Modelling and Forecast of Air Pollution Concentrations during COVID

Pandemic Emergency with ARIMA Techniques: the Case Study of Two

Italian Cities

D. ROSSI1, A. MASCOLO1, S. MANCINI2, J. G. CERON BRETON3, R. M. CERON BRETON3,

C. GUARNACCIA1

1Department of Civil Engineering, University of Salerno, via Giovanni Paolo II 132, 84084, Fisciano,

SA, ITALY

2Department of Information and Electric Engineering and Applied Mathematics, University of

Salerno, via Giovanni Paolo II 132, 84084, Fisciano, SA, ITALY

3Chemistry Faculty, Universidad Autónoma del Carmen, Calle 56 n. 4, Col. Benito Juárez, C.P.

24180, Ciudad del Carmen, Campeche, MEXICO

Abstract:- An efficient and punctual monitoring of air pollutants is very useful to evaluate and prevent possible

threats to human beings’ health. Especially in areas where such pollutants are highly concentrated, an accurate

collection of data could suggest mitigation actions to be implemented. Moreover, a well-performed data

collection could also permit the forecast of future scenarios, in relation to the seasonality of the phenomenon.

With a particular focus on COVID pandemic period, several literature works demonstrated a decreasing of

pollutant concentrations in air of urban areas, mainly for NOx, while CO and PM10, on the opposite, has been

observed to remain still, mainly because of the intensive usage of heating systems by the people forced to stay

home (on specific regions). With the present contribution the authors here present an application of Time Series

analysis (TSA) approach to pollutants concentration data of two Italian cities during first lockdown (9 march –

18 may 2020), demonstrating the possibility to predict pollutants concentration over time.

Key-Words: - COVID pandemic, Air pollutant, Environmental monitoring, Time Series modelling, ARIMA

Received: December 19, 2022. Revised: January 31, 2023. Accepted: February 13, 2023. Published: February 24, 2023

1 Introduction

Among all the environmental hazards, air pollutants

are the most dangerous, and represent a serious

threat to people’s health, especially in areas where

such pollutants are highly concentrated, [1].

Constantly high levels of pollutants can lead, in fact,

to severe cardiovascular and respiratory problems

and mortality, both in short-mid then in long term,

[2]. To preserve inhabitants’ health, it is then

mandatory to implement large, effective and prompt

monitoring networks to control and register

pollutants’ concentration over time, [3]. With an

accurate data collection it is possible to calibrate and

validate models able to predict pollution severity

and immediately alert the population, take actions

and controls on pollutants sources, and track

changes in relation to the seasonality, [4], [5], [6]. A

prompt identification of pollutants’ concentration is,

on the other side, crucial for whatever effective

mitigation action to be implemented, [7].

Many scientific works have covered the subject,

and lately literature has focused the attention on the

COVID pandemic period, [8], [9], observing a

general decrease of pollutants in the air of urban

areas, mainly for NOx. CO and PM10, on the

opposite, have been observed to remain still, mainly

because of the intensive usage of heating systems of

people forced to stay home (in specific regions).

Nevertheless, the mentioned studies offer a deep and

yet retrospective statistical analysis of the

phenomenon. Other noticeable works present the

application of ARIMA models and other deep

learning models to implement a pollutant prevision

in Bangladesh, [10], and in Turkey, [11].

With the present contribution, the authors

propose the results obtained by applying the “Time

Series Analysis” (TSA) approach to pollutants

concentration data of two Italian cities during the

first lockdown (9 March – 18 May 2020), when we

observed an unpredictable situation regarding air

pollutants.

The aim of this study is twofold: on one hand, to

compare the observed concentrations of pollutants

with respect to concentrations measured in the same

months of 2018 and 2019; on the other hand, to

provide a reliable model to forecast the pollutant in

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.13

D. Rossi, A. Mascolo, S. Mancini,

J. G. Ceron Breton, R. M. Ceron Breton,

C. Guarnaccia

E-ISSN: 2224-3496

151

Volume 19, 2023

urban area by analyzing previously measured

concentrations of the same pollutant. The combined

effect of the continuous monitoring and the forecast

could then provide a secure monitoring network, to

profit from the implementation of early decisions

regarding the principal pollutants emitters (cars,

industries). Collected data have been described in

detail, then compared with the values of previous

years and finally used to calibrate a predictive

model. The TSA approach applied on such datasets

is widely documented in literature, [12], [13], [14],

[15], assuring the goodness of the adopted

methodology. A preliminary analysis of the datasets

and the models used in this paper has been

published in [16]. In this paper, the complete

validation of the models, as well as the forecast and

the residuals analysis, will be presented.

2 Material and Methods

Basically, the analysis of a Time Series is the

observation and study of the slope of a selected

variable over time, in terms of trend and seasonal

patterns. Since these techniques can be applied only

to continuous datasets, if any problem occurs during

the measurements, resulting in a hole in the time

series, it’s necessary to impute the missing data. The

imputation can be performed in many ways. In

particular, the most used are related to regression

techniques or modelling imputation, as reported in

[17]. Whereas the variable is the only one present,

the time series is called univariate, and the most

used approaches are based on a deterministic

decomposition or on Auto-Regressive Integrated

Moving-Average (ARIMA) procedures, [18], [19].

The ARIMA(p,d,q) general formula is reported in

equation 1:

𝜙𝑝(𝐵)(1 − 𝐵)𝑑 𝑌

𝑡= 𝜃𝑞(𝐵) 𝑒𝑡

(1)

where Yt is the observed variable, B the delay

operator, 𝜙p the autoregressive polynomials, θq the

moving average polynomials, et the residual

(difference between the observed values and the

predicted ones at time t). p, d and q are the model

hyperparameters, being respectively the

autoregression, differentiation and moving average

orders.

In this paper, the calibration and validation of

two ARIMA models applied on air pollutants

concentrations data is presented. Hyperparameters

and coefficient estimation has been performed with

“R” software, [20], [21], by means of Akaike

Information Criterion (AIC) and Bayesian

Information Criterion (BIC) criteria optimization.

These criteria optimize the balance between

likelihood maximization and number of parameters

minimization, in order to fulfil the parsimony

principle. The AIC and BIC are defined respectively

in equations 2 and 3:

𝐴𝐼𝐶 = −2 ln(𝐿)+ 2(𝑘)

(2)

𝐵𝐼𝐶 = −2 ln(𝐿)+ln (𝑛) ∙ (𝑘)

(3)

where L is the likelihood function, k is the number

of estimated parameters in the model and n is the

sample size.

3 Case Studies and Dataset

Presentation

The case studies presented in this paper are the two

Italian cities of Nocera Inferiore and Solofra, both of

them in the Campania region (Fig. 1). Data used in

the application was obtained from two fixed

monitoring stations settled and maintained by

ARPAC (Agenzia Regionale per la Protezione

dell’Ambiente Campania, i.e. Regional Agency for

Environmental Protection in Campania) - which is

the regional agency taking care of environmental

protection. The pictures of external and internal

views of the stations are reported in Figure 2.

ARPAC recently announced that the

dissemination of air quality data, in the form of

daily bulletins, has been resumed on the agency

website. The publication of the bulletin, in its

traditional form, had been suspended following the

hacker attack that hit the Agency's servers in August

2022. However, a periodic summary of the data was

published uninterruptedly, and the monitoring

stations continued to operate, without any loss of

data, even during the months when the daily bulletin

was not published.

Fig. 1: Sites location highlighting the Campania

region (in red) and the provinces, [16].

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.13

D. Rossi, A. Mascolo, S. Mancini,

J. G. Ceron Breton, R. M. Ceron Breton,

C. Guarnaccia

E-ISSN: 2224-3496

152

Volume 19, 2023

Fig. 2: External, [22], and internal, [23], views of

the monitoring stations installed by ARPAC.

The two selected monitoring stations

continuously record levels of Benzene, NO2, SO2,

PM10, PM2.5 and CO, together with values of

humidity and temperature.

The first monitoring station is positioned in

Solofra, in the province of Avellino, which is a city

having a long history of leather production and

tanning, with many industries present and active. As

known, leather production and processing,

especially tanning processes, involve a massive

production of pollutants: Volatile Organic

Compounds (VOCs), Particulate Matter (PM),

Hydrogen sulphide (H2S) - responsible for the

peculiar bad smell. On the other hand, such

industries also engender CO and NOx, and that’s

because of the large amount of hot water needed to

ensure the tanning procedure. Finally, pollutants

coming from industrial sources must be obviously

summed to the ones coming from the surrounding

sources – private properties, other factories, road

traffic.

The monitoring station of Nocera Inferiore is, on

the opposite, situated near to a highway and some

city roads having high car presence, therefore

primarily collecting road traffic pollution levels. In

such regard, it is worthy to point out that Nocera

Inferiore is one of the most polluted cities of the

Campania Region: in 2020, for instance, PM10

concentrations exceeded the allowed threshold 67

times over the 35 permitted by law, [24]. The

monitoring station is positioned on a residential

downtown, also having many houses and buildings

in the surroundings.

From both the monitoring stations authors have

collected, for the presented analysis, PM10 and CO

daily concentrations, in a timespan going from

February to May 2020.

According to the study on the air quality over the

time span 2015-2021, provided by ARPAC in 2022,

[25], in Campania region PM10 concentrations are

mainly due to non-industrial combustion plants that

contribute more than 67% in 2016. Transport roads

account for about 13% of PM10 emissions. The

agriculture sector is responsible for more than 9% of

emissions and Industrial processes without

combustion for about 4%. A non-negligible

contribution comes from forest fires, with a 3%

share.

As for CO, the same document, [25], reports that

the main carbon monoxide emissions in Campania

region are from vehicle exhausts, while other

emission sources are heating systems and industrial

processes. However, the continuous development of

the technologies has made it possible to minimise

the presence of this pollutant in the air. In 2016,

emissions of CO were mainly due to the road

transport sector for over 48% and non-industrial

combustion plants for about 45%.

At first, the selected datasets for PM10 and CO

levels have been compared with those measured in

the same time span of 2018 and 2019. To visualize

the trends, data were organized using a bar plot after

aggregating them by month of collection (Fig. 3). A

line plot has been subsequently plotted, with

aggregation by week (Fig. 4). In detail, the authors

highlighted in green the period of public restrictions,

going from the 4th of March when public schools

were closed and the first phase of lockdown started,

up to the 17th of May.

For a better comprehension of the succession of

events during the mentioned period, the dates and

description of containment measures imposed by

law on the Italian population during COVID

pandemic burst are reported in Table 1.

In Figures 3 and 4 it can be noticed that, even if

during lockdown road traffic drastically decreased, a

substantial lowering of pollutant concentration has

not been recorded for CO and PM10. The reason is

maybe due to the fact that people forced to stay

home extensively used heating systems, contributing

to a higher level of CO. All the people staying

home, in fact, by using boilers, which are often

based on old functioning systems, with large

emissions and gas consumption, increased the

absolute value of pollution sources.

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.13

D. Rossi, A. Mascolo, S. Mancini,

J. G. Ceron Breton, R. M. Ceron Breton,

C. Guarnaccia

E-ISSN: 2224-3496

153

Volume 19, 2023

Fig. 3: Average monthly levels of PM10 and CO registered in Nocera Inferiore and Solofra from

February to May 2018, 2019 and 2020.

Fig. 4: Average daily levels of PM10 and CO registered in Nocera Inferiore and Solofra from February

to May 2018, 2019 and 2020 with evidence of the different restriction phases. The red dashed line in

PM10 plots is the threshold for daily concentration that the Italian regulation allows to overcome 35

days per year.

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.13

D. Rossi, A. Mascolo, S. Mancini,

J. G. Ceron Breton, R. M. Ceron Breton,

C. Guarnaccia

E-ISSN: 2224-3496

154

Volume 19, 2023

Table 1. Timeline of the Italian governmental restriction from February to May 2020

Phase

Decree of the Government

Adopted containment measures

4th of March 2020

Suspension of all educational activities (all levels schools and universities)

1bis

8th of March 2020

Total lockdown of all the cities

1ter

23rd of March 2020

Closure of all activities excepted essential industrial and commercial ones

17th of May 2020

End of restriction on displacement among cities and regions

During the same periods of 2018 and 2019 large

part of the population was in working places and

schools, which are energetically more efficient,

from the pollution point of view, than private

residential units. As an example, University of

Salerno uses photovoltaic roofs and plants to

produce 30% of the energy needed for daily

activities of the about 1000 professors and

researchers, 500 technicians and administrative

staff, plus all the people working in the lab and the

35000 students, [26].

For the aforementioned reasons the authors

decided to calibrate and test (validate) the chosen

model with the datasets on CO for the city of

Solofra and on PM10 datasets for Nocera Inferiore.

Two different approaches have been used for

imputing missing data: for CO datasets we choose to

impute with mean value between precedent and

successive values. For PM10 we used, instead, the

“cold neck” technique, meaning that missing values

have been substituted with the values coming from

concentrations observed in the same period of

previous years. In such a way we were able to

preserve the mean and the standard deviation of the

whole data distribution, as visible in Tables 2 and 3.

4 Results and Discussion

In this work, the authors implemented three TSA

models, which have been tuned and validated in “R”

software. All the models are based on ARIMA, and

they have been calibrated by minimizing AIC and

BIC criteria, according to the parsimony principle.

4.1 PM10 Concentrations in Nocera Inferiore

After checking the autocorrelation and partial

autocorrelation (Fig. 5), an AR(1) model is

suggested for the PM10 concentrations in Nocera

Inferiore, having only order 1 autoregressive

component. Moreover, the routine “auto.arima”

implemented in the forecast package of “R”

software, also hinted at such a choice.

Figure 6 shows the existing overlap between

PM10 values measured and simulated, observing a

one-day delay in the prediction.

Figure 7, instead, reports on the left a scatter plot

correlating observed and simulated level of PM10

concentrations. It is remarkable that 80.2% of the

simulations lie in the area determined by average of

the observation  one standard deviation. Especially

in the low concentration range, the plot shows a

certain number of overestimations in the simulated

values of PM10, while in the high concentration

range underestimated simulated values are present.

On the right side of Figure 7, the histogram of the

residuals of the model, i.e. the difference between

observed and simulated values, is plotted, while the

summary statistics of the distribution of the

residuals are reported in Table 4. The obtained

kurtosis index is a positive value, which indicates

that the distribution is leptokurtic. This is consistent

with what can be discerned by observing the

histogram: the trend of the residuals has a more

“pointed” shape than a normal Gaussian

distribution. Furthermore, the positive skewness

index is properly substantiated by the evidence that

a rightward tail is present in the histogram.

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.13

D. Rossi, A. Mascolo, S. Mancini,

J. G. Ceron Breton, R. M. Ceron Breton,

C. Guarnaccia

E-ISSN: 2224-3496

155

Volume 19, 2023

Table 2. Summary statistics of the CO concentrations measured in Solofra.

Calibration dataset

Mean

[mg/m3]

Std. Dev.

[mg/m3]

Median

[mg/m3]

Skew

Kurt

Observed

0.45

0.32

0.33

0.71

-0.77

Reconstructed

0.45

0.32

0.35

0.72

-0.79

Table 3. Summary statistics of the PM10 concentrations measured in Nocera Inferiore.

Calibration dataset

Mean

[µg/m3]

Std. Dev.

[µg/m3]

Median

[µg/m3]

Skew

Kurt

Observed

33.44

17.13

29.75

0.81

0.23

Reconstructed

33.53

17.04

29.50

0.80

0.19

Fig. 5: Autocorrelation and Partial autocorrelation for PM10 observed in Nocera Inferiore, [16].

Fig. 6: Plot of observed and simulated PM10 concentrations in Nocera Inferiore in the calibration

phase.

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.13

D. Rossi, A. Mascolo, S. Mancini,

J. G. Ceron Breton, R. M. Ceron Breton,

C. Guarnaccia

E-ISSN: 2224-3496

156

Volume 19, 2023

Fig. 7: Scatterplot of observed and simulated PM10 concentrations in Nocera Inferiore in the calibration phase

and histogram of the residuals.

Table 4. Summary statistics of the residuals of AR(1) model for PM10 concentrations - Nocera Inferiore

Mean

[µg/m3]

Std. Dev.

[µg/m3]

Median

[µg/m3]

Skew

Kurt

Residuals AR(1)

-0.25

13.99

-2.22

0.62

0.36

4.2 CO Concentrations in Solofra

The same analysis implemented for PM10

concentration in Nocera Inferiore has been

produced by using the data of CO concentration in

Solofra, obtaining the same graphs.

The series generated with this dataset is

nonstationary, thus a differentiation became

necessary in order to work with a smoother time

series. By looking at the autocorrelation and the

partial autocorrelation plots (Fig. 8) the authors

decided to choose a ARIMA(14,1,14) model,

which was tested together with the ARIMA(0,1,1)

simple model suggested by the BIC criterion (a

ranking has been obtained with the “arimaId”

function of the “ast” package in the “R” software).

Figure 9 shows the slope of the measured CO

concentrations overlapped with the two ARIMA

models results. Both the models are good enough in

fitting CO concentrations curve, but ARIMA(0,0,1)

has a certain delay in the process.

ARIMA(14,1,14), on the contrary, does not exhibit

the delay, but its implementation requires a higher

computational effort due to the large number of

parameters. In Figure 10 it can be appreciated how

the entirety of the simulations obtained both with

ARIMA(0,1,1) and ARIMA(14,1,14) have a high

level of accuracy, since they are in the region

outlined by average of the observation  one

standard deviation.

The summary statistics of the distribution of

residuals for both models are reported in Table 5

and their histograms are plotted in Figure 11.

In the ARIMA(0,1,1) model, both the skewness

index and the kurtosis index are positive values.

Indeed, this model can be described as a

leptokurtic-type distribution of residuals, with

positive skewness. In the ARIMA(14,1,14) model,

on the other hand, the negative skewness index and

positive kurtosis index identify a leptokurtic

distribution, but with negative skewness.

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.13

D. Rossi, A. Mascolo, S. Mancini,

J. G. Ceron Breton, R. M. Ceron Breton,

C. Guarnaccia

E-ISSN: 2224-3496

157

Volume 19, 2023

Fig. 8: Autocorrelation and Partial autocorrelation for CO a) observed series and b) differenced

series in Solofra, in calibration phase, [16].

Fig. 9: Plot of observed and simulated CO concentrations in Solofra, with ARIMA(0,1,1) and

ARIMA(14,1,14) during the calibration phase.

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.13

D. Rossi, A. Mascolo, S. Mancini,

J. G. Ceron Breton, R. M. Ceron Breton,

C. Guarnaccia

E-ISSN: 2224-3496

158

Volume 19, 2023

Fig. 10: Scatter plot of observed and simulated CO concentrations in Solofra, during the calibration

phase, with ARIMA(0,1,1) on the left, and ARIMA(14,1,14) on the right.

Fig. 11: Histograms of the residuals of the ARIMA(0,1,1) model on the left and the ARIMA(14,1,14)

model on the right for CO concentrations – Solofra.

Table 5. Summary statistics of the ARIMA(0,1,1) and ARIMA(14,1,14) models residuals for CO

concentrations – Solofra.

Mean

[mg/m3]

Std. Dev.

[mg/m3]

Median

[mg/m3]

Skew

Kurt

Residuals ARIMA(0,1,1)

-0.01

0.13

-0.01

0.51

3.45

Residuals ARIMA(14,1,14)

-0.02

0.1

-0.02

-0.31

1.56

5 Forecast Results

The ARIMA models can be used to forecast future

variations of the analyzed pollutants, to be then

compared with actual collected data. By accurately

choosing hyperparameters p, d and q, in section 4

different models were found to describe PM10 and

CO time slope. Thus, hereafter the procedure and

results of forecasting on the same pollutants by

using the selected models, are reported.

5.1 AR(1) Model Forecast for PM10

The forecast of PM10 values in Nocera Inferiore has

been generated for the 10 days after the last day

used for calibration (31st of May). This interval has

been selected since it ensures a quite large time

range for possible mitigation actions. The forecasts,

in fact, can be used to support policy makers and

local governments in the decision process, allowing

to prevent large numbers of exceedances of the safe

thresholds. In Figure 12 a forecast plot is

represented, showing PM10 concentration as a

function of “future” days.

The forecasted values strictly lie above the

measurements, indicating a general overestimation

of the model. This is confirmed by the statistical

values of the errors reported in Table 6. Even if this

slight overestimation could be interpreted as a

limitation of the model, in a practical application,

overestimating a pollutant is a safe approach, since

the possible mitigation actions driven by such

forecasts would follow a precautionary principle.

5.2 ARIMA Models Forecasting for CO

Data of CO values recorded in Solofra have also

been forecasted for the first ten days after 31st of

May 2020, i.e. the days immediately subsequent to

the removal of lockdown restrictions. Again, the

forecast interval has been chosen to provide useful

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.13

D. Rossi, A. Mascolo, S. Mancini,

J. G. Ceron Breton, R. M. Ceron Breton,

C. Guarnaccia

E-ISSN: 2224-3496

159

Volume 19, 2023

information on the pollutant concentration slope

over time. A 10 days’ time range is large enough to

observe the increasing or decreasing trend in the

data and to decide if any intervention is needed.

Results of both ARIMA(0,1,1) and ARIMA

(14,1,14) are reported respectively in Figure 13. In

this case the predicted values provided by both

models exhibit a general underestimation, even

though the increasing trend is detected by the

ARIMA(14,1,14). The error summary statistics are

reported in Table 7.

The good agreement shown by both the models

at the very first periods (2 days) suggests that these

models can be used to provide useful information

for the decision process of the policy makers in a

short time range. The continuous measurements

collected by the monitoring stations allow to

recalibrate the models day by day, making it

possible to test the daily forecast and to move

further the predictions.

Fig. 12: Plot of observed and forecasted PM10

concentrations in Nocera Inferiore.

Table 6. Summary statistics of the errors of AR(1)

model for PM10 concentrations - Nocera Inferiore.

Errors

Mean

[µg/m3]

Std. Dev.

[µg/m3]

Median

[µg/m3]

AR(1)

5.83

2.00

6.06

Fig. 13: Plot of observed and simulated CO

concentrations in Solofra in the forecast phase, for

ARIMA(0,1,1) model (green line) and

ARIMA(14,1,14) model (red line).

Table 7. Summary statistics of the errors of

ARIMA(0,1,1) and ARIMA(14,1,14) models for

CO concentrations - Solofra.

Errors

Mean

[mg/m3]

Std. Dev.

[mg/m3]

Median

[mg/m3]

ARIMA(0,1,1)

0.07

0.05

ARIMA(14,1,14)

-0.10

0.06

0.12

6 Conclusions

Italy has been hardly affected by COVID-19

explosion, and was one of the first nations to

implement a drastic containment policy to limit

virus spread and contagion. Trying to mitigate the

pandemic, in fact, many governmental restrictions

were adopted starting from February 2020.

By analysing the outcomes of such restrictions,

the presented work investigated the variations of

CO and PM10 levels in the Campania region and

how ARIMA models could perform good

simulation of data. In order to compare how

governmental restrictions altered such pollutants’

concentrations, two selected cities of the region

were chosen. The first is Solofra, in the province of

Avellino, an industrial city where leather tanning

processing daily takes place. The second one is

Nocera Inferiore, in the province of Salerno, which

has a high population density (about 2200

inhabitants/km2) and a widespread and busy road

network (also a highway barrier). By choosing the

two presented cities the authors wanted to evaluate

how the pandemic affected two main sources of

pollution: the industrial activities and the road

traffic together with the population activities

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.13

D. Rossi, A. Mascolo, S. Mancini,

J. G. Ceron Breton, R. M. Ceron Breton,

C. Guarnaccia

E-ISSN: 2224-3496

160

Volume 19, 2023

(Nocera Inferiore is also characterized by the close

presence of residential areas, schools, shops,

highways and small industries).

To evaluate the pollution variations in the

considered areas, data recorded from two

environmental monitoring stations have been used,

and data of the February-May period over three

different years have been compared: 2018, 2019,

2020. By using proper physic-mathematical

models, the variation of the temporal trend of

pollutants before (2018, 2019) and during COVID-

19 lockdown (2020) has been assessed through the

calibration and validation of models on interesting

selected series: CO for Solofra and PM10 for

Nocera Inferiore. ARIMA models applied showed

good performance in the simulation of data. As a

result, the authors observed how restriction policies

did not significantly contribute to reducing PM10

and CO concentrations in air, also compared to

previous years. ARIMA models also permit to

implement a prediction of the pollutants levels over

time, and the authors specifically performed a 10-

days forecast of both PM10 and CO concentrations,

respectively in Nocera Inferiore and Solofra.

ARIMA models selected for the forecasting gave

results approximately good in a very short

prediction range, as documented in literature.

The main limitation of this study relies on the

fact that going further from the start of the forecast

period, the simulated concentrations start to be

significantly different from the observed values.

This means that a possible improvement can be the

automatic recalibration of the model, day by day,

taking into account the latest measurements of the

pollutants under study and their possible slope

variations. In future works, a parameter sensitivity

analysis could be performed to achieve an

estimation of the maximum range of prediction that

can be used before the need to recalibrate the

model. Once the optimal time range is set, the

recalibration procedure will provide a reliable

predictive model, constantly updated, that could

help decision makers in implementing temporary or

permanent actions to mitigate the pollutants

concentrations, to fulfil the thresholds imposed by

the national regulations and to protect human

beings’ health.

References:

[1] Thieriot H., Myllyvirta L., Air pollution

returns to European capitals: Paris faces

largest rebound, Centre for Research on

Energy and Clean Air (CREA), 2020.

[2] Brunekreef B., Holgate S. T., Air pollution

and health, Lancet, Vol. 360(9341), 2002,

pp. 1233-1242.

[3] Cabaneros S, Calautit J K, Hughes B R, A

review of artificial neural network models for

ambient air pollution prediction.

Environmental Modelling & Software, Vol.

119, 2019, pp. 285-304.

[4] Achcar J. A., Rodrigues E. R., Guadalupe T.,

Using non-homogeneous Poisson models

with multiple change-points to estimate the

number of ozone exceedances in Mexico

City, Environmetrics, Vol. 22, N. 1, 2011,

pp.1-12

[5] Guarnaccia C, Lenza TLL, Mastorakis NE

and Quartieri J, A comparison between

traffic noise experimental data and predictive

models results, International Journal of

Mechanics, Vol. 5 (4), 2011, pp. 379-386

[6] Liao K, Huang X, Dang H, Ren Y, Zuo S,

Duan C, Statistical Approaches for

Forecasting Primary Air Pollutants: A

Review. Atmosphere, Vol. 12, 2021, 686.

[7] Cerón Bretón J. G., Cerón Bretón R.M.,

Morales S.M., Kahl J.D.W., Guarnaccia C.,

del Carmen Lara Severino R., Marrón M.R.,

Lara E.R., de la Luz Espinosa Fuentes M.,

Chi M.P.U., Sánchez G.L., Health risk

assessment of the levels of BTEX in ambient

air of one urban site located in Leon,

Guanajuato, Mexico during two climatic

seasons, Atmosphere, Vol. 11, N. 21, 2020.

[8] Rovetta A. The Impact of COVID-19

Lockdowns on Particulate Matter Emissions

in Lombardy and Italian Citizens’

Consumption Habits. Frontiers in

Sustainability, Vol. 2, 2021, 649715.

[9] Bray CD, Nahas A, Battye WH, Aneja VP.

Impact of lockdown during the COVID-19

outbreak on multi-scale air quality. 2021,

Atmospheric Environment, Vol. 254, 1994,

118386.

[10] Shahriar S., Kayes I., Hasan K., Hasan M.,

Islam R., Awang N.R., Hamzah Z., Eh Rak

A., Salam M.A., Potential of ARIMA-ANN,

ARIMA-SVM, DT and CatBoost for

Atmospheric PM2.5 Forecasting in

Bangladesh, Atmosphere Vol. 12, 1, 2021,

100.

[11] Aladağ E., Forecasting of particulate matter

with a hybrid ARIMA model based on

wavelet transformation and seasonal

adjustment, Urban Climate Vol. 39, 2021,

100930.

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.13

D. Rossi, A. Mascolo, S. Mancini,

J. G. Ceron Breton, R. M. Ceron Breton,

C. Guarnaccia

E-ISSN: 2224-3496

161

Volume 19, 2023

[12] Guarnaccia C, Quartieri J, Tepedino C.,

Deterministic decomposition and seasonal

ARIMA time series models applied to airport

noise forecasting, Proc. of the Int. Conf. on

Applied Mathematics and Computer Science,

AIP Conference Proceedings 1836, 2017, pp.

1-7.

[13] Guarnaccia C, Quartieri J, Mastorakis NE

and Tepedino C, Development and

application of a time series predictive model

to Acoustical noise levels WSEAS

Transactions on Systems, Vol. 13, 2014, pp.

745-756.

[14] Guarnaccia C, Quartieri J, Rodrigues ER and

Tepedino C, Acoustical noise analysis and

prediction by means of multiple seasonality

time series model, International Journal of

Mathematical Models and Methods in

Applied Sciences, Vol. 8, 2014, 384-393

[15] Guarnaccia C, Cerón Bretón J G, Quartieri J,

Tepedino C, Cerón Bretón C R., An

Application of Time Series Analysis for

Forecasting and Control of Carbon Monoxide

Concentrations, International Journal of

Mathematical Models and Methods in

Applied Sciences, Vol. 8, 2014, pp 505-515.

[16] Mancini S., Francavilla A.B., Graziuso G.,

Guarnaccia C, An Application of ARIMA

modelling to air pollution concentrations

during covid pandemic in Italy, Journal of

Physics: Conference Series, Vol. 2162, 2022,

012009.

[17] Guarnaccia C., Quartieri J., Tepedino C.,

Petrovic L., A Comparison of Imputation

Techniques in Acoustic Level Datasets,

International Journal of Mechanics, Vol. 9,

2015, pp. 272-278.

[18] Guarnaccia C., Ceron Breton J. G., Ceron

Breton R. M., Tepedino C., Quartieri J.,

Mastorakis N. E., ARIMA models application

to air pollution data in Monterrey, Mexico, in

AIP Conference Proceedings Vol. 1982, No.

1, 2018, p. 020041.

[19] Guarnaccia C, Mancini S, Quartieri J, Ceron

Breton JG, Ceron Breton RM, Prediction of

CO Concentrations in Monterrey, Mexico, by

means of ARIMA Models, WSEAS

Transactions on Environment and

Development, Vol. 14, 2018, pp. 653-661.

[20] Cryer P D, Chan K. Time Series Analysis,

with applications in R. 2008.Second Edition,

Springer.

[21] R Core Team (2018). R: A language and

environment for statistical computing. R

Foundation for Statistical Computing,

Vienna, Austria.

[22] ARPA Campania, Qualità dell’aria in

Campania: monitoraggio mai interrotto,

riprende pubblicazione bollettini, available

online: https://www.arpacampania.it/-

/qualit%C3%A0-dell-aria-in-campania-

monitoraggio-mai-interrotto-riprende-

pubblicazione-bollettini, in Italian, last

accessed on February 10, 2023.

[23] ARPA Campania, Qualità dell’aria: dati

orari su polveri sottili in altre 20 stazioni di

monitoraggio in Campania, available online:

https://www.arpacampania.it/-/qualita-dell-

aria-dati-orari-su-polveri-sottili-in-altre-20-

stazioni-di-monitoraggio-in-campania, in

Italian, last accessed on February 10, 2023.

[24] Gazzetta Ufficiale della Repubblica Italiana

(2010). Ministerial Decree August 13, 2010

n.155. Available online at:

https://www.camera.it/parlam/leggi/deleghe/t

esti/10155dl.html .

[25] ARPAC (2022), La qualità dell’aria in

Campania, 2015-2021, available online:

https://www.arpacampania.it/web/guest/relaz

ioni-e-report, last accessed on February 10,

2023.

[26] University of Salerno, Photovoltaic Park of

the Campus of Fisciano, available online:

https://web.unisa.it/en/campus-

life/campus/photovoltaic-park, last accessed

on February 10, 2023.

Contribution of Individual Authors to the

Creation of a Scientific Article (Ghostwriting

Policy)

Conceptualization: C. Guarnaccia, J.G. Ceron

Breton, R.M. Ceron Breton

Data curation: D. Rossi, A. Mascolo, C. Guarnaccia

Methodology: D. Rossi, C. Guarnaccia, J.G. Ceron

Breton, R.M. Ceron Breton

Software: D. Rossi, A. Mascolo, C. Guarnaccia

Supervision: C. Guarnaccia

Visualization: D. Rossi, A. Mascolo

Writing - original draft: D. Rossi

Writing - review & editing: all the co-authors

Corresponding authors: D. Rossi, C. Guarnaccia

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.13

D. Rossi, A. Mascolo, S. Mancini,

J. G. Ceron Breton, R. M. Ceron Breton,

C. Guarnaccia

E-ISSN: 2224-3496

162

Volume 19, 2023

Sources of Funding for Research Presented in a

Scientific Article or Scientific Article Itself

No funding was received for conducting this study.

Conflict of Interest

The authors have no conflicts of interest to declare

that are relevant to the content of this article.

Creative Commons Attribution License 4.0

(Attribution 4.0 International, CC BY 4.0)

This article is published under the terms of the

Creative Commons Attribution License 4.0

https://creativecommons.org/licenses/by/4.0/deed.en

_US