Working papers SIET 2015 – ISSN 1973-3208

Prediction of air pollution peaks generated by urban transport networks Margaret Bell1, Angela S. Bergantino 2, Mario Catalano1 , Fabio Galatioto1 1

Transport Operations Research Group (TORG), School of Civil Engineering and Geosciences, Newcastle University. 2

Department of Economics, Management and Law (DISAG), University of Bari.

Abstract This paper illustrates the first results of an ongoing research for developing novel methods to analyse and simulate the relationship between trasport-related air pollutant concentrations and easily accessible explanatory variables. The final scope of the analysis is to integrate the new models in traditional traffic management decision-support systems for a sustainable mobility of road vehicles in urban areas. This first stage concerns the relationship between the mean hourly concentration of nitrogen dioxide and explanatory factors like traffic and weather conditions, with particular reference to the prediction of pollution peaks, defined as exceedances of normative concentration limits. Two modelling frameworks are explored: the Artificial Neural Network approach and the ARIMAX model. Furthermore, the benefit of a synergic use of both models for air quality forecasting is investigated. The analysis of findings points out that the prediction of extreme pollutant concentrations is best performed by the integration of the two models into an ensemble. The neural network is outperformed by the ARIMAX model in foreseeing peaks, but gives a more realistic representation of the relationships between concentration and wind characteristics. So, it can be exploited to direct the ARIMAX model specification. At last, the study shows that the ability at forecasting exceedances of pollution regulative limits can be enhanced by requiring traffic management actions when the predicted concentration exceeds a threshold that is pretty high but lower than the normative one. Keywords: air quality forecasting, exceedances of pollutant concentration limits, nitrogen dioxide, artificial neural network, ARIMAX model, ensemble techniques.

1 Research question and review of literature It is broadly demonstrated that air pollution in urban areas is mainly due to the intense use of motorized transport for travelling, with particular regard to private cars and heavy goods vehicles. This is a top priority issue for transportation planners and public authorities, given the harmful effects of pollution to human health and the environment. Numerous studies (Heinrich et al., 2005; Zhang et al., 2012) argue that acute exposure to air pollutants may cause serious temporary health concerns such as eye irritation, breathing difficulty, cardio-vascular problems, while chronic exposure may lead to damages to the body’s immune, neurological, reproductive and respiratory systems, cancer and even premature death. In November 2014 the British Committee on the Medical Effects of Air Pollutants reported that air pollution may be responsible for as many as 60,000 early deaths in Britain each year. Also the environment is affected in terms of global climate change and adverse effects for plants and eco-systems (Seinfeld and Pandis, 2006; Zhang et al., 2012). To protect human health and the environment, various national contexts throughout the world have issued guidelines and regulations. The United States Environmental Protection Agency (EPA) has set national ambient air quality standards for six pollutants: sulfur dioxide (SO 2), nitrogen dioxide (NO 2), carbon monoxide (CO), ozone (O3), lead (Pb) and particulate matter (Seinfeld and Pandis, 2006).



The contribution of Mario Catalano was carried out during his visiting in 2014 and 2015 at the Transport Operations Research Group of the University of Newcastle, School of Civil Engineering and Geosciences.

Working papers SIET 2015 – ISSN 1973-3208 In Europe, over the last decades, the European Union has adopted an ample range of environmental measures to improve the quality of life for the Community's citizens. The final step of this legislative process is the Directive 2008/50/EC (EU, 2008), which has integrated an extensive body of laws establishing health-based concentration standards for a number of pollutants in outdoor ambient air. The European Commission has the task of ensuring that environmental law is applied by the Member States through infringement procedures. Long term measures like mode switch policies in favour of mass transit and public regulation on road use are pretty effective in abating atmospheric pollution in cities, but pollution peaks and the consequent exceedance of regulative concentration thresholds are often caused by substantial fluctuations of mobility patterns and weather conditions around their expected behaviours. Hence, air quality protection needs to be fine-tuned through the introduction in the local policy portfolio of further tools and actions to forecast extreme pollution events and manage traffic over short-term periods in order to prevent the predicted concentration peaks. Given the above, this research has been started to investigate traffic-related air pollution modelling with the final aim of developing a real-time decision-support system for a more sustainable mobility of road vehicles in urban areas. In more detail, four main challenges will be addressed: 1. to develop a model to predict accurately the density of those airborne pollutants subject to normative standards for hourly or daily state of concentration, so as to permit local authorities to prevent their occurrence by real-time traffic management; 2. to explore the benefits in terms of accuracy and geographical transferability of models based on panel data, which are novel in the air quality modelling literature; 3. to build a method for predicting well in advance the yearly average of hourly mean concentrations, thus providing Local Authorities with a powerful tool to determine if and when to act in order to respect the environmental law, when this limits the behaviour of pollution over an annual period; 4. to experiment the synergic interplay between pollutant concentration forecasting and vehicular mobility microsimulation for an enhanced traffic management system based also on air quality targets. By addressing the identified challenges, this research might be of strategic importance for many national contexts. To have an idea of the worldwide scale of atmospheric pollution problems, one could examine the 2014 version of the WHO (World Health Organization) Ambient Air Pollution database consisting mainly of urban air quality data, notably annual means of PM10 and PM2.56 concentration for about 1,600 cities of 91 countries in the 2008‐2013 period (WHO, 2014). As can be seen in Fig. 1, the world's annual mean levels of PM10 by region range from 26 to 208 µg/m3; the world's average is 71 µg/m3 against the value of 20 µg/m3 recommended by the WHO air quality guidelines (WHO, 2014). Particular concern is associated to the East side of the planet, where countires like China, India, Nepal, Bangadlesh, Mongolia and, in the Mediterranean Area, Egypt, Iran, Jordan, Afghanistan, Pakistan far exceed the world's yearly mean density of PM10.

PM10 (microg/m3) 250 208 200

150 100

128 87

78 51

50

71

49 26

0 Afr

6

Amr LMI Emr LMI Eur LMI

Particles with diameter smaller than 10 and 2.5 microns, respectively.

Sear

Wpr LMI

HIC

World

Working papers SIET 2015 – ISSN 1973-3208 Fig. 1. PM10 levels by region, for the last available year in the period 2008‐2012. Amr: America, Afr: Africa; Emr: Eastern Mediterranean, Sear: South-East Asia, Wpr: Western Pacific; LMI: Low- and middle-income; HI: highincome.

The 2014 Air Quality in Europe Report (European Environment Agency, 2014) states that, in EU cities, exposure to atmospheric pollution levels exceeding the WHO air quality limits (in general stricter than the EU standards) is significantly high for various chemical agents. This over limit exposure regards 64% and 92% of the total EU-28 urban population in 2012 for PM10 and PM2.5, respectively. Moreover, in the case of ozone, in the same year, the exposure incidence rises to even 98% of people living in towns. There has been a clear decreasing trend, instead, for NO2 concentration in many European countries over the last decade7, but, in the United Kingdom, the NO 2 levels have exceeded the relative WHO and EU target values persistently. This is confirmed by the fact that, in the early part of 2014, the European Commission launched legal proceedings against the UK for its failure to cut excessive levels of nitrogen dioxide (EU Press Release Database, 2014). Lastly, while exposure of the Europeans to CO concentrations above the EU and WHO thresholds is negligible, in the case of benzene (C6 H6), around 10% of the EU-28 urban population is subject to pollution beyond the WHO levels and the percentage takes on the value of 37% in the case of SO2. This paper presents the early stage of our ongoing research, which refer to nitrogen dioxide, a toxic gas emitted by road vehicles, shipping, power generation, industry and households, which, even in the case of short term exposures (from 30 minutes to 24 hours), may cause adverse respiratory effects in healthy individuals. Furthermore, it is the main precursor for ground-level ozone, that is very harmful to human health. For NO2, the EU environmental legislation sets two types of standard: the hourly mean concentration cannot go beyond the level of 200 µg/m3 more than 18 times each calendar year; the annual average of hourly concentrations is not allowed to exceed 40 µg/m3. In particular, we have modelled the relationship between NO 2 hourly concentration and potential explanatory variables such as transport-related attributes, that influence emissions, and weather conditions, that are responsible for dispersion and transformation of pollutants. As regards the scientific background of the research, few studies have appeared in the scientific literature on real time air quality forecasting near urban arteries, amongst which some are particularly interesting for this work, since they investigate the relationship between nitrogen oxides levels and meteorological and transport-related variables (Kukkon et al., 2003; Ming et al., 2009; Nagendra and Khare, 2006; Perez and Trier, 2001; Viotti et al., 2002). The leitmotiv of these studies is to consider the Neural Network, from the domain of Artificial Intelligence science, the most effective tool to predict air quality in urban areas. In some cases, this methodology is compared with other approaches, but they are usually linear regression models or deterministic models simulating the relevant physical processes. If also other pollutants are considered, it is possible to find air quality modelling works applying further statistical methods. For example, Arwa et al. (2014) tackle the issue of predicting particular matter concentration evaluating different models: multiple linear regression, quantile regression, generalised additive models and regression trees. Baur et al. (2004) compare the performance of quantile regression with multiple linear regression for predicting ozone concentrations. Kaushik and Melwani (2007) adopt the Seasonal Autoregressive Integrated Moving Average (ARIMA) model to forecast the daily levels of sulphur dioxide, nitrogen dioxide and suspended particulate matters. Generally speaking, parametric and non-parametric statistical methods are more suitable for the description of complex relations between concentrations and potential predictors, and often present a higher accuracy, as compared to deterministic (physically-based) models, which are, furthermore, computationally expensive. However, statistical techniques are usually confined to the conditions occurring during the measurements and cannot be generalized to other areas with different chemical and meteorological characteristics. In addition, they fail to forecast concentrations during periods of unusual emissions and/or weather conditions that deviate significantly from the historical record (Zhang et al., 2012). Given the above, in an attempt to develop an effective tool for predicting exceedances of NO2 hourly concentration thresholds set by the EU, this research explores the statistical approach. For the particular case of nitrogen dioxide, a gap in the scientific literature has been identified in relation to the comparison between non-parametric statistical techniques, like the popular neural network, and sophisticated parametric methods as the Auto-Regressive Integrated Moving Averages with eXogenous inputs (ARIMAX) model (Hamilton, 7

Between 2003 and 2012, in EU-28, the ambient air NO2 annual mean concentration dropped by 18% on the average. Only 8% of the EU-28 citizens live in areas where the annual WHO and EU thresholds for NO2 were exceeded in 2012.

Working papers SIET 2015 – ISSN 1973-3208 1994). Hence, the neural network and the ARIMAX framework have been compared with respect to NO2 concentration forecasting through a set of indicators for missed exceedances and false alarms. In addition, based on successful experience in other fields 8 (Bishop, 1995; Re and Valentini, 2012; VV. AA., 2008), the effectiveness of multimodel approaches have been evaluated. So, the forecasts from different models have been combined and the resulting impact on prediction accuracy has been quantified. The remainder of the paper is structured as follows: section 2 describes the site where the data used in this work have been collected along with the dataset itself in terms of descriptive statistics and statistical properties of the involved time series; section 3 illustrates the theoretical foundations, specification and estimation of the models employed to forecast air quality; section 4 performs a comparative analysis of the predicition models based on statistical and catagorical metrics; in the end, section 5 draws conclusions and points out new topics for future research.

2 Case study This section gives details about the study area with respect to its geometric characteristics and the technologies used to collect air quality and traffic data. In addition, the dataset employed to derive insights into the research problem has been analysed with descriptive statistics and time series analysis techniques.

2.1 Air quality monitoring site The study area chosen to perform the analysis is Marylebone Road in the City of London (see Fig. 2) 9. Marylebone Road has three lanes each way, with the nearside lanes in both directions reserved for buses and taxis. The traffic along the corridor is controlled by a demand responsive signal control system termed SCOOT, Split Cycle Offset Optimization Technique (Hunt et al, 1981). The cabin for air quality measurement is located on the southern side of the road, on a spot where the road is characterised by a canyon H/W ratio of 0.8610 (see inset in Fig. 2). For Marylebone road, a reach dataset over a ten year period (1998-2007) is available, which contains traffic (flow and speed for each lane), weather and air quality data at an hourly resolution.

Fig. 2. The study area, Marylebone road in London, and the air quality monitoring site.

8

Such as machine learning science, astronomy, astrophysics, computer network intrusion detection, early diagnosis of diseases, face recognition. 9 For a preliminary analysis see: Bell et al. 2015. 10 The ratio of average buildings' height to road width (Tartaglia, 1999).

2.2 Descriptive analysis of the dataset The analysis has concerned the relationship between the mean hourly concentration of NO2 and explanatory factors like traffic and weather in a central spot of London, Marylebone Road, throughout the year 2007. Table 1 shows the main statistics describing the set of data collected in 2007 within the study-area. As can be seen in the table, the average of all hourly NO2 concentrations is quite high (102.5); furthermore, it has been calculated that the 200 µg/m3 EU threshold was exceeded 457 times during 2007, against a maximum regulative limit of 18 times in a year. This makes the considered study-area a source of severe nitrogen dioxide air pollution, that needs long term policy-actions for sustainable transport, but also effective prediction models to support real-time traffic management. Such a problematic atmospheric condition may be probably ascribed to two main reasons: the high traffic volume which, from 7:00 to 21:00 o'clock, is greater than 4000 passenger car units/hour for 72% of time; the quasi-canyon layout of the monitored street, since the ratio of average buildings' height to road width is around 0.9 (Tartaglia, 1999). Table 1. Descriptive statistics for the set of air quality, transport and weather data collected in 2007 within the studyarea. Variables

Observations Average

Min

Max

Standard Deviation

Hourly mean concentration of NO2 (µg/m3)

8584

102.5

4

329

53.1

Total traffic volume* (passenger car units/hour)

8734

3505.7

303.7

5161.5

1173.2

Hourly mean speed of wind (km/h)

7993

2.8

0.1

15.1

2.1

Hourly mean direction of wind (North degrees)

7993

203.9

2.9

358.1

82.4

Hourly mean temperature (centigrade degrees)

7993

13.5

-0.8

29.7

5.2

*

The orginal data on traffic were disaggregated by six vehicle classes: Motorcycle, Car or Light Van (length < 5.2m), Car and trailer, Rigid Lorry, Heavy Van (length ≥ 5.2m) or mini-bus, Articulated Lorry, Bus or Coach. The measured flows for each category have been turned into passenger car equivalents through the following multiplicative parameters (Lavecchia et al., 2007): 1 for motorcycles and cars, 1.5 for light vehicles (5m