Monitoring and evaluation of Terni (Central Italy) air quality through spatially resolved analyses

: A study of spatial variability of PM 10 elemental components was conducted in Terni city (Central Italy), situated in an intramountain depression characterized by the presence of several particulate matter emission sources. The meteorological conditions of Terni basin limit the dispersion and enhance the accumulation of the atmospheric pollutants. Thanks to the utilization of new samplers ( Smart Sampler), used for the first time and working in parallel at 23 sampling sites, spatially resolved data were obtained. Localizations of the samplers were chosen in order to evaluate the impact of different local PM 10 sources. Chemical composition of the samples was determined in combination with a chemical fractioning procedure, that allowed us to discriminate water-soluble and residual fractions of analyzed elements in which proved to be a valuable approach for increasing selectivity of elements as source tracers. Spatial variability of elements underlined the contribution of local emission sources and the different dispersion capacity of each element. Terni city resulted to be an ideal area to test and validate a new experimental method for the acquisition of spatially resolved data providing the possibility to properly evaluate the spatial variability of PM 10 and its chemical components.


Introduction
Particulate matter air pollution is an air-suspended mixture of solid and liquid particles that vary in size, shape, chemical composition, solubility, and origin [1]. Its direct and indirect effects on human health and on the environment have led to numerous studies focusing upon its complex composition and toxicology [2]. One of the most efficient tools for the formulation of better control strategies for the atmospheric pollutant is the PM source apportionment, relevant to identify the role of different particulate matter emission sources. Typically, in air pollution control, the dispersion of air pollutants released from different emission sources are estimated through atmospheric dispersion mathematical models, without using wide and expensive monitoring networks. However it is important to consider the limitations of mathematical models [3][4]. Reducible error results from inadequate air quality data inputs and from difficulties of models to describe such a complex issue like PM dispersion, due the large number of different sources and the complexity of PM transport and transformation processes [5][6]. The space distribution of particulate matter, in fact, strictly depends on factors such as the location and emission rate of its sources and the dispersion capacity of each single contribute, which varies as a function of the particle dimension, shape, chemical composition and density [7]. The purpose of this study is to elaborate a new experimental procedure for source apportionment studies, exceeding the limits of modeling approach and the high costs associated to the realization of an extended monitoring network. The site of the study is Terni city, one of the most polluted urban and industrial site in Central Italy [8]. Terni is situated in an intramountain depression, delimited by the Apennine mountain range. This area is characterized by the presence of typical urban PM emission sources such as vehicular traffic, domestic heating and industrial emission sources such as a power plant for waste treatment and a steel plant [9]. Peculiar geomorphological and meteorological conditions of the Terni basin limit the dispersion and enhance the accumulation of the atmospheric pollutants. For these reasons Terni city results to be an ideal area to test and validate a new experimental method for the evaluation of the spatial variability of PM10 and its chemical components through the acquisition of spatially resolved data. Thanks to the utilization of innovative, low cost, automatic and self-powered samplers (named Smart Sampler), a wide monitoring network was designed across Terni in order to represent the contributions to the total PM10 of each local emission source. The present study represents the first application on field of the Smart Sampler and it is also proposed to contribute to their validation and optimization through a real monitoring campaign. The aim of the study was the evaluation of the spatial variability of the chemical components of PM10, some of which can be used as efficient source tracers. For increasing selectivity of the elements as source tracers, a chemical fractionation procedure, previously optimized and validated [10][11][12], was applied to the collected samples. Chemical fractioning based on elemental solubility, allows to get information on the chemical form in which the element is released and this may be typical of its emission source [13]. Moreover the use of the chemical fractioning yields useful information for an estimate of the environmental mobility of the elements. Biogeochemical distribution and bioaccessibility, health and environmental effects are in fact strongly related to the chemical form of the elements [14][15][16]. The use of the Smart Sampler, in combination with the chemical characterization and fractioning procedure applied on the samples, allowed to analyze the spatial variability of the elemental components of PM10 in Terni. Through the acquisition of data with high spatial resolution, we could identify the source profiles of the different local emission sources present in the area and experimentally evaluate the dispersion capacities of each analyzed elemental component.

Sampling sites and sampling equipment
The first phase of this study required great commitment and large amount of time for the identification and the choice of the best locations for the installation of the 23 sites. At first, in the selection of locations, a particulate matter dispersion grid already developed by the regional agency for environmental protection (ARPA Umbria) was used as a reference system (Fig. 1). It was also very important to guarantee that each sampler was positioned at the same height (within 4 meters) and with the same exposition (southern exposition) to ensure homogeneous sampling conditions. Particular attention was also paied to the representativness of sites with respect to each local emission source of PM: power plant for waste treatment located in the West of the city; vehicular traffic due to the presence of streets with heavy traffic; domestic heating; very extensive steel plant in the East of the city. Tab. 1 reports the geographical coordinates of the 23 selected sites. BR, AR (North of the city), PI and LG (South of the city) are sites located just outside the high density urban areas; MA, RI, GI, FR, CR, HG, FA and CB are situated in the West of the city, in an industrial area near the power plant for waste treatment (Terni En.A.) where domestic biomass heating systems are used; RO, PR, OB and CP are industrial sites close to (RO and PR) and in proximity (OB and CP) of the steel plant, in the East of the city (Terni AST); CO is a site located between the steel plant and the city centre; CZ, HV, SA, UC, CA and PV are sites situated in high density urban areas and/or near heavy traffic streets.  In these 23 selected sites, Smart Sampler equipped with PM10 sampling head were installed ( Smart Samplers are low cost, automatic, self-powered with a photovoltaic panel and characterized by a low flow (0,5 l/min). Due to low flow condition data obtained during this monitoring campaign had a monthly time resolution with a substantial reduction of associated costs. Samples of PM10 were collected by using Teflon membrane filters (PTFE membranes, 37 mm diameter, 2 µm pore size, PALL Corporation, U.S.A). All the membrane filters were weighed before and after the sampling, in order to gravimetrically determine the PM10 mass concentration. Smart Sampler are equipped with a SD card from which it was possible to download the volumes of sampled air (m 3 ), used for the calculation of the mass concentration of PM10 and its elemental components. The lack of PM10 mass concentration values and elemental concentration values for some collection sites is due to the progressive extension of the monitoring network during the monitoring campaign and sometimes to the malfunction of some samplers. The monitoring campaign was carried out from December 2016 to April 2017.

Analytical procedure
The analytical procedure for the elemental chemical fractionation, previously optimized and validated [10][11][12], was carried on each PTFE membrane collected during the campaign. This procedure allows to discriminate watersoluble from insoluble fraction of each analyzed element and resulted to be useful in increasing the selectivity of each element as source tracer [10]. First, the supporting polymethylpentene rings were removed from each sampled PTFE membrane. During the first phase of the chemical fractionation procedure, the sampled PTFE membranes were extracted in ultrasonic bath in 10 mL of deionized water (produced by Arioso UP 900 Integrate Water Purification System) for 30 minutes. The xtracted solutions were then filtered on a cellulose nitrate filter (Merck Millipore Ltd, 0.45 µm pore size). After the extraction step, both the cellulose nitrate filter and the original PTFE filter were acid-digested in microwave oven (Ethos Touch Control with Q20 rotor, Milestone, Italy) by using 2 mL of HNO3 (67% Promochem) and 1 mL of H2O2 (30% Suprapur, Merck). In the beginning of the digestion programme, the temperature was linearly increased to 100 °C in 5 minutes. In the second step, the temperature was linearly increased to 180 °C in 12 minutes. In the third step, the temperature was kept at 180 °C for 15 minutes. After 30 minutes of ventilation and cooling down, the digested solutions were diluted in 50 mL of deionized water. These solutions were filtered with syringe filters (GVS Filter Technology 25 mm diameter, 0.45 µm pore size, U.K.). The concentrations of 34 elements (Al, As, B, Ba, Bi, Ca, Cd, Ce, Co, Cr, Cs, Cu, Fe, Ga, La, Li, Mg, Mn, Mo, Na, Nb, Ni, Pb, Rb, Sb, Sn, Sr, Ti, Tl, U, V, W, Zn and Zr) were determined in both extract and mineralized residue fractions obtained from each sample by using inductively coupled plasma mass spectrometry (ICP-MS, Bruker 820-MS, U.S.A.).

Statistical analysis
Multivariate statistical computations were performed by the statistical software: R-3.0Gui (32-bit) for Windows (R-project for statistical computing). Multivariate analyses were performed on the data yielded by the analytical determinations on the elemental components of PM10 at each collection site. Two principal component analyses (PCA) were performed; the first one was performed for the 23 monitoring sites by including the averages of the 34 monthly total elemental concentration data (sum of the extracted and residual fractions) recorded at each collection site; the second one was performed for the 23 monitoring sites by including the 34 total elemental concentration data (sum of the extracted and residual fractions) recorded each month at each collection site. The first PCA was performed to separate or group the monitoring sites depending on the type and on the concentration of the source tracers recorded at each site in order to identify the main emission source tracers and their action at each site. The second PCA was performed in order to assess the monthly action of the source tracers at each site and to evaluate the monthly variations of the elemental components of PM10 in the monitored area. Thanks to the choice of the best locations for the 23 Smart Sampler, the 23 monitoring sites are well differentiated for type and location, allowing a detailed discussion about differences in source nature and monthly contributions. In the data set used some outliers were excluded after having assessed that their presence increased unstable results of PCA.

Element solubility percentages
Solubility percentages of PM10 elemental components were calculated by dividing the soluble fraction of each element analyzed to its total fraction (sum of the extracted and residual fractions) and multiplying by 100. The monthly solubility percentage for each element analyzed in the entire monitored area was obtained making each month an average of the solubility percentages calculated for each element at all the 23 monitoring sites. Standard deviations were calculated to evaluate the contribution of local emission sources to the variability of the elements solubility in the entire monitored area.  The data above demonstrates that a clear increase in PM concentration occurred in all the sites during the winter. The highest PM10 mass concentration (96 µg/m 3 ) was recorded in January, during the colder period, at CO located between the heavily trafficked streets of the city and the steel plant. The lowest PM10 mass concentration (6,2 µg/m 3 ) was recorded in April, during the warmer period, at the monitoring station FR located in the West of the city in an industrial area near the power plant, far from the city centre. The data in Fig. 3 show that the concentration of PM10 was driven by the contribution of different PM local emission sources as well as by the meteorological conditions. In general, the PM10 mass concentration reflects the strength of the contribution from traffic-related sources at these sites [10]. In fact the highest values of PM10 were recorded during the winter at sites located in high density urban areas and near heavy traffic streets: CO, CA and UC. The increase of PM10 mass concentrations during the colder months (December and January) can be explained by a combination of two factors: the less efficient mixing of the lower atmosphere during the winter and the strength of some seasonal PM sources, which include domestic heating [14][15]17]. The meteorological conditions of Terni basin limit the dispersion and enhance the accumulation of the atmospheric pollutants. Hence the local circulation is often weak and this leads to a more stable atmosphere which is generally responsible for a concentration increase in all PM chemical species. Particularly during the winter, the meteorological conditions favors the development of temperature inversions, leading to extended periods of high atmospheric stability. The lack of PM10 mass concentration values for some collection sites is due to the progressive extension of the monitoring network during the monitoring campaign and sometimes to the malfunction of some samplers.

Elemental concentration
Thanks to the spatially resolved data obtained by sampling in parallel at 23 collection sites, it was possible to properly evaluate the spatial variability of PM10 as well as of some PM10 chemical components, and in particular of tracers of the principal local sources. For example Fig. 4 reports the monthly values of the total concentrations (sum of the extracted and residual fractions) of Ni, Cu and Rb in PM10 collected from the 23 collection sites during the 5-month monitoring campaign.   4 shows how, also for the elements, the highest elemental concentrations, at all the 23 monitoring sites, were recorded during the colder months, due to the presence of additional PM sources and/or to the lower mixing of the atmosphere [14]. The concentration of most elements increased with the level of urbanization at the collection site and with the proximity of the site to one of the main emission sources of the city: the steel plant. Cr, Mn, Mo showed the same behaviour as Ni (upper panel of Fig. 4) with a relevant increase of concentrations at the sites in proximity of the steel plant: RO, PR, OB, CP and CO. The highest concentrations of Ni were individuated in February at RO: 93 ng/m 3 . It is worth noting that this value is almost five times higher than the European Union (EU) target value (20 ng/m 3 ), calculated on yearly averages (EU 2008/50/CE). Ni, Cr, Mo and Mn were identified as the main source tracers of the steel plant emission source. Cr and Ni are the basic composition of stainless steel which allows them to be the known tracers of the industry. Cr enables the steel to become resistant to chemical oxidation; Ni functions to create properties such as strength, ductility and toughness. Mn and Mo are secondary components of the stainless steel. Mo increases the corrosion resistance of the steel; Mn is used to increase toughness and tensile strength [18]. Even thought the general behaviour of the four tracers was similar, the results showed that Ni and Mo resulted less dispersed than Cr and Mn. The limited dispersion of these elements in the monitored area suggests that particles containing Ni and Mo had a lower diffusion capacity than those containing Cr and Mn. The low spatial dispersion of this steel plant related contribute suggest that, at least in part, mechanical and abrasive processes are more relevant than combustion ones. In fact, it is well known that fine particles, mainly produced by combustive processes, are easily carried by the wind while coarse ones, mainly produced by mechanical and abrasive processes, tend to remain within the local emission area [19]. Each emission source is responsible for the release of particles having different dimensional ranges and containing chemical species having different solubility. Fig. 5 reports the concentration of Ni and Mo measured in their soluble and residual fractions during the month of February; it is possible to observe that the two elements showed quite the same dispersion profiles, but they had a very different solubility distribution: Mo was emitted mainly as soluble species while Ni was almost entirely contained in the residual fraction. Hence it is possible to suppose that the steel plant released course particles cointaining mainly water-soluble species of Mo and insoluble species of Ni. Fe concentration (mainly present in the insoluble fraction) also reflected the contribution from the steel plant, but it showed a different spatial profile respect to Ni, Cr, Mo and Mn. It was identified as a source tracer of both steel plant and vehicular traffic emission sources; concentration peaks of Fe (>1000 ng/m 3 ) in PM10 were individuated during the winter at CA which is in proximity of heavy traffic streets, at sites close to the steel plant (RO and PR); at CO, which is located between the steel plant and the high density urban area and at sites in proximity of the railway (GI, CR and HG). As it is possible to see from the second panel of Fig. 4, Cu resulted to be a more representative traffic-related source tracer. It showed concentration peaks in December only at sites located in proximity of heavy traffic streets (CA, SA and CO) and at sites in proximity of the railway (GI, CR and HG). Cu, Fe and Sb showed a similar distribution pattern at sites HG, CA and CO in which their concentrations are relatively high because they are in proximity of the most trafficked roads of the city.
Differently to what was previously described for the steel plant tracers, for Cu and Sb the differences in terms of concentration between the monitoring sites resulted less marked, highlighting the role of vehicular traffic as the most common and distributed emission source of the city. The concentration of Pb demonstrated a consistent distribution throughout the city. Although it is known that Pb is mainly emitted by fuel combustion, at the site PR, close to the steel industry, the Pb concentrations resulted especially high. This suggests that Pb must also be emitted by the steel plant possibly when it is added to carbon-steels in order to improve their machinability. Finally, as it is possible to note from the third panel of Fig. 4, Rb was emitted in higher concentrations during the winter months of December and January at sites in the West of the city (left part of the graph). Rb (main in the water-soluble fraction) was emitted in highest concentrations in December at sites RI and FR which are located in an industrial area in proximity of the power plant where domestic biomass heating systems are used. Hence, Rb (water-soluble fraction) was identified as the most representative source tracer of biomass burning. It is worth noting that Rb concentration during the month of December also reflected the contribution of New Year's fireworks, togheter with Bi, Sr and Mg. The analyses of the spatial variability in concentrations of other elements such as V did not reveal significant differences between the sites, without defining clear source profiles. The spatial homogeneity of V in the PM10 within the city is due to the lack of influence of local sources responsible for its emission. In summary, three main important source profiles were identified from East to West of the city: steel plant: Ni, Cr, Fe (principal components of the steel), Mo, Mn (secondary components of the steel); vehicular traffic: Fe, Cu, (from mechanical abrasion of train and vehicle brakes), Sb (product of motor combustion); biomass burning: Rb (from domestic biomass heating systems and power plant combustion processes).

Source tracers identification by principal component analysis
Results from the two principal component analyses performed in this study were evaluated and reported in Fig.  6; all the total concentrations of the elemental components of PM10 recorded were considered to identify the source tracers. The monitoring sites were represented in the left panels of Fig. 6 with different colors depending on the type and location of each site: -  An overview of all sources affecting the 23 monitoring sites used for the multivariate statistical analysis was obtained by performing a first preliminary PCA (upper panels of Fig. 6) including the averages of all the 34 monthly total elemental concentration data (sum of the extracted and residual fractions) recorded at each collection site. Thanks to the first PCA it was possible to separate or group the monitoring sites (scores) in the score plot depending on the type and on the concentration of the source tracers (loading) recorded at each site (upper panels of Fig. 6). Hence, it was possible to properly identify the main source tracers (loadings) of the monitored area and their action at each site. The 63,1% of the total variance was explained by the first two principal components (PC1 and PC2). The sites located outside the high density urban areas (green color), were positioned in the center of the score plot (left upper panel of Fig. 6) because at these monitoring sites, lower concentrations of all the elemental components of PM10 were recorded. The sites in the industrial area near the power plant (blue color), were separated from these on the PC2 by Zr, Sb, Bi, Tl, Sn, Ba and Rb, which were classified as biomass burning source tracers. The sites within the high density urban areas (brown color) resulted well distributed on the two principal components confirming the role of vehicular traffic as most common and distributed emission source of the city. The sites CZ, PV and HV which are located in the city centre but quite distant from the heavy traffic streets, were separated from UC, SA and CA by Sb, Cd, Fe, Cu, Ba and Zn which are typical source tracers of vehicular traffic, by Rb, tracer of domestic heating and by Bi and Sr which are fireworks burning source tracers. The sites at the East of the city, closer to the steel plant, were separated from all the other monitoring sites on the PC2 by Ni, Cr, Mo and Mn which were already identified as steel plant source tracers and by Ti, Nb, W, Co, Mg, Pb. It is interesting to note that PR was separated from RO on the PC1 by Pb, Mg, Mn, and Ti. This suggests the presence of different steel plant local emission sources. The sites in proximity of the steel plant or between the steel plant and the city centre (black color) were positioned exactly between the sites of the high density urban areas and the sites closer to the steel plant, confirming the action of both vehicular traffic and steel plant emission source at site CO. V did not reveal significant differences between the sites confirming the spatial homogeneity of V in the PM10 within the city. From the second PCA performed, it was possible to assess the monthly action of the source tracers at each monitoring site and to evaluate the monthly variations of the elemental components of PM10 at each collection site. As it can be seen from the lower panels of Fig. 6, the first two components explained the 60,6% of the total variance. The number associated to the name of the monitoring sites in the left lower panel of Fig. 6 correspond to the month of sampling at each site: December: 12, January: 1, February: 2, March: 3, April: 4. The second PCA confirmed the progressively decrease of PM10 elemental components' concentrations at each site from December to April, due to the presence of additional PM sources and/or to the lower mixing of the atmosphere during the colder months. The months of sampling at each collection site were separated on the PC1 from the right part to the left part of the score plot, following the monthly sequence of sampling, from December to April. The different types of monitoring sites were maintained separated on the PC2 confirming the efficiency of the source tracers identified from the first PCA performed.

Seasonal variability of element solubility percentages in PM10
The monthly solubility percentages obtained for each element analyzed in the entire monitored area (Tab. 2) allowed to evaluate the seasonal variability of the element soluble fractions in PM10 and to estimate the environmental mobility and the bio-accessibility of each element [14]. From the standard deviations calculated it was also possible to evaluate the contribution of local sources to the variability of the elements solubility in the entire monitored area. Cd, B, Rb, Sr, Mo, Cs and Tl showed the highest solubility percentages and bio-accessibility during the 5-month monitoring campaign while Al, Ti, Cr, Fe, Ni, Zr, Nb, Sn, La, Ce, Pb and Bi resulted released mostly in the residual fraction. The solubility percentages of the elements in PM10 did not show a clear seasonal variability. From December to March the elements solubility percentages resulted constant, while a slight, and in some cases, considerable increase was recorded for the majority of the elements in the month of April. This increase might be due to the variation of some local emission surces which released particles containg mostly insoluble chemical species.  Table 2. Means and standard deviations of the elements solubility percentages (%) calculated monthly for each element at all the 23 monitoring sites.

Conclusions
The results obtained proved the efficiency of the new experimental procedure elaborated for the evaluation of the spatial variability of PM10 and its chemical components through the acquisition of spatially resolved data. In particular the innovative Smart Sampler, used for the first time in this campaign, allowed to build an extended monitoring network, with low associated costs, which was able to represent the different emission source contributes to total PM10 in the monitored area. From the acquisition of spatially resolved data, it was possible to properly evaluate the dispersion capacity of PM10 and its elemental components. The elemental analyses associated with the chemical fractionation procedure and the statistical treatment of the data allowed to identify the source tracers of each PM10 local emission source which will be used for future source apportionment studies. Chemical fractionation, by providing information on the chemical form, improved the selectivity of different source tracers. Spatial variability of Ni, Cr, Mn (insoluble fraction) and Mo (water-soluble fraction) concentrations showed the steel plant role in the emission of PM10. Spatial variability of Fe (insoluble fraction) resulted to be correlated not only to the steel plant emission but also to the influence of vehicular traffic sources. The role of this emission source was also confirmed by the spatial variability of element such as Sb and Cu, which showed high concentrations at the sites near high-traffic streets and the railway. Rb (soluble fraction) was confirmed to be a good tracer of biomass combustion processes. This experimental approach, based on the acquisition of spatially resolved data for analyzing spatial variability of PM10 and its chemical components, will be effective for the validation of dispersion models without the high costs associated to an air quality monitoring network.