3.1 Spatial distribution
Figure 2 shows the PM₂.₅ and PM₁₀ mean concentrations across the three years at the monitoring stations. All of the stations measured PM₁₀, but only four stations monitored PM₂.₅. The interannual trend of PM₂.₅ was similar at the Politecnico, Rubino, and Rebaudengo stations. However, the Lingotto station showed a persistent decreasing trend. The concentrations at the Rebaudengo site were higher than those of the other stations. Unlike the other stations, Rebaudengo is a traffic monitoring station, adjacent to intensely busy urban streets. Its average concentration was highest by 31.56% in the first year, 12.39% in the second year, and 11.65% in the third year.
Among the six stations that measured PM₁₀ concentrations, the trends at the Grassi and Rebaudengo stations, both traffic stations, over the three years were similar. A progressive reduction in concentration occurred. The Politecnico, Rubino, and Lingotto stations showed similar trends of second-year concentrations lower than the concentrations of the other two years. The Consolata station had an increasing trend, with a peak in the second year.
To analyze the spatial differences in concentrations more effectively, a descriptive statistical analysis using boxplots is shown in Fig. 3.
Overall, the three-year mean PM₂.₅ concentrations were not statistically different among the Politecnico (19.93 µg/m³), Rubino (19.82 µg/m³), and Lingotto stations (19.77 µg/m³) (ANOVA test, p > 0.05). The three-year mean at the Rebaudengo station (23.51 µg/m³) was statistically different from those of the other stations (ANOVA, p < 0.01). All of the monitoring points had a mean lower than the annual limit, 25 µg/m³ according to 2008/50/EC. The box plots in Fig. 3 show similarities between the Politecnico and Lingotto stations, both background stations, in quartiles and upper and lower limits. However, the Rebaudengo station had a three-year mean close to the legal limit. Its boxplot reveals higher concentration levels in both means and quartiles than at the other stations.
The PM₁₀ concentrations of two background stations and the two urban stations were not statistically different. The mean values were 28.69 µg/m³ at Politecnico, 28.53 µg/m³ at Lingotto (t-test, p > 0.05), 29.61 µg/m³ at Rubino and 30.78 µg/m³ at Consolata (t-test, p > 0.05). The two traffic stations showed different trends. In particular, the mean at the Grassi station was 38.45 µg/m³, but the mean at the Rebaudengo station was 33.99 µg/m³. The quartiles and the upper and lower extremes of these stations were significantly higher than those of the other stations. Although both stations are traffic stations, the Grassi station had higher overall concentrations because of the different surrounding traffic conditions than at Rebaudengo. Combining the PM₁₀ concentrations for the same kinds of stations revealed statistically significant differences (ANOVA, p < 0.05) between the background (28.61 µg/m³), urban (30.21 µg/m³), and traffic (36.16 µg/m³) contexts. This result confirms the observations of (Boanini et al., 2021; Lonati and Trentini, 2019) about concentrations measured in different spatial contexts.
Concentration distributions for the background, urban, and traffic locations were constructed from the data of the six stations by averaging their daily data (Fig. 4). Each curve is based on data derived from two same-type stations. The frequency distribution was calculated by grouping the data into 20 constant-step classes for PM₂.₅ and 24 classes for PM₁₀. The size of the classes for both was 5 µg/m³.
The concentrations of PM₂.₅ and PM₁₀ showed a left-skewed distribution in all boundary conditions, as also shown by studies carried out at other locations (Fan et al., 2020; Ma and Jia, 2016). The distribution shape is attributable to the impact of the summer period, which is characterized by low concentrations, on the distribution. Furthermore, dilution by rain and wind tend to reduce concentrations, contributing to a high frequency in the lower classes (Ouyang et al., 2015). On the other hand, persistent pollution phenomena and atmospheric stability help to increase the frequency of the higher classes, thus lengthening the tail of the distribution (Galindo et al., 2018). Among the PM₂.₅ concentration distributions, the background and urban trends are similar. The distribution mode occurs in the same class: 6.9 µg/m³ for background and 7.9 µg/m³ for urban conditions. Based on the three years of observations, the probabilities of exceeding the daily limit (25 µg/m³) for the urban environment and background were 21.58% and 21.56%, respectively. The distribution mode in the traffic condition (10.2 µg/m³) was more centered than in the other conditions. .Traffic conditions produce higher probabilities in the higher concentration classes than urban or background conditions. The probability of exceeding the legal limit was 27.15%. The PM₁₀ concentrations showed three different distributions for background, urban, and traffic conditions. In background conditions, the mode occurred at 12 µg/m³. Furthermore, for concentrations greater than 50 µg/m³, the daily threshold value, the probability was lower than for urban or traffic conditions. In urban conditions, the mode was 17.48 µg/m³; in traffic conditions, it was 19.88 µg/m³. After its peak, the traffic distribution remained constantly above the other two distributions. The probabilities of exceeding the threshold value under background, urban, and traffic conditions were 14.31%, 17.59%, and 23.14%, respectively. Unlike the case of PM₂.₅, for PM₁₀, there was a noticeable difference between the background and urban concentration distributions. This difference could be attributable to the more significant influence of local urban conditions on PM₁₀ than PM₂.₅, as the authors evidenced in (Boanini et al., 2021).
3.2 Temporal variation
Monthly variation of pollutants
Figure 5 shows the monthly variations of PM₁, PM₂.₅, and PM₁₀ concentrations and the monthly ratios of PM1/PM₁₀, PM₂.₅/PM₁₀, and PM1/PM₂.₅. The data were derived from the mean of the background station at Turin Polytechnic in the period between September 2018 and September 2021. The overall means over the three years for PM₁, PM₂.₅, PM₁₀ were 16.96 µg/m³, 19.93 µg/m³, and 28.69 µg/m³, respectively. In all three years, the maximum limit of daily PM₁₀ exceedance (35 days/year for PM₁₀ > 50 µg/m3) was surpassed. The mean PM1/PM₂.₅ ratio was 0.85, and the PM₂.₅/PM₁₀ ratio was 0.69.
The figure reveals a noticeable V-shaped variation in concentrations and ratios over the months. This trend is similar for all three particulate fractions. The highest concentrations were recorded in the winter months, at the beginning and end of each year. The peaks occurred in January and February, which are generally characterized by haze pollution due to atmospheric conditions favorable to accumulation in the lower layer of the atmosphere (Maurizi et al., 2013). Furthermore, in these months, the PM₂.₅/PM₁₀ ratio is generally higher, reflecting the difference in sources between summer and winter conditions (Choi et al., 2013). Several studies have evaluated the impacts of typical winter heating sources, such as domestic boilers, in the area under study (Pognant et al., 2017) and the entire Po Valley (Gilardoni et al., 2020). The concentration was lowest in the summer months,, especially in May. The spring months showed a gradual reduction in concentrations, whereas the autumn months showed an increase. The same trend was also found by (Chen et al., 2016; Xu et al., 2017). The monthly variations of the PM1/PM₁₀ and PM₂.₅/PM₁₀ ratios were more sensitive than the variation of the PM1/PM₂.₅ ratio. The latter had an almost constant trend throughout the months, with only a slight reduction in the summer. The first two ratios showed a summer month reduction with respect to the annual mean of 17.11% and 15.62%, respectively. However, the PM1/PM₂.₅ ratio had a reduction of only 4.76%. The differences in winter and summer particulate sources and atmospheric conditions have a greater impact on the coarse fraction than the fine fraction (Pecorari et al., 2013; Pernigotti et al., 2012). To provide additonal details, the correlations of PM₁ with PM₂.₅ and PM₁₀ were studied on the basis of different seasons.
In all seasons, the correlation between PM₁ and PM₂.₅ was greater than 0.95 (p < 0.01), as shown in Fig. 6. Moreover, the regression line had a similar slope in winter, autumn, and spring. However, in summer, the regression line had a lower slope. Overall, PM₁ concentrations were better correlated with PM₂.₅ than with PM₁₀. In fact, the PM₁–PM₁₀ correlation coefficient varied from a maximum of 0.91 in winter to a minimum of 0.88 in summer. In all seasons, the PM1/PM10 ratio was lower than the PM1/PM₂.₅ ratio. The PM₂.₅/PM₁₀ ratio reached its maximum value during winter. Autumn and spring showed a similar slope, but summer had the minimum value (0.41, p < 0.01). The latter was attributable to the greater involvement of fine particles in photochemical reactions (Carbone et al., 2010; Wang et al., 2016). In autumn and winter, some points deviated from the global trend. Some points had high concentrations of PM₁₀ in correspondence with low values of PM₁. This is a typical behavior of a large pollution circumstance, as also noted in (X. Li et al., 2017; Xue et al., 2020).
Hourly and seasonal variations of pollutants
Several natural and anthropic factors affect the concentration levels of particulate matter during the day. The most important are direct emissions, secondary particulate formation, dilution or removal processes, and variation of the height of the PBL (Maurizi et al., 2013; Pecorari et al., 2013; Sullivan et al., 2016). Based on the three years of observations, the hourly data through the day were processed to obtain evidence of these processes affecting concentration. In addition, the season-based variation was studied to highlight the hourly evolution of PM concentration.
For the particulate fractions shown in Fig. 7, there was significant hourly variation throughout the day. The variation was more pronounced for PM₁₀ than for PM₂.₅ or PM₁. In particular, the values of PM₂.₅ and PM₁ were substantially stable at night, and their morning increase was reduced compared to that of PM₁₀. The nocturnal decrease of PM₁₀ is attributable to dry deposition, as suggested by (Li et al., 2019), whereas the concentration increase from 7:00 AM to 9:00 AM is typical of the traffic schedule (Chen et al., 2020). For all fractions, after the peak, there was a decline in the afternoon due to the height of the planetary limit state (Du et al., 2013; Su et al., 2018). The dilution peak occurred at 17:00. Subsequently, the evening increase in transport and nighttime stability caused concentrations to increase (Chen et al., 2016).
The PM1/PM₁₀ and PM₂.₅/PM₁₀ ratios shown in Fig. 7 had a fairly similar shape through the day. As confirmation, the PM1/PM₂.₅ ratio had a very narrow fluctuation throughout the day, with an amplitude of 0.05, while the PM1/PM₁₀ and PM₂.₅/PM₁₀ ratios had amplitudes of 0.16 and 0.13, respectively. In comparison to the hourly concentration trend curves, the ratio trends are out of phase. More precisely, the absolute peaks for PM1/PM₁₀ and PM₂.₅/PM₁₀ were reached at 6:00 AM during the relative minimums of PM₁ and PM₂.₅ concentration. This confirms a greater nocturnal removal of coarse fractions than fine particles (Galindo et al., 2018). Finally, the ratio decreased from 7:00 AM to 9:00 AM and reached a maximum at about 13:00. The maximum seemed to be attributable to the higher dilution rate of coarse particles than fine particles, as proven by (Lestari et al., 2003). Furthermore, this peak occurred at the time of maximum solar radiation, which affects the secondary formation processes of particulate matter. Typically, secondary training involves fine fractions to a greater extent (Squizzato et al., 2017; Sullivan et al., 2016; Wang et al., 2016).
To perceive the daily concentration variation more thoroughly, Fig. 8 illustrates the differences in trends during the four seasons and Fig. 9 shows the differences between workdays and weekend days.
According to Fig. 8, the highest concentrations occurred in winter, the lowest occurred in summer, and autumn values were higher than spring values. For PM₂.₅ and PM₁₀ concentrations, the hourly autumn trend was correlated with the hourly-mean annual trend, with correlation coefficients of to 0.92 (p < 0.01) and 0.95 (p < 0.01), respectively. At the same time, the hourly autumn concentrations were less correlated with the hourly annual mean, with a coefficient of 0.88 (p < 0.01).
During the day, the intensity fluctuations varied according to season. Winter had the greatest fluctuation, and summer had the least. The daily percentage variations of PM₁, PM₂.₅, and PM₁₀ were 33.1%, 30.44%, and 24.8% in winter and 25.4%, 20.4%, and 20.9% in summer. This indicates that the height of the PBL, which is lower in winter than in summer, strongly affects the daily concentration fluctuation (Maurizi et al., 2013). As subsidence tends to increase concentrations, radiance produces daily variation in the accumulation and dilution of contaminants (Chen et al., 2016). The fluctuations in spring and autumn were minor in comparison to those in winter but greater than those in summer. The daily oscillations in PM₁, PM₂.₅, and PM₁₀ were 26.8%, 23.9%, and 21.7% in spring and 29.5%, 27.4%, and 23.9% in autumn, respectively. Autumn and winter had similar inter-day variation behaviors, as did spring had to summer, confirming the observations of (Chen et al., 2016; R. Li et al., 2017; Zhao et al., 2018). One of the most important aspects of these trends is the diurnal concentration peak. In the winter and autumn, the peak occurred at 11:00. In summer and spring, it occurred at 9:00 AM.
Differences between weekend and workday trends
An important aspect of the analysis of temporal variation, the differences between weekend days (Saturday, Sunday, and holidays) and workdays (all other days), are illustrated in Fig. 9. The mean daily concentrations of PM₁ and PM₂.₅ were similar for weekend day and workdays. PM₁ had a 0.35 µg/m³ lower concentration on weekend days (19.02 µg/m³)than on workdays (19.37 µg/m³). For PM₂.₅, the difference between workdays (21.51 µg/m³) and weekend days (21.06 µg/m³) was 0.45 µg/m³. On the other hand, for PM₁₀, there was a significant reduction on weekend days (28.75 µg/m³) compared to workdays (30.63 µg/m³). The difference between the two classes of days was 1.88 µg/m³.
For all fractions, the hourly trends through a day showed a greater amplitude on workdays than on weekend days. The morning hour increase and afternoon decrease are more marked for workdays (orange curve). Unlike the results of (Chen et al., 2016), the concentration peak was about two hours ahead on weekends. Furthermore, for PM₂.₅ and PM₁, the relative and absolute minimum values were advanced by one hour.
The mean PM1/PM₁₀ ratio was 0.67 on weekend days and 0.62 on workdays. The mean PM₂.₅/PM₁₀ ratio was 0.75 on weekend days and 0.70 on workdays. Hence, PM1/PM₂.₅ was 0.89 and 0.86 on weekend days and workdays, respectively. In each ratio, the weekend days had a higher value than the workdays. Generally, there is a prevalence of fine particles at night on workdays. In the other hours, the weekend day curve is higher. This means that the weekend days are affected by a higher proportion of fine particles in comparison to PM₁₀ than the workdays. An interesting element of this evaluation is the different trends for the two cases, which was verified by all three ratios considered. From 9:00 AM to 11:00 AM, there was an increase in workday ratios. This did not occur on weekend days, for which the line is U-shaped with a minimum in the afternoon.
The differences in the trends can be attributed to the different traffic flows on weekend days and workdays (Giugliano et al., 2005; Lonati et al., 2011). Traffic is the only source that is reduced during the weekend (Chang et al., 2015b). As was observed recently by (Filigrana et al., 2020) in a traffic study in the Po Valley, and also in the present case, the weekend-day traffic reduction provides significant reductions in concentrations and substantial changes in the relationships between fine and coarse fractions.
Figure 10 shows the percentage differences from the weekly mean value for each day of the week for each of the four seasons. Considering the differences between weekend day and workday concentrations, the objective was to verify how the concentrations were distributed through the week and which days were more polluted than the seasonal mean pollution.
In all of the seasonal charts, the PM₁ and PM₂.₅ curves are very close. PM₁₀ concentrations had a similar pattern to PM₁ and PM₂.₅ fractions in summer and spring. However, the PM₁₀ concentrations in autumn and winter formed a different trend. In summer, Thursday had the highest mean values: +13.64% for PM₁₀, + 11.91% for PM₂.₅, and + 13.69% for PM₁. Saturday and Sunday had values below the mean, but the Sunday reductions were about 10% for PM₁ and PM₂.₅ and 14.35 for PM₁₀. Mondays showed a substantial reduction of PM₁ and PM₂.₅, but the coarse fraction concentration was consistent with the global mean. In spring, Thursdays and Fridays had increases compared to the mean of more than 15%. On the other hand, in addition to weekend days, Mondays and Tuesdays had concentrations 10% lower than the mean. In winter, the concentration peaks of PM₁ and PM₂.₅ occurred on Sunday (+ 12.3% and 9.6%), followed by a linear reduction until Wednesday, when changes of − 8.8% and − 9.5% were recorded. The PM₁₀ fraction had fewer variations during the week, with two days at higher concentration. Similarly, in autumn, the minimum occurred on Tuesday, and Thursday and Friday had the highest values.
Globally, summer and spring had lower values on weekends and higher values on workdays. In autumn and winter, the minimum was in the middle of the week, and the highest values occurred on the weekend. As in Fig. 8, similar trends were observed in autumn and winter and in spring and summer. A similar trend result was obtained by (Zhao et al., 2018). Furthermore, (Xue et al., 2020) found the same result with PM0.1, the ultrafine fraction of particulate matter.
3.3 Source direction
According (Tiwari et al., 2017), assessing the PM concentration with wind direction highlights the contributions of local and global emissions such as combustion from industries, biofuel burning, vehicular emissions, and dust transportation along plains. To do this, statistical tools such as the conditional probability function (CPF) and bivariate conditional probability function (CPBF) are used. Such models were introduced by (Ashbaugh et al., 1985; Kim et al., 2003) and applied in different contexts by (Heo et al., 2009; Squizzato and Masiol, 2015; Tiwari et al., 2014). With the CPF tool, the probability of exceeding a limit value is evaluated for each direction to identify preferential transport directions. The statistical model is based on the following formula:
$$CP{F}_{\varDelta \theta }=\frac{{m}_{\varDelta \theta |c\ge x }}{{n}_{\varDelta \theta }}$$
1
According to this formula, for each angular sector Δθ, the CPF is equal to the ratio between the occurrences m of concentration greater than a limit value x and the number of overall values in the interval n. From a methodological point of view, 16 angular sectors with an amplitude of 22.5° were selected. For the calculation of probabilities, all concentration values corresponding to a wind speed lower than 0.5 m/s were excluded (Tiwari et al., 2017). This was done because low wind speeds typically have isotropic characteristics of direction (Ashbaugh et al., 1985). The selected concentration limit value corresponded to the 75th percentile: 21.8, 26.0, and 36.6 µg/m³ for PM₁, PM₂.₅, and PM₁₀, respectively.
The results are shown in Fig. 11. As shown, similar configurations were obtained for all three particulate classes. There is a low probability of concentrations greater than the 75th percentile in the N–NE–E angular sectors. High probabilities occur in the opposite S–SW–W angular sectors. This result has particular interest as these are typical directions of for Saharan particle transport. For high concentration values, there are no significant influences in the direction of the Po Valley (NE–E). This confirms that the phenomenon of major pollution due to subsidence is homogeneous and involves the entire Po Valley (Arvani et al., 2016; Diémoz et al., 2019).
The CBPF was applied as an implementation of the CPF (Jain et al., 2020; Tiwari et al., 2017). In this configuration, wind speed is added to the system as an additional variable. The probabilistic relationships combined with wind speed and direction are then utilized to deepen understanding of the spatial distribution of sources. The relationship for determining CBPF is as follows.
$$CBP{F}_{\varDelta \theta , \varDelta u}=\frac{{m}_{\varDelta \theta , \varDelta u |y\ge c\ge x }}{{n}_{\varDelta \theta ,\varDelta u}}$$
2
For each combination of Δθ and Δu intervals, the CBPF is equal to the ratio between the occurrences m of the concentration between the y and x limit values and the overall values n in the interval.
As suggested by (Rai et al., 2016; Uria-Tellaetxe and Carslaw, 2014), four concentration ranges (the four main quartiles: 1–25%, 25–50%, 50–75%, and 75–99%) were selected for PM₁, PM₂.₅, and PM₁₀. To exclude outliers, the extreme percentiles were not considered in the analysis (Uria-Tellaetxe and Carslaw, 2014).
The graphs in Fig. 12 show the combinations of wind speed and direction at the measurement site. The color intensity indicates the concentration recurrence probability for each selected concentration range. As shown, there were significantly different results for the three analyzed particulate fractions. The finer fractions (PM₁ and PM₂.₅) showed a prevalence of low concentrations (first quartile) when the wind blows from the east with high intensity, as in the 4–6 m/s range. This phenomenon could be attributed to the wind dilution of the PM concentrations in the Po Valley (Diémoz et al., 2019), which is in an easterly direction with respect to the measurement point. On the other hand, the PM₁₀ prevalent in the first quartile occurred when the wind was from the NW–W with varying speeds (2–8 m/s). For the upper quartiles, the highest probabilities were gathered around the origin, at low speeds (< 2 m/s) and with a slight prevalence of NE direction.
For the last two quartiles, there was a slight probability for the NE–SW line for PM₂.₅ and PM₁₀. These directions represent the two major openings of the city of Turin toward the Po Valley. In fact, in the E–SE direction, the city is separated by a hill; on the opposite side, it is surrounded by the Alps. The S–SW direction is therefore the main arrival direction of particulates during Saharan events (fourth quartile) (Diémoz et al., 2021; Tositti et al., 2014). This also confirms the result illustrated in Fig. 11 regarding the preponderance of concentrations in the fourth quartile in the S–SW direction.