Introduction

Port activities have a major impact on the surrounding urban areas. In terms of emissions, naval traffic makes a non-negligible contribution to municipal emissions both for the main pollutants (NOx, SOx, PM and NMVOC) and for greenhouse gases (CO2). In addition to maneuvers within the port area, substantial contributions derive from the housing of ships at the docks in most cases with auxiliary engines running.

In this work, an estimate of the spatial distribution of NO2 is presented starting from data measured with passive samplers in 10 locations in the municipality of Livorno in Tuscany (Italy). The measurements were carried out as part of the Aernostrum project on air quality in ports and at the port-city interface. Aernostrum is an Interreg project that involved several Mediterranean ports to estimate the impacts of port activity and identify possible mitigation measures [Aernostrum Interreg Maritime n.d.]. The data collected with the passive samplers were integrated with those from the fixed air quality stations and mobile measurements carried out within the port, as part of the project. The data was processed to produce a map of the spatial distribution of NO2 in a domain within the municipality, to represent the contribution of the port to the interface with the city. For each site a ß index (Janssen et al. 2008), (Janssen et al. 2012), (Piersanti et al. 2013) was calculated; this index is used to remove at each point the trend due to the local contributions of the pressure sources in order to carry out a Ordinary Kriging on the detrended data and spatialize the measured data. This work was focused on defining the appropriate factors to build the beta index for the domain and context studied. In particular, specific first guess coefficients have been developed for traffic, for port and industrial emissions and also for urban areas. In particular, according to the optimization results, it was evaluated how to manage the traffic contribution; the optimization is the process that linearizes the relationship between the beta index and the measured levels by acting on the first guess coefficients, and comparison between first guess coefficients and optimized coefficients also gives important information.

Materials and methods

Beta index

The beta index is a parameter that can be calculated at each point of the domain and associated with each monitoring site. It represents the characteristics of a site by summarizing the information on land use and local emission pressures in a single value.

The beta index was proposed and used in Belgium for the evaluation of the spatial representativeness of the stations (Janssen et al. 2012), and for the interpolation of air quality data (Janssen et al. 2008). It is calculated in a buffer around the measurement site according to the formula:

$$\beta ={\text{log}}\left[1+\frac{\sum\limits_{i}{a}_{i}\bullet {n}_{RCLi}}{\sum\limits_{i}{n}_{RCLi}}\right]$$
nRCLi:

number of pixels (or area) in the i-th Corine Land Cover class.

ai:

coefficient related to emissions for the i-th class, used to weigh the contribution of a particular class of land use on the concentrations of pollutants in the air.

The radius of the vicinity buffer is a free parameter in the methodology, but for Janssen et al. 2008, it was adopted a radius of 2 km. The radius chosen can depend also on the type of pollutant considered and on the variability of land use in the area of the study. When the buffer is chosen too small, it becomes difficult to have more than one of the CLC classes included in the buffer (essential for a good optimization), and in this way it is also considered a very local contribution. When the buffer is too large, the site-specific character of the CLC class distribution disappears and the distribution evolves to a general spectrum with very little discriminating power. In this study as better explained in the next section a buffer radius of 1 km was chosen.

The logarithm function, given by the methodology, is related to the fact that the beta index is a synthetic index of pressures on air quality for which a linear relationship with air quality data is assumed.

The beta index is then specific for each pollutant of interest. In order to establish a relationship between the polluting potential and the land use classes, the CLC classes are grouped into 11 categories and then combined with the EMEP sectors of the activities that produce emissions into the atmosphere, same approach used in Janssen et al. (2008). In the Tuscany Region the beta index was previously applied to estimate the representativeness of the regional network stations, limited to PM10 background stations (Regione Toscana Rappresentatività 2015), as indicated in the methodology proposed at national level by ENEA (Piersanti et al. 2013). Coefficients optimization is made with the R software [R Core Team 2022] and in particular the optim routine, using the L-BFGS-B methodology (Byrd et al. 1995). This methodology allows you to define appropriate variability constraints (maximum and minimum value) for each of the independent variables of the optimization process. The constraint imposed for this study is that the coefficients, representing an emission pressure associated with a given land use, are all positive.

NO2 measure

The measurements with passive samplers were made with Radiello® Rad66 for NOx and SOx, with 15 days sampling and according to the UNI EN16339 method. The sampling period was repeated four times during the year covering different seasons and for this work the average of all sampling was used as an estimate of the annual average in each location. The NOx data from the fixed and mobile stations are measured by Teledyne API T200, Teledyne API 200A or Thermo Electron 42 analysers, in compliance with the technical standard UNI EN 14211 and D.Lgs 155/2010, which in Italy is the transposition of the European Directive for air quality. The sampling periods are shown in Table 1, while the sampling sites are shown in Fig. 1.

Table 1 Sampling periods
Full size table
Fig. 1
figure 1

Sampling sites

Full size image

Calculating the composite uncertainty with the maximum uncertainties for each step of the method provided by the EN 16339:13, we obtain for the results of the passive samplers an expanded uncertainty of about 30%. Continuous analyzers used for monitoring in this study comply with UNI EN 14211:2012 as indicated by Legislative Decree 155/2010, and thus have expanded uncertainty of less than 15 percent. To verify the comparability between the two methods, passive sampling was performed in parallel with Teledyne at the LI-La Pira site and, limited to the August campaign, at the Bengasi site. The results are shown in Table 2.

Table 2 Comparison between passive samplers and continuous sampling
Full size table

Representativeness on an annual basis is ensured by coverage of different seasonal periods as required by Legislative Decree 155/2010 which, for indicative campaigns, prescribes that the measurement be carried out on variable days of the week and distributed throughout the year for a total of eight weeks. In addition, to verify representativeness on an annual basis, the results of the radiello on Li La Pira were compared with those of the regional LI-La Pira station, which has full annual coverage. The results are shown in Table 3.

Table 3 Comparison on annual mean basis
Full size table

To verify that sampling periods are representative also from a meteorological point of view the wind regime was examined, and wind roses for each of the periods in Table 3 are shown in Fig. 2. The city of Livorno is subject to breeze regimes and the most frequent winds are those on the northeast/southwest axis that are also the directions of strongest winds.

Fig. 2
figure 2

Wind rose comparison in 2021 (a), 2022 (b) and in the sampling periods (c)

Full size image

The wind profile results similar in all periods.

The domain is a rectangle of dimensions 10 × 6,7 km containing the urban area of the Municipality of Livorno, the port area and the industrial area of Livorno Nord, extended to the municipality of Collesalvetti; the domain also contains the Calambrone area, a fraction of the Municipality of Pisa, which is a natural area and constitutes an impact zone for both the industrial area and the port. All cartographic processing were carried out with QGis (3.26 Firenze) (QGis Project n.d.). For the Ordinary Kriging geoprocessing was made using the plugin Smart Map (Pereira et al. 2022).

The emission data, classified according to SNAP97 codes, are extracted from the IRSE Regional Inventory of Emission Sources of the Tuscany Region for the year 2017 (IRSE Database 2017).

Additional traffic data were downloaded from Open Street Maps [Open Street Maps Project (n.d.)], available with Open Database License, for the study domain.

Data of Land Use are from Corine Land Cover 2018 (Corine Land Cover n.d.).

Results and discussion

On the 13 points considered, the distribution of land use over a 1 km buffer was examined. The radius, smaller than that used for PM10, is motivated both by the lower spatial representativeness that can be expected for NO2 and by the dense measurement grid with a pitch of approximately 1 km.

Before choosing the buffer radius, several possibilities were explored (Fig. 3a, b, c).

Fig. 3
figure 3

Buffer radius 500 m (a), 1000 m (b), 2000 m (c)

Full size image

RIO classes are highly aggregated so land use in many buffers with a radius of 500 m would be of a single type, thus not allowing a good optimization of the relationship between beta and NO2 value. A 2-km buffer on the chosen domain, on the contrary, does not seem suitable to adequately represent a contribution to beta that could differentiate local specificities.

To choose the study domain, the distribution of land use in the municipality of Livorno and neighboring municipalities were examined (Fig. 4a, b and Fig. 5). The study domain was selected to focus on the urban area, excluding all the countryside around the city but including all the industrial areas further north, in the neighboring municipalities of Collesalvetti and Pisa.

Fig. 4
figure 4

Satellite view and Land use view of the Municipality and domain of study

Full size image
Fig. 5
figure 5

Comparison of Land use distribution in the municipality of Livorno (a) and in the Domain of the study (b)

Full size image

The entire domain area includes two land use classes that do not fall into any of the buffers: class 142, urban green, with a percentage of 0.5% and class 132, landfills, with a percentage of 0.3%.

For the purposes of this paper for beta calculation, class 142, where present, is assimilated to RIO CL2 (discontinuous urban fabric) and class 132 is assimilated to RIO CL3 (Industrial or commercial units).

Four sites (Calambrone, Cappiello, Regina and via Toti) have a smaller buffer area because they are less than 1 km from the sea, and the marine body of water is not considered. In a first approximation the marine body is not considered because it is assumed that the sea does not bring an emission contribution on the city. Actually, the sea routes and the roadstead could have a non-negligible contribution but the emission data available at the moment are very uncertain. This issue could be further investigated in the future.

RIO Land use category for each site in a buffer radius of 1 km are shown in Fig. 6. Continuous urban areas are found exclusively in the city center and are in high percentage only in the Province site (63%). Continuous urban areas are also present in the buffers of the sites ARPAT (24%) and Bastia (16%) and in lower percentages in Regina (8%) and Carducci (6%). Apart from the Bengasi site (which is purely port and industrial and where the urban share between continuous and discontinuous does not reach 1%), all the sites have a significant share of discontinuous urban land use.

Fig. 6
figure 6

RIO Land use category for each site in a buffer radius of 1 km

Full size image

The ENI Stagno site, located in the municipality of Collesalvetti in front of the refinery, is characterized by a rather low percentage of urbanization (11%, the lowest after Bengasi) and by the presence of a high percentage of industrial land (14%) surrounded by agricultural areas (permanent crops 30%) and fields (arable land 25%). Road infrastructures has maximum coverage at this site. Land use with road infrastructures is also present in Bastia (2.5%), Carducci (10%), Gemini (1.5%) and La Pira (2.5%), but further investigations have been carried out for this type of land use. The industrial character is prevalent in the sites of Enriques (91%) and La Pira (63%), but is also present in the sites in the northern area of ​​Livorno, Bastia (13%) and Carducci (8%), in ENI-Stagno (14%) and near the port area in the Bengasi site (20%). Only Calambrone contains in its buffer forests and seminatural areas.

A first guess of coefficients is carried out according to the Table 4 of association of SNAP emission categories with Rio land use classes.

Table 4 First guess coefficient composition
Full size table

Emissions from domestic heating and urban (only diffuse) traffic emissions are associated with continuous urban fabric land use. For discontinuous fabric, to consider the lower population density, a coefficient composed of 70% of heating and 50% of urban traffic was considered. The industrial areas include macro-sectors 01, 03, 04 and 09 (excluding landfills and the burning of pruning waste, the latter attributed to agricultural areas).

The entire sector 07 of road transport is attributed to the CL4 code, with some specific modifications illustrated below.

The CL5 code includes all port activities in macro-sector 08 (080401). Maritime routes, fishing and boating are excluded. Codes CL6 and CL7 are not present in our domain.

Agricultural activities, arable land or permanent crops, include the related emissions from macro-sector 10, off-road vehicles in agriculture and the agriculture heating sector (0203). For arable land, the entry relating to the burning of stubble is added, while for permanent crops 0907, burning of prunings and agricultural residues, is added. Wooded land and semi-natural environments are associated with a minimum amount of residential heating (1%) and sector 0807, off-road in forestry. Finally, the humid environments in the study domain represent:

  • the marinas for pleasure boats to which emissions from boating (080405) have been associated;

  • the body of water inside the port of Livorno, which is instead assimilated to the port land area.

In our set of measurements there is an urban traffic station, LI-Carducci, which has among the highest NO2 levels in Tuscany, lower only than those of the FI-Gramsci urban traffic station, as in Annual report in ARPAT website (ARPAT Rapporto 2023). So the traffic contribution was deepen by extracting the roads with traffic volumes greater than 100 vehicle per hour, (trafficvol > 100) from the Open transport map layers falling within our domain. For each buffer around the stations, total kms of roads with trafficvol > 100 and the total volume of traffic from these roads were calculated (Table 5).

Table 5 Roads data relative to the monitoring site
Full size table

The two indicators are almost equivalent except for Carducci and Gemini, for which the value increases significantly if the traffic volume is considered, and for Eni Stagno for which, on the contrary, the value decreases (Fig. 7).

Fig. 7
figure 7

Roads data

Full size image

At this point a factor dependent on the distance of the monitoring site from the road was added.

The length/distance indicator was chosen, having already made a selection on the volume of roads, as only those with volumes greater than 100 vehicles per hour are considered. However, both these indicators seem closer to reality and also more coherent with each other (Fig. 8).

Fig. 8
figure 8

Roads data relative to the monitoring sites distance

Full size image

This indicator could be used in two alternative ways:

  1. 1.

    As a proxy to replace the CL4 variable with a surface proportional to that of the CL4 in the Carducci buffer, i.e.:

    $$\text{Cl}4\_\text{new}(\text{i})=>\mathrm{CL}4\;(\text{c})\ast\lbrack\text{L}(\text{i})/\text{d}(\text{i})\rbrack/\lbrack\text{L}(\text{c})/\text{d}(\text{c})\rbrack$$

  • CL4_new: new area assignment for the i-th site

  • L(i)/d(i) length over distance indicator calculated for the i-th site

  • L(c)/d(c) length over distance indicator calculated for Carducci

  • CL4 (c) RIO CL4 class area in the Carducci station buffer

To keep the total area of the buffer unchanged, the difference between CL4 and CL4_new for each buffer is redistributed (or scaled if negative) in the other RIO land use classes in a proportional way to their presence in the area.

  1. 2.

    After interpolation of background values only, as a proxy to attribute a further contribution to NO2 levels to buffers containing selected roads

In this work the procedure in point 1 was used to calculate the beta indexes for each site with the first guess coefficients. Then the optimization was carried out with and without the Carducci station. Finally, based on the results, the method was chosen.

The coefficients thus attributed, and normalized with respect to the maximum contribution, give a value of the beta index ranging from a minimum of 0.024 for the Calambrone site to a maximum of 0.26 for the Bengasi site.

With the first guess coefficients, the relationship between beta and measured NO2 values gives an R2 of 0.25 if all stations are considered and an R2 of 0.74 if the only traffic station is excluded (Fig. 9).

Fig. 9
figure 9

Relationship between beta and NO2 before optimization with all stations and with only background

Full size image

After the optimization, considering all the stations the R2 goes from 0.25 to 0.74, while with only the background stations it goes from 0.74 to 0.90. Furthermore, by including the traffic station, an important difference is observed between the first guess coefficients and the optimized ones; in fact, to linearize the traffic value, the routine emphasizes the coefficient relating to traffic (Fig. 10).

Fig. 10
figure 10

Optimization results and coefficients comparison with complete dataset and only background

Full size image

For this reason, the traffic station was excluded from the optimization and the contribution of the main roads added a posteriori in a 10 m buffer, as a proportional contribution to the NO2 measured in Carducci based on the relative flows.

The beta values ​​were used to remove the trend from the measured data. For detrending a constant concentration of 70 µg/m3 was taken as reference. Values of NO2 calculated on the basis of beta for all the points of measure were subtracted from this value and then added to measured values. Results are shown in Fig. 11.

Fig. 11
figure 11

Values and detrended values vs ß

Full size image

Normalization with respect to the variance is not considered here because for each point of passive samplers there are only four measure and variances do not show a trend. Detrended values are interpolated using the Smart Map Plugin (Pereira et al. 2022) in QGis with variogram settings suggested by the application, that is a maximum distance of 5064.575 m and a lag of 1695.941 m. The model selected is linear to sill with the sill value of 6.726 µg/m3, very similar to the sample variance. The plugin also provides the Moran index of the set of data which varies from -1 to 1 and is a measure of spatial correlation. A value of 0 would indicate a spatial pattern not unlike a random phenomenon. For the detrended values, the value of the Moran index is 0.136, indicating that most of the variance is removed by detrending data, so most of the variance in our measures seems to be explained by the beta index.

The resulting map is shown in Fig. 12 and the standard deviation estimated is 2.76 µg/m3.

Fig. 12
figure 12

Detrended values spatial interpolation results

Full size image

On the same grid of the interpolation the ß index is calculated. Beta index is here considered as a property of each point of the domain, influenced by the surroundings in a circular buffer of 1 km. The detail of calculation is better explained in Fig. 13.

Fig. 13
figure 13

Calculation of ß on a grid of 100 m x 100 m

Full size image

The shift of the area form one point to the other is given by a 12% variation. For example, see points highlighted in figure. Passing from the left to the right the yellow area (6%) is lost whereas the green area (6%) is gained. In terms of land use the change for the case in the example is given in Fig. 14.

Fig. 14
figure 14

Comparison of Land Use distribution for two next points on a grid of 100 m × 100 m

Full size image

In terms of beta index the variation in this case is from 0,6870 for the left point to 0,6601 for the right point as the effect of substituting harbor areas with urban discontinuous and industrial areas.

Beta distribution on the grid is shown in Fig. 15.

Fig. 15
figure 15

Beta distribution on the domain

Full size image

Once ß is calculated for each point of the grid, the retrending of interpolated values was made (Fig. 16). A comparison was also made between NO2 calculated through the optimized linear relationship between beta and measured concentrations, and NO2 retrended (Fig. 16). The two estimates are very similar and this fact seems to confirm that almost all the variance is removed by the ß. In Fig. 17 results from both methods are shown for measurement points.

Fig. 16
figure 16

NO2 Retrended maps

Full size image
Fig. 17
figure 17

Comparison between the two methods of estimating NO2

Full size image

Map in Fig. 18 shows the background levels estimated for NO2 in the city of Livorno, highlighting the contribution of the port, the port-city interface and peripheric and background values.

Fig. 18
figure 18

Results for data spatialization: the different background levels and traffic along main roads

Full size image

Roads, with a traffic volume over 100 vehicles per hour, are added to complete the representation assuming that street levels can be estimated a posteriori for each arch on the base of their relative volume and data measured in the traffic station Carducci taken as reference.

Conclusions

In this work, NO2 measurements were carried out with passive samplers on a grid of points approximately 1 km apart from each other and with mobile campaigns representative of the calendar year. Some of these points were chosen within the port and others to represent adequately the port-city interface. By integrating the information obtained with the data from the monitoring stations, which represent the urban background and traffic in the city of Livorno, the spatialization of the data was carried out.

After examining various possibilities, it was decided to interpolate only the background data and, once the trend of the local contribution was removed through the use of the beta index, to treat the contribution of road traffic (limited to selected roads) as a linear source capable of influencing a buffer of a few meters around the road edge.

In the process of detrending, interpolating e retrending data, ß index is treated as a property of each point on a grid of 100 m × 100 m. The calculation of ß through his linear relationship with NO2 and the retrended map are very similar. This is probably due to the fact that detrended values vary very slightly in the domain and all the variability of data seem to be explained by beta. Even with few points, but on a well-distributed grid (1 km), for a relatively limited domain area, the ß index presents good linearity with the values of the urban background even in a context where there are numerous important sources of emissions (port and heavy industry, urban), while road traffic remains the most complicated data to manage. Excluding the traffic station and treating the “on-the road” contribution apart could be a good compromise and it is probably the reason for the good relationship between ß and the measured data also before the optimization process.