Air Quality Monitoring and Advanced Bayesian Modeling

By Yongjie Li, Ka In Hoi, Kai Meng Mok and Ka Veng Yuen

Air Quality Monitoring and Advanced Bayesian Modeling introduces recent developments in urban air quality monitoring and forecasting. The book presents concepts, theories, and case studies related to monitoring methods of criteria air pollutants, advanced methods for real-time characterization of chemical composition of PM and VOCs, and emerging strategies for air quality monitoring. The book illustrates concepts and theories through case studies about the development of common statistical air quality forecasting models. Readers will also learn advanced topics such as the Bayesian model class selection, adaptive forecasting model development with Kalman filter, and the Bayesian model averaging of multiple adaptive forecasting models.

• Covers fundamental to advanced applications of urban air quality monitoring and forecasting
• Includes detailed descriptions and applications of the instruments necessary for the most successful monitoring techniques
• Presents case studies throughout to provide real-world context to the research presented in the book

• Language: English
• Publisher: Elsevier Science
• Release date: Jan 14, 2023
• ISBN: 9780323902670

Yongjie Li

Yongjie Li is an Associate Professor in the Department of Civil and Environmental Engineering at the University of Macau. He obtained his Ph.D. degree in Environmental Engineering from The Hong Kong University of Science and Technology (2010), after receiving a B.Sc. degree in Chemistry from Peking University (2004). He was a postdoctoral fellow at Harvard University from 2014 – 2015 before joining the University of Macau. His research interests include air pollution measurements and atmospheric chemistry. He has been working on mass spectrometric techniques for real-time air pollution measurements and chemical reactions leading to secondary pollution formation, which resulted in over 100 SCI journal articles on these topics. He teaches one undergraduate course, Environmental Engineering, and two postgraduate courses, Air Pollution Meteorology and Chemistry and Air Pollution Control. He was the recipient of the Asian Young Aerosol Scientist Award in 2022 and the China Aerosol Young Scientist Award in 2019.

Book preview

9780323902670_FC

Air Quality Monitoring and Advanced Bayesian Modeling

First Edition

Yongjie Li

Department of Civil and Environmental Engineering, University of Macau, Avenida da Universidade, Taipa, Macau

Ka In Hoi

Department of Civil and Environmental Engineering, University of Macau, Avenida da Universidade, Taipa, Macau

Kai Meng Mok

Department of Civil and Environmental Engineering, University of Macau, Avenida da Universidade, Taipa, Macau

Ka Veng Yuen

State Key Laboratory on Internet of Things for Smart City, Department of Civil and Environmental Engineering, University of Macau, Avenida da Universidade, Taipa, Macau

Table of Contents

Cover

Title page

Copyright

Chapter 1: Introduction

Abstract

1.1: Clean versus polluted air

1.2: Sources and impacts of air pollutants

1.3: Air quality monitoring strategies

1.4: Modeling and forecasting of air pollution

1.5: About this book

References

Chapter 2: Current air quality monitoring methods

Abstract

2.1: Methods for criteria air pollutants

2.2: Real-time chemical composition monitoring

2.3: Conclusions

References

Chapter 3: Emerging air quality monitoring methods

Abstract

3.1: Low-cost sensors

3.2: Mobile measurement platforms

3.3: Conclusions

References

Chapter 4: Traditional statistical air quality forecasting methods

Abstract

4.1: Multiple linear regression (MLR)

4.2: Classification and regression tree (CART)

4.3: Multilayer perceptron

4.4: Support vector regression (SVR)

4.5: Case study

References

Chapter 5: Advanced Bayesian air quality forecasting methods

Abstract

5.1: Overview of technique limitations and advanced topics for improvement

5.2: Bayesian model class selection of linear regression model

5.3: Kalman filter-based adaptive air quality model

5.4: Time-varying multilayer perceptron

5.5: Adaptive Bayesian model averaging of multiple time-varying regression models

5.6: Case study

References

Index

Copyright

Elsevier

Radarweg 29, PO Box 211, 1000 AE Amsterdam, Netherlands

The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom

50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States

Copyright © 2023 Elsevier Inc. All rights reserved.

No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Details on how to seek permission, further information about the Publisher’s permissions policies and our arrangements with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions.

This book and the individual contributions contained in it are protected under copyright by the Publisher (other than as may be noted herein).

Notices

Knowledge and best practice in this field are constantly changing. As new research and experience broaden our understanding, changes in research methods, professional practices, or medical treatment may become necessary.

Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information, methods, compounds, or experiments described herein. In using such information or methods they should be mindful of their own safety and the safety of others, including parties for whom they have a professional responsibility.

To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions, or ideas contained in the material herein.

ISBN: 978-0-323-90266-3

For information on all Elsevier publications visit our website at https://www.elsevier.com/books-and-journals

Unlabelled Image

Publisher: Candice G Janco

Acquisitions Editor: Jennette McClain

Editorial Project Manager: Naomi Robertson

Production Project Manager: Rashmi Manoharan

Cover Designer: Vicky Pearson

Typeset by STRAIVE, India

Chapter 1: Introduction

Abstract

Air pollution is one of the most serious environmental issues we are facing. We introduce here what are the components in the air and what are the pollutants that we are concerned about. The sources and impacts of those pollutants are also briefly described. To facilitate effective management of air quality, monitoring of air pollutants is an essential step, and the strategies of air quality monitoring are briefly introduced, laying out a context for more detailed description in later chapters. On the other hand, forecasting on air quality is also important for policy makers and general public to be informed about the future variation of ambient air quality. Thus, modeling approaches for air quality prediction are also briefly described.

Keywords

Atmospheric composition; Air pollutants; Air quality monitoring; Air quality modelling; Air quality forecasting

Contents

1.1.Clean versus polluted air

1.2.Sources and impacts of air pollutants

1.3.Air quality monitoring strategies

1.4.Modeling and forecasting of air pollution

1.5.About this book

References

When we say a volume of air is polluted, we mean that it contains unwanted substances that might affect the species living in that volume of air or cause some other effects to the ecosystem. As the key information conveyed to the public about the quality of air, an array of indexes is used to describe the levels of pollutants in the volume of air that we concern about. The indexes, normally called air quality index (AQI), are derived from numerical expressions scalable to the pollutant concentrations. For instance, numeric index from 0 to 100 (or 0 to 10 in some countries) is commonly used to indicate the AQI, with 0 being excellent, though not achievable, and 100 being bad. There are, however, situations where AQI values can reach 500 or more, which is normally labeled as severe or hazardous. The AQI values are normally determined from the hourly concentration levels of several criteria pollutants automatically measured by the environmental protection agencies. Therefore, before one can derive these indexes, the concentrations of the air pollutants have to be measured first; if one wants to know what will be the anticipated AQI tomorrow, then forecasting models have to be invoked to predict the concentrations of the pollutants. These two topics are the focus of this book. In addition, of course, for regulatory and research purposes, much more detailed information on air pollutants is needed than just those covered by AQI; such information is also obtained from different measurement methods, which will also be covered by this book.

1.1: Clean versus polluted air

The question of how polluted the volume of air is, however, requires quantitative assessment of the concentration levels of that particular pollutant. Before that, perhaps some even more fundamental questions have to be answered first:

–What is the unpolluted clean air supposed to contain?

–Has the composition of clean air been the same in the geologic time since the formation of our atmosphere?

–What substances can be considered pollutants and need to be measured?

The first atmosphere (c. 4.6 billion years ago) of the Earth is believed to consist of mainly hydrogen (H2) and helium (He), which were swept away by solar wind or lost due to gravitation escape (Jacobson, 2002). Then the outgassing of carbon dioxide (CO2), water vapor, and assorted gases from the Earth's mantle formed the second atmosphere, which is referred to prebiotic atmosphere. The biotic atmosphere, emerged after the appearance of living organisms (c. 3.5 billion years ago), mainly consisted of methane (CH4), molecular nitrogen (N2), sulfur dioxide (SO2), and CO2. After living organisms capable of photosynthesis appeared (c. 2.3 billion years ago), oxygen and ozone were produced; the oxygen buildup resulted in aerobic respiration, which led to more efficient production of N2, making it the major component in the atmosphere today.

Today, the dry air is mainly composed of N2 (78.1% by volume) and O2 (20.9% by volume). The other relatively abundant gas is argon (Ar, 0.9% by volume), which is a noble (and extremely nonreactive) gas. Other nonreactive gases in clean air include neon (Ne), He, krypton (Kr), H2, and xenon (Xe), which have volume fractions of (1–200) × 10−7. All these are considered nonreactive and relatively invariable constituents of clean gases, thus are not considered pollutants. Another gas that is relatively nonreactive (with a long atmospheric lifetime of about 15 years) and invariable gas, CO2 (currently with a volume fraction of around 0.04%, or 400 parts per million, ppm), is an important greenhouse gas that causes global warming; it is thus considered as an air pollutant from anthropogenic activities (e.g., burning of fossil fuels). This invariable characteristic of CO2 is, however, a relative description. Its concentration did increase from about 280 ppm before the Industrial Revolution to over 400 ppm nowadays due to anthropogenic input; it shows a clear annual pattern with lows in summer (efficient uptake by plant growth) and highs in winter (release from decays of plant litters). Water vapor is also an important constituent in the atmosphere and it is highly variable, with volume fractions ranging from a few parts per million all the way to approximately 4%. Water vapor is generally not considered as an air pollutant because it occurs mostly naturally, but its presence and the dynamics of it greatly affect the hydrological cycle and the meteorological phenomena.

Most air pollutants we concern about are in trace amounts and their concentrations are highly variable (Jacobson, 2002). For instance, carbon monoxide (CO) has a volume mixing ratio of 0.04–0.2 ppm in clean environments, but it can reach 2–10 ppm in polluted environments. The mixing ratio of surface ozone can be as low as 10 ppb (parts per billion) in clean environments, but as high as 350 ppb in polluted regions. Sulfur dioxide (SO2) mixing ratios are normally less than 1 ppb in clean areas but can be as high as 30 ppb in polluted areas. Nitrogen oxides (mainly nitric oxide, NO, and nitrogen dioxide, NO2) have mixing ratios of <0.3 ppb in clean environments but can be up to 300 ppb in polluted areas. Ammonia (NH3) normally has sub-ppb mixing ratios in clean areas but can be above 10 ppb near sources. Volatile organic compounds (VOCs) are normally present in the atmosphere in sub-ppb level, except for methane (CH4), whose mixing ratio can be 1–2 ppm even in clean environments. In polluted regions, however, some VOCs can have mixing ratios of >10 ppb. Particulate matters (PM) are suspended liquid or solid particles (Seinfeld and Pandis, 2016) that have sufficiently long lifetimes in the atmosphere to exert various effects on the environment. PM are normally classified by a cut-point diameter, with PM10 denoting particles with aerodynamic diameters < 10 μm and PM2.5 and PM1 of <2.5 and <1 μm, respectively. Typical PM2.5 mass concentration in clean environments is normally less than 10 μg m−3, but it can be way above 100 μg m−3 in polluted urban areas.

Most of these air pollutants have relatively short atmospheric lifetimes, ranging from 1 day to approximately a month, making their concentrations highly variable and spatially heterogeneous. Some of the emission sources of these air pollutants are also not continuous, at least not emitted at a constant strength. For instance, the CO from vehicular exhausts has much higher emissions during rush hours, while the oxidation of it by OH radicals during noon time contributes to the removal of this air pollutant. As such, most of these air pollutants show significant dynamics in concentrations and spatial distributions, making the monitoring of them a challenging task.

1.2: Sources and impacts of air pollutants

Most of the air pollutants of concern are more or less related to combustion processes, with only a few exceptions. Burning of fossil fuel is a process to harvest the energy released from the breakage of chemical bonds in hydrocarbons in the fuel, yielding ideally CO2 and H2O in complete combustion. Incomplete combustion, which is almost inevitably, actually releases a substantial amount of CO, black carbon (BC), and fly ash. While CO is a gaseous pollutant that can inhibit the capacity of blood to carry oxygen to organs and tissues, both BC and ash are important components in PM2.5 and may cause adverse health effects and alter energy balance due to absorption of solar radiation. Some fossil fuel, especially coal, contains a significant amount of sulfur, and the combustion of it releases SO2 as the by-product. If not properly treated, emission of SO2 in high amount will have health implications because it can cause respiratory symptoms in sensitive population. In addition, SO2 is also an important precursor for the sulfuric acid (H2SO4) that is responsible for acid rain formation; even after being neutralized by alkaline species such as NH3, the sulfuric acid is converted to ammonium sulfate, (NH4)2SO4, which is one of the major inorganic components in PM2.5.

Nitrogen oxides (e.g., NO and NO2, often termed collectively as NOx) are formed from high-temperature reaction of N2 and O2 in the air (termed as thermal NOx) or the oxidation of organically bound nitrogen in fuels (termed as fuel NOx). In particular, NO2 can also affect lung function in persons with asthma, and oxidation of it can form nitric acid (HNO3), another strong acid that contributes to the formation of acid rain. Similarly, neutralization of HNO3 by NH3 forms ammonium nitrate, NH4NO3, which also contributes to PM2.5 mass substantially. NH3 is generally not considered as a criteria air pollutant, but its participation in neutralization of the aforementioned inorganic acids contributes to the formation of secondary PM components. Therefore the most important secondary inorganic aerosol (SIA) components include ammonium (NH4+), sulfate (SO4²−), and nitrate (NO3−). The sources of NH3 include naturally occurring decomposition of amino acids in organic waste. Anthropogenic sources of NH3 include livestock farming, fertilizer usage, and biomass burning. More recently, NH3 emission from vehicles has become a great concern. This unexpected source of NH3 is related to the treatment of other air pollutants of vehicles, NO and NO2, by after-the-pipe catalytic reduction technology. Specifically, NH3 can be formed from the reduction of NO by H2 in gasoline-powered three-way catalytic convertors or from the reduction of NO and NO2 by urea in diesel-powered selective catalytic reduction devices (Farren et al., 2020).

Unlike other gaseous pollutants, O3 is not directly emitted by any sources. It is a secondary pollutant formed from various reactions of gaseous components with the right mixture and ripe condition. In the stratosphere (about 12 km above surface), high O3 concentration is desired because it helps block the short-wavelength ultraviolet (UV) lights in the solar radiation from penetrating to the Earth's surface and affecting living organisms (including us) down here. Up there, O3 formation proceeds mainly via the Chapman mechanism (Chapman, 1930), in which a O2 molecule is split into two triplet state O(³P) atoms upon irradiation by UV lights with wavelengths less than 240 nm; the O(³P) combines rapidly with another O2 molecule to form O3, with collision of a third molecule (normally denoted as M and should be N2 or O2 because of their high abundance) to remove excess energy. The high-O3 layer up there in the stratosphere is perturbed by some other species that can undergo catalytic destruction of O3, including H2O, NOx, and chlorinated and fluorinated carbons (CFCs). The perturbation by the last one, CFCs, was found to be promoted by the polar stratospheric clouds (PSC) in spring time over Antarctica (Molina and Rowland, 1974), resulting a hole with extremely low concentrations of O3 in the stratosphere over the South Pole.

In the troposphere (from the surface to about 12 km), NO2 plays a central role in the formation of O3, whose presence in high concentration is not desired as it can damage respiratory tissues in animals and also tissues in plants. Different from the Chapman mechanism that requires formation of O(³P) from photolysis of O2 molecules (short-wavelength UV lights are needed), O(³P) can also be formed down here in the troposphere by photolysis of NO2, which only requires lights with wavelengths less than 380 nm (Finlayson-Pitts and Pitts, 2000). NO2 is converted to NO in this reaction. The presence of an array of reduced gaseous pollutants in the troposphere, however, amplifies this O3 formation mechanism. First, reduced gases such as CO, CH4, and other VOCs are oxidized by OH radicals to form peroxy radicals, including HO2 and RO2. Then HO2 and RO2 can convert NO back to NO2, which can photolyze to generate O(³P) for O3 formation again. Therefore O3 formation in the troposphere is manifested by the mixture of NOx and reduced gases such as CO, CH4, and other VOCs, with the help of sunlight that triggers the radical reactions. In addition, O3 in the troposphere will also be affected by occasional downward transport of the high concentrations of O3 in the stratosphere.

PM has multiple sources, complex composition, as well as many effects on the environment. Some PM can be produced naturally, for example, biological aerosols that include intact microorganisms (such as bacteria, fungi, and viruses) or debris of plants and animals. Wind-induced dust suspension and bubble bursting on the ocean surface are also processes that produce PM in large quantities. Volcano eruption is also a natural phenomenon that will input a large amount of PM into the atmosphere in a short period of time. Anthropogenic activities, on the other hand, also release substantial amounts of PM. These activities include combustion of various fuels, including fossil fuel and biomass used in some household settings, as well as other industrial processes. The composition of PM is highly complex, which includes primary (directly emitted) components such as BC, crustal materials, and sodium chloride (NaCl, in sea spray aerosols); some PM components are secondarily formed, which include SIA components such as sulfate, nitrate, and ammonium, as well as secondary organic aerosol (SOA) components from the oxidation of VOCs. There are several adverse effects of PM, but in some occasions, it might also bring some benefit to the environments. First, PM can cause respiratory and cardiovascular diseases in human beings. Second, high PM concentration can reduce visibility greatly via light scattering. Third, some PM components, for instance BC and brown carbon (BrC, light-absorbing organic compounds), can absorb solar radiation and warm the climate. On the same subject matter, some other components (e.g., sulfate) can cool the climate by light scattering (direct effect) or cloud formation (indirect effect). Finally, transport of large amounts of dust particles, which contain micronutrients such as iron, from the desert to the oceans is beneficial for the marine ecosystem.

The most abundant VOC in the atmosphere is methane (CH4), with a mixing ratio of 1–2 ppm. This mixing ratio of CH4 nowadays is approximately 2.5 times that of preindustrial era (about 0.8 ppm). On the other hand, the reaction rate constant between CH4 and OH radical (kOH) is on the order of 10−14 cm³ mol−1 s−1, two orders of magnitude lower than those between most other VOCs and OH radical. With high abundance and a slow reaction rate constant with OH (as the main cleansing agent in the atmosphere), CH4 is normally separately discussed from other VOCs. The main source of methane is from the fermentation process of anaerobic bacteria, which are widely distributed in wetland, rice paddy, ruminating animals, etc. Anthropogenic sources of CH4 include burning of fossil fuel and biomass, landfill, and leakage from gaseous fuels. The annual CH4 emission is over 500 Tg yr−1 (teragrams per year). Although CH4 is generally not considered as a serious air pollutant, its effect on climate has attracted growing attention in recent years because of its relatively higher warming potential (>20 times than CO2 on a molecular basis).

The VOCs other than CH4 are normally termed as nonmethane hydrocarbons (NMHCs). In terms of sources, there are biogenic and anthropogenic NMHCs, with the former emitted by vegetation on land or phytoplankton over the oceans, and the latter from human activities. Terminology of biogenic VOCs (BVOCs) and anthropogenic VOCs (AVOCs) is also commonly used in literature to refer to these two important categories of NMHCs. During the growth or under stress, plants will emit various terpenoids, among which isoprenes (C5H8), monoterpenes (C10H16), sesquiterpenes (C15H24), and their slightly oxygenated derivatives are common compound classes. Emission of BVOCs, however, is a natural phenomenon and BVOCs participate in the chemistry of the atmosphere, but is not necessarily considered one type of air pollutants. Anthropogenic activities, such as transportation and various industries, emit a large amount of aromatic (e.g., benzene, toluene, and xylenes) and aliphatic (short- to medium-chain alkanes, alkenes, and alkynes) NMHCs, which are categorized as AVOCs and could (and should) be controlled. There are mainly two impacts NMHCs have on the atmospheric environment. First, NMHCs participate in the formation of tropospheric O3, via the help of NOx and sunlight. Second, NMHCs are also precursors of SOA, which currently contribute substantially to PM2.5 mass in some locations. The SOA is formed via oxidation of various NMHCs, producing products that are more oxygenated thus less volatile; the products so formed are converted to particle phase by either condensation or nucleation.

1.3: Air quality monitoring strategies

There are a number of rationales behind the measurement and monitoring of trace gaseous and particulate pollutants in the air. The first one is to continuously monitor the ambient air quality that affects the health of human population exposed to it. This is usually done by environmental protection agencies and includes a handful of criteria pollutants with well-established linkage to harmful effects on human health. The ambient monitoring serves a number of purposes. It can provide continuous assessment of the extent of air pollution, which is the information that should be accessible by the general public in a timely manner. In addition, frequent evaluation on the trends and improvements of air quality before and after control measures can help evaluate the effectiveness of the policy implemented and provide critical information on amendment of the policy if necessary.

The second one is to sporadically (sometimes continuously as well) measure the emission strength of air pollutants from certain sources for regulatory purposes or compilation of emission inventories. This is usually done by regulatory bodies or research institutes and might cover more pollutants than routine ambient monitoring. The emission measurements are also essential to regulate the amount of air pollutant emitted from certain sources, such as factories and vehicles, operating or new, such that the ambient air quality will not be deteriorated. The compilation of the emission inventories from such emission measurements is also of vital importance in atmospheric models that use a bottom-up approach to assess air pollution levels in the past and in the future.

The third one is to intermittently (ideally continuously too) investigate the physical and chemical processes occurring in the atmosphere by measuring the temporal and/or spatial variations of air pollutants. This is usually done by research institutes and encompasses a wide range of atmospheric trace constituents that are important in regulating the abundance and the impacts of them. The research measurements are also important to understand the dynamics, physical or chemical, of the air pollutants emitted in the atmosphere. Therefore measurements of air pollutants in different scales and scopes are essential for the improvement of ambient air quality and the understanding of the global change of our atmosphere.

The three aspects mentioned before are not completely isolated. For instance, source measurement results can be used to assess ambient monitoring data via modeling, while atmospheric research via advanced measurements can also push the boundary of conventional monitoring strategies and discover emerging pollutants that affect air quality or atmospheric processes. It is the integration of these different approaches conducted by various entities that continuous improvement of air quality can be achieved.

Most traditional air pollution measurements are based on offline sampling of the pollutants and subsequent analysis back in the laboratory. For example, PM can be sampled by drawing a large volume of particle-laden air through a filter, where the particles are intercepted and collected; the PM sample on the filter will be analyzed with various techniques (e.g., chromatography and spectrometry) after solvent extraction and other proper pretreatment procedures. These methods can usually provide very detailed chemical information of the complex pollutants such as PM and VOCs. The drawbacks are, however, they usually have a very low time resolution (hours to days) and might suffer from artifacts during sampling and transport of the samples. These methods are still widely used nowadays when the purpose is to investigate the air pollutants on a molecular level. Since offline analysis is not the focus of the current book, interested readers are directed to other resources for sampling and measurement methods (Forbes, 2015; Lodge, 1988; Wight, 1994).

Although the traditional offline measurements of air pollutants can provide detailed chemical information of the trace atmospheric constituents, the main disadvantage is the low time resolution, which is on the order of 24 h or sometimes more (Li et al., 2017). The dynamics of the air pollution in the atmosphere is a feature that requires fast-responding instruments to capture the rapid changes of the emission as well as the chemistry that occurs after emission. For instance, vehicular emissions peak at rush hours cannot be revealed accurately using traditional offline measurements; the photochemistry-induced secondary formation in the afternoon will also be missed if only low-time-resolution measurements are available. Without such temporally refined measurement results, subsequent modeling or policy making will not be accurately reflecting the emission strength and chemical conversion of air pollutants, thereby missing the keys for the formulation of abatement strategies.

Even with more and more real-time monitoring stations for criteria pollutants, as well as those for other trace constituents, there is still a large gap in our understanding of the air pollution problem—the spatial coverage. Current air monitoring stations, either routine or research oriented, are sparsely located at a handful of carefully selected sites to represent different environments, e.g., urban, rural, or hotspot (Kuhlbusch et al., 2013). The datasets obtained, however, cannot meet the increasing demands in high spatial coverage and resolution. These demands are essential in exposure studies to link public health data and environmental stressors, as well as in validation of models and satellite observation, the latter of which becomes more readily available and can provide unprecedented coverage of the air pollution measurements. There are, to this end, two directions to increase the spatial coverage and resolution of air pollution measurements. The first one is to make measurement platform mobile and more flexible, instead of stationary, such that measurements can be made at different locations in a short period of time to reflect spatial variations of air pollutants. The second one is to develop low-cost sensors and sensor networks to cover a large area with a reasonably large number of measurement nodes. Although there are some studies on these two aspects, there is still a great need to improve the data quality using these two approaches for more comprehensive spatial measurements of air pollutants.

1.4: Modeling and forecasting of air pollution

Air quality modeling and forecasting aim to fulfill the needs of citizens and government departments to know about the historical and future variation of the ambient air quality at a given place over a specific time horizon. For the citizens, especially the sensitive groups with heart diseases or respiratory diseases, they may need to know the future air quality index for planning of outdoor activities during days of bad air quality. As for the government departments releasing this public information, it is obliged to develop an air quality forecasting system which predicts the variation of the concentrations for criteria air pollutants (O3, PM2.5, PM10, NO2, SO2, CO) over the next 24 h or even further. There are two main approaches to develop the air quality forecasting system. The first approach is the statistical modeling approach. In this approach, an empirical model or an ensemble of models is developed to predict the pollutant concentration based on past concentrations of the target pollutant, precursor concentrations, and other relevant meteorological variables that can influence the transport, dispersion, and removal of the target pollutant (e.g., wind velocity, rainfall, temperature, and relative humidity). The second approach is the chemical transport modeling approach. The chemical transport model is a 3-D Eulerian/Lagrangian air quality model which simulates the pollutant concentration for the model domain or the following air parcel based on the emission model, the meteorological model and the chemical model (Seinfeld and Pandis, 2016). Table 1.1 presents the comparison of both modeling approaches. There is no definite answer to say which approach is better. Different modeling approaches can be adopted according to available resources and the needs of the users. However, the introduction of the forecasting techniques in this book is oriented to the statistical air quality modeling and forecasting.

Table 1.1

Traditional statistical models of ambient air quality forecasting for pollutants such as PM2.5 and other gaseous pollutants include the multiple linear regression, the classification and regression tree (also called the decision tree), and the multilayer perceptron (artificial neural network). These techniques were widely applied to air quality forecasting since the 1970s and the 1990s, respectively (Ryan, 2016; Li et al., 2022). The multiple linear regression models the pollutant concentration based on a linear combination of the input variables. The classification and regression tree performs a recursive partitioning of the input space so that different partitions can use their own models (e.g., multiple linear regression models with different model coefficients). The multilayer perceptron is a type of feedforward artificial neural network that models the pollutant concentration based on the linear combination of the network outputs from the hidden neurons in the network. Each hidden neuron describes a particular nonlinear relationship between the input variables and the pollutant concentration. The aforementioned models require training with the input–output data. The coefficients in the model are adjusted in order to achieve the least square modeling error between the measured output and the predictions for the given inputs. The trained model is applied to forecast the air quality based on the model parameters estimated or learnt from the training data.

In a statistical air quality forecasting model, the air pollutant concentration to be forecasted can be related to many potential meteorological factors. When the choice of the model is too simple, it tends to underfit the training data with a large fitting error. On the contrary, a complex model may overfit the training data with little fitting error but gives inaccurate predictions during validation with unseen data. Forecasters need to decide the appropriate combination during the model development. Apart from that, another challenge is usually met even when the input variables are systematically selected, i.e., the models are nonadaptive. The model parameters are obtained based on a given set of training data and will be fixed in the operational forecasting. The drift in the model coefficients during operational forecasting can introduce forecast error (Ryan, 2016). In more complex situations, the air pollutant can be related to different factors throughout the year. This means different adaptive seasonal models or transitional model across the seasons are suitable to describe the pollutant behavior in different periods of the year. To tackle these challenges, the Bayesian approach is used in this book to select an efficient and robust input combination in the forecasting model. Apart from that, the adaptive statistical air quality model will be employed to tackle the problem of model drift (also called concept drift in machine learning) by changing the model coefficients adaptively at each time step. Finally, the adaptive Bayesian model averaging will be used to produce a weighted forecast from different model candidates (e.g., seasonal models and transitional model) according to the importance of their predictions.

1.5: About this book

This book starts with a brief introduction of the air pollution problem and the currently available strategies in air quality monitoring, as well as a brief account on modeling and forecasting of air pollution (this chapter). Chapter 2 deals with measurement methods for criteria pollutants, focusing on the real-time methods that are widely adopted in regulatory measurements stations, as well as methods for chemical composition of the more complex particulate matter (PM) and volatile organic compounds (VOCs). Chapter 3 provides some details in the emerging air quality monitoring methods, including low-cost sensors and sensor networks, as well as mobile measurement platforms in different scales (on-road, manned aircraft, and unmanned aircraft, etc.). Chapter 4 provides the overview and comparison of the statistical air quality models (including multiple linear regression, the classification and regression tree, the multilayer perceptron, and the support vector regression) widely used in the field over the last few decades. Chapter 5 presents the advanced topics that deal with the selection of input variables in the linear regression-based air quality model, the automation of model retraining with the dynamic linear model and the time-varying multilayer perceptron, and finally combining forecasts from the dynamic linear models with the Bayesian model averaging.

References

Chapman S. A theory of upperatmospheric ozone. Mem. Roy. Meteor. 1930;3:103–125.

Farren N.J., Davison J., Rose R.A., Wagner R.L., Carslaw D.C. Underestimated ammonia emissions from road vehicles. Environ. Sci. Technol. 2020;54:15689–15697.

Finlayson-Pitts B.J., Pitts J.N. Chemistry of the Upper and Lower Atmosphere: Theory, Experiments, and Applications. San Diego: Academic Press; 2000.

Forbes P.B.C. Monitoring of Air Pollutants: Sampling. 2015 Sample Preparation and Analytical Techniques.

Jacobson M.Z. Atmospheric Pollution: History, Science, and Regulation. Cambridge: Cambridge University Press; 2002.

Kuhlbusch T.A.J., Quass U., Fuller G., Viana M., Querol X., Katsouyanni K., Quincey P. Air pollution monitoring strategies and technologies for urban areas. In: Viana M., ed. Urban Air Quality in Europe. Berlin Heidelberg, Berlin, Heidelberg: Springer; 2013:277–296.

Li Y.J., Sun Y., Zhang Q., Li X., Li M., Zhou Z., Chan C.K. Real-time chemical characterization of atmospheric particulate matter in China: a review. Atmos. Environ. 2017;158:270–304.

Li Y., Guo J.E., Sun S., Li J., Wang S., Zhang C. Air quality forecasting with artificial intelligence techniques: A scientometric and content analysis. Environ. Model. Software. 2022;149:105329.

Lodge J.P. Methods of Air Sampling and Analysis. third ed. Routledge; 1988.

Molina M.J., Rowland F.S. Stratospheric sink for chlorofluoromethanes: chlorine atom-catalysed destruction of ozone. Nature. 1974;249:810–812.

Ryan W.F. The air quality forecast rote: Recent changes and future challenges. J. Air Waste Manag. Assoc. 2016;66:576–596.

Seinfeld J.H., Pandis S.N. Atmospheric Chemistry and Physics: From Air Pollution to Climate Change. third ed. New Jersey: Wiley; 2016.

Wight G.D. Fundamentals of Air Sampling. first ed. CRC Press; 1994.

Chapter 2: Current air quality monitoring methods

Abstract

We describe in this section the methods for monitoring criteria air pollutants, including carbon monoxide (CO), sulfur dioxide (SO2), nitrogen oxides (NOx = NO + NO2), ozone (O3), and particulate matters (PM10 and PM2.5). While the most widely used federal reference method (FRM) or federal equivalent method (FEM) are described in detail, other methods of less popularity or those developed in research community as custom-built instruments will only be briefly described. For the latter ones, readers are directed to the literature cited for more details in the principles and configuration.

Keywords

Criteria air pollutants; Air pollution monitoring; Spectroscopy; Mass spectrometry; Interferences

Contents

2.1Methods for criteria air pollutants

2.1.1Carbon monoxide (CO)

2.1.2Sulfur dioxide (SO2)

2.1.3Nitrogen oxides (NO and NO2)

2.1.4Ozone (O3)

2.1.5Particulate matters (PM10 and PM2.5)

2.2Real-time chemical composition monitoring

2.2.1Particulate matters

2.2.2Volatile organic compounds

2.2.3Other real-time techniques

2.3Conclusions

References

Among many trace species in the atmosphere, several pollutants are considered as criteria pollutants and their concentrations are normally continuously monitored by environmental protection agencies in different countries and regions. This is because there is mounting evidence showing that these criteria pollutants are associated with certain adverse health effects. These criteria pollutants include carbon monoxide (CO), sulfur dioxide (SO2), nitrogen oxides (NOx = NO + NO2, in which only NO2 is a criteria