Data Study Group Network Final Report: Bristol City Council

Data Study

Group Network

Final Report:

Bristol City

Council

Get Bristol moving: tackling air

pollution in Bristol city centre

5–9 August 2019

________________________________________________________________

https://doi.org/10.5281/zenodo.3775497

This work was supported by Wave 1 of The UKRI Strategic Priorities

Fund under the EPSRC Grant EP/T001569/1, 'AI for Science and

Government' programme at The Alan Turing Institute

The Alan Turing Institute

Jean Golding Institute - University of Bristol

Get Bristol moving: tackling air pollution in

Bristol city centre

Authors

Carlos Mougan

Siddharth Dixit

Caitlyn Robinson

Hyesop Shin

Qian Fu

Ella M. Gale

Jojeena Kolath

Laurens Geﬀert

Huan Tong

Ahmad Abd Rabuh

Exploring air quality data in Bristol

Contents

1 Executive Summary 3

1.1 Challenge overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3 Data overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.4 Main objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.5 Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.6 Main conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.7 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.8 Recommendations and further work . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Dataset overview 6

2.1 Data description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.1.1 Air quality data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.1.2 Weather data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.1.3 Holiday data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.1.4 Traﬃc data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.1.5 Additional useful datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.2 Data quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.2.1 Metadata . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.2.2 Data Integrity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.2.3 Data Schema . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.2.4 Naming Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.2.5 Dataset information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3 Experiments and Results 10

3.1 WP1: Analysing the temporal distribution of air quality . . . . . . . . . . . . . . . . 10

3.1.1 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3.1.2 Results and visualizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3.2 WP2: Analysing the spatial distribution of air quality . . . . . . . . . . . . . . . . . 13

3.2.1 Task description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.2.2 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.2.3 Results and visualizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3.2.4 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.3 WP3: Estimating the importance of diﬀerent drivers of air quality . . . . . . . . . . 15

3.3.1 Task description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.3.2 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.3.3 Results and Visualizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

4 Conclusions 19

5 Future work and research avenues 19

5.1 Statistical Hypothesis Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

5.2 Future datasets and data collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

5.3 Expanding work on traﬃc volumes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

5.4 Geographically weighted modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

Exploring air quality data in Bristol

5.5 Log Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

5.6 Winsorization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

5.7 More Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

5.8 Hierarchical/Conditional Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

6 Participant proﬁles: 22

List of Figures

1 Location of Bristol area air pollution monitoring site . . . . . . . . . . . . . . . . . . 6

2 Daily aggregated concentrations of NO

over 2018-2019. . . . . . . . . . . . . . . . . 11

3 Holiday verus term box plot for NO

. . . . . . . . . . . . . . . . . . . . . . . . . . . 11

4 Monthly average temporal variation of NO

concentration for all stations over 2018-

2019. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

5 Visualization of the yearly evolution of the NO

concentration . . . . . . . . . . . . . 13

6 The spatial relationships between NO

distribution and transport type to work for

residents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

7 Concentrations of NO

in diﬀerent measurement spots versus day hours. . . . . . . . 15

8 Feature importance in the ElasticNet model. In order of appearance (from left to

right): temperature sd, wind speed sd, wind speed mean, rainfall sd, temperature

mean, humidity sd, hour, rainfall mean, humidity means, wind direction means, wind

direction sd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

9 Model predictions on holdout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

Exploring air quality data in Bristol

1 Executive Summary

1.1 Challenge overview

Air quality is an increasing public health concern in UK cities, and Bristol is no exception breaching

annual targets for nitrogen dioxide

. In Bristol, ﬁve people die each week as a result of poor air

quality

. Like most UK cities, in Bristol the main cause of air pollution is traﬃc. To reduce

the negative impacts of air pollution, and meet increasingly stringent European Union pollutant

standards, Bristol is currently consulting on a Clean Air Strategy

. A range of policies is proposed

including increased public transport, clean air zones and road closures. The city council currently

collects a wealth of data (about air quality, traﬃc ﬂows and weather) for analyzing the distribution

of air quality and the relationship with various drivers. Such understanding has the potential to

feed into the design and evaluation of air quality policies. However, the council has limited capacity

for analyzing and interpreting such large and complex datasets. This project mines this data to

understand both the spatial and temporal distribution of air quality. the driving factors of air

pollution in diﬀerent parts of the city, with a particular focus upon traﬃc.

1.2 Background

• There is ample evidence of inverse relationships between NO

and children’s risk of health

e.g. exposure, asthma, headache, and mental health.

– Roosbroeck et al.(2007)

measured personal exposure to traﬃc-related air pollution in

Utrecht, the Netherlands. From 54 young participants, the personal exposure to NO

was 37% lower on near-road schoolers compared to background schoolers.

– Kim et al. (2004)

conducted 10 school-based surveys from students who walk to school

to compare NO

exposure levels based on their residential location in San Fransisco,

USA. Using a logistic model, children who lived within 300m to major roads and at a

downwind location have higher exposure levels(OR: 1.05±0.1) compared to those who

live far and upwind of major roads.

– Robert et al. (2019)

analyzed a longitudinal cohort study of London-based teenagers

and discovered that the 18-year olds who have had depression was associated with high

pollution exposure when they were 12 years old (Age-12 pollution exposure was not

associated with age-12 mental health problems).

• School zones in Bristol that have a high volume of traﬃc during drop-oﬀ and pick-up times

are facing a serious risk of pollution exposure, and children whose schools are closer to the

road will be more susceptible to these disease symptoms than the schools distant from the

road.

• In line with the Bristol clean air zone proposals

, the city council is in consideration of closing

roads outside schools (Bristol local news

<https://www.cleanairforbristol.org/what-is-air-pollution/what-is-air-quality-like-in-bristol/>

<https://www.claircity.eu/bristol/city-shockers/air-pollution-map-of-bristol/>

<https://www.cleanairforbristol.org/>

<https://www.sciencedirect.com/science/article/pii/S1352231006012878>

<https://www.atsjournals.org/doi/full/10.1164/rccm.200403-281OC>

<https://www.sciencedirect.com/science/article/pii/S016517811830800X>

<https://airqualitynews.com/2019/07/01/bristol-launch-clean-air-zone-consultation/>

<https://www.bristolpost.co.uk/news/bristol-news/roads-outside-schools-could-closed-3001742>

Exploring air quality data in Bristol

• However, we argue that the proposed plans can only take into further consideration once a

critical exploration of pollution, weather, and traﬃc is done.

1.3 Data overview

We build upon a wide range of exciting and open datasets provided by Bristol City Council. This

includes continuous air quality datasets, point observation, and route observation traﬃc datasets,

weather observations and other spatial datasets.

1.4 Main objectives

The projects aims were twofold:

• A1. To understand the spatial and temporal distribution of pollutants and drivers

• A2. To examine the relationship between air quality and its drivers (e.g. traﬃc and weather)

In working with the council’s data to understand this relationship, the project also feeds into the

Our Data Bristol

, an exciting initiative that allows local people and organizations to access, use

and beneﬁt from a wide range of open-source datasets and technology. Experience of working with

the datasets during the project will allow recommendations to be made about how datasets could

be more accessible and comprehensive, and how to transport datasets could be made available in

the future.

1.5 Approach

The project was divided into three work packages (WP). WP1 and WP2 address our ﬁrst aim.

WP3 addresses our second aim.

WP1. Analyzing the temporal distribution of air quality

WP2. Analyzing the spatial distribution of air quality

WP3. Estimating the importance of diﬀerent drivers of air quality

1.6 Main conclusions

This study ﬁrstly (WP1) explored the temporal features of pollutants (NO

), then examined Parson

Street School as a case study. Our initial ﬁnding was that most stations had a daily and seasonal

oscillation of NO

throughout the whole period ranging from less than 20g/m

to over 1000g/m

based on hourly measurement. Looking into an averaged hourly NO

, there was a clear trough

around 20g/m

at 3-5 am but peaked at around 60g/m

after 10 am. However, the concentration

was always higher in the city center sites.

Our second ﬁnding (WP2) was that the NO

concentration between holidays and non-holidays

(school holidays and bank holidays 2018-2019) was on average less than 5g/m

, however, varied by

locations.

We also were able to ﬁnd what are the most important drivers in diﬀerent geospatial locations

(WP3). We saw that some meteorological conditions were important in some places and that they

were less important in others. One of the most important drivers that we found was the standard

<https://www.bristol.gov.uk/data-protection-foi/open-data>

Exploring air quality data in Bristol

deviation over time of temperature, we interpreted this as the change of temperature in any given

hour. This goes along with the intuition that changes in the meteorology have an impact on the

pollutant concentration.

1.7 Limitations

In this project we found several limitations that did not allow us to proceed further and ﬁnd better

results:

Data Limitations: Most of the data sets were not pre-processed enough or they didn’t have

enough information to answer accurately the questions proposed. For example the diﬀusion tubes

limited the spatial understanding of the pollution. It was also hard to work with the traﬃc data,

and to extract solid conclusions due to diﬃculties while combining the tables (not a common and

index). In order to improve that, we suggest trying to aggregate all features in one tabular dataset,

where data is easier to understand and manipulate. Even if this is not achievable due to the task

requirements just orientating the dataset in this way allows is for faster pre-processing.

Approach Limitations: With the data provided, we could not ﬁnd the association between

the diﬀusion tubes and the NO

. This blocked us from trying more algorithms and eventually

getting better results. Some of the approaches we wanted to try but we couldn’t either for lack

of time or lack of quality data are statistical hypothesis testing, geographically weighted modeling,

hierarchical modeling, and k-nearest neighbors for spatial classiﬁcation of tubes. On the other hand

when we tried to answer if there was any diﬀerence in holiday vs term but we were only able to do

visualization and we were unable to proceed with further statistical analysis.

1.8 Recommendations and further work

In the last section, we provide diﬀerent ideas of how we would have approached the problem given

more time and resources. Some of them are more focused on the data and others on mathematical

modeling.

• For the data: improving the datasets, gathering more data, expanding the work on traﬃc

volumes.

• For the mathematical modeling: the ﬁrst improvement would be to do statistical hypothesis

testing. Other recommendations that we think that would be convenient is to do geograph-

ically weighted modeling, as this will allow us to see how the driver’s behaviour changes in

a spatial way. Once we have this, we suggest doing hierarchical modeling to see if there is a

station that behaves similarly. Some further recommendations in the Data engineering tech-

niques: winsorization

, logarithmic transformation and adding meteorological features such

as geopotential height, dew point . . .

• For determining the diﬀerence between holiday and term-time, it would be necessary to do a

statistical analysis called hypothesis testing, without these tests, we cannot say whether there

was a statistically signiﬁcant diﬀerent or not. One idea that we were not able to experiment

with was making a model that predicted if a certain date was a holiday or not.

<https://en.wikipedia.org/wiki/Winsorizing>

Exploring air quality data in Bristol

2 Dataset overview

The following section provides information about the datasets used during the project, how these

were processed and the dataset limitations.

2.1 Data description

A wide variety of datasets was available to the group which is explained in detail below, including

information about how to access the processed datasets.

2.1.1 Air quality data

Bristol has been monitoring nitros oxides NO

, which includes NO and NO

) from a single site

since January 1998, and has expanded it to multiple stations around the city center.

At present, the city has seven stations, in the city center, that monitors NO

, NO

, NO,

and PM

(see Figure 1). Each station monitors a diﬀerent type of environment: Colston avenue

(detects exposure in city center), AURN St Pauls (detects background exposure in residential area),

Brislington depot (freight-caused exposure), Fishponds road (residential and shopping), Parson

street school (residential area and school zone), Wells road (continuous traﬃc in and out of city

center).

Figure 1: Location of Bristol area air pollution monitoring site

Exploring the continuous air quality data we can see that the stored value for many measure-

ments of air pollutants was NA, indicating that the data was missing or it was not logged. We

decided to retain these in the dataset for now though in case we wanted to use them in a later

analysis.

Exploring air quality data in Bristol

Diﬀusion tube data measuring annual averages of NO

were also made available. Diﬀusion

tubes/Diﬀusive samplers are widely used for indicative monitoring of ambient nitrogen dioxide

(NO

)

. Although much less accurate than the continuous air quality measurements, diﬀusion

tube data was available at approximately 127 sites (varying depending upon the year) across the

city making it useful for spatial analyses.

At the recommendation of the council, we primarily focused upon NO

because the pollutant

more accurately reﬂects t raﬃc p atterns. S ome o f t he a nalyses a lso e xplored N O

w hich i s useful

for the council when considering regulatory compliance.

2.1.2 Weather data

There was weather data available from the challenge data and the Open Data Bristol

website.

However, of the two stations available, one went oﬄine in 2015

. The other was still up and run-

ning but was missing some essential variables

such as rainfall. We, therefore, obtained an external

dataset from a metoﬃce weather station

. The data was downloaded from the website manually (one

month at a time) and combined using the code in "products/01/data /clean/data/reader /weather.R".

CSV formatting seemed to change as a function of the operating system of the downloading com-

puter, which resulted in the necessity to use the two slightly diﬀerent import functions that you see

in the R script. It is worth noting that since there was only one weather data station available,

we assume in our analysis that the weather conditions are the same across the entirety of Bristol.

2.1.3 Holiday data

To understand whether and how school traﬃc impacts air quality, we used school holidays and bank

holidays as control group observations. Data on holiday days was not provided in the challenge

data, so we created our own dataset. School term dates were taken from the local council website

and bank holiday dates were taken from the national government website

. From this information,

we created Boolean variables for weekends, school holidays, and bank holidays for the school year

2018/19 (August to July).

2.1.4 Traﬃc data

Four traﬃc datasets were provided: ﬂow d etector d ata, l ink t ravel t ime d ata, c amera p late data,

and SCOOT link data. Flow detector data is generated by detectors measuring vehicles passing

a certain point in the road. Camera plate data is generated by average speed cameras, which can

monitor the time that has passed between two sightings of the same number plate, however, stops

along the road are not taken into account. This means that the journey time recorded will be an

upper bound but should be used with caution, we may see high-value outliers.

<https://laqm.defra.gov.uk/diffusion-tubes/diffusion-tubes.html>

<https://opendata.bristol.gov.uk>

<https://opendata.bristol.gov.uk/explore/dataset/meteorological-data-create/information/>

<https://opendata.bristol.gov.uk/explore/dataset/met-data-bristol-lulsgate/table/?sort=date_time>

<https://wow.metoffice.gov.uk/observations/details/20190806qbpnbopacce6ucrdyyb96sczue>

<https://www.bristol.gov.uk/schools-learning-early-years/school-term-and-holiday-dates>

<https://www.gov.uk/bank-holidays>

Exploring air quality data in Bristol

2.1.5 Additional useful datasets

In addition to data about air quality, travel and weather, we also used several datasets to pro-

vide context to our analyses: area-based demographic datasets and boundaries from the Oﬃce for

National Statistics

and Ordnance Survey Open Roads

2.2 Data quality

Many of the datasets available to the group required pre-processing cleaning and aggregation. Here

we give a list of issues identiﬁed when importing the data, explain how we addressed each issue,

and provide cleaned versions of the datasets.

2.2.1 Metadata

Some of the metadata is not easily available. For example, geolocations of air quality measurement

stations are only available in the full continuous measurement ﬁle, not just the coordinates in a

separate ﬁle or table.

2.2.2 Data Integrity

Parts of the data contained unintended artifacts. For example, the traﬃc data contained repeated

header rows pasted into the data rows of the CSV. Our data importing functions take care of

this and create cleaned versions of these datasets. Another issue we encountered was inconsistent

formatting of date ﬁelds but we were able to address this with the excellent conversion functions of

the lubridate R package.

2.2.3 Data Schema

There seems to be no distinction between FACT (a table that containes

measurements, metrics or

facts about a business process.) and DIM data (companion table to the FACT table that contains

descriptive attributes to be used as query constraining.) in the data schema. It could be useful to

use a star schema

and group small DIM tables around larger FACT tables. Examples for possible

DIM tables could be locations, setup times, names, and types of traﬃc sensors or weather stations,

which can then be referred to by ID in the larger FACT tables such as traﬃc measurement and

air quality data measurement tables. This would both make DIM data more easily accessible and

reduce the size of the FACT tables. Besides, this will ensure data consistency in the future when

DIM values are updated centrally.

2.2.4 Naming Conventions

In several cases, we decided to rename columns from the names they had in the original datasets

provided. We are listing some examples of our renaming policy in case this can inform a future

database design:

<https://www.ons.gov.uk/searchdata?q=neighbourhood%20statistics>

<https://www.ordnancesurvey.co.uk/business-and-government/products/os-open-roads.html>

<https://www.guru99.com/fact-table-vs-dimension-table.html>

<https://en.wikipedia.org/wiki/Star_schema>

Exploring air quality data in Bristol

Remove spaces and special characters - Spaces and special characters can make it diﬃcult

to access columns when working in python. We replaced all spaces with an underscore and removed

special characters where possible.

Remove upper-case-letters - (unless from acronyms) Uppercase adds cognitive load for the

user and can lead to inconsistencies in data processing pipelines. We converted all uppercase letters

with lowercase letters, apart from acronyms or chemical formulas such as NO

Use hierarchical semantic taxonomies - Informative short words with hierarchical

group-ing sequences could be used to create a variable name hierarchy. We used this to make

it eas-ier for the user to identify similar variables and to use command-line autocomplete for

the de-sired result. For example, wind variables in the weather file were

named_wind_direction_mean,wind_direction_sd,wind_speed_mean,wind_speed_sd,wind_gust

_mean, wind_gust_sd to keep similar variables together.

Ensure consistency across datasets - Consistency can be important when datasets are oﬀered

together. Examples for this that oﬀer the potential for improvement are the coordinate variables

which are sometimes called lat/lon, sometimes Eastings/Northings, and sometimes are misspelled

(e.g. Longitude in traﬃc metadata). We renamed all of these, making it easier for the user to find

variables that overlap and can be used to combine datasets.

2.2.5 Dataset information

Transport-related datasets utilised during the experiments can be located using the following:

• A - Weekly Cemara Plate - ("CSV CameraPlateWeekly")-

• B - Detector Flow - ("CSV DetectorFlow")-

• C - Link Travel Time -("CSV LinkTravelTime")-

• D - SCOOT Link Data -("CSV SCOOTLinkData")-

Understanding and cleaning the transport-related data is a signiﬁcant challenge faced by the

team.

• Geographical information was missing/unavailable in both the data sets of A and B, so that

the group was unable to identify/locate where the information about traﬃc volumes was

sourced.

• There was a lack of some common attribute(s) that might allow the team to merge these data

sets.

• Identiﬁers of the links (i.e. "road sections" - to the understanding of the team) in both data

sets C and D are diﬀerent.

• In data set D, it happens frequently that traﬃc information (e.g. average speed) on any given

link remains constant along the time.

• Finally, traﬃc datasets do not have a long history as the oldest goes back to 2018-12-05 as

opposed to air pollution and weather data that are, in one station, go back to 1998-01-10.

Exploring air quality data in Bristol

3 Experiments and Results

Given the issues in the available data sets, only dataset C was further processed. The team deﬁned

an area, referred to as a "buﬀer zone", for each AQ monitoring station. For the sake of testing the

idea, the buﬀer zone was designed as a circle for this task, with the location of the station being

the center. The team examined diﬀerent sizes of buﬀer zones, using the radii of 50, 100, 500 and

1000 meters, respectively, and extracted relevant road links within and/or intersecting the buﬀer

zones. This can facilitate geo-coordinates transformation.

3.1 WP1: Analysing the temporal distribution of air quality

Addressing aim A1, WP1 explores the temporal features of NO

across Bristol monitoring sites.

Besides, the temporal features of meteorological factors are examined. Due to the focus upon

temporality, we chose to focus upon a case study of air pollution at Parson street primary school.

3.1.1 Experiments

There have been similar studies in other UK cities, for example, using air pollution measurements

from a background site in Central London, Bigi and Harrison (2010)

analyze the seasonality of

recording minimum values in June/July and maximum values during Winter. Annual patterns

of NO

were found to be similar to that of CO and NO. Meanwhile, the NO

weekly pattern is

weak, with the mean concentration remaining steady during weekdays with only a small drop during

weekends. NO

is a relatively stable pollutant.

3.1.2 Results and visualizations

In general, all stations have a daily and seasonal oscillation of NO

. However, there was a variation

by stations. The averaged NO

was the highest on Rupert Street (city center) at 93.1µg/m

followed by Colston Avenue and Temple Meads station at 65.7, 63.2 respectively.

From all the stations, Parson Street school has been monitored since 2002 had an average of

47.7µg/m

, but the hourly concentration of NO

exceeded the legal limit on 154,189 data points.

Although it has declined in recent years, the high counts are very much a concern. Overall, the

averaged NO

tends to soar rapidly between 7am and 10am, which roughly peaks at twice the

amount than the concentration at 5am. It remains the concentration until 6pm, then gradually

decreases late at night. For example, Parson street school has the lowest NO

concentration just

above 20µg/m

at 4am, then rose up to 54µg/m

at 10am.

The boxplot in ﬁgure 2 used the monthly average of NO

at 7 stations in 2018-2019. Overall, the

weekday concentration was higher than on weekends. Amongst all stations, Colston avenue (city

center) had the highest amount of NO

that ranged 50 − 70µg/m

during weekdays and fallen 60

µg/m

on Saturdays, then decreased further on Sundays at around 50µg/m

. Parson Street school

just managed to go below the national NO

limit of 40µg/m

during weekdays.

<https://www.sciencedirect.com/science/article/abs/pii/S135223101000155X>

Exploring air quality data in Bristol

Figure 2: Daily aggregated concentrations of NO

over 2018-2019.

School Holiday

Overall, while there is a small diﬀerence between holidays and non-holidays (5µg/m

), it has a

distinction by locations.

• Colston Avenue a consistently high concentration of NO

above 50 regardless of holidays, but

the concentration varied during holidays. Similar patterns were seen on Temple Way.

• During term time, children who were commuting to schools near Parson street primary school

or passing Fishponds road might have experienced a high level of NO

since the concentration

ranges up to 80µg/m

Figure 3: Holiday verus term box plot for NO