Air quality citizen awareness: An explorative study on what to measure, where to measure and how to present it?

(1)

Master Thesis

HALMSTAD

UNIVERSITY

Master's Programme (60 credits) in Applied Environmental Science 60 credits

Air quality citizen awareness: An explorative study on what to measure, where to

measure and how to present it?

Degree Project in Environmental Science 15 credits

Halmstad 2020-06-23

Lesley Zvikomborero Hakunavanhu

(2)

1 | P a g e

Air quality citizen awareness: An explorative study on what to measure, where to measure and how to

present it?

Halmstad University: School of Business, Engineering and Science Masters in Applied Environmental Science

28 May 2020

by, Lesley Zvikomborero Hakunavanhu

Supervisor: Sylvia Waara

(3)

2 | P a g e

Abstract

Air quality in general is a hot topic that is significantly linked to respiratory diseases that have been causing a decrease in life expectancy around the globe. Over the past three decades, European cities have done air quality monitoring using one or two air quality instruments per 249 thousand inhabitants. The instruments are big, rigid and expensive. This thesis focuses on new ways of air quality monitoring using air quality sensors that are small, cheap and flexible. An investigatory approach was used through taking insights from citizens to get an understanding of what air pollutants to measure, where to measure and how to present the information in ways that are clear and easy to understand. From a survey of 82 participants, results showed that 74% of the people are concerned about air quality and they would like to receive information in ways that are easy to understand like colours (red for bad or green for good), numbers (µg/m³) and graphs. The citizens identified carbon dioxide (40.6%), particulate matter (25%), nitrogen oxide (18.8%) and ozone (9.4%) as pollutants that need critical attention which can be measured in places that they spend a lot of time in for example playgrounds and parks as well as near sources of pollution like busy roads and industries. The survey was followed by an analysis of eight sensors that were put in the urban environment measuring PM1, PM2.5 and PM10 together with environmental factors (temperature and humidity) to find how the information can be used for citizen air quality awareness and if was accurate enough. The sensors could accurately detect particulate matter variations from all the places. All environmental factors (temperature, humidity and wind) significantly affected particulate matter (p ≤ 0.05) variations. However, associations with particulate matter were weak to moderate (r = 0.02 to 0.46) which were influenced by the surroundings in the locations of the sensors. A strong correlation with municipal refence instruments and the ability to detect pollution variations is enough accuracy for sensors to be used for awareness which the sensors did. In conclusion, the sensors can be used for air quality citizen awareness and for correct sensor placement it is important to do a background study of the area including where people spend most time and sources of pollution if they are to be used. Therefore, using sensors will bring more awareness in air quality monitoring by measuring air pollution concentrations in localised places.

Key words: air quality, citizen awareness, explorative

(4)

3 | P a g e

Acknowledgements

First and foremost, I would like to express my gratitude to my professor Silvia Waara who supervised me through the thesis up to the final submissions. I thank her for her patience and motivation, I could not have imagined having a better mentor for the thesis.

My gratitude also goes to the company I was working with on the thesis for providing all materials and data required for the thesis. I thank them for the opportunity to work with them and the experience gained is priceless;

not only did they improve the quality of my thesis but also helped me develop critical thinking skills as an environmentalist.

Special mention goes to the Swedish Institute for awarding me a scholarship to study in Sweden. This opportunity was life changing to me, I gained a lot of skills that I can use to contribute to Sustainable Development in my home country and the world at large.

Lastly, I would like to thank my family, friends and my better half Natasha Pembere, for their emotional support as I went through the crazy deadlines during the thesis, their contribution is worth mentioning.

(5)

4 | P a g e

1. Introduction

Air pollution poses as one of the major threats to human health in the world, causing respiratory complications that lead to 4.2 million fatalities every year (WHO, 2019). Researchers have found positive and statistically significant associations between various pollutants like particulate matter (PM), Volatile Organic Compounds (VOCs), ozone (O3) and oxides of nitrogen (NOx) and health complications like asthma, cancer, lung problems, osteoporosis, metabolic dysfunctions and type 2 diabetes just to mention a few (Alderete et al., 2018, Chen et al., 2017, Brook et al., 2010 &Im et al., 2018). These health problems have been more severe in children as they are closer to the ground and have respiratory systems that function faster than adults’ thus, they are more susceptible to infections including pneumonia, bronchitis and asthma (Bates., 1995). Europe has been adversely affected by pollution related diseases, causing European countries to spend an average of EUR 1 billion per year on health-related problems from air pollution (European Commission report on air quality., 2018). In a survey done in Eurobarometer (2019) with 27.5 thousand respondents from 20 member states, results showed that more than 50% of the respondents thought that respiratory diseases from air pollution are a very serious problem in their countries. More so, 54% of the respondents felt that there is lack in awareness towards EU air quality problems in their countries (Eurobarometer., 2019). Therefore, it is important that research and innovations should focus on air quality monitoring to analyse the pollutants that are affecting people and put out the information so that people can take more informed decisions and feel confident with their health in relation to pollution.

It is in urban environments where air pollution is having more adverse effects because of the ongoing advances in industrial production and urbanization. Dandotiya (2019) concluded that air pollution in urban environments is mainly coming from traffic, combustion, domestic fuel burning, natural dust, and industrial activities. Furthermore, European countries exceed PM2.5

air quality standards outlined by the air quality directives, with 77% of the urban population exposed to concentrations higher than the WHO air quality guidelines. Therefore, countries have taken the initiative to monitor air pollution through being legally bound by regulations like the EU Ambient Air Quality Directive which has limits and target values for concentrations of major air pollutants (Judith et al., 2017). The directive states that monitoring and measuring air pollution should be done by at least one or two air quality stations per 249 thousand inhabitants which are mostly large stationary municipal air quality monitoring stations (European Environment Agency., 2019). These have helped reduce emissions for the past three decades through the adoption of the first EU air quality directive. However, one air quality monitoring station in an urban environment with 249 thousand inhabitants or less does not detect events in which inhabitants are exposed to levels higher that WHO air quality guidelines.

Further to this, European countries have been less ambitious in taking strict measures towards PM1 and also the air quality directives do not have regulations for very fine particles like PM1

which is also the same for regulations in other parts of the world (European Commission., 2018). Munir et al., (2019) observed that urban environments have spatial variability in air pollution levels and the stationary stations would be difficult use for a dense network setup for monitoring air pollution. It would be difficult to do comprehensive air quality monitoring in urban areas considering that there are various sources of air pollution and only one station

(7)

6 | P a g e would not be enough therefore there is need in improvement by monitoring using more than one station.

Traditional air quality monitoring stations are known for their limitations of being big, expensive, and prone to failure to identify micro level drivers of air pollution in urban environments. They are therefore not practical to use when conducting pollutant measurements in localised areas like the city centre or playgrounds or schools (Santiago et al., 2013). Cities have therefore started to find other ways of monitoring air quality systematically using air quality monitoring networks, air quality modelling and data assimilation. They are using different types of technologies which have brought new and advanced ways of collecting air pollution data (Kaivonen & Ngai., 2020, Moore et al., 2012 & Kamal et al., 2019).

Chikhooreeah & Khedo (2017) describe an air quality networks as a connection of sensors that are spread over a geographical area to record environmental conditions, such as air pollution.

Air quality networks have gained popularity because of their flexibility and efficiency. They then pave a way for air quality modelling which involves the prediction of air pollution variations through data assimilation and, in this case, air quality data assimilation refers to observations from ground and satellite stations to predict trends in pollution levels that can be presented on maps, graphs or tables (Mailler et al. 2017). Recent studies on air quality monitoring systems have been using sensors to measure emissions in urban environments placing them on roadside, kerbside, buses, bus stops, traffic lights and buildings (Ellermann et al., 2018 & Moore et al., 2012).

The history of air quality monitoring using sensors can be traced back to 1800s, but their intensive use started in Britain in the 1980s when they were used in mines to detect carbon dioxide, carbon monoxide, and methane (Tim Dye., 2017). However, throughout the 19^th and 20^th centuries, their uses were more popular in companies and as personal devices (Tim Dye., 2017) until in the last decade when they were now being popular for air quality mapping in cities for example in smart city projects like the Array of things (2013) and CITI-Sense (2016).

Since the last decade, cities and governments research on the use of sensors for air quality monitoring in urban environments and presently, the use of sensors is a new technical approach to air quality monitoring that needs a lot of attention. For example, the United States of America government endorsed a legislation supporting the use of sensors for public health and environmental management goals in urban environments (California Legislative Information., 2017). Because of the above-mentioned factors, it is important to explore the subject of air pollution monitoring using sensors in urban environments.

The aim of this thesis is to find out where exactly to place the sensors, what to do with the information and the needed level of accuracy for the sensors to be used for air quality awareness. Therefore, the study intends to assess the usefulness of the employed network of air quality sensors for measurement of particulate matter of different sizes. During the study environmental factors (temperature and humidity) will be measured and relationship between air pollutants and the environmental conditions will be explored. This master thesis is part of an ongoing exploratory study made in a department working with concept development and will seek to answer the following questions in a generalized approach to find indicative answers, that will work as eye openers and as a basis for future work.

(8)

7 | P a g e

1.1 Research questions

1. Which pollutants are of interest to urban inhabitants, and what is their interest in air quality awareness?

2. How should this information be presented to the inhabitants, and what accuracy is expected?

3. What are levels, and their variation, of PM1, PM2.5 and PM10 as measured by the sensors at 8 sampling sites within the city?

4. In which urban environments could the sensors be placed to give interesting air quality information to the inhabitants’?

(9)

8 | P a g e

2. Material and Methods

The company which will remain unnamed had prior to this thesis work placed eight particulate matter sensors in an urban environment, and there existed different ideas concerning the use of information from these sensors. This thesis is a part of an ongoing project in the company, and I decided to work on the citizen awareness use case. The thesis used an exploratory approach in defining and understanding problems whilst seeking inspiration from users or other developers then testing different solutions to come up with more refined ideas. During the time of the thesis there where attempts to add other sensors as well, but due to technical problems this study became limited to study measurements of particulate matter, relative humidity and temperature. However, other pollutants were included in questionnaires and a literature survey to give a more complete picture.

The project started with meetings that were backed by extensive research on how other cities and companies had concluded similar research. The research and inspiration from other case studies showed the importance of citizen involvement and thus it was concluded to explore the use of air quality information for citizens awareness. Therefore, there was a need to conduct research on what people want to know about air quality. Questionnaires were designed to get feedback from representative users and from these questionnaires useful insights oriented towards air quality awareness were received.

2.2 Study area

The study was conducted in an urban environment covering about 22km² with a population of

~100 000 inhabitants in Southern Sweden. It experiences an oceanic climate with warm summers averaging high temperatures of 23°C and winters with temperatures around −1 to 3

°C. The area has mixed residential and commercial areas, with raised buildings.

2.3 Initial research, investigation & preparation for inspiration

Limited experience and prior knowledge in the field of the use of sensors in air quality monitoring made initial research and preparation crucial. In the first two weeks, there was intensive research and discussions around how cities are embracing the subject of air quality monitoring using sensors. One of the first steps was to investigate current use cases in the targeted field as well as related areas. Inspiration was drawn from smart cities projects relating to air quality monitoring. The research was focused on understanding what other cities and researchers are doing in air quality monitoring using sensors. After the two weeks of deep research and analysis, seven articles in Table 2 and ten case studies in Table 1 were useful as starting points to what the thesis should focus on. Based on this, I decided to focus on the citizen awareness use case. All results and discussions in this thesis will then be focused on how air quality information would be used in order to make citizens aware of what they are inhaling and take informed decisions that would benefit them.

(10)

9 | P a g e Table 1. Case studies used for initial inspiration

Question Case

What are the new ways of monitoring in smart cities?

Array of things., (2013) Village Green Project., (2019) Dubai Air Quality Network., (2011) City flow., (2020)

London Government (2019

What can we do with the air quality information?

Clean rider mapper., (2015)

Smart & Connected - City of Stockholm., (2017) Hackair., (2018)

London Government (2019 See the air., (2019)

Aeroqual outdoor air pollution case studies., (2020)

The case studies in Table 1 proved to be useful tools for identification of other citizen awareness projects but varied greatly in technical complexity and size. These case studies were used to support and discuss my answers to the research questions.

The seven selected articles were found through a search process and the search scope included these terms:

• Network of sensors measuring particulate matter

• Ambient air pollution in an urban environment

• Environmental factors such as wind speed and direction, humidity or temperature

• Placement of sensors

After considering the desired characteristics, searches were made in Web of Science, One Search and Google Scholar. Selection of articles to reduce the number of results was based on, environmental monitoring subject, air quality subject, urban environment, air quality awareness strategies and studies which were highly cited. For this selection, highly cited articles were those considered the most recent in 10 years of publications and are regarded as articles of scientific excellence with top performance. To narrow down to the choice of articles in Table. 2, the abstract, methodology and parts of results were read checking for the desired characteristics. The European air quality directive 2008/50/EC was searched and downloaded on the European Commission website.

(11)

10 | P a g e Table 2. Search results and selection of articles used for initial inspiration

Database Word search Results Selected

Web of science Air quality, OPC, urban environments 154 Penza et al., (2017) Azimi et al., (2018)

Shafran-Nathan et al., (2019)

One search Network of sensors, OPC, urban 82 Munir et al., (2019) Malings et al., (2020)

Google scholar Searched by title Ramos et al., (2018)

Lemeš., (2018)

2.5 Air quality awareness survey

It was important to know what the users would be interested in when it comes to air quality awareness. Two digital questionnaires were sent to a small selected group of people who represented the anticipated target audience; see appendix. This was done to gain insights as well as possibly confirming expected needs of the target group. The sample consisted of 87 participants who were selected on a voluntary basis using the stratified random sampling method which involves dividing the target population into groups with the same characteristics (Ding, 2020). In this case, work colleagues, classmates and schoolmates were divided into two groups of people with environmental background and people without environmental background to give insights on what they would like to know about air quality. It was also believed that this questionnaire could potentially yield some insights or inspirations for the product which could add value.

2.6 Questionnaire data collection

Data was gathered from online questionnaires after the development of two questionnaires and was based on user feedback questions as defined by Portigal, (2014) who explained the use of these questions as a guide on how to uncover compelling insights from users. The online questionnaire was created using the google survey tool and contained both open ended and closed questions. The questions were sent out online for data collection and it was open for the whole period of the study. The questionnaires were designed to cater for people with environmental background (18 questions) and those without environmental background (15 questions) which made it possible to present different views of the topic in a detailed way, see appendix A and B. The group with an environmental background was selected from people who had studied or worked with environmental science for more than a year. People with an environmental background were given more technical questions whilst people without environmental background were given more general questions. This was done to be able to get the best information according to one’s knowledge of air quality and to compare the views of people with and without environmental background.

(12)

11 | P a g e

2.7 Field measurement methods

Prior to this thesis, the company placed eight Optical Particle Counters (OPC) measuring inhalable particulate matter (PM) with diameters of 1 micrometre, 2.5 micrometres and 10 micrometres which are referred to as PM1, PM2.5 and PM10 throughout the study. The OPCs were placed around the city (Table 3) and were used to collect data for a period of two weeks, between 1 April to 17 April 2020. OPCs are small sensors that are based on the principle of light scattering from particles. They are instruments that are typically used to measure particles ranging between 0.35μm to 40μm in diameter and can measure concentrations up to 2000μg/m3. The instruments were factory calibrated thus there were no further calibrations made. One sensor (sensor 4) was placed next to the municipality reference instrument station and was used as a benchmark sensor for the analysis of other sensors.Reference data for PM10

was obtained from the municipality air monitoring station located in the city center. The municipality PM meter utilizes the Tapered element oscillating microbalance method (TEOM) which is a technique used to continuously measure the weight of air particles. It makes use of a size selective inlet that lets in PM10 particles and is mounted at a height of 4m above the ground. The selection of sites for placement of the particulate matter sensors was done by the company in collaboration with possible future stakeholders. The company’s selection criteria for sites were places likely exposed to the highest levels of air pollutants, however, the ease of sensor mounting was also taken into account. Please see Table 3 for an overview of the sensor placements.

Table 3. Location of the 8 sampling points

Sensor number Location

1 3m above the ground at a parking space fence, 30m close to a highway (~40 000 vehicles daily) that is 3.9m above the ground

2 On an office window 6m above the ground, 1m from a parking space ~300m from a highway

3 3m above the ground at a parking lot fence in the city centre, 40m away from a railway

4 At a sports field fence, 4m above the ground and 2m away from a busy road

5 Parking space 3m above the ground and 50m from the highway (~40 000 vehicles daily).

There are tall trees between the sensor and the highway

6 At the window of a building 6m above the ground and close to a parking space.

7 At an open space on a fence 3m high, 32m from the highway. The ground is not even so it is ~5m above the road

8 3m above the ground in a school courtyard 100m from the city park

(13)

12 | P a g e

2.8 Environmental conditions

Environmental parameters of wind, temperature and relative humidity of the ambient air were measured for the study. Temperature and humidity sensors were put in the box with the PM recording data 10 times per hour. The sensors have a relative humidity range of 0% to 100%

and a temperature range of -40^oC to 125^oC. Data for wind speed and wind direction during the period of the study was also obtained from Swedish Meteorological and Hydrological Institute website and they served as reference data when analysing variations in concentrations.

2.9 Data analysis

Data analysis was done in IBM SPSS 26 and Microsoft Excel to explore the resulting data set in the following ways: (1) patterns of relative differences of measured parameters; (2) box plots and summary statistics for each place; (3) descriptive statistics for frequencies and (4) statistical significance testing of measurements compared to place and records of wind speeds, temperature and humidity.

To evaluate the statistical significance of the location comparisons, one way anova (Tyrrell., 2008) and post hoc test were used to make paired comparisons of each simultaneously measured parameter (e.g., PM1, PM2.5 and PM10 measured at either per hour or per day intervals) across the eight places. To evaluate the statistical significance of comparisons between parameter measurements and place, wind speeds, and wind directions, the nonparametric Spearman rank correlation coefficients (Tyrrell., 2008) was calculated between hourly averages of variables. Outliers were identified using SPSS descriptive statistics from the explore function and from boxplots. It was not possible to show graphs with high outlier values, thereby the method of replacing an outlier with the mean was used to have better graphical presentation (Lakshmanan., 2019).

The google reports platform that displays statistics, percentages and graphs was used to analyse the surveys and produce results for the desired questions.

(14)

13 | P a g e

3. Results

This section reports the results of the analysis from the questionnaires with the support of initial case reviews that led to the insights which managed to narrow down the focus of the study.

Following is a presentation of PM1, PM2.5and PM10 together with environmental data recorded between 1 April and 18 April 2020. Firstly, there is the presentation of temperature and humidity data for the period of the study is presented as box lots (Figure 3) that show the interquartile range (IQR) as boxes, median values (lines within boxes) and upper and lower adjacent values. Outliers were excluded for clear graphical presentations. Wind data is shown in form of a wind rose (Figure 4). Following is a section showing the systems’ ability to correctly detect particulate matter concentrations (Figure 8 & Figure 9) and its ability to predict pattern variations in different places (Figure 10 & Table 6).

3.1 Target audience

A total of 87 people answered the questionnaires, (55 without environmental backgrounds and 32 with environmental background). 73.5% of these people were students aged between 18 to 34, 22% were employed and 4.5% were either self-employed or unemployed. 92.5% of the respondents live or work in urban Sweden whilst the remaining 7.5% were from urban areas in Europe and Africa.

3.2 What should be measured?

The question of what pollutant should be measured was posed to people with environmental background who were in beginning and advanced stages in their careers with 1 to 2 years’

experience (56.3%), 3 to 5 years’ experience (28.1%) and over 5 years’ experience (15.6%).

This question was structured in such a way that the respondents could select the most important pollutant and suggest their own such that there was no limitation to the answers that they would give. Figure 1 Shows that 40.6% of the respondents believed that carbon dioxide was the most important pollutant to monitor followed by 25% who believed particulate matter was important to monitor. Nitrogen and ozone received concern from 18.8% and 9.4% respectively whilst 2 respondents believed that every pollutant is equally important to be monitored and that air quality monitoring depends with what you want to use the information for (e.g. road mapping, citizen awareness, town planning, etc.).

Figure 1. Results of questionnaire: Pollutants that should receive attention and more monitoring according to the people asked.

(15)

14 | P a g e

3.3 Level of air quality concern

Overall, respondents were concerned about the environment but people with environmental background had deeper concern and deeper interest in the environment and this was expected as people working in a subject area tend to be more knowledgeable and interested in it (Figure 2). 62.5% of the respondents without environmental background had never checked air quality information and 54.2% never knew that they could access air quality information. However, 77.1% would like to know more about air quality. Opposite to people without environmental background, 85.7% of respondents with environmental background had access to air quality information. All respondents showed interest in getting information about air quality for health purposes and for avoiding polluted areas.

a)

b)

Figure 2. Level of interest for air quality information by (a) people without an environmental background and (b) people with an environmental background.

3.4 Where to measure?

In the multiple answers 70.3% of the respondents marked industrial areas as places where there should be more air quality monitoring. This was followed by monitoring near busy roads which was marked by 40.6% of the respondents; however, this was influenced by the group of people with an environmental background in which 71.9% marked monitoring near busy roads whilst 9.3% of respondents without environmental background marked monitoring near busy

(16)

15 | P a g e roads. Parks and playgrounds received a lot of attention also with 47.9% of respondents being concerned about monitoring at these places meaning that they are concerned about air quality around areas in which they spend a lot of time in and places of interests might also include city centers. Below is a table with a ranking of the measured places (Table 4) whilst keeping in mind that information from the sensors will be used for air quality awareness to the citizens.

All findings from this thesis was used to make the ranking and rank 1 is the highest rank in order of importance.

Table 4. Ranking of the measured places in order of importance and recommendations of improvement for air quality awareness.

Rank Sensor Reason Recommendation

1 8 Showed high levels of PM10 in a school courtyard were kids spend most of their days in. Surpassed the PM10 EU Hourly Air Quality limit once

A camera can be installed at this place to investigate what is going on if extreme values show (presents of school children in this place gives it high priority).

2 4 At a sports field were people spend more than 2 hours inhaling at high rates and the municipal reference instrument showed levels surpassing the PM10 EU Hourly Air Quality limit

Pollution is influenced by high traffic volumes; recommendations can be made to try and reduce traffic flow.

3 5 Surpassed the PM10 EU Hourly Air Quality limit twice and is close to office areas

Highly influenced by traffic volumes, more awareness should be emphasised during peak hours.

4 1 Surpassed the PM10 EU Hourly Air Quality limit once

Highly influenced by traffic volumes, more awareness should be emphasised during peak hours.

5 2 & 6 In transitional places near roads The sensors showed a strong correlation r² = 0.95 which was good for the analysing of sensor functionality, however, the other sensor can be removed and placed in another area.

6 3 Showed the second lowest PM1, PM2.5 and PM10

volumes throughout the study

The sensor is in a confined place, there is a park nearby, this sensor can be more useful if placed in a park.

7 7 Out of the city, in an open space and showed the lowest PM1, PM2.5 and PM10 volumes throughout the study

This can be used as a sensor measuring urban background and should be placed downwind where pollutants from various sources would have mixed.

3.5 Presentation of air quality information

It is important that people should understand air quality information otherwise they cannot make informed action towards what they are inhaling. 64.8% of respondents without an environmental background did not have access to any air quality information and 43.3% of them do not understand any air quality information. This led to the question on how they would appreciate the presentation of air quality information and 48.1% preferred colors (e.g. red for bad or green for good) whilst 33.3% preferred numbers in µg/m³. Only 7.4% preferred information to be presented using emojis whilst some suggestions came from individuals

(17)

16 | P a g e preferring both colors and number and as illustrations (graphs). Unlike respondents without an environmental background, respondents with environmental background preferred numbers in µg/m³ (43.8%) whilst 37.5% of the group preferred colors. Individual suggestions from respondents with an environmental background were also attractive infographics and both numbers and colors.

3.6 Environmental conditions during the period of the study

Temperature and humidity recorded at each of the places during the period of the study is presented by Figure 3 (2 weeks’ time series data for temperature and humidity are shown in Figure 5). The temperature recorded from all places during the period of the study ranged from 2.4ôC to 26.4ôC with an average of 11.5ôC. The temperature difference from the average amongst the places was +/- 5ôC. The average relative humidity recorded from all places during the period of the study was 51.5% with 20.8% and 82.6% recorded as the lowest and highest values. Sensor 3 experienced the lowest relative humidity average of 64,3% followed by sensor 8 with 74.5% whereas the rest of the sensors had close averages between 78% and 83%. There were prevailing westerly and north-westerly winds during the period of the study as shown by Figure. 4. The most common wind was the strongest (8-10m/s) which blew 23% of the study period followed by the winds with 6-8m/s windspeed which blew 19% of the study period.

Winds with 2-6m/s blew the least making 15% of the study period. It was windy during the study and no calm days were recorded. Hourly averages of PM1, PM2.5 and PM10 were correlated against environmental conditions to investigate how these conditions might have affected variations in the measured parameters (Table 5)

PM1, PM2.5 and PM10 were selected as dependent variables whilst temperature, humidity, windspeed and wind direction were selected as independent variables. All environmental factors had significant effects on the variations of PM1, PM2.5 and PM10 (p < 0.05) with temperature and humidity showing positive association whereas windspeed and wind direction showed negative relationships. However, the relationship between the parameters and temperature were weak r = 0.093 (PM1), r = 0.025 (PM2.5) and r = 0.084 (PM10). Unlike temperature, humidity showed a more moderate association with r = 0.388 (PM1) and r = 0.463 (PM2.5) although the relationship with PM10 was weak (r = 0.183). Association with windspeed was also weak for all the parameters showing r = -0.183 (PM1), r = -0.155 (PM2.5) and r = - 0.047 (PM10). Likewise, wind direction had weak associations of r = -0.277 and r = -0.121for PM2.5 and PM10 respectively but had a moderately strong association with PM1 (r = 0.375).

(18)

17 | P a g e a)

b)

Figure 3. Temperature and humidity measured on each of the 8 places from 1 April to 17 April. The middle line represents the median, the boxes show the 25th and 75th percentiles whilst the lines show upper and lower adjacent values. (outliers were excluded for graphical presentation)

(19)

18 | P a g e Figure 4. Wind data recorded during the period of the study. The colors (blue, orange, grey & yellow) represent windspeed in m/s and wind direction whereas the numbers in percentages represent the prevalence of the wind.

Figure 5. Hourly averaged temperature (black) and humidity (red) recorded from 1 April to 17 April for all the 8 sensors.

Table 5. Correlation and significance testing between measured variables and environmental conditions.

Numbers in brackets show p-values and numbers that are not in brackets show R-values.

Parameter Temperature Humidity Windspeed Wind direction

PM1 0.093 0.388 -0.183 -0.375

(0.001) (0.001) (0.001) (0.001)

PM2.5 0.025 0.463 -0.155 -0.277

(0.007) (0.001) (0.001) (0.001)

PM10 0.084 0.183 -0.047 -0.121

(0.001) (0.001) (0.008) (0.001)

(20)

19 | P a g e

3.7 Accuracy in detecting particulate matter variations

Figure 6 Shows a linear relationship between PM10 concentration of the municipality reference station and sensor 4 which were at the same position (~1,5m apart). There is a strong positive association (r² = 0.648) between the two instruments. The reference sensor managed to follow the flow variations which were recorded by the municipality reference station (Figure 7a).

Sensor 4 showed a larger difference in PM10 concentration volumes on 3, 6, 7, 8, 14 and 15 April during the period of the study. During these days there were offsets of +/-40µg/m³ in PM10 concentrations shown on the reference sensor compared to the municipality reference station but the trend was the same. For quality control purposes, sensor 2 and sensor 6 were put close together (4m apart) in an open area without the interference of buildings of trees. This was done to show that difference sensors measure the same particulate matter variations if put in a localized area with the same air quality conditions. Results showed a strong positive correlation (r = 0.96) and the sensors followed the same pattern of PM10 concentration during the period of the study (Figure 7b)

Figure 6. Relation between hourly PM10 recorded from the municipal reference station and the sensor 4 showing correlation between 1 April to 17 April. The black dots are PM10 values and the black line represents a regression line (r2 = 0.648).

a)

(21)

20 | P a g e b)

Figure 7. a) Sensor 4 (black) follows PM10 variations depicted by the municipal reference station (red) managing to measure and display variations for the period between 1 April and 17 April, b) Sensor 2 (black) and sensor 6 (red) (34m apart) manage to follow the same PM10 flow variations.

3.8 Prediction of variations that significantly affect human health

Based on the assumption that peaks show instances with high particle concentration (and not noise), it can be assessed whether the sensors could correctly detect concentrations that significantly affect human health. This assumption can be made since the dependency of humidity, temperature and wind was shown not to give large deviations in particle concentration, please see section 3.6. It was important to assess whether the sensors can be used to pollution peaks to give awareness if there are high pollution levels in an area. Thus, test with PM10 against EU Air Quality Limitation of 50µg/m³ were done. The sensors can show PM10 hourly variations that exclude the stipulated EU air quality limit value of 50µg/m³ (Figure 8) which is then useful in awareness purposes. Sensor 1, 5, 7 and 8 were used as examples to show days were extreme PM10 peaks were recorded during the study. The sensors recorded 100µg/m³, 58µg/m³, 65µg/m³, 80µg/m³ and 204µg/m³ on different days (8, 9, 13 & 16 April) in different places during the study. These days and places were randomly chosen as examples to show the functionality of the sensors. Overall, the highest hourly concentration recorded during the study was the 204µg/m³ recorded from sensor 8 on the 16th of April. However, there was a point on the 6^th of April when sensor 4 could not detect a peak which significantly affect human health whereas the reference instrument managed to do so (Figure 9).

(22)

21 | P a g e

(23)

22 | P a g e Figure 8. Sensor 1, 5,7 and 8 successfully predicted extreme variations in concentration for PM10 hourly averages that significantly affect human health on different days.

Figure 9. The municipal reference station managed to show a peak that was above the EU hourly air quality standard that could significantly affect human health but the sensor 4 could not show this peak.

(24)

23 | P a g e

3.9 PM

1

, PM

2.5

and PM

10

concentrations across the measured places

The sensors managed to show different values of particulate matter concentrations from the measured places. The average PM1 concentration across the places during the period of the study was 1.4µg/m³, 3.5µg/m³ for PM2.5 and 9µg/m³ for PM10. The minimum concentrations were 0.1 µg/m³, 0.3 µg/m³ and 0.95 µg/m³ whilst the maximum concentrations were 6.8 µg/m³, 14.6 µg/m³ and 204 µg/m³ for PM1, PM2.5 and PM10 respectively. There are significant differences of PM1, PM2.5 and PM10 concentrations between the measured places (PM1 showing p = 0.001, F = 3.6 and df = 7) (PM2.5 showing p = 0.001, F = 3.7 and df = 7) (PM10 showing p

= 0.001, F = 4.8 and df = 7). The p value is a number which tells us how likely it is that our result is just random (p stands for probability, p = 0.05 means 5% probability, p < 0.05 means lower than 5% probability that the result is random) (Tyrrell., 2008). F represents how data varies from the mean (Tyrrell., 2008). Lastly, df means degrees of freedom which is the number of variables in a sample that free to vary calculated as number of variables minus 1 (Tyrrell., 2008). Sensor 1, 2, 5 and 8 recorded the highest amounts of PM10

concentrations whilst sensor 7 recorded the lowest PM10 concentrations over the period of the study. Although sensor 8 recorded one of the highest PM10 concentrations, it showed one of the lowest PM1 and PM2.5 values whilst sensor 1, 2 and 5 recorded the highest PM1 and PM2.5

concentrations as expected; see Table 6.

Table 6. Means of PM1, PM2.5 and PM10 concentrations for all measured places from 1 April to 17 April.

Means of PM1, PM2.5 and PM10 concentrations for all measured places from 1 April to 17 April.

Sensor PM1 (µg/m³) PM2.5 (µg/m³) PM10 (µg/m³)

1 1.52 4.38 11.82

2 1.62 4.07 9.42

3 1.29 2.51 6.37

4 1.38 3.71 9.12

5 1.57 3.96 10.4

6 1.54 3.60 8.7

7 0.78 1.98 5.3

8 1.38 3.43 10.9

The sensors from all the 8 places showed similarities in daily patterns of PM1, PM2.5 and PM10

as shown by Figure 10. Fridays during the period of the study experienced PM1, PM2.5 and PM10 increases (3, 10 & 17 April) that decreased as the day progressed. All sensors showed low values of PM1 followed by PM2.5 then the highest values in PM10 at any point in time during the period of the study which had averages of 1.4µg/m³, 3.5µg/m³ and 9µg/m³ respectively.

This is expected since the count of PM10 accounts for both PM1 and PM2.5 and PM2.5 includes PM1 whilst PM1 can only be counted as it is.

The highest peak for PM1 were all recorded on the 5th of April in the morning (3:00am – 7:00am) from sensor 5 which ranged from 5.9µg/m³ to 6.8µg/m³ whilst the lowest PM1

(25)

24 | P a g e volumes where all recorded from sensor 7 ranging from 0.1µg/m³ to 0.14µg/m³ on the 13th and 14th of April at between 9:00am and 11:00am. PM2.5 highest volumes were recorded by sensor 1, 8 and 5 with values ranging from 11.7µg/m³ to 14.5µg/m³ on the 16th of April at 2:00pm (sensor 8), 8th and 10th of April between 7:00am and 8:00am (sensor 1) and on the 5th of April between 4:00am and 5:00am (sensor 5). However, all the lowest PM2.5 where all recorded from sensor 7 on the 13th and 14th of April (10:00am and 2:00am) ranging from 0.2µg/m³ to 0.4µg/m³. Lastly, the highest PM10 volumes were again recorded from sensor 1, 5 and 8 on the 16th of April at 2:00pm (sensor 8), 10th of April between 7:00am and 8:00am (sensor 1) and on the 13th and 17th of April at 2:00pm and 8:00am (sensor 5) ranging from 72µg/m³ to 204µg/m³. The lowest PM10 volumes were recorded from sensor 3, 4 and 7 ranging from 0.9µg/m³ to 1.3µg/m³ on the 1st of April at 3:00am (sensor 3), 13th of April at 8:00am (sensor 4) and on the 13th of April at 8:00am and between 2:00pm and 4:00pm.

(26)

25 | P a g e

(27)

26 | P a g e Figure 10. Differences in daily patterns of PM1 (blue), PM2.5 (red) and PM10 (black) concentrations measured from 1 April to 17 April amongst 8 selected places. Outliners for sensor 1, 5, 7 and 8 were excluded for graphical clarity.

(28)

27 | P a g e

4. Discussion

4.1 Level of awareness and what should be measured

Insights from the questionnaires recognized CO² as the most important pollutant to measure.

Particulate matter, nitrogen oxides and ozone followed as the most critical pollutants to measure. These pollutants are also critical as recognized by the European Union (2014) Clean Air Act and the European Environment Agency (2008) report which emphasized that 90%

European city dwellers were exposed to particulate matter, nitrogen oxide and ozone concentrations that are higher than regulated limitations. Because of the high traffic volumes, high energy usage and abundance of industrial activities in urban environments, there are increased concentrations of CO², particulate matter, NOx and ozone which can be the reason why the respondents from the questionnaires were concerned about these pollutants.

Karagulian et al. (2015) made a global review with 419 records from 51 countries linking sources of PM2.5 to heavy traffic (25%), industrial activities (15%), domestic fuel burning (20%) and natural (18%). Also, the concern on CO² might have been influenced by strong media coverage about climate change as CO² is one of the biggest drivers of climate change (Eurostat 2020). More so, 92.5% of the respondents reside in Sweden which is regarded as one of the PM2.5 free environments in the world (number 93 out of 98) according to the IQAir report (2019) thus if the survey was conducted in another region or country the interests in CO² might have been different. The results on the acknowledgement of critical pollutants and the presents of many sources of pollution in urban areas will ultimately link with the high concern of air quality as shown in Figure 2. However, the level of concern is higher amongst people with an environmental background as shown in the results and this is also supported by Fredriksson et al. (2018) who explained the significant influence of environmental lobby groups in environmental policy making. From the questionnaires it can be concluded that people are concerned about air quality and there is need for improvements in air quality monitoring which can be enhanced through the use of air quality sensors to monitor in localised areas. Also, to get more informed answers on which pollutants to measure, a bigger sample should be used in answering the questions and consider the air quality history of the city.

4.2 Presentation of information and measurements level of accuracy

Information is useful when it is understood and results from the insights show that simple visuals like colors are good enough for users to understand what is meant. To accommodate those citizens who are more concerned about numbers, a presentation that shows a color and the concentration level could be implemented. Due to increased digitalization, smart cities projects find it easy to present the information with mobile applications or websites as used by the 10 case studies in Table 1. Creative ways of awareness bring out the best results in awareness. For example, the article Toxic Toby in the streets of London by Ach (2018), explained how a teddy bear mounted on street poles that coughed as pollution levels reached levels that affect human health, gave vivid awareness to the citizens of London. Other visuals can also include streets lights, towers or trees that change color according to the level of pollution. These creative visualizations are easily understood by citizens and can be utilized for awareness purposes. Colors can also be a huge success in presenting air quality information.

This was observed by Ramos et al. (2018), who reviewed air quality indices of government departments and research communities, namely the US EPA, Canada, Common air quality

(29)

28 | P a g e index, UK Defra, Irish EPA, Spain, France, Singapore, Cairncross et al. (2007), Stieb et al.

(2008), Kyrkills et al. (2007), Sicard et al. (2011) and Munera (2004). The result from the review showed all indices using colours with the France ATMO going a step further to use giraffes and colours. The only exception was Kyrkills et al. (2007), who had no colour presentation. In summary, it is important to note that presentation of air quality information should be clear such that citizens can use it and take informed decisions.

Pollutant measurements using sensors do not have to be 100% accurate, devices that are correlated to the reference instruments and follow variations of peaks and low-level readings are good enough for air pollution awareness. Piedrahita et al. (2013), Gao et al. (2015) and Mukherje et al. (2017) assessed air quality sensors using correlations with reference instruments and concluded that accuracy shown by moderate to strong associations (r² ≥ 0,6) with offsets of 2µg/m³ to 4µg/m³ can still be used as qualitative measurements for awareness purposes. My study evaluated accuracy by how sensor 4 managed to correlate PM10 variations with the municipal reference station readings and a correlation of sensor 2 and sensor 6 measuring PM1, PM2.5 and PM10 from the same location (Figure 5). EU Air Quality Guidelines (2000) reviewed that there is a significant risk to respiratory diseases and mortality in every 10µg/m³ increase in PM2.5 and PM10. Sensor 4 had a moderate positive relationship with the municipal reference sensor (r² = 0.64) however the offsets on 3, 6, 7, 8, 14 and 15 April (+/- 40µg/m³) were too high to be used for awareness purposes. This might have been that sensor 4 was not calibrated to the municipal reference station however, there were strong associations with +/-4µg/m³ offsets on some days during the study, for example on 1, 4 and 17 April. The association between sensor 2 and sensor 6 (r² = 0.93) proved that the sensors perform the same if put in the same environment. Therefor it can be concluded that all sensors used performed with the same accuracy for all the measured places. Thus, in this case the sensors can be used for awareness purposes because they can predict variations that are predicted by the reference sensors however calibrations are needed.

Another factor that was used to access accuracy was the performance of the sensors when subjected to humidity and temperature variations. Particulate matter sensors have been associated with failing to predict variations at low level ambient air concentrations and can easily be affected by temperature and humidity (Kelly et al. 2017 and Rai et al. 2017). Jayaratne et al. (2018) concluded that the sensors tend to increase readings if exposed to fog and sometimes can be affected by relative humidity as low as 50%. When correlated to the reference station (Figure 2), sensor 4 followed the flow in particulate matter variations throughout the study even during days like the 1st and the 17th of April when humidity was as high as 80%.

This suggests that the optical particulate matter counters used in this study were not affected by variations in humidity and temperature which would give false positive or false negative values that can lead wrong interpretations for citizens’ awareness.

4.3 Particulate matter patterns and variations

Variations in measured parameters were assessed in relation to environmental conditions and physical surroundings in the places where the sensors were placed. The sensors were placed around an area of ~9,1 km² and Figure 3 shows how the temperature and humidity around the places can be regarded as similar. However, sensor 3 and sensor 8 have experienced slightly higher temperatures than the rest of the sensors. This can be explained by the fact that these two sensors are surrounded by high raised buildings which created micro environmental

(30)

29 | P a g e conditions that are warmer than the surroundings as explained by Abdeslam (2016). The localized conditions within microenvironments shows the importance of measuring pollutants within those microenvironments.

Although the results show that environmental conditions affect the particle matter concentration, the influence of environmental conditions probably also depends on the physical surroundings. For example, sensor 3, 4, 8 and 5 were surrounded by buildings and trees whereas sensor 7 was in an open space (Table 1). Thus, wind could easily clear the air at sensor 7, but not as easy at sensor 3,4,8 and 5. Another factor that might have affected pollutants variations is rainfall. Rainfall has a negative impact on particulate matter, as suggested by Zalakeviciute et al. (2018). In this thesis, some drops in particulate matter on the 1st of April between 5:00am and 6:00am and 12th of April between 2:50am and 3:20am where likely cause by rainfall during this time (Figure 10). Trees are at times used as measures to reduce pollution concentrations because of their ability of capturing particulate matter with their leaves and blocking wind, as proven by Baldacchininet al. (2019). This also could have affected variations in this study especially for sensor 5 which had trees in-between its location and the highway.

These factors represent some variables that significantly affect particulate matter variations in an urban environment but were not taken into consideration in this thesis. Therefore, there were unexpected variations that couldn’t be explained without taking the surroundings of the sensors into account.

Sensors 1, 2, 5 and 6 had the highest averages of PM1 (1.54µg/m³ to 1.62µg/m³) during the period of the study whilst sensor 7 had the lowest PM1 average (0.78µg/m³). The possible explanation to these values can be the fact that sensors 1 and 5 are along a highway receiving high volumes of traffic during the peak hour. Sensor 1 is 30m away from the highway whilst sensor 5 is 50m away from the highway. Peaks in PM1 were mainly recorded in the morning to up to 7am which will be periods of heavy traffic flow. Similar patterns were also observed by Ferm & Sjöberg (2005) who went on to conclude that higher traffic flow increases PM2.5

and PM10 concentrations. Sensor 2 and 6 are near a busy road in an office area thus high PM1 levels were also expected during early mornings. Another explanation that can relate to the variations is that from 1am to 5am on the day of the peak (5 April) the wind was calm (0m/s) which saw the accumulation of PM1 in these areas. Likewise, Zhou et al. (2018) observed increases in PM1 during calm days where accumulation of PM1 was easily noticeable in localised areas. Highest averages and peaks of PM2.5 and PM10 were recorded from sensor 1, 5 and 8. For sensor 1 and 5 it was expected since particulate matter levels from these positions are influenced by traffic flow as explained before. However, for sensor 8 it was mainly affected by huge peaks on the 16th of April that recorded 14.5 µg/m³ and 204 µg/m³ of PM2.5 and PM10, respectively. These peaks could have been caused by an independent factor like construction works or someone smoking near the sensor. Such a result would need further investigations on the happenings of that day. Thus, the use of cameras on the sensors would be useful in investigating cases like these and avoid giving false alarms of bad air quality to people in such a location. On the other hand, sensor 7 showed low values of PM1, PM2.5 and PM10 during the period of the study. This might have been because of its location that it is after an exit to the highway thus there would be less vehicles on the road which is just 32m away. Also, the sensor is placed in an open space and the wind would blow from the west were the sensor is towards the city therefor it would record less concentrations. This type of location would be good in monitoring urban background pollution, but the location should be shifted to downwind of the

(31)

30 | P a g e city, as concluded by Keuken et al. (2013). They suggest that urban background is best monitored 2km downwind of the city, when pollution from all sources are mixed up.

The measurements of particulate matter concentrations were evaluated in order to understand if the sensors can detect particulate matter peaks that significantly affect human health at the different measured places. This was done by comparing the PM10 hourly averaged concentrations against the EU Air Quality standards. An analysis of PM1 and PM2.5 could not be done because the EU Air Quality standard does not have regulations for PM1 and only have annually averaged limitations for PM2.5. There were differences between the municipality air quality station readings and sensor 4 and these are referred to as offsets. Offsets of up to 10µg/m³ can be acceptable since a 10µg/m³ hourly averaged PM10 difference between concentrations was acceptable as per the EU Air Quality Guidelines (2000). However, difference of more than 10µg/m³ is huge enough to affect human health. The EU Air Quality Guidelines (2000) included a review that observed 1.001 to 1.012 relative risk in hourly fluctuations per 10µg/m³ that significantly lead to respiratory diseases or mortality within 24 hours. In the guideline, relative risk was calculated using the back-calculation of logistic regression coefficients to estimate the relative risk over a 10 µg/m³ range; thus, a relative risk of 1.0074 is 0.74% increase. However finer particles like PM1 and PM2.5 penetrate deeper in the respiratory system where mucociliary clearance mechanisms maybe insufficient hence becoming more fatal (Bakand et al., 2012). Because there is a lack of regulations for the hourly limits of PM1 and PM10, there could not be an assessment of PM1 and PM10 peaks that significantly affect human health. When looking at the ability to detect hourly averaged PM10

variations ≥50µg/m³ that significantly affect human health according to EU Air Quality standards, sensor 1, 5, 7 and 8 successfully showed that they could detect such variations (Figure 7a). The municipal reference station also showed the ability to predict these peaks (Figure 7b). However, on the day of the peak, sensor 4 had a 40µg/m³ offset, which is unacceptable and would give a false positive of good air quality of an hourly average of 20µg/m³. This kind of an offset can be corrected by calibration of the benchmark sensor to the municipal station, as done by Gao et al. (2015). Overall, the sensors show the ability to detect air quality drops that significantly affect human health and can alert residents of bad air quality.

4.4 Placement of sensors

Through observations and research in this thesis it is clear that the placement of sensors is critical, especially if the information is used for awareness purposes. As a first step, demographical analysis is imperative to identify the most common places were people spend time at home. A demographic study would involve the dynamics of populations in the urban environment. As a second step, the sensors should be placed in locations of interest, meaning where people spend time away from home. Sources of air pollution, especially busy roads in an urban environment, can also be taken into account in order to identify hotspots, as done by Penza et al. (2017). All of the 10 cases of smart city projects (Table 1.) that were analysed emphasized placing sensors in localised areas where people spend most of their time in.

Development of such technological initiatives allow air quality monitoring in places where the traditional reference monitoring stations cannot be located (Dubai Air Quality Network, 2011).

The U.S. National Science Foundation conducted similar research through the Array of Things., (2013); a smart city project in which a network of up to 500 sensors were installed in the city centre on streetlights and on sides of buildings. However, the U.S Environmental