temporal variation of the aforementioned differences may be useful for identifying months or periods of the day when the NO2 loading departed from the expected variability, which in turn may help in finding the cause for the outliers seen in Figure 3. Values lying in the positive/negative part mean that the NO2 concentration were higher/lower than the average at that particular time.
A clear pattern could be observed for equinox seasons: the afternoon peak was higher during colder months (March and October) than during warmer months (April and September) for each year. This was not true for the morning peak or for solstice seasons. The month of May, which is part of the summer season, was the least regular; the NO2 content in 2010 was much lower, while in 2011 it was much higher than the average. The unusual increase during the second part of 2010, shown in   Figure 4 for July 2010 and, partially, for June 2010. The explanation might relate to the construction works previously described, which resulted in significant alterations of the traffic flow during 2010-2011. In general, the major contributor to the summer mean for all years came from the NO2 content in August, since the departures from the seasonal mean were positive for all years except 2011.

Monthly Deviation of NO 2 Concentration from the Seasonal Mean
Obviously, the diurnal variation of NO 2 was not the same for each year. Figure 4 shows the difference between the hourly NO 2 concentrations at ground level and the hourly seasonal mean during each month of each year. Hourly seasonal means are averages of hourly NO 2 values measured during 2009 and 2014 over those months that define a season, i.e., November-February for winter, March-April for spring, May-August for summer and September-October for fall.
The analysis of the temporal variation of the aforementioned differences may be useful for identifying months or periods of the day when the NO 2 loading departed from the expected variability, which in turn may help in finding the cause for the outliers seen in Figure 3. Values lying in the positive/negative part mean that the NO 2 concentration were higher/lower than the average at that particular time.
A clear pattern could be observed for equinox seasons: the afternoon peak was higher during colder months (March and October) than during warmer months (April and September) for each year. This was not true for the morning peak or for solstice seasons. The month of May, which is part of the summer season, was the least regular; the NO 2 content in 2010 was much lower, while in 2011 it was much higher than the average. The unusual increase during the second part of 2010, shown in Figure 2, is confirmed by the higher values seen in Figure 4 for July 2010 and, partially, for June 2010. The explanation might relate to the construction works previously described, which resulted in significant alterations of the traffic flow during 2010-2011. In general, the major contributor to the summer mean for all years came from the NO 2 content in August, since the departures from the seasonal mean were positive for all years except 2011.
The NO 2 content in winter also depended on month and year. November seemed to be the month with the highest contribution to the winter seasonal average of NO 2 concentration for the first part of the interval (2009-2011). In December, the NO 2 diurnal variability changed with the year: the NO 2 content was lower during the first three years and higher afterwards. We assume that the explanation lies in a combination of meteorological conditions and important variations of traffic and industrial emissions during these particular years. During the analyzed period, the distribution of thermal energy and hot water in Braila municipality was mainly controlled by the heating operator SC "CET" SA. Starting with 2012, the lack of investments in modernizing the heating system affected the control of emissions: severe losses and the evolution of the methane gas tariff led to an increase of classical housing heating systems, whose impact is important especially during the cold season [39].

Correlation between NO 2 Concentration and Meteorological Parameters
The effect of meteorological parameters on the variability of NO 2 concentration for urban sites is, still, a conundrum. Table 2 shows some examples of correlations between NO 2 and meteorological factors for different locations [13][14][15][16][17][18][19][20]. The correlation between temperature and humidity, on the one side, and NO x , on the other side, was positive, negative or insignificant. The correlation with the wind speed was more consistent, i.e., is negative for all sites, which is rather normal, since a stronger wind will disperse the NO 2 at a local urban site. This is the reason for investigating whether meteorological parameters may be linked to the variability of the NO 2 in a relatively small city, with an average level of pollution [31] and whether this relationship depends on seasons or on time of day.  Figures 2-4 show that during about 07:00 and 21:00 LT, the NO 2 variability was strongly influenced by traffic, which is also confirmed by [8,37,[40][41][42]. The correlation coefficients were computed separately for the full 24 h time (black bars), for daytime (8-21 LT, red bars) and for the nighttime (22-7 LT, blue bars). The separation between daytime and nighttime was done because the influence of the road traffic should be lower during nighttime, and thus meteorological parameters may play a different role in modulating the NO 2 concentration during the night. Obviously, this is not valid for radiation, which is absent during nighttime. However, one should keep in mind that the NO 2 content is largely controlled by traffic, especially during the day, and this will definitely affect the correlation with meteorological parameters (or lack thereof). However, we assumed that the traffic pressure does not change for the analysis period. Figure 5 shows the variation of the direct bivariate correlation between NO 2 and temperature (left), and humidity (right) with time. Correlation coefficients calculated for the full day are shown by black bars, while for day (night) these are shown by red (blue) bars. Only coefficients that were significant at 90% are shown. There was no clear NO2 dependence on temperature, since correlation coefficients were positive for March, May, August and November, while for February and September these were negative. During daytime most correlations were negative; however, this was already discussed as being artificially induced by rush-hour traffic. Correlations did not change from day to night, except for May, when the correlation changed from negative for the full-day to positive for the full day and night. The full 24 h correlation was rather small or insignificant for most months and changed from positive to negative during months that had similar meteorological characteristics, e.g., March (positive), April (negative) and May (positive). Unsurprisingly, NO2 and temperature were anticorrelated during the day, but this was already discussed as being mostly artificial. Significant correlation during the night was positive in March, May, August and November, and negative in January, February and September. All in all, temperature seemed to play no clear role in the variability of NO2.
Negative coefficients were found for the NO2-temperature dependence by [14,[16][17][18], without a clear association with seasons. The NO2 lifetime is higher during winter; thus the "night" analysis may be less relevant during winter because of the lower concentrations of the N2O and N2O5 species. These species are key factors in the removal of NO2 during nighttime and in its transformation into nitric acid [43]. The amount of NOy species is higher during warmer seasons when bacteria and agriculture activities are intensified [44]. However, [15] and [19] found that an increase in temperature would be followed by an increase in NOx.
The correlation of the NO2 concentration with the humidity, when significant, was positive for most months. The only exception was May, during which the NO2 variability departed significantly from the expected behavior (Figure 4). The correlation did not change from day to night except in September.
In the following the intervening effect on direct correlations is analyzed, to see whether the existing correlations were spuriously induced, especially for the link to humidity. The partial There was no clear NO 2 dependence on temperature, since correlation coefficients were positive for March, May, August and November, while for February and September these were negative. During daytime most correlations were negative; however, this was already discussed as being artificially induced by rush-hour traffic. Correlations did not change from day to night, except for May, when the correlation changed from negative for the full-day to positive for the full day and night. The full 24 h correlation was rather small or insignificant for most months and changed from positive to negative during months that had similar meteorological characteristics, e.g., March (positive), April (negative) and May (positive). Unsurprisingly, NO 2 and temperature were anticorrelated during the day, but this was already discussed as being mostly artificial. Significant correlation during the night was positive in March, May, August and November, and negative in January, February and September. All in all, temperature seemed to play no clear role in the variability of NO 2 .
Negative coefficients were found for the NO 2 -temperature dependence by [14,[16][17][18], without a clear association with seasons. The NO 2 lifetime is higher during winter; thus the "night" analysis may be less relevant during winter because of the lower concentrations of the N 2 O and N 2 O 5 species. These species are key factors in the removal of NO 2 during nighttime and in its transformation into nitric acid [43]. The amount of NO y species is higher during warmer seasons when bacteria and agriculture activities are intensified [44]. However, [15,19] found that an increase in temperature would be followed by an increase in NO x .
The correlation of the NO 2 concentration with the humidity, when significant, was positive for most months. The only exception was May, during which the NO 2 variability departed significantly from the expected behavior ( Figure 4). The correlation did not change from day to night except in September.
In the following the intervening effect on direct correlations is analyzed, to see whether the existing correlations were spuriously induced, especially for the link to humidity. The partial correlation was used to identify whether the effects of meteorological parameters (temperature, relative humidity and solar radiation) on NO 2 concentration were interconnected. The main and intervening variables (described in Section 2) are shown in Table 3. The results are shown only for four out of the six possible combinations, because these are the most relevant for identifying possible intervening effects. The correlation between NO 2 and temperature was inconsistent and when assessing the intervening effect of radiation on correlation with humidity one can infer that the intervening effect of humidity was similar.  Figure 6. Partial correlation between NO2 concentration and meteorological parameters. Full bars describe the bivariate correlation and empty bars describe the difference (D) between the partial and bivariate correlation. Correlation coefficients for the whole day are shown in black, while red/blue correspond to day/night, respectively-see text for details. Coefficients are shown for the >90% confidence level. Partial correlation between NO2 and humidity is shown in (a) (temperature constant) and (b) (radiation constant). Partial correlation between NO2 and radiation is shown in (c) (temperature constant) and the effect of radiation on the correlation with temperature is in (d). Table 3. Meteorological parameters as main/intervening variables, for the partial correlation analysis.

Variable 1 (X) Variable 2 (Z)
Intervening Variable (Y) Plot in Figure 6  NO2 relative humidity temperature a NO2 relative humidity radiation b Figure 6. Partial correlation between NO 2 concentration and meteorological parameters. Full bars describe the bivariate correlation and empty bars describe the difference (D) between the partial and bivariate correlation. Correlation coefficients for the whole day are shown in black, while red/blue correspond to day/night, respectively-see text for details. Coefficients are shown for the >90% confidence level. Partial correlation between NO 2 and humidity is shown in (a) (temperature constant) and (b) (radiation constant). Partial correlation between NO 2 and radiation is shown in (c) (temperature constant) and the effect of radiation on the correlation with temperature is in (d). Figure 6 shows the results of the partial correlation analysis for the combinations in Table 3. Full bars describe the direct correlation, while empty bars stand for the differences between partial and direct coefficients. According to Table 1, when these two are opposite, the correlation is partially mediated by the intervening variable. When these both have the same sign, the correlation is real.
No consistent direct correlation between NO 2 and humidity was found for the whole 24 h period (Figure 6a, full black bars are completely absent). Notably, D NO(t)H was rather high and positive for the whole 24 h day during May, June and December. This means that the temperature suppressed a potentially positive NO x -humidity correlation. During daytime, NO 2 and humidity were positively correlated (red full bars) for a large part of the year, except for February, August and November. The differences between partial and direct correlation, D NO(t)H , albeit small, were of opposing signs for most cases. This suggests that the temperature was partially responsible for the NO 2 -humidity correlation.
The important role played by the humidity in the variation of NO 2 is supported by Figure 6b, where the effect of radiation on the same correlation (NO 2 -humidity) is shown only for daytime. The solar radiation partially induced a positive correlation during some months, but the effect was not important, since all D NO(R)H were pretty small. Our results agree with [19] or [20], but contradict [13], who showed that NO 2 was negatively correlated with relative humidity. They argued that NO 2 concentrations are slightly higher at a lower relative humidity because the reactions between NO 2 and OH are less frequent, and thus the NO 2 persists more in the atmosphere. However, this was not confirmed by our results.
The NO 2 concentration was negatively correlated with solar radiation during the entire year. A higher direct radiation implies, usually, a higher air temperature; thus the intervening effect of temperature and radiation was analyzed. Figure 6c shows that the temperature did not artificially induce the anticorrelation with solar radiation (small or absent empty bars), except for May and September, when the effect was, however, small. This holds for the intervening effect of solar radiation on the anticorrelation with temperature, (Figure 6d), which was also small.
Based on observations in [45], which showed that, for a site in India, O 3 correlated negatively with both NO x and the humidity during all seasons, we suggest that the positive correlation between NO x and humidity may be an indirect result of the photochemical effect of solar radiation and humidity on ozone. Unfortunately, ozone measurements were not available to confirm this hypothesis. This also may partly explain the observed anticorrelation between NO 2 and solar radiation. Increased solar radiation favors the production of O 3 , which, in turn, reduces the NO 2 loading in the atmosphere [45].
In general, comparisons with other studies are not straightforward, since we are not aware of similar investigations of the intervening effect of meteorological parameters. Additionally, most studies did not consider monthly changes. Moreover, correlations between the NO 2 concentration and the meteorological parameters should be different for different cities, since the anthropogenic landscape and microclimate change significantly from one urban location to another [10,31].

Conclusions
This paper describes the diurnal, monthly, seasonal and annual evolution of the NO 2 concentration for 2009-2014, measured in situ by an urban traffic station in southeast Romania. The role that meteorological parameters might play in modulating the NO 2 variability was investigated in an attempt to separate the anthropogenic effects (which are well-known) from the effect of the local microclimate.
As expected, the NO 2 was higher during the cold season, except for one year, 2010, when summer NO 2 levels were highest. This suggests that natural factors, such as the effect of temperature on the NO 2 lifetime, are less important than the anthropogenic ones at urban sites. Some annual variation also existed, with low values at the beginning of the interval (2009-2010), most likely caused by a severe reduction of industrial activity.The summer minima and winter maxima have both anthropogenic and natural causes and the departure from these may relate to temporary changes of the local traffic and/or construction activities. The NO 2 diurnal variability was clearly shaped by the local transport: two diurnal peaks were observed, one around 8-9 LT and another one around 20-21 LT, and both were associated with increased road traffic, confirming previous observations at other urban sites. The afternoon peak was higher during the colder months (March and October) than during the warmer months (April and September) for each year. An irregular diurnal variation of the NO 2 concentration was seen in May and December. The most consistent season was autumn, with a relatively similar diurnal variation in all years.
Additionally, we found that over Braila, space observations of OMI followed the in situ observations during most of the selected interval (R = 0.4).
The analysis of the correlations between the NO 2 concentration and temperature, relative humidity and radiation has shown that the association with temperature is the least relevant. The correlation changed from positive to negative throughout the year without a clear pattern.
Obviously, the contribution of traffic cannot be disregarded and may mask or suppress the impact of temperature variations on the NO 2 concentration. Our assumption that during the night, the situation may change due to traffic disappearance, was not confirmed. The correlations with the humidity and radiation, on the other hand, were notably consistent: the NO 2 concentration correlated with the relative humidity and was anticorrelated with radiation for almost the entire year. Moreover, most of these relationships were real and the intervening effect of the other meteorological parameters was small.
Our results showed that finding a link between meteorological parameters and NO 2 variability for an urban site is a difficult task. Attempts to predict the NO 2 behavior based on meteorological data, even combined with traffic flux data, cannot be very successful at the local or regional scale or on a short-term basis, since landscape, infrastructure, traffic, local activities, and population clearly affect the NO 2 concentration. However, this may be useful for assessing trends or long-tern variability at a large scale, since these are averaging over a large area, thus reducing local and short-term contributions.