Popular in Formal Sciences

Formal Sciences

Diameter rigidity for Kähler manifolds with positive bisectional curvature

Indian Institute of Science Ved Datar

Formal Sciences

Rank-Based Multivariate Sarmanov for Modeling Dependence

McMaster University Anas Abdallah

Formal Sciences

The Hybrid High-Order Method for Polytopal Meshes

Monash University

Formal Sciences

Financial Informatics: An Information-Based Approach to Asset Pricing

University of Surrey

Formal Sciences

Multiplicative orders of Gauss periods and the arithmetic of real quadratic fields

University of Newcastle

Formal Sciences

Some Numerical Methods for the Hele-Shaw Equations

University of Massachusetts at Amherst

Formal Sciences

Geometric stabilisation via p-adic integration

University of Toronto

Data-driven framework for understanding and predicting air quality in urban areas

View more insights in Sciences - Computer Science - Formal Sciences

View more figures from Anglia Ruskin University

View the page of Lakshimi Babu Saheer

Read the Study

Lakshimi Babu Saheer proposes a cost-effective air quality modeling framework using artificial intelligence techniques to address climate change.

Image courtesy of interviewee. February 12, 2024

Generating Workspace

Export as PDF

This video video provides an overview of air quality and its implications for human health and climate change. It defines air quality as the presence of pollutants in the air—specifically, gas pollutants (like CO2 and NOx) and particulate matter (such as PM10 and PM2.5)—and stresses the importance of monitoring air pollution due to its association with respiratory diseases, lung cancer, and even premature death. The discussion highlights the contribution of air pollutants to climate change through carbon emissions.

This video video explains that pollutants come from various sources, including transportation, industry, and households. To address air quality issues, regulatory standards are set by organizations like the WHO, while recent advancements allow for more accurate monitoring and data-driven predictions of air quality, utilizing historic data and weather factors in modeling.

Vegetation is noted as a significant factor in air quality, with studies indicating both positive and negative effects on pollutant levels. This video video emphasizes the need for improved modeling that incorporates these factors and discusses various statistical and machine learning models used to predict air quality.

A case study from Colchester explores the air quality differences between biking routes and car routes. Findings suggest that biking routes generally have better air quality, but proximity to busy traffic can compromise this benefit. Recommendations include using tree canopies to shield cyclists from pollutants.

Lastly, the text advocates for the development of digital twins—simulations that utilize AI to model environmental factors—to test policies in a virtual setting before real-world implementation. Public education is stressed as vital for encouraging community involvement in combatting climate change and improving air quality.

Language:

Export as PDF

Export as PDF
Here are some key quotes from the video regarding air quality:

1. **Definition and Importance of Air Quality**:
- "Air quality, as the name stands, it's the quality of the air, and it's mainly the pollutants in the air that we're worried about."

2. **Types of Pollutants**:
- "There are different types of pollutants in the air, uh, the gas pollutants, like carbon dioxide, carbon monoxide, sulfur oxide or nitrous oxides and ozone, or even volatile gases."
- "But there are also particulate matter, which are a bit more bigger particles of different sizes. So PM 10, pm 2.5, and things like that."

3. **Health Effects**:
- "Air quality has very adverse effects on human health and the pollutants [...] can cause a lot of respiratory diseases, even lung cancer, and sometimes even due to mortality in some cases."

4. **Climate Change Connection**:
- "The carbon contents in these air pollutants contribute to the CO2 equivalent that we talk about, which is a main contributor to climate change."

5. **Monitoring Air Quality**:
- "In the past decade, air quality was not really monitored or measured. It was estimated [...] recently it has been moved to monitoring or even prediction of air quality."

6. **Factors Affecting Air Quality**:
- "Some of the studies [...] show that vegetation has both negative and positive effects [...] the leaves actually absorb most of the pollutants."

7. **Digital Twins and Predictive Modeling**:
- "The next solution is to look at technology like digital twins, where we model all these factors [...] and predict all these outputs and pollutants based on these factors."

8. **Public Engagement**:
- "It's important to educate the public to make them understand [...] and maybe inform them of the actions they need to take to contribute."

9. **Collective Responsibility**:
- "Only if all the public and the citizens contribute to the effective solutions of climate change, it can happen."

These quotes encapsulate the central themes of the video, including the definition and significance of air quality, the health impacts of pollutants, the relationship with climate change, the evolution of monitoring techniques, and the importance of public engagement and technological advancements in addressing air quality issues.

Export as PDF

X-TIMESTAMP-MAP:LOCAL=00:00:00.000,MPEGTS=900000

00:00.205 --> 00:01.785
To give an introduction on air quality.

00:01.925 --> 00:04.505
Air quality, as the name stands, it's the quality

00:04.645 --> 00:08.265
of the air, and it's mainly the pollutants in the air

00:08.265 --> 00:09.385
that we're worried about.

00:09.485 --> 00:12.705
So there are different type of pollutants in the air,

00:13.085 --> 00:17.185
the gas pollutants, like carbon dioxide, carbon monoxide,

00:17.245 --> 00:22.185
or sulf sulfur, oxide or nitrous oxides and ozone,

00:22.565 --> 00:24.105
or even volatile gases.

00:24.245 --> 00:25.945
But there are also particulate matter,

00:26.075 --> 00:30.225
which are a bit more bigger particles of different sizes.

00:30.405 --> 00:33.345
So PM 10, pm 2.5, and things like that.

00:33.975 --> 00:37.465
It's been, an important topic recently, mainly

00:37.465 --> 00:40.905
because air quality has very adverse effect on human

00:40.905 --> 00:42.665
health and the pollutants.

00:42.665 --> 00:47.065
Stem cells can cause a lot of, respiratory diseases,

00:47.095 --> 00:50.545
even lung cancer, and sometimes even due

00:50.545 --> 00:52.025
to mortality in some cases.

00:52.765 --> 00:54.545
So at the same time,

00:55.005 --> 00:58.025
the air quality has been a great part in the climate change,

00:58.075 --> 01:01.185
which is a, recent topic of interest.

01:01.645 --> 01:04.225
So the carbon contents in these air pollutants,

01:04.225 --> 01:07.105
they contribute to the CO2 equivalent that we talk about,

01:07.515 --> 01:09.745
which is a main contributor to the climate change.

01:09.885 --> 01:14.525
So, these pollutants actually come out of,

01:15.065 --> 01:18.765
different sources, like transport or industry

01:18.785 --> 01:20.925
or even households in case of cities.

01:21.105 --> 01:25.645
And so, we have to understand, how,

01:25.715 --> 01:27.565
what the sources are,

01:27.595 --> 01:29.205
what their concentrations are in there

01:29.305 --> 01:30.485
and how to control them.

01:31.065 --> 01:33.885
Uh, so actually, WHO

01:33.905 --> 01:35.485
and even other national,

01:36.045 --> 01:40.245
standards have been put forward, to the maximum

01:41.055 --> 01:42.245
quantities of these

01:42.345 --> 01:45.325
or concentrations of these pollutants in the air to keep,

01:45.465 --> 01:46.485
to have a healthy living.

01:46.945 --> 01:49.805
In the past decade, air quality was not really

01:50.075 --> 01:51.325
monitored or measured.

01:51.325 --> 01:54.645
It was estimated. So there could be physical chemical models

01:54.795 --> 01:57.125
that understands the atmospheric interactions

01:57.145 --> 02:00.685
and different, chemical, conditions

02:00.705 --> 02:01.765
or the cloud conditions,

02:01.765 --> 02:05.765
and they would estimate the air, pollutant quantities

02:05.785 --> 02:08.085
or concentrations at different levels or heights.

02:08.465 --> 02:13.125
And then recently it has been moved, to monitoring

02:13.225 --> 02:17.565
or even prediction of air quality, that,

02:17.565 --> 02:20.725
could happen from, the historic data.

02:20.825 --> 02:22.645
So we have access to historic data now,

02:22.665 --> 02:24.485
and we have monitoring stations now,

02:24.745 --> 02:26.685
and we could have actually a data driven

02:26.955 --> 02:28.245
statistical prediction.

02:28.945 --> 02:32.205
Now, the thing here is most of these traditions, they try,

02:32.425 --> 02:35.565
try to see air quality as a linear time series,

02:35.825 --> 02:38.805
and we don't accommodate other factors

02:38.835 --> 02:40.965
that could actually affect the air quality.

02:41.225 --> 02:44.845
And some of the studies actually started looking at,

02:44.915 --> 02:46.805
weather as an important factor.

02:47.185 --> 02:50.285
So both weather and vegetation from literature

02:50.285 --> 02:53.325
and press studies we have seen has a great effect on air

02:53.325 --> 02:55.805
quality, mainly due to dispersion

02:55.905 --> 02:59.565
or absorption, for vegetation in in particular.

03:00.305 --> 03:04.655
So these studies actually are a bit controversial as well,

03:04.655 --> 03:06.215
because some of these studies show

03:06.215 --> 03:08.935
that vegetation has having both negative

03:09.075 --> 03:11.815
and positive effects, so we can, understand

03:11.815 --> 03:14.375
that the leaves actually absorb most

03:14.375 --> 03:15.695
of the geese pollutants.

03:16.125 --> 03:20.175
They're also, places where these PM particles,

03:20.175 --> 03:23.095
the bigger pollutant contents actually,

03:23.095 --> 03:24.775
they accumulate on these leaves

03:24.835 --> 03:27.255
and they stay there so they can take it out of the way.

03:27.835 --> 03:31.855
So we usually put up tree canopies if you have seen,

03:31.925 --> 03:34.975
next to these, highways and major roads,

03:35.155 --> 03:38.335
and actually that keeps our shields the rest

03:38.355 --> 03:42.455
of the world from these pollutants that's been dis

03:42.565 --> 03:45.175
sourced from the transport that's on the main road.

03:46.035 --> 03:48.975
Now, the problem there is if you have a canopy,

03:48.975 --> 03:51.375
maybe perhaps the pollutants,

03:51.795 --> 03:53.735
remain there like a canyon effect.

03:54.115 --> 03:57.295
So it, it, it has effects of microclimate,

03:57.295 --> 04:00.135
problems when you have, when you are not,

04:00.195 --> 04:01.455
modeling it, right?

04:01.755 --> 04:06.165
So that's why we, we, we started looking at the importance

04:06.165 --> 04:09.525
of vegetation and try to understand how this can be related

04:09.665 --> 04:10.725
to air quality modeling

04:10.725 --> 04:14.845
and how can we improve the prediction of the models,

04:15.015 --> 04:18.565
based on the other factors including, vegetation

04:18.745 --> 04:21.925
and maybe the factors of weather, like the seasonality

04:21.985 --> 04:24.525
that's not being taken into account in the past.

04:25.305 --> 04:30.085
We have actually, as I said, monitoring stations from,

04:30.085 --> 04:32.925
most of the councils or local authorities, especially UK

04:32.985 --> 04:34.045
or all over the world.

04:34.585 --> 04:37.245
The city's councils have been trying to,

04:37.435 --> 04:40.245
keep the quality of the air, at least monitor the quality

04:40.305 --> 04:43.165
of the air as much as per the regulations,

04:43.185 --> 04:44.405
the WH regulations.

04:44.945 --> 04:48.675
Uh, so the data from these monitoring stations,

04:48.825 --> 04:50.475
even though it's only like three

04:50.735 --> 04:54.395
or maybe at most five, 10 stations in a big city,

04:54.575 --> 04:56.435
is available online publicly.

04:56.815 --> 04:59.555
And so we have all the historic data that we need

05:00.055 --> 05:02.995
to build a, for example, a statistical machine learning

05:02.995 --> 05:03.995
or an AI model.

05:04.535 --> 05:08.755
Uh, so the place we had to, work around was

05:08.775 --> 05:10.955
to find the vegetation data,

05:10.985 --> 05:13.235
even the weather monitoring stations are available.

05:13.375 --> 05:15.875
So we have some information from the, from them.

05:16.255 --> 05:19.675
Uh, the vegetation information in particular was a tricky

05:19.775 --> 05:20.995
one to get hold of.

05:21.095 --> 05:24.755
So when we start initially with our work, there was a,

05:25.095 --> 05:27.835
repository of record of trees

05:28.425 --> 05:31.475
from the London authorities, which was not up to date.

05:31.475 --> 05:34.755
So we started looking at at least the relation of trees,

05:34.755 --> 05:37.555
number of trees or the placement of the trees

05:37.585 --> 05:41.155
with air quality in London, main city of London.

05:41.535 --> 05:43.355
And then we saw there is some correlations,

05:43.355 --> 05:45.595
but we wanted to correctly understand

05:46.385 --> 05:48.275
what is the exact relation.

05:48.375 --> 05:52.755
So we tried to, build AI models to detect trees,

05:53.175 --> 05:54.955
the, exact location of the trees,

05:55.015 --> 05:58.285
or sometimes even species of the trees from,

05:58.595 --> 05:59.765
maps or aerial images.

05:59.905 --> 06:02.645
So we, downloaded Google maps

06:02.645 --> 06:03.645
or Google Earth images,

06:04.105 --> 06:06.525
and then we tried to build our own models

06:06.665 --> 06:08.325
to estimate these three locations

06:08.325 --> 06:11.485
and vegetation, which we are taking forward to make,

06:11.705 --> 06:13.125
to understand the terrain as well.

06:13.125 --> 06:14.565
So terrain identification

06:14.565 --> 06:16.845
or terrain classification from these images.

06:17.425 --> 06:21.405
So once we have these information, actually we can,

06:21.675 --> 06:25.205
accommodate that as the features for our modeling,

06:26.275 --> 06:29.645
including, other, variations in,

06:29.665 --> 06:30.885
weather, for example.

06:31.385 --> 06:34.925
Uh, we could look at the seasonality, the timings

06:34.925 --> 06:36.525
of the days or the weekdays

06:36.545 --> 06:39.805
and other things into account while we are modeling,

06:39.985 --> 06:41.565
the air quality or the pollutants.

06:41.625 --> 06:46.045
So we put these all together, including, the tree counts

06:46.045 --> 06:49.365
or vegetation, which is, at different radius within the,

06:49.625 --> 06:53.165
emission or the source of the pollutants along

06:53.165 --> 06:56.725
with these other, weather, factors.

06:56.865 --> 06:58.965
And then we can model these more effectively.

06:59.305 --> 07:00.925
We, we try to use different models.

07:00.985 --> 07:03.245
So traditionally, people use,

07:03.275 --> 07:06.045
statistical models like Arima, Sima, they,

07:06.665 --> 07:09.165
and even multivariate, arima,

07:09.165 --> 07:12.165
where you can have multiple, features involved.

07:12.635 --> 07:13.645
Then we moved on

07:13.645 --> 07:16.405
to actually machine learning regression techniques like

07:16.405 --> 07:18.165
linear regression, poly normal regression

07:18.185 --> 07:20.845
or even support vector machines,

07:21.115 --> 07:22.925
with different types of kernels.

07:23.385 --> 07:26.525
And we moved, we also tested on the latest deep learning

07:26.525 --> 07:30.765
models like the recurrent neural networks, RNs, mainly

07:31.315 --> 07:33.525
ltms, long short-term memory models.

07:33.945 --> 07:36.045
So these are supposed to,

07:36.075 --> 07:38.525
work better when we have a lot, lots of data.

07:38.665 --> 07:41.805
And, our initial studies, we didn't have a lot

07:41.805 --> 07:44.325
of data, especially around vegetation, as you say.

07:44.385 --> 07:45.885
Uh, see, we have number

07:45.885 --> 07:48.125
of trees within like multiple radiuses,

07:48.125 --> 07:50.325
which are like just one or two counts.

07:50.665 --> 07:53.885
Uh, so it's, IM important that we have more data points.

07:54.145 --> 07:57.165
So these bigger models didn't perform very well.

07:57.225 --> 07:59.565
It was the standard machine learning models,

07:59.595 --> 08:01.325
like the support vector machines, that

08:01.555 --> 08:02.565
that worked really well.

08:02.625 --> 08:06.565
And we could get, performance close to like 0.8

08:06.625 --> 08:07.965
for the R two score, if

08:08.025 --> 08:10.805
that's the standard regression measure that you look at.

08:11.425 --> 08:15.685
Uh, what what we did was actually model each location

08:15.785 --> 08:19.485
of monitoring stations separately, so that actually,

08:19.805 --> 08:21.805
improved the performance rather than a generic model.

08:22.265 --> 08:25.085
Uh, and that's the similar studies have shown,

08:25.225 --> 08:29.405
in the literature that, these, if you have enough data

08:29.505 --> 08:30.885
or maybe large amount of data,

08:31.315 --> 08:33.365
deep planning models could perform better.

08:33.555 --> 08:37.125
Otherwise, the traditional Sima models even work,

08:37.585 --> 08:38.965
reasonably well.

08:39.555 --> 08:41.525
What I've been talking about is mainly

08:41.655 --> 08:43.245
about temporal modeling.

08:43.625 --> 08:45.925
So what we are saying is we have a location

08:45.935 --> 08:47.845
where we have a source of pollutant,

08:47.865 --> 08:50.165
and we want to see how we can predict

08:50.235 --> 08:52.845
what will be the amount of pollution at this point

08:53.465 --> 08:54.525
in, in time.

08:55.225 --> 08:58.205
Now, this has to be expanded to, spatial

08:58.205 --> 09:00.005
or spatial temporal aspect

09:00.005 --> 09:03.085
because we need to understand different,

09:03.765 --> 09:04.845
locations separately.

09:04.905 --> 09:07.965
And we can't afford to have monitoring stations every five

09:07.965 --> 09:09.005
meter or 10 meters.

09:09.385 --> 09:12.765
So there have been, integration techniques to see

09:12.865 --> 09:15.725
how we can interpolate these values

09:16.145 --> 09:18.045
to the other locations within the area

09:18.505 --> 09:22.005
and also have to predict temporal aspect of these locations.

09:22.465 --> 09:26.085
So that's what we had been, we had been thinking about.

09:26.225 --> 09:28.485
And then we started a case study with,

09:28.625 --> 09:30.925
the Colchester council, where we looked at

09:31.545 --> 09:32.605
two different routes.

09:32.605 --> 09:36.565
So they had a school around which we had a biking route

09:36.565 --> 09:39.525
and, a car route, motor route.

09:39.545 --> 09:43.605
So the idea is to, instill, citizen behavior

09:43.705 --> 09:44.805
around biking

09:45.065 --> 09:47.005
and walking rather, green modes of

09:47.565 --> 09:49.165
transport rather than, cars

09:49.165 --> 09:50.765
because it was causing,

09:50.875 --> 09:52.965
have in tra traffic in that area.

09:53.545 --> 09:57.475
So we tried to measure the air quality monitor with our own.

09:57.775 --> 09:59.955
Um, so instead of the big monitoring stations,

09:59.955 --> 10:02.235
we had handheld industrial grade devices.

10:02.855 --> 10:05.275
We even, built our own custom monitoring devices

10:05.305 --> 10:07.355
with raspberry pie and chips.

10:07.615 --> 10:11.915
So we used these to measure, air quality different

10:12.555 --> 10:15.835
locations within these two route, these two parallel routes.

10:16.215 --> 10:18.915
Uh, and we found that, mostly,

10:19.055 --> 10:21.155
the air quality is better in bike routes.

10:21.315 --> 10:22.875
I mean, especially the NO two.

10:23.135 --> 10:26.755
Um, sometimes the particular matter is bit higher in the,

10:26.855 --> 10:29.155
pa around the parks because there's trees

10:29.295 --> 10:30.755
and other dust particles there.

10:31.415 --> 10:34.955
But overall, the main problem we saw was

10:34.985 --> 10:38.595
that the bike routes, they are not isolated bike routes.

10:38.745 --> 10:42.275
They're next to the roads where there is heavy traffic.

10:42.455 --> 10:47.155
So it's not ideal to have, bike next

10:47.175 --> 10:49.835
to the, cars where the cars are passing next

10:49.835 --> 10:51.195
to the, the main roads.

10:51.535 --> 10:54.075
At least. If we can build, shields

10:54.075 --> 10:55.835
and canopies, as I talked about earlier,

10:56.225 --> 10:57.955
that would actually isolate them

10:57.955 --> 11:00.475
and help them, to improve their quality care

11:00.495 --> 11:04.235
and actually, encourage more citizens to bike

11:04.335 --> 11:07.035
and walk, if, if they have a cleaner air.

11:07.575 --> 11:09.755
So, as I said,

11:09.815 --> 11:12.955
we have been looking at these different factors that affect,

11:13.335 --> 11:15.995
so it's important to understand these factors,

11:16.095 --> 11:19.435
to understand the microclimate of the cities, so

11:19.435 --> 11:23.115
that we can bring in effective solutions, even rules,

11:23.385 --> 11:24.915
regulations, and policies.

11:25.335 --> 11:27.395
So these have been adopted in different

11:27.395 --> 11:28.515
cities around the world.

11:28.755 --> 11:32.915
I dunno if you've heard of gt a city in Belgium where,

11:32.935 --> 11:35.675
people are, dividing the city into zones,

11:35.675 --> 11:37.875
and there is no throw through traffic allowed

11:37.875 --> 11:41.875
between the zones, only bikes and cars can go, bikes

11:41.995 --> 11:44.435
and pe pedestrians can walk within the zones,

11:44.655 --> 11:46.755
and then if you have to go to a next zone, you have

11:46.755 --> 11:48.635
to go out to a ring road and then come back.

11:49.255 --> 11:51.555
So these kind of regulations might work,

11:52.095 --> 11:54.165
but it's important to test them as well.

11:54.305 --> 11:56.205
So it's, really expensive

11:56.225 --> 11:58.285
to test them practically in a city.

11:58.465 --> 12:02.245
So what, is the next solution is to look at,

12:02.245 --> 12:04.125
technology like digital twins,

12:04.455 --> 12:08.085
where we model all these factors, using AI

12:08.265 --> 12:10.165
and, other deep learning models.

12:10.345 --> 12:13.125
And then we can predict all these outputs

12:13.125 --> 12:16.685
and, pollutants based on these factors.

12:17.225 --> 12:20.565
We can even test a, specific policy

12:20.785 --> 12:23.085
or regulation in the digital twin world,

12:23.105 --> 12:25.725
in the virtual world, and see if it's effective

12:25.725 --> 12:27.525
or how much effective it is,

12:28.465 --> 12:30.685
and test the efficacy of these solutions

12:30.745 --> 12:32.685
before implementing them in the real world.

12:33.265 --> 12:37.605
Uh, so it's, it's, I believe this is the next step towards,

12:37.785 --> 12:41.845
improving air quality and, climate change.

12:42.065 --> 12:45.645
Uh, all in all in one, I think it's important to,

12:46.475 --> 12:50.445
educate the public to, to make them understand,

12:50.875 --> 12:54.245
what are the cause, what are the problems that's associated,

12:54.425 --> 12:58.805
and maybe inform them of the actions they need to take

12:59.305 --> 13:00.365
to contribute.

13:01.145 --> 13:02.725
So only if all the public

13:02.825 --> 13:06.605
and the citizens contribute to the, effective solutions

13:06.945 --> 13:09.245
of, climate change, it can happen.

13:09.425 --> 13:13.845
So I think it's, it's important that these technology,

13:14.095 --> 13:17.125
helps to inform the public about these, solutions.

00:00 - To give an introduction on air quality. Air quality, as the name stands, it's the quality of the air, and it's mainly the pollutants in the air that we're worried about. So there are different type of pollutants in the air, the gas pollutants, like carbon dioxide, carbon monoxide,

00:17 - or sulf sulfur, oxide or nitrous oxides and ozone, or even volatile gases. But there are also particulate matter, which are a bit more bigger particles of different sizes. So PM 10, pm 2.5, and things like that. It's been, an important topic recently, mainly because air quality has very adverse effect on human health and the pollutants. Stem cells can cause a lot of, respiratory diseases, even lung cancer, and sometimes even due

00:50 - to mortality in some cases. So at the same time, the air quality has been a great part in the climate change, which is a, recent topic of interest. So the carbon contents in these air pollutants, they contribute to the CO2 equivalent that we talk about, which is a main contributor to the climate change. So, these pollutants actually come out of, different sources, like transport or industry or even households in case of cities.

01:21 - And so, we have to understand, how, what the sources are, what their concentrations are in there and how to control them. so actually, WHO and even other national, standards have been put forward, to the maximum quantities of these or concentrations of these pollutants in the air to keep, to have a healthy living.

01:46 - In the past decade, air quality was not really monitored or measured. It was estimated. So there could be physical chemical models that understands the atmospheric interactions and different, chemical, conditions or the cloud conditions, and they would estimate the air, pollutant quantities or concentrations at different levels or heights. And then recently it has been moved, to monitoring or even prediction of air quality, that,

02:17 - could happen from, the historic data. So we have access to historic data now, and we have monitoring stations now, and we could have actually a data driven statistical prediction. Now, the thing here is most of these traditions, they try, try to see air quality as a linear time series, and we don't accommodate other factors that could actually affect the air quality. And some of the studies actually started looking at,

02:44 - weather as an important factor. So both weather and vegetation from literature and press studies we have seen has a great effect on air quality, mainly due to dispersion or absorption, for vegetation in in particular. So these studies actually are a bit controversial as well, because some of these studies show that vegetation has having both negative and positive effects, so we can, understand that the leaves actually absorb most

03:14 - of the geese pollutants. They're also, places where these PM particles, the bigger pollutant contents actually, they accumulate on these leaves and they stay there so they can take it out of the way. So we usually put up tree canopies if you have seen, next to these, highways and major roads, and actually that keeps our shields the rest of the world from these pollutants that's been dis sourced from the transport that's on the main road.

03:46 - Now, the problem there is if you have a canopy, maybe perhaps the pollutants, remain there like a canyon effect. So it, it, it has effects of microclimate, problems when you have, when you are not, modeling it, right? So that's why we, we, we started looking at the importance of vegetation and try to understand how this can be related to air quality modeling and how can we improve the prediction of the models,

04:15 - based on the other factors including, vegetation and maybe the factors of weather, like the seasonality that's not being taken into account in the past. We have actually, as I said, monitoring stations from, most of the councils or local authorities, especially UK or all over the world. The city's councils have been trying to, keep the quality of the air, at least monitor the quality of the air as much as per the regulations, the WH regulations.

04:44 - so the data from these monitoring stations, even though it's only like three or maybe at most five, 10 stations in a big city, is available online publicly. And so we have all the historic data that we need to build a, for example, a statistical machine learning or an AI model. so the place we had to, work around was to find the vegetation data, even the weather monitoring stations are available.

05:13 - So we have some information from the, from them. the vegetation information in particular was a tricky one to get hold of. So when we start initially with our work, there was a, repository of record of trees from the London authorities, which was not up to date. So we started looking at at least the relation of trees, number of trees or the placement of the trees with air quality in London, main city of London. And then we saw there is some correlations,

05:43 - but we wanted to correctly understand what is the exact relation. So we tried to, build AI models to detect trees, the, exact location of the trees, or sometimes even species of the trees from, maps or aerial images. So we, downloaded Google maps or Google Earth images, and then we tried to build our own models to estimate these three locations

06:08 - and vegetation, which we are taking forward to make, to understand the terrain as well. So terrain identification or terrain classification from these images. So once we have these information, actually we can, accommodate that as the features for our modeling, including, other, variations in, weather, for example. we could look at the seasonality, the timings of the days or the weekdays

06:36 - and other things into account while we are modeling, the air quality or the pollutants. So we put these all together, including, the tree counts or vegetation, which is, at different radius within the, emission or the source of the pollutants along with these other, weather, factors. And then we can model these more effectively. We, we try to use different models. So traditionally, people use, statistical models like Arima, Sima, they,

07:06 - and even multivariate, arima, where you can have multiple, features involved. Then we moved on to actually machine learning regression techniques like linear regression, poly normal regression or even support vector machines, with different types of kernels. And we moved, we also tested on the latest deep learning models like the recurrent neural networks, RNs, mainly ltms, long short-term memory models.

07:33 - So these are supposed to, work better when we have a lot, lots of data. And, our initial studies, we didn't have a lot of data, especially around vegetation, as you say. see, we have number of trees within like multiple radiuses, which are like just one or two counts. so it's, IM important that we have more data points. So these bigger models didn't perform very well. It was the standard machine learning models,

07:59 - like the support vector machines, that that worked really well. And we could get, performance close to like 0.8 for the R two score, if that's the standard regression measure that you look at. what what we did was actually model each location of monitoring stations separately, so that actually, improved the performance rather than a generic model. and that's the similar studies have shown, in the literature that, these, if you have enough data

08:29 - or maybe large amount of data, deep planning models could perform better. Otherwise, the traditional Sima models even work, reasonably well. What I've been talking about is mainly about temporal modeling. So what we are saying is we have a location where we have a source of pollutant, and we want to see how we can predict what will be the amount of pollution at this point

08:53 - in, in time. Now, this has to be expanded to, spatial or spatial temporal aspect because we need to understand different, locations separately. And we can't afford to have monitoring stations every five meter or 10 meters. So there have been, integration techniques to see how we can interpolate these values to the other locations within the area

09:18 - and also have to predict temporal aspect of these locations. So that's what we had been, we had been thinking about. And then we started a case study with, the Colchester council, where we looked at two different routes. So they had a school around which we had a biking route and, a car route, motor route. So the idea is to, instill, citizen behavior around biking and walking rather, green modes of

09:47 - transport rather than, cars because it was causing, have in tra traffic in that area. So we tried to measure the air quality monitor with our own. so instead of the big monitoring stations, we had handheld industrial grade devices. We even, built our own custom monitoring devices with raspberry pie and chips. So we used these to measure, air quality different locations within these two route, these two parallel routes.

10:16 - and we found that, mostly, the air quality is better in bike routes. I mean, especially the NO two. sometimes the particular matter is bit higher in the, pa around the parks because there's trees and other dust particles there. But overall, the main problem we saw was that the bike routes, they are not isolated bike routes. They're next to the roads where there is heavy traffic. So it's not ideal to have, bike next

10:47 - to the, cars where the cars are passing next to the, the main roads. At least. If we can build, shields and canopies, as I talked about earlier, that would actually isolate them and help them, to improve their quality care and actually, encourage more citizens to bike and walk, if, if they have a cleaner air. So, as I said, we have been looking at these different factors that affect,

11:13 - so it's important to understand these factors, to understand the microclimate of the cities, so that we can bring in effective solutions, even rules, regulations, and policies. So these have been adopted in different cities around the world. I dunno if you've heard of gt a city in Belgium where, people are, dividing the city into zones, and there is no throw through traffic allowed between the zones, only bikes and cars can go, bikes

11:41 - and pe pedestrians can walk within the zones, and then if you have to go to a next zone, you have to go out to a ring road and then come back. So these kind of regulations might work, but it's important to test them as well. So it's, really expensive to test them practically in a city. So what, is the next solution is to look at, technology like digital twins, where we model all these factors, using AI

12:08 - and, other deep learning models. And then we can predict all these outputs and, pollutants based on these factors. We can even test a, specific policy or regulation in the digital twin world, in the virtual world, and see if it's effective or how much effective it is, and test the efficacy of these solutions before implementing them in the real world. so it's, it's, I believe this is the next step towards,

12:37 - improving air quality and, climate change. all in all in one, I think it's important to, educate the public to, to make them understand, what are the cause, what are the problems that's associated, and maybe inform them of the actions they need to take to contribute. So only if all the public and the citizens contribute to the, effective solutions of, climate change, it can happen. So I think it's, it's important that these technology,

13:14 - helps to inform the public about these, solutions.

Faculti AI can make mistakes