Robust Machine Learning Algorithmic Rules for Detecting Air Pollution in the Lower Parts of the Atmosphere

Kassim Mwitondi; Hugo Wai Leung Mak

doi:10.5334/dsj-2025-027

Figures & Tables

Table 1

Layers of our atmosphere.

LAYER	DESCRIPTION & RELEVANCE TO HUMANITY
Troposphere	Closest to our habitat–stretching up to 10 km above earth. Its temperature decreases inversely with distance from the centre of the earth (approx. ${6.5}^{\circ} C$ per kilometre) (Omrani et al., 2022).
Stratosphere	Consists the majority of atmospheric ozone, which absorbs ultraviolet radiation and protects us from potential health risks. It is characterised by high temperatures over summer and lowest over the winter period (Xu et al., 2023).
Mesosphere	The temperature varies inversely with vertical height above ground (Laštovička, 2023).
Thermosphere/Ionosphere	Absorption of energetic ultraviolet & X-ray radiation from the sun, thus temperature increases with vertical height. They also vary between night and day as well as between seasons. It reflects and absorbs radio waves, allowing global radio wave transmission (Goncharenko et al., 2021).
Exosphere	Contains mainly oxygen and hydrogen atoms, but they rarely collide - they follow “ballistic” trajectories under the influence of gravity (Janches et al., 2021).
Magnetosphere	The outer region surrounding the earth, where charged particles spiral along the magnetic field lines, with the earth behaving like a huge magnet (Lu et al., 2022).

Table 2

Selected key air pollutants.

DATA ITEM	DESCRIPTION	DIMENSION & COMPLETENESS
${PM}_{10}$ (FSPMC)	Fine Suspended Particulates (FSP)	$2952 \times 27$ samples: $3.17 %$ missing
NO_2	Nitrogen Dioxide	$2952 \times 28$ samples: $2.18 %$ missing
$NO_x = NO+NO_2$	Nitrogen Oxides–in Hong Kong	$2952 \times 19$ samples: $2.67 %$ missing
O_3	Ozone	$2952 \times 114$ samples: $4.39 %$ missing
${PM}_{2.5}$ (RSPMC)	Respirable Suspended Particulates (RSP)	$2952 \times 27$ samples: $7.9 %$ missing
Time	Daily and hourly recording	From 00:00hrs to 23:00hrs of 1–31 January 2023, 1–30 April 2023, 1–31 July 2023 and 1–31 October 2023 (inclusive)
DayTimes	Discretised time periods of day	Night: 00:00-06:00hrs Morning: 06:00-11:00hrs Day: 12:00-18:00hrs Evening 19:00-23:00hrs
Period	Monthly weather periods	January, April, July and October

Data sources in southern China including Hong Kong and Macau.

Hourly averaged concentrations for all sampled pollutants.

Table 3

Daily and monthly averages.

START HOUR	END HOUR	HOURS RANGE	CATEGORIES OF DAY
00:00hrs $01^{st}$ -Jan-2023	23:00hrs $31^{st}$ -Jan-2023	$1^{st} - 744^{th}$ hour	Night, Morning, Day, Evening
00:00hrs $01^{st}$ -Apr-2023	23:00hrs $30^{th}$ -Apr-2023	$745^{th} - 1464^{th}$ hour	Night, Morning, Day, Evening
00:00hrs $01^{st}$ -Jul-2023	23:00hrs $31^{st}$ -Jul-2023	$1465^{st} - 2208^{th}$ hour	Night, Morning, Day, Evening
00:00hrs $01^{st}$ -Oct-2023	23:00hrs $31^{st}$ -Oct-2023	$2209^{th} - 2952^{th}$ hour	Night, Morning, Day, Evening

Table 4

Daily and monthly average concentrations of each pollutant.

AVERAGE TIME	${PM}_{10}$	${NO}_{2}$	${NO}_{x} = NO + {NO}_{2}$	$O_{3}$	${PM}_{2.5}$
PERIODS	$\symbf μ {g/m}^{3}$	ppb	ppb	ppb	$\symbf μ {g/m}^{3}$
Day	20.08	15.77	38.00	39.51	35.52
Evening	21.51	15.99	35.85	30.29	38.41
Morning	19.96	13.61	34.79	23.70	34.68
Night	19.60	11.05	22.37	25.47	35.35
January	28.36	15.30	39.38	26.06	45.01
April	19.51	15.93	33.47	34.17	41.77
July	9.05	11.69	35.56	29.44	23.05
October	20.14	14.49	29.79	35.79	33.84

Fine suspended particulates monthly and daily averages and variations.

Density distribution of pollutants across day time periods.

Pollution data points on a multidimensional scaling.

Table 5

Centroids of the selected clusters formed.

POLLUTANT	2 CLUSTER CENTRES	3 CLUSTER CENTRES	4 CLUSTER CENTRES
Fine Suspended Particulates ( ${PM}_{10}$ )	12.11 25.85	11.79 21.44 25.55	11.53 20.38 20.48 28.54
Nitrogen Dioxide (NO_2)	11.79 16.70	10.93 20.61 15.07	10.88 20.26 17.99 13.56
Ozones (O_3)	22.12 39.53	22.49 26.54 41.50	22.51 23.37 55.29 30.69
Nitrogen Oxides ( $NO_x = NO+NO_2$ )	29.77 34.85	25.51 58.79 28.43	25.44 60.33 35.99 25.34
Respirable Suspended Particulates ( ${PM}_{2.5}$ )	25.15 48.14	24.85 37.94 48.59	24.42 36.08 40.39 53.02

Actual and estimated ${PM}_{10}$ (Left) and ${PM}_{2.5}$ (Right).

Correlations among paired pollutant variables.

Table 6

Component loadings–contribution of each pollutant in each component.

	LOADINGS
POLLUTANT	COMPONENT 1	COMPONENT 2	COMPONENT 3	COMPONENT 4	COMPONENT 5
${PM}_{10}$	0.399		0.334	0.822	0.221
Nitrogen Dioxide ( ${NO}_{2}$ )	0.162	–0.200		0.203	–0.944
Ozones ( $O_{3}$ )	0.536	0.199	–0.816
Nitrogen Oxide ( ${NO}_{x}$ )	0.258	–0.935			0.225
${PM}_{2.5}$	0.679	0.203	0.469	–0.524

${PM}_{10}$ associations with daily and annual periods.

Table 7

Row points vs Principal Dimension 1.

	HIGH	LOW	MEDIUM
Day	484	4	373
Evening	352	3	260
Morning	369	20	349
Night	351	17	370

Table 8

Columns vs Principal Dimension 1.

	HIGH	LOW	MEDIUM
January	587	0	157
April	618	0	102
July	58	4	682
October	293	40	411

Table 9

Row points vs PD 1 for Day Times.

	DIMENSION 1	DIMENSION 2
Day	26.50596	3.576712
Evening	23.52921	3.377988
Morning	23.04545	49.011043
Night	26.91938	44.034257

Table 10

Columns vs PD 1 for Annual Periods.

	DIMENSION 1	DIMENSION 2
April	26.778387	2.672733
January	17.240575	4.196079
July	51.795246	22.524998
October	4.185793	70.606190

${PM}_{10}$ and ${PM}_{2.5}$ associations from Stations 1 and 8.

Robust Machine Learning Algorithmic Rules for Detecting Air Pollution in the Lower Parts of the Atmosphere

Figures & Tables

Table 1

Table 2

Figure 1

Figure 2

Table 3

Table 4

Figure 3

Figure 4

Figure 5

Figure 6

Table 5

Figure 7

Figure 8

Table 6

Figure 9

Table 7

Table 8

Table 9

Table 10

Figure 10

Paradigm

My account