This section presents a temporal and spatial analysis for the pollutants of interest in this study (i.e., PM10 and O3). Also, the classification of the TSS, and their possible relation with PM10 and O3 concentrations in the GMA, are presented as well.
3.1. Temporal and Spatial Analysis of PM10
Figure 3 shows the hourly (a), weekly (b), and daily (c) mean concentrations of PM
10 (in μg/m
3) from the eight measurement stations for the study period.
Figure 3a shows a bimodal feature of the PM
10 hourly mean concentrations. They oscillate around 50 μg/m
3 with a minimum mean concentration of 32.89 μg/m
3 at 4 am (local time). The first maximum of 76.61 μg/m
3 is found at 8 am in the morning, while a secondary maximum is reported at 8 pm in the late afternoon. These peak hour schedules coincide with the hours of highest vehicular traffic when the PM
10 emissions rise. Also, the extreme values, shown with the plus sign in the figure, are found to reach up to 400 μg/m
3.
From
Figure 3b, it can be seen that PM
10 mean concentrations remain relatively high during weekdays and decrease on weekends as a consequence of the decay of the traffic and industrial activity at the end of the week [
30]. The average daily concentrations of PM
10 in
Figure 3c range around 50.13 μg/m
3 with a standard deviation of 21.76 μg/m
3 and extreme values as high as 173.79 μg/m
3 and as low as 7.31 μg/m
3. Such high variance makes that the Mexican standard and the amount recommended by the WHO for daily concentrations of PM
10 are frequently exceeded. One can also notice that these concentrations are related to the seasonal pattern, reaching the highest values in the dryer wintertime.
Furthermore, the monthly concentrations of PM
10 in
Figure 4a begin to rise in October, reaching its maximum value in December and maintaining high amounts until May. This behavior is directly related to the rainy season in the GMA. From May to October, precipitations are more common; therefore, PM
10 concentrations remain low. During the period from November to April, PM
10 concentrations rise in correspondence with the time of low rainfall (see
Figure 2a). A slight increase of these values in August can also be noted, which is related to the midsummer drought.
On the other hand, the annual average of PM
10 concentrations exceeds the Mexican annual standard from 1996 to 2014, as seen in
Figure 4b. It is worth noting that the annual trend points to the decrease of the annual mean concentrations of PM
10, which could be related to the modernization of the vehicle fleet. Furthermore, a series of regulations on the quality of petroleum products have been implemented (NOM-016-CRE-2016 is the most recent one), as well as immediate contingency plans and mandatory vehicle verification programs [
31]. Also note that
Figure 4b shows a definite trend of decrease in the PM
10 concentrations, contrary to what Sánchez et al. [
10] found for the period 2000–2005.
One of the most interesting features that
Figure 4b shows is the peak of PM
10 concentrations reached during 2011. This peak was also observed for O
3 levels (see
Section 3.2). This increase, of both pollutant concentrations, is related to a severe drought that affected the region from Nebraska in the United States (USA) to Central Mexico that year [
32].
Figure A1 shows PM
10 and O
3 monthly concentrations anomalies; it also indicates monthly precipitation anomalies for 2011. Very high negative precipitation anomalies during the rainy season, especially, during June, August, and September can be observed in
Figure A1. Consequently, PM
10 concentrations rose well above mean values during the following, more dry months (September–December). It is worth mentioning that these dryer-than-usual months typically increase forest fires in the region, which are a significant source of PM
10 during the dry season. It should also be worth mentioning that an extensive wooded area called Bosque de la Primavera is located on the immediate west of the GMA. The response for O
3 concentrations to precipitation is rather direct. The lesser the rainfall during the rainy season, the higher the O
3 levels during those months. In other words, as radiation is already favorable for ozone formation in the summertime (rainy season), the absence of precipitation during these same months increase the probability of finding O
3 concentrations higher than usual. During March, which is presented in
Figure 2a as a month that is even dryer than February in the GMA, positive anomalies of PM
10 and O
3 were also found, which could be related (especially for PM
10) with February having registered a negative precipitation anomaly. However, note that positive anomalies of ozone concentrations were found for all of the months of that year.
Finally,
Figure 5 shows the spatial behavior of the mean PM
10 concentrations in the GMA; the image was created using the objective mapping interpolation method with decorrelation scales, so the mean values on the measurement stations (red circles) are conserved. The errors associated with the interpolation method are shown in white solid lines. Here, it can be seen that the largest concentrations are located southeast of the city, which is where most of the point-emitting sources can be found [
4]. Also, this part of the metropolis is less urbanized and keeps many unpaved roads. In contrast, the north and northwest regions report lower mean values of PM
10. This result agrees well with Ramírez-Sánchez et al. [
10] in their analysis of pollutants in the GMA between 2000 and 2005.
3.2. Temporal and Spatial Analysis of O3
The average hourly concentrations of O
3 are shown in
Figure 6a. They fluctuate around 0.026 ppm, while the maximum point values can exceed 0.175 ppm. It can be noted that higher mean O
3 values occur during the hours of higher solar radiation (1 pm to 4 pm), which coincide with the higher temperature intervals in
Figure 2b. Also, the local minimum reached between 7–8 am is linked to the depletion of O
3 during the previous nighttime hours when the absence of the photochemical production of ozone and its destruction on solid surfaces prevail [
6]. Although mean values remained slightly above 0.05 ppm between 1–4 pm, the maximum permissible of hourly concentrations of O
3, as established by the Mexican norm, were frequently exceeded.
Eight-hour mean ozone concentrations are calculated for each hour by averaging the observed O
3 level of the hour, and the values of the previous seven hours. Subsequently, the maximum concentration of all of these averages is determined for each year, and so it is reported as the eight-hour moving average of O
3 concentrations; see
Figure 6c. In the 21-year period studied here, this variable appears to always exceed the established Mexican norm, excelling the year 1996 with the highest value. In this year, several particular episodes of very high concentrations of O
3 were reported [
9].
Figure 6b presents the day-of-week average ozone and nitrogen dioxide concentrations in the GMA for the period of study. It might appear that the O
3 levels behave somewhat odd, since the bar chart shows two local maxima: a maximum during the weekdays and, let’s say, an unexpected peak during the weekend. The average concentrations of O
3 are above 0.0265 ppm during the weekdays and increase on the weekends, reaching their maximum mean value of the week (~0.0270 ppm) on Sundays. However, this behavior is known as the weekend effect [
5,
33,
34,
35]. This effect is produced because the formation of ozone is a highly non-linear process with the rates of production depending on the concentrations of its precursors (NO
x and VOC). Therefore, the weekend effect is often attributed to a reduction of 20% to 30% of the industrial activity and the flow of vehicles during the weekends, as compared with the weekdays, when the phenomenon is limited by VOC [
36]. It can also be noted in
Figure 6b that the NO
2 concentrations decrease drastically toward Sunday, which lets one believe that the excess of NO
2 on weekdays is keeping the ozone formation low. During the weekends, the NO
x concentration falls as the O
3 concentration rises. This result can be explained by the titration reaction (1). Kanda et al. [
5] further explain this topic.
The monthly mean levels of O
3 from 1996 to 2016 in the GMA are presented in
Figure 7a. Here, the ozone concentrations begin to rise in January, reaching their maximum in May; then, they decrease gradually back to January levels. The maximum monthly mean value corresponds to the maximum monthly mean temperature, and to the start of the rainy season; see
Figure 2a. After this maximum, the levels decrease because of two main factors: the decrease in temperature that can be translated into a reduction of the solar radiation received due to cloudiness, and an increase in precipitation that cleans the atmosphere.
On the other hand, the annual trend of the mean concentrations of O
3 in the GMA for the period of study can be divided (as shown in
Figure 7b) into three time intervals. The first time lapse was from 1996–1999, in which ozone levels decreased from 0.03908 ppm to 0.02234 ppm. Next, from 1999 to 2011, concentrations increased gradually, reaching a mean value of 0.03493 ppm in 2011. Benítez-García et al. [
6] noted this same trend to increase the ozone levels and suggested that some other factor, besides temperature, should be causing this long-term behavior. The last period (2012–2016) shows a rapid reduction of ozone levels toward the year 2014 with a minimum value of 0.02038 ppm relative to the whole dataset. Of particular interest are the two maximum peaks in the time series (1996–1997 and 2010–2011). Not much has been said in the currently available literature regarding this feature of the yearly mean ozone concentrations in the GMA. Interannual comparisons of monthly mean ozone levels showed no informative differences between years with extreme values. This comparison is not shown here, because their behaviors are similar to those presented in
Figure 7a. The only noticeable difference between them is that for the years when the maximum is reached, mean O
3 concentrations remain relatively higher during the whole year. Also, no evident relations were found with temperature anomalies nor with the frequency (in cases per year) of a particular TSS; see
Section 3.3 for identified TSS. However, we did note an inversely proportional relation between the years of maximum mean ozone concentrations and the number of sunspots (see
Figure A2). Indeed, its relation with the total column ozone has been previously documented [
37,
38,
39]. Although most of the ozone is found in the stratosphere, stratospheric air intrusions caused by tropopause folding constitute a considerable source of tropospheric ozone [
40,
41]. In addition, a series of studies have been devoted to analyzing the signal of the 11-year sunspot cycle in the troposphere [
42,
43,
44]. On the other hand, Chandra et al. [
45] found that in the tropics, the changes in the tropospheric ozone are out of phase with the stratospheric ozone changes on a time scale of a solar cycle. Therefore, further analysis of the implications of the 11-year sunspot cycle on the lower-troposphere ozone levels is recommended in the area of study.
Similar to the spatial analysis performed for PM
10 mean concentrations,
Figure 8 shows the spatial distribution of mean ozone levels over the GMA. It is observed that the higher values are produced in the north and center of the GMA, in agreement with the most-emitting areas of nitrogen monoxide and VOC [
5].
3.3. TSS Classification
Following the procedure explained in
Section 2.3, and using a correlation coefficient of 0.6, over 100 groups (108) with different synoptic patterns were determined. Subsequently, the average sea level pressure maps of each of them were visually analyzed and reduced to expert judgment to only six groups; that is, six TSS. This reduction was possible because when examined individually, many of them referred to the same category. They were automatically classified as different, merely because the centers of action were displaced. The identified TSS in this study show a high level of agreement with the Isobaric Surface Configuration Types of Mosiño [
18].
Table 2 specifies the frequency of each identified TSS for the period 1996–2016.
Furthermore,
Figure 9 shows the mean fields of sea level pressure for the six TSS identified in this research. TSS I (
Figure 9a) consists of an area of low pressure over Baja California that is related to the heating of the Gulf of California and the dorsal of the Azores-Bermuda anticyclone extending over Central Mexico. The TSS II is a low-pressure system centered on the United States of America (USA). In this TSS, it may appear as a center of high pressure located in the eastern United States; see
Figure 9b. With this pressure field, relatively high pressures can be found over central Mexico. TSS III in
Figure 9c consists of an area of low pressure over northeastern Mexico. This type is recognized when the low center is close to the Gulf of Mexico or already on it, and it may be considered as a particular case of TSS II. The TSS IV stands for those configurations of the baric field that occurs when a trough extends over the east of the Sierra Madre Oriental to south of the Gulf of Mexico; see
Figure 9d. This synoptic situation differs from TSS II regarding the intensity and extension of the trough. TSS V in
Figure 9e shows the influence of high migratory pressures on the entire Mexican Republic. The center of the high-pressure system may be located in the central part of the USA, on the western Atlantic Ocean, or over the Gulf of Mexico. Lastly, the TSS VI is related to a weak subtropical anticyclonic influence, or to the existence of weak troughs or low-pressure systems over Mexico that favor convection, which is typical of the rainy season (
Figure 9f).
Table 3 shows the means and standard deviations of hourly concentrations of PM
10 and O
3 relative to each TSS. To determine whether there are differences between these averages, an ANOVA (one-way) analysis was performed. The ANOVA test has two critical assumptions that must be satisfied for the associated
p-value to be valid: the observations within each group (TSS) must be normally distributed, and their standard deviations should be comparable (homoscedastic assumption). In this case, both conditions are satisfied. Also, one may say a priori that, for the PM
10 in TSS I and TSS VI, their means and standard deviations are already different from the rest of the TSS.
The statistical null hypothesis to test significant differences is that the averages of the hourly PM
10/O
3 concentrations are the same for the different TSS; the alternative hypothesis is that they are not all the same. If the
p-value is less than or equal to the chosen significance level (0.05), the null hypothesis is rejected, and it is concluded that not all of the population means are equal. In both cases (i.e., PM
10 and O
3), their means were significantly heterogeneous: (one-way ANOVA,
F5, 77511 = 1288.3,
p < 0.001) for PM
10 concentrations and (one-way ANOVA,
F5, 58314 = 420.4,
p < 0.001) for O
3 levels. In both cases,
Fk−1, n−k is the
F-value of the test for
k−1 and
n−
k degrees of freedom at 95% significance level, while
p stands for the associated
p-value from the F-distribution; see
Table 1. Therefore, the null hypothesis is rejected, knowing that at least one pair of means has statistically significant differences.
The ANOVA analysis of the hourly concentrations of PM
10/O
3 was performed after subtracting its corresponding hourly means. Also, this analysis used observations that are valid from Monday to Thursday in the case of PM
10, and from Tuesday to Thursday for O
3 levels, in order to eliminate hourly and daily trends related to the anthropogenic activity. This way, only the local effects of the associated TSS on PM
10 and O
3 concentrations are considered. See
Figure 3b and
Figure 6b.
Afterward, Tukey’s test of multiple comparisons was applied to determine which pairs of PM
10 and ozone mean concentrations were statistically different. In this sense,
Figure 10a,b show the results of the post-hoc analysis for PM
10 and O
3, respectively. These figures present the confidence intervals in error bars for all of the compared pairs of mean concentrations relative to each TSS. Whenever the error bars do not cross the y = 0 (discontinuous red) line, averages are said to be statistically different (red error bars). Otherwise, we cannot reject the null hypothesis that there are no differences (blue error bars) between each pair of elements.
For the PM
10 concentrations,
Figure 10a shows (as expected) that TSS I and TSS VI are different from the rest. Among TSS II, III, IV, and V, there are no significant differences. Also, note that although TSS I and TSS VI are different from each other, they tend to be more similar, since its confidence interval is closest to y = 0. A visual comparison between
Figure 9a,f could also support this finding, since their baric fields are alike. It can be seen in
Table 3 that PM
10 averages for TSS I and TSS VI are the lowest of all six situations. This issue is easily explained: in both TSS, convergence will prevail in the GMA, which will favor the dispersion of the pollutants and also will increase the likelihood of precipitation on the region. It is evident that these two factors contribute to the cleaning of the atmosphere.
The spatial distributions of mean hourly concentrations of PM
10 in the GMA are shown in
Figure 11a,b for TSS V and TSS I, respectively. In the GMA, the highest levels of PM
10 are found in TSS V (
Figure 11a). During this TSS, thermal inversions are likely, which can occur on 78% of all days during the dry season [
10]. During TSS I, the lowest levels of PM
10 are reported in the GMA, as shown in
Figure 11b. Its spatial distribution, as in the case of TSS V, seems to be responding to the location of the main emitting sources of PM
10.
In the case of O
3 concentrations, in
Figure 10b, all of the pairs of TSS results were statistically different except for the comparison between TSS I and TSS III. From
Table 3, it can be seen that the ozone levels for TSS II, III, and VI are higher than those reported for TSS I, IV, and V. This result is somewhat surprising. One would expect low O
3 levels for TSS VI, because dispersion is more common during this TSS. Similarly, TSS V reported low ozone concentrations, even though it has been linked with thermal inversions. However, Comrie and Yarnal [
46] found that ozone concentrations are sensitive to seasonality. Therefore, different behaviors in the O
3 levels for one TSS can be observed depending on whether it is the rainy or the dry season. Following this reasoning,
Table A3 shows the hourly mean concentrations of ozone in the GMA for both seasons (dry and wet) within each TSS. It can be noted that TSS VI and TSS IV change their behavior with the season. While TSS VI presents higher values during the dry season, TSS IV shows higher concentrations in the dry season. However, TSS V reports low concentrations of O
3 in both periods. We did not relate these low values with an erroneous classification of the synoptic pattern, since this unusual behavior was only reported for ozone concentrations, and not for PM
10. Maximum values and the interannual variability of the O
3 means were analyzed (not shown here) to try to explain this unexpected feature. Those results did not allow us to justify the unexpected ozone concentrations during non-dispersive atmospheric conditions. These issues require further analysis; hence, it is recommended, as future work, to characterize the local circulations in the GMA for different TSS during the dry and wet seasons.
Furthermore, the spatial distribution of mean hourly concentrations of O
3 in the GMA are shown in
Figure 11c,d for TSS II and TSS I, respectively. According to
Figure 11c, during TSS II O
3, concentrations remain high all over the city with maximum values toward the north. This synoptic situation imposes weak horizontal pressure gradients with a clear sky and low wind speeds over the GMA. Tereshchenko and Filonov [
9] studied some poor air quality episodes of high ozone levels in Guadalajara and linked them to a very similar TSS. In contrast,
Figure 11d shows that the mean O
3 concentrations remain low in the city during TSS I.