Surface PM 10 Air Pollution of Karachi,Pakistan: Spatial-Temporal Statistical Modeling Using Aerosol-Optical-Depth Remote Sensing

Abstract

The levels of particulate matter (PM₁₀) in Karachi, Pakistan, are hazards to public health and environmental degradation. In this study, two statistical techniques, land use regression (LUR) and Pearson Correlation Coefficient (PCC), have been applied to detect PM₁₀ and Moderate Resolution Imaging Spectroradiometer (MODIS) along with four meteorological parameters. The average values of observed PM₁₀-MODIS aerosol optical depth (AOD)-predicted PM₁₀ have been analyzed with PCC correlation. Results, the M4 and M7 models are more reliable, where coefficient of determination (R² > 0.6) and root mean square error (RMSE) = 2–14. PCC showed a strong positive correlation with significant levels among observed PM₁₀-MODIS AOD-Predicted PM₁₀. The concentrations of PM₁₀ and AOD approximately increased by 182%, and 208%, respectively, and dropped by 28% to 30%, during COVID-19 (2020). In conclusion, utilize remote sensing with hybrid modeling to mitigate and monitor air pollutants in regions lacking ground-based air quality data resources.

Introduction

Respirable airborne dust particles with aerodynamic diameters of less than 10 μm particulate matter (PM₁₀) have a significant impact on the atmosphere (Huang, 2023), ecosystem (Sohrab et al., 2023), and human wellbeing (Sasmita et al., 2022). Its two main sources like anthropogenic and natural activities. Like vehicle movement, incomplete fossil fuel burning, relict dust from construction residuals (Gao et al., 2022), along with airborne pollutants including sea salt, soil, and dust, all of which can contribute greatly to the production of PM₁₀ (Wei et al., 2021). PM₁₀, or fine inhalable particles, makes up most of the mass concentration of PM, with diameters 10 ≤ μm, e.g., burning particles, carbon-based compounds, metals). It is well known that this diverse mass of particles can enter human lungs quickly, resulting in respiratory issues such as bronchitis and emphysema that increase hospital admissions and premature death(s) (Kumar et al., 2023). Consequently, PM₁₀ is quite concerning to all aspects of life, including the atmospheric environment and human health (Edwards et al., 2022). Considering its possible role in respiratory illnesses (Bălă et al., 2021), cancer (Smotherman et al., 2023), and heart diseases (Idrees et al., 2023), especially in industrial cities of underdeveloped countries such as Pakistan (Pakistan Population, 2023; World Bank, 2020).

Pakistan is today the world’s fourth-most polluted country (Macías-Hernández et al., 2023) and ranked fifth populated in the list of countries, equal to 2.83% of the global populace (Pedde et al., 2022). According to World Health Organization (WHO) research studies based on the air quality health index for Pakistan, air pollutants reduce the life expectancy of an individual Pakistani by 4.3 years, and PM₁₀ concentrations in the air reduce life expectancy by over 7 years in the most polluted cities (Fan et al., 2023). Due to feasibility and economic considerations, air quality continuous monitoring stations (CAMSs) must be present in Karachi City, Pakistan. The field measurements’ PM₁₀ small exposure to Karachi’s atmosphere may not represent the city’s various regions, which could lead to bias in epidemiological and environmental research (Ceamanos et al., 2023; Gao et al., 2022).

Determination of the mix source gets more complex in metropolitan settings, and identifying the main emission sources of ambient PM needs a great deal of work. The difficulty of identifying the predominant sources has become more pressing because of the dense dispersion of unstructured fugitive and nonducted area sources and the dearth of information regarding regional and long-range high emission sources (Bojjagani et al., 2023). Regarding reclamations of aerosol optical depth (AOD) from many platforms, remote sensing via satellite has become more accessible for PM₁₀ monitoring studies throughout a vast spatial domain, as well as for small regions and big cities (She et al., 2022). AOD measures the attenuation of light by suspended particles in the atmosphere and is a substitute for the atmospheric pollution level (Nguyen et al., 2024). Different platforms have recently been employed for PM₁₀-AOD research, such as the Multiangle Imaging SpectroRadiometer (Liang et al., 2023), the Spinning Enhanced Visible and Infrared Imager (Álvarez-Mendoza et al., 2023), the Medium Resolution Imaging Spectrometer (MERIS) (Gharibzadeh and Abadi, 2022; Ranjan et al., 2021), and Landsat (Payra et al., 2023; Zhang et al., 2022b). Moderate Resolution Imaging Spectroradiometer (MODIS) is extensively used in AOD-PM₁₀ research (Zhang et al., 2022b). Among the numerous remote sensing platforms, the MODIS has more precise tools than other satellite-predicted resources and has been used by many researchers and academics from 2002 to now (Ahmed et al., 2022; Kalisa et al., 2023; Wei et al., 2021).

Generally, Land Use Regression (LUR) models are employed in small regions and big cities where there is a shortage of CAMSs and a scarcity of ground measurements (Feng et al., 2023; Zoran et al., 2023). Therefore, spatial geographical information system (GIS) interpolation techniques cannot be applied to AOD databases, because of limited spatial coverage, such as inverse distance weighting, kriging, and natural neighbor (Hameed et al., 2021; Shareef and Hashmi, 2020). In the case of insufficient ground-based observed (Obs) PM₁₀ data or discontinuity of AOD data, LUR modeling is a helpful strategy for screening investigations and predicting datasets to fill gaps and improve modeling presentations (Sipra et al., 2021). This led to the development of LUR modeling based on weather studies and the annual PM₁₀ concentration of 10 monitoring locations. Using the covariance method, the Pearson Correlation Coefficient (PCC) assesses the degree and statistical significance of the association between two relevant continuous variables. Many researchers investigated the pearson correlation among atmospheric contaminations globally (Das et al., 2022).

The study period spans 2013 through 2021 and incorporates the COVID-19 epidemic’s consequences. The worldwide COVID-19 pandemic has presented unimaginable difficulties, disturbing routine activities, and production systems everywhere (Index of/archive/rgb/MODAL2_E_AER_OD). Researchers in Karachi examined the impact of lockdown protocols during the COVID-19 pandemic on air quality (Caido et al., 2022). This thinks about pointed to gauge the levels of the ground-based PM₁₀ and remotely detecting AOD and how the city’s air was influenced by lockdowns brought on by pandemics.

Utilizing MODIS AOD datasets from satellite and ground-based meteorological factors, such as temperature (TEMP), relative humidity (RH), wind speed (WS), and dew point, the goal of this work is to establish a link between PM₁₀ and AOD in Karachi. Parameters utilizing univariate and multivariate linear regression techniques to assess the surface PM₁₀ over the think-about region for nine a long time (2013–2021). Subsequently, the developed models were rigorously validated and evaluated.

The available studies on estimations of PM₁₀ concentrations, particularly those employing remote sensing hybrid modeling, still need to be explored in the study area. However, such methods often need to pay more attention to the complexities within specific cities, where localized emissions are crucial. This study, focusing on Karachi City, addresses these gaps by exploring the factors affecting PM₁₀ concentrations in a densely populated urban environment. The discoveries can upgrade future investigations, improving Karachi’s air quality assessments’ scope and exactness and aiding in developing targeted strategies to mitigate health risks associated with airborne PM in the region.

After a careful audit of concurrent writing and understanding the gaps, the display ponder was attempted with the taking after targets in Karachi are: a)

Study MODIS AOD data, a visualization-assisted tool that integrates with ground-based PM₁₀;

Statistical application to convert AOD data to predicted PM₁₀ values;

Estimate the expected error percent (EE %) between PM₁₀-AOD;

COVID-19 lockdown impacts on the atmosphere of Karachi;

Trend dynamics of observed PM₁₀ relationship with predicted PM₁₀.

The significance of this study is to analyze the negative impacts of PM₁₀, which are expected to worsen Karachi’s air quality and are heavily influenced by the city’s growing health burden and economic activities.

Materials and Methods

The proposed methodology is based on a study designed for a single ground-based air contamination PM₁₀ concentration analysis, integration with remote sensing satellite data in Metropolitan City Karachi, Sindh, and Pakistan. The ground-based ambient PM₁₀ preliminary analysis requires data ingestion, visualization, and deriving insights. The high-tech geostatic platforms are already geared with all possibilities to PM₁₀ data analytics and aid in environmental management decision-making and action. To achieve the above objectives for Karachi City, the construct real PM₁₀-AOD inversions to visualize distribution and variability trend levels into simple plots such as spatial-temporal maps, line charts, and bar charts.

Study area

With a population of over 24 million, Karachi is the largest city in Pakistan and the 12th most populated city worldwide. The focus of this particular study is on the quality of the air in Karachi. Karachi lies at the southernmost point of the nation, between latitudes 24°47'N and 25°12'N and longitudes 66°50'E and 67°37'E, along the Arabian Sea Coast, and its current area is 3,640 km² (Rahman, 2023). Karachi has an arid climate; moderate TEMP includes a low precipitation rate.

Figure 1 depicts the 10 sampling points located in urban areas of Karachi, which together make up 1500 km². Therefore, it depicts the regional features of urban cities in calm residential, commercially active, high-traffic zones, and industrial areas. The PM₁₀ observation stations in the study area cover large residential and commercial complexes including the five industrial zones of North Karachi, F.B. Area, Landhi, Korangi, and SITE.

FIG. 1.

Base map and data locations map of the study area.

Research framework

The research endeavors to contextualize and enhance the understanding of Karachi’s air quality dynamics. To achieve the goal of improving prediction LUR and PCC, the spatial-temporal analysis of PM₁₀-AOD along with meteorological variables. More details are given as follows by introducing the structure of this study, uses air quality ground-based measurements and Earth observations (EO) of atmospheric composition and historical trends, statistical prediction modeling (SPSS 22), and finally, all datasets analyzed and mapped at different spatial, temporal scales with ArcGIS (ArcMap).

PM₁₀ data extraction

The study required extensive ground-based PM₁₀ sampling (pollutant concentrations), geographic data, meteorological data, and preprocessed satellite remote sensing data. We used the field-based database from published provincial and federal agency data. The PM₁₀ concentration data of 10 selected sites in Table 1, from 2013 to 2018, were gathered from the Council of Scientific and Industrial Research in Pakistan. Yu et al., (2022), the Sindh Environmental Protection Agency, and Environmental Research Center collected the data for the remaining 3 years (2019–2021).

Table 1.

Data Location Points (Loc, ID, Long-Lat, and on Field Observations)

Sr. No.	Location	ID	Latitude	Longitude	(industrial, commercial, residential, and traffic flow)
1	Karimabad	Krmb	24.92191	67.06168	Commercial, terrific thickly populated residential.
2	Tibet Center	TbtC	24.86287	67.02161	High commercial and residential areas, dense traffic
3	Liaquatabad	Lqtd	24.90025	67.04492	Junction of traffic, heavy residential, and commercial
4	PIB Colony	PIBC	24.89035	67.05222	Thickly populated and commercial, high-traffic
5	Nizamabad	Nzmd	24.91037	67.03067	Commercial, densely populated, heavy traffic.
6	Civic Center	CvcC	24.90061	67.07317	Residential and commercial, high traffic flow.
7	SITE Siemens	Seim	24.90205	67.00034	Commercial and residential, heavy traffic flow.
8	SITE Nuras	Nurs	24.90554	67.01569	Heavy industrial, densely populated, and heavy traffic.
9	Singer Korangi	SngC	24.84606	67.16061	Industrial, commercial active, high traffic, and residential.
10	Shan Korangi	ShnC	24.83837	67.12036	Heavy industrial, densely populated, heavy traffic.

Level 2 Tera/Aqua MODIS AOD (Aerosol-L2-3 Km) data of 1-month average (54 observations) images for 6 months from March to August 2013–2021 were downloaded from the NASE Earth Observations (NEO). NEO MODIS (Tera/Aura) AOD data have collected a spatial resolution of 0.1^° × 0.1,^° which is a merged 0.55 μm combination of dark target and deep blue for the Karachi region study https://neo.gsfc.nasa.gov/ (Shin et al., 2020; Tella and Balogun, 2021).

The four meteorological parameters data included TEMP, RH, WS, and dewpoint (DP) retrieved from the Pakistan Meteorological Department’s page and Wunderground.com (Nakada and Urban, 2020). The historical weather data of Karachi for 9 years in Table 3, from 2013 to 2021, were obtained corresponding to 10 locations and periods of PM₁₀ measurements at PM₁₀ at ground level.

PM₁₀-AOD data diffusion

The diffusion between satellite-based AOD and in situ PM₁₀ concentrations changes dramatically with regional climate variables, local emissions, and chemical composition. To consider these variations, all the prediction factors were determined per stations, time, and stack predators in 6 months, according to the time period of all PM₁₀ observations. Along with the univariate basic formula equation applied on M1 (PM₁₀-AOD), multivariant (M2–M7) linear regression formulas are used to integrate the PM₁₀ with AOD and selected meteorological variables (Zhang et al., 2022b). The performance of linear regression models output results is estimated in terms of intercept (β) and slopes (αAOD, αTEMP, αRH, αWS, αDP, and αPBLH) of predictors, respectively. Initially, the statistical model was built with just one variable as the dependent variable (PM₁₀) and dependent variables AOD, and meteorological factors stepwise one by one in sequence as the independent variables according to the model formula in Table 2 and vice versa.

Table 2.

Regression Models and Speciﬁcation

Model	Predictors description
M1	PM₁₀ = β + α_AOD AOD
M1	AOD as the only predictor
M2	PM₁₀ = β + α_AOD AOD + α_TEMP TEMP.
M2	AOD and one meteorological variable as predictors
M3	PM₁₀ = β + α_AOD AOD + α_TEMP TEMP + α_RH RH.
M3	AOD and two meteorological variables as predictors
M4	PM₁₀ = β + α_AOD AOD + α_TEMP TEMP + α_RH RH + α_WS WS.
M4	AOD and three meteorological as predictors
M5	PM₁₀ = β + α_AOD AOD + α_TEMP TEMP + α_RH RH + α_WS WS + α_DP DP.
M5	AOD and four meteorological variables predictors
M6	PM₁₀ = β + α_AOD AOD + α_ER ER%
M6	AOD and Error% as the predictors
M7	PM₁₀ = β + α_AOD AOD + α_TEMP TEMP + α_RH RH + α_WS WS + α_DP DP + α_ER ER%.
M7	AOD, error%, and four meteorological variables from synoptic stations as predictors

Table 2 represents the simulation carried out using the seven LUR models to predict PM₁₀ in Karachi City. These LUR models predicted the air quality of the atmosphere from the optimum data of metrological profiles (Zhang et al., 2022b). The results of these regression models simulation are collected in terms of intercept β and slopes, respectively, αAOD, αTEMP, αRH, αWS, and αDP. The statistical analysis of the seven LUR model’s performance was assessed by subset coefficients: that is, coefficient of determination (R²), adjusted regression coefficient R², root mean square error (RMSE), Durbin Watson (DW), and EE%.

Figure 2. presents the data diffusion workflow of statistical simulation and spatial analysis of atmospheric variables of air quality dynamics of Karachi region insights using LUR and PCC modeling techniques for ground-based PM₁₀ and MODIS AOD.

FIG. 2.

Workflow of LUR and PCC modeling for GB PM₁₀-AOD data diffusion. LUR, land use regression; PCC, Pearson correlation coefficient; PM₁₀-AOD, particulate matter-aerosol optical depth.

The EE% is the difference between the PM₁₀ actual value and the expected value of AOD. For the dispersion measurement of AOD-PM₁₀ statistically to calculate and may predict error (Zuyi et al., 2022). $Accuracy Error = (1 - AOD / {PM}_{10})$ $Expected % = (100 \times accuracy error)$ (1)

The ratio of expected error in the percentage of PM₁₀ AOD is represented as EE%. The LUR models provide correctness of data by counting the number of pixels from NEO-AOD repossessions with ArcMap.

Results

The results of the present study were acquired by applying seven LUR models to estimate the surface PM₁₀ concentrations variable from 2013 to 2021. The hybrid integrates the observed PM₁₀ data as a dependent variable with AOD derived from MODIS imaginary, including TEMP, RH, WS, and DP as independent variables. These models are developed to predict the surface PM₁₀ levels to compare with observed PM₁₀. These five independent parameters are crucial in predicting PM₁₀ concentrations over the specified period.

Table 3 shows the data values of observed (Obs:) PM₁₀, MODIS-AOD retrievals, and metrological variables during the study period. The minimum and maximum values for the same time and space were displayed at each of the 10 selected data sites. In this study, the criteria to consider PM₁₀ as the dependent variable and six independent variables indicated in Table 3 as predictors to predict near-surface PM₁₀. The outcomes demonstrated in Table 3 are that the variation in surface PM₁₀ concentration are consistent with AOD variance and display data values for PM₁₀ and AOD that have increased by nearly 40% over the past 9 years. The MODIS AOD considers only those datasets with at most 10% cloud density. The regression analysis produces the coefficients in constants and slopes between PM₁₀-AOD and all other inputs.

Table 3.

Annual Min–Max Observed PM₁₀, AOD, and Metrological Parameters

Year	Range	PM10 μg/m³	AOD unitless	Error%	TEMP ^oC	WS m/sec	RH %	DP ^oC
2013	Min	112	121	46	28	4	46	17
2013	Max	139	134	57	30	5	49	19
2014	Min	120	80	55	28	4	49	17
2014	Max	135	97	57	30	5	51	19
2015	Min	168	98	56	31	4	48	17
2015	Max	201	112	62	33	6	50	19
2016	Min	129	101	49	29	5	49	17
2016	Max	169	108	58	31	6	52	20
2017	Min	145	63	56	27	4	49	17
2017	Max	187	76	60	31	5	51	19
2018	Min	187	70	48	28	5	49	17
2018	Max	231	73	56	31	6	53	19
2019	Min	193	71	44	28	4	50	16
2019	Max	224	83	60	32	5	53	19
2020	Min	167	52	48	26	4	49	17
2020	Max	201	61	56	31	5	52	19
2021	Min	213	55	42	28	5	46	18
2021	Max	229	67	48	32	6	49	19

Figure 3 demonstrated a correlation between the observed PM₁₀ with AOD, and meteorological parameters were evaluated, and the results identified the influence of different meteorological parameters (TEMP, RH, WS, and DP) on PM₁₀-AOD concentrations. The study period clearly shows that the TEMP trends are in phase (positive correlation) and RH-DP are out of phase (negative correlation) with PM₁₀-AOD concentrations. A negative correlation was found between RH-WS-DP and PM₁₀-AOD concentrations, indicating the predominance of local climatic patterns. RH-DP trend showed a very low negative correlation with both PM₁₀-AOD due to the near homogeneity observed in the daily RH values. It revealed a significant influence of TEMP on PM₁₀ compared with the concentration of RH-WS-DP parameters in the Karachi City atmosphere (Zhang et al., 2022b).

FIG. 3.

Averaged PM10-AOD with meteorology variables during the study period (2013–2021).

Statistical model validation

The LUR quantitative studies have consistently revealed that greater PM₁₀ prediction accuracy can only be attained when the various predator data are included as inputs. Nevertheless, evaluating how much each climatic parameter is improving the prediction performance is also crucial. Nevertheless, the most effective predictive models are M4 and M7, the years of modeling were 2014, 2018, and 2020, which were repeated with multiple iterations, keeping each alternate input parameter used and analyzed.

Table 4 shows the regression analysis produces the coefficient results in the form of constants and slopes between PM₁₀-AOD with all other inputs. The univariant and multivariate linear regression output parameters of modeling in terms of coefficient of determination “R²,” “adjusted R²,” “RMSE,” Durbin-Watson. The AOD has a very positive coefficient correlation with PM₁₀ in all models and all years of analysis. The regression modeling coefficient analysis of all climatic variables was randomly associated with predicted (Pred:) PM₁₀ over Karachi City locations, as observed by recent studies in other parts of the world in recent years (Hameed et al., 2021). The TEMP, RH, and WS have the majority of positive values, showing that PM₁₀ is better associated with TEMP and WS, while it is more sensitive to RH and DP. Hence, higher RH and dew points lead to a significant drop in PM₁₀ pollution, and TEMP and WS increase the PM₁₀ concentrations (Maleki et al., 2022).

Table 4.

Regression Output Summaries, R²; Adjusted R²; RMSE; Durban Watson

Model	Range	Year	R ²	Year	Adj: R²	Year	RMSE	Year	D/W
M1	Min	2017	0.065	2017	−0.52	2017	4.12	2017	0.595
M1	Max	2014	0.447	2014	0.378	2014	13.91	2014	2.33
M2	Min	2019	0.128	2015	−0.212	2014	3.452	2018	0.577
M2	Max	2014	0.634	2014	0.575	2018	14.76	2014	1.88
M3	Min	2015	0.178	2019	−0.212	2014	3.45	2018	0.765
M3	Max	2014	0.715	2016	0.565	2018	14.76	2013	2.18
M4	Min	2019	0.211	2018	−0.421	2014	3.66	2020	0.955
M4	Max	2018	0.715	2014	0.51	2015	15.01	2013	2.303
M5	Min	2019	0.226	2019	−0.716	2014	0.934	2019	1.22
M5	Max	2014	0.986	2014	0.543	2016	13.34	2018	2.53
M6	Min	2018	0.996	2019	0.994	2014	0.183	2013	1.53
M6	Max	2014	0.999	2014	0.999	2018	10.57	2019	2.371
M7	Min	2017	0.997	2017	0.991	2014	0.234	2013	1.795
M7	Max	2014	1	2014	0.998	2017	14.37	2019	2.79

Figure 4 demonstrates the distribution of the annual mean of observed PM₁₀ at Karachi City compared with the two most favorable models, predicted PM₁₀ for 9 years. It estimated with the help of models M4 and M7, that both multivariable linear regression models showed better coherence with observed PM₁₀ and indicate that the variations in concentration of PM₁₀ are consistent with AOD variance.

FIG. 4.

Mean annual observed PM₁₀ with Predicted PM₁₀ (M4 and M7).

The EE% was calculated among the observed PM₁₀-AOD using equation no. (1). The computation between observed PM₁₀ and AOD has the EE% value of 53 ± 4% (47%–56%). Therefore, this study’s EE% computing accuracy is much better, and the EE% gave better results in predicting near-surface PM₁₀ contaminations. It is a common practice to estimate the EE% to validate the Aeronet-AOD modeling performance EE% and recently were estimated at 59.15% and 57.03% (Zhang et al., 2022b). In the computation of EE% with MODIS AOD as independent variables in model M7, values of R² were 1 or very near to 1 and DW nearly 2; approximately 100% model fitting suggests the accuracy of EE% estimation between PM₁₀ and MODIS AOD data.

Statistical model performance

Fixed effects intercepts (β) and coefficients (α) present the average values for the duration of the study, while the random effects depict the PM₁₀-AOD relationship’s time-varying properties. We run seven models with different independent variables to check their impact on PM₁₀ in year-wise capacity. In all models, a constant value is highly positive except in 2016 and 2018. The coefficient sign of AOD is also positive in all years except 2019 and 2020. AOD is the sole variable that provides a minimum effect on PM₁₀ in comparing the rest of the independent variables. The RH coefficient is also positive in all years except 2013, 2018, and 2020.

Table 5 presents the annual minimum and maximum values of AOD slop (αAOD) and intercept (β) for the seven LUR models. Analyzing all the models, the β value is generally positive except for 2016 and 2018. The α of the AOD is positive except for 2019 and 2020. Examining the regression coefficients in Table 5, positive signals show a favorable link between PM₁₀ and AOD, while negative values point to the reverse relationship. Analyzing the three tables collectively, the all-models constant value is mostly positive except in 2016 and 2018, and a Coefficient sign of AOD is also positive in all years except 2019 and 2020.

Table 5.

Model Performance Intercept (β) and AOD Slope (αAOD)

Model	Range	Year	β	Year	α_AOD
MI	Min	2018	−12.03	2020	−1.05
MI	Max	2019	280.03	2016	2.718
M2	Min	2013	−77.89	2020	−1.63
M2	Max	2020	621.62	2016	3.5
M3	Min	2014	−39.32	2016	−3.81
M3	Max	2020	646.9	2018	2.185
M4	Min	2013	−58.8	2016	−3.92
M4	Max	2020	621.2	2018	1.78
M5	Min	2013	−66	2016	−2.37
M5	Max	2020	898.1	2017	2.477
M6	Min	2018	−225	2019	1.64
M6	Max	2017	13.54	2020	2.363
M7	Min	2013	−399	2019	1.62
M7	Max	2019	52.13	2020	2.33

Figure 5 illustrates the annual average variations in PM₁₀-AOD concentrations in Karachi over the observation period. It shows that the R² in both series is almost 70%, with positive intercepts (β) and slopes (α) of 10 and 7, respectively. The regression analysis of observed PM₁₀ shows the best-fitting results and is identical to AOD outputs Figure 5. The trends of both the averaged datasets are identical and should validate the ground measurements and remote sensing AOD retrievals. Figure 5 shows a comparison between observed PM₁₀ and predicted PM₁₀ and shows depression during the COVID-19 lockdown period from 2019 to 2021. The annual averaged PM₁₀-AOD concentrations in Figure 5 varied from 125 to 220 μg/m³ and 56–124 (unitless). In more than 70% of the observations, levels of PM₁₀ were significantly higher than Pakistan’s National Ambient Air Quality Standards (150 μg/m³) and the WHO’s (50 μg/m³).

FIG. 5.

Annual averaged PM10-AOD comp is, intercept, R², slope, and COVID-19 anomaly.

Pearson correlation and comparison among Obs: PM₁₀-MODIS AOD-Pred: PM₁₀

PCC results reveal that the AOD concentration level in the atmosphere is related to the PM₁₀ contamination formation at the surface and near the surface of the Karachi area.

Table 6 revealed that Obs: PM₁₀, Pred: PM₁₀, and MODIS AOD depicted strong positive correlations with r values ranging more than 90% with highly significant values p = 0.01. The atmospheric concentration of AOD observed PM₁₀ and predicted PM₁₀ were all highly consistent. Thus, a strong correlation was found at a 0.01 significance level among the concentration of observed PM₁₀-AOD (r = 0.919, p = 0.01), observed PM₁₀-predicted PM₁₀ (r = 0.998, p = 0.01), and predicted PM₁₀-AOD (r = 0.922, p = 0.01)

Table 6.

Correlation Among the Obs: PM₁₀, MODIS AOD, and Pred: PM₁₀

		Obs: PM₁₀	AOD	Perd: PM₁₀
Obs: PM₁₀	Pearson correlation	1	0.919^a	0.998^a
AOD	Pearson correlation	0.919^a	1	0.922^a
Perd: PM₁₀	Pearson correlation	0.998^a	.922^a	1

Correlation is significant at the 0.01 level (two-tailed).

Figure 6 demonstrates the ground-based observed PM₁₀ measurements of 9 years (2013–2021) at Karachi City are compared with corresponding MODIS AOD and Pred: PM₁₀ values. Three parts of Figure 6 depicted the comparison of Obs: PM₁₀-Pred: PM₁₀ (R² = 0.9954, RMSE = 0.27 μg/m³)_, Obs: PM₁₀-MODIS AOD (R² = 0.845, RMSE = 21.2 μg/m³) and Pred: PM₁₀-MODIS AOD (R² = 0.8492, RMSE = 21.5 μg/m³). The best comparison among the three was observed and predicted PM₁₀ concentrations. High values in all three univariant regression analyses of the R² indicated demonstrated in Figure 6 (a, b, and c) that PM₁₀-AOD measurements and predicated PM₁₀ concentration have at the same time and with the same relative magnitude during the analysis period.

FIG. 6.

Comparison among the observed PM₁₀, MODIS AOD, and predicted PM10: (a) Obs: PM₁₀-Perd: PM₁₀, (b) Obs: PM₁₀-MODIS AOD, and (c) Perd: PM₁₀-MODIS AOD.

COVID-19 impact on PM₁₀-AOD of Karachi air

WHO announced the COVID-2019 (SARS-COVID-19) on March 11, 2020, and the lockdown was deemed the most effective tool in ceasing the spread of the devastating coronavirus infection globally. The study’s key findings displayed that the AOD levels in Karachi urban areas greatly reduced (∼28%) through the lockdown times (March 25 to July 15, 2020) related to the overtime mean AOD level from 2013 to 2019. However, the increase in AOD concentration levels was ∼30% during the same period of 2021. The reduction in ground base PM₁₀ levels due to COVID-19 in Karachi, Pakistan, during March–April 2020 related to the same duration in 2019 was decreased by ∼17% in 2019, and an increase of ∼24% concentrations in 2021 was observed during the lockdown over Karachi due pandemics.

Discussion

This study explores the correlation between PM₁₀ levels and AOD from MODIS satellite imagery in Karachi City, Pakistan. It also incorporates ground-based meteorological parameters such as WS, dew point, TEMP, and RH into the analysis. Using univariate and multivariate linear regression methods, PM₁₀ concentrations are forecasted across the study area over 9 years (2013–2021). It is evident from the results that models M4 and M7 consistently yield favorable results with consistent signs of independent variables. These models exhibit favorable statistical metrics, including Durbin-Watson, R², adj R², and RMSE, as depicted in Figure 4. Notably, models M4 and M7 demonstrate R² values exceeding 60% across all years, along with Durbin-Watson statistics approaching two in nearly all instances.

Moreover, the RMSE indicates a minimal mean square error for these models, underscoring their reliability. Consequently, we endorse using these models for PM₁₀ prediction, aligning with recent research findings. Despite employing statistical equations (Table 2), for near-surface PM₁₀ estimation, model summaries and coefficient analyses affirm the efficacy and significant coherence of formulas M4 and M7 with observed PM₁₀ levels. Figure 4 compares observed and predicted PM₁₀ using multivariate linear regression models M4 and M7, revealing their superior performance and robust correlation. Notably, the observed–predicted PM₁₀ relationship exhibits fluctuations during the COVID-19 lockdown period from 2019 to 2021.

Recent studies globally have integrated ground-based PM₁₀ measurements with MODIS AOD retrievals to develop hybrid models. The authors compared AOD retrievals with observed PM₁₀ pollution levels in Quito, Ecuador, employing various meteorological parameters and environmental indices as predictors to formulate an optimal equation for citywide PM₁₀ estimation. Zhang et al. (2022a) applied nonlinear LUR models based on generalized additive models to predict air pollutant concentrations and AQI in Beijing, achieving adjusted R² levels ranging from 70% to 90%. The researcher utilized a hybrid model combining a chemical transport model and geographic time-weighted regression to estimate PM_2.5 and PM₁₀ concentrations in mining cities of China, with an R² of 0.926 and RMSE of 16.67 μg/m³ for PM₁₀ (Zhang et al., 2022a). Munir et al. (2021) also generated NO₂ maps and estimated concentrations using LUR models, and RS data fused with measurements from low-cost sensors in Sheffield, U.K. The authors explored 10 years of data in Ahvaz, Iran, constructing linear and nonlinear multivariable regression models using PM₁₀-AOD data and meteorological variables to estimate PM_2.5 and PM₁₀ concentrations, achieving a forecast accuracy of 64% for PM₁₀ concentrations.

During the COVID-19 pandemic, various studies across different regions consistently reported significant reductions in PM₁₀ concentrations, including in Karachi City, Pakistan. In comparison, studies of COVID-19, the following hybrid studies, were carried out over the globe during the COVID-19 lockdown. In Brazil, authors observed PM₁₀ concentration decreased from −24% to −62% (Martins et al., 2024); Malaysia’s Klang Valley found a reduction in PM₁₀ concentration from −51.8% to +14.2% (Rahim et al., 2023), and concluded PM₁₀ concentration decreased from approximately −56.6% to −51.9%. Finally, combined air pollutants monitoring data with satellite data in Singapore (Mailepessov et al., 2024), revealing a −23.0% decrease in PM₁₀ concentration values. This reduction in air pollution emphasizes the environmental impact of such interventions and the significance of long-term practices in addressing air quality challenges. Multiple cities across India found consistent reductions in PM₁₀ during the same period in 2019–2022, ranging from −24% to −62% (Bojjagani et al., 2023; Kumar et al., 2023). These studies, many of which were done in neighboring countries, clearly demonstrate that PM₁₀ concentrations reduced significantly during the COVID-19 pandemic. The resulting reduction in air pollution emphasizes the environmental effects of these interventions and the importance of long-term strategies in resolving issues with air quality.

From a future perspective, the results of this study indicate that the population and environmental ecology of Karachi City will be significantly impacted by the application of hybrid modeling for PM₁₀-AOD using LUR in conjunction with meteorological data. By adding meteorological factors, the hybrid model can more correctly estimate PM₁₀ concentrations, allowing for targeted efforts to lessen hotspots for air pollution. This is particularly crucial for Karachi City, as the ecosystem can benefit from improved air quality management, yet excessive PM concentrations can have detrimental effects on human health. Future generations’ well-being and Karachi City’s ecological balance will be enhanced as the research’s conclusions are put into practice. In order to improve the model’s prediction power and expand its suitability for Karachi’s varied urban surroundings, governmental and commercial institutions ought to support and incentivize researchers to carry out ongoing studies in this area.

The future directions of research revealed that the integration of social and geographical features with meteorological elements in hybrid LUR models holds significant potential to improve efficiency and accuracy in future research paths related to PM₁₀-AOD modeling in Karachi City and has the potential to foster a healthier urban environment, safeguarding both humans.

Geographical indices (cover and topography) and social parameters (population density, industrial activities, GDP, and vehicular traffic) are pivotal in shaping air quality in urban areas such as Karachi. Utilize more remote satellite sensors such as MLS, CALIPSO, MERIS, SCIAMACHY, VIIRS, OMI, or Tropomi with MODIS for PM₁₀-AOD integration to enhance the efficiency of LUR hybrid modeling in Karachi as a future perspective in PM₁₀-AOD modeling. This multisensory approach allows for a more detailed understanding of the spatiotemporal variability of PM, considering both local and regional sources. The synergy of these satellite datasets not only refines the accuracy of predictions but also aids in identifying specific emission hotspots and understanding the dynamics of air quality in the urban environment.

Conclusion

This study quantitatively analyzed to predict and correlate with strength near surface PM₁₀-AOD in Karachi urban through LUR (seven models) and PCC techniques between 2013 and 2021. It is concluded the most consistent results with the same sign of independent variables are found in models M4 and M7. DW, R², adj R^2, and RMSE (μg/m³) results are quite good for models M4 and M7. In these models, R² in all years is greater than 60%, and DW is also near two in almost all years. RMSE also indicates the lowest mean square error for such models. From this analysis, it is feasible to draw conclusions: (1) traditional LUR models can be replaced with empirical models that are built using solely remote sensing data as predictors, (2) the outcomes of the linear regression models comparison show that good and most robust result are produced by model number M4 and M7, (3) PCC correlation among the three datasets are very high with strong strength levels, (4) the overall average values increase in nine years, observed PM₁₀ and AOD 182% and 208% respectively, (5) EE% have very similar temporal patterns (PM₁₀-AOD) in average 53 ± 4% is observed in the case of Karachi, can be used to estimate PM₁₀ at locations without PM₁₀ data coverage and also can fill in-gaps in AOD, (6) generally, in all years of the month of April have nearly 15% to 20% higher values with compare to other months in AOD data (TEMP affect), (7) MODIS data pixel point reading it was a general trend that the pixel values near the coastal areas have less values with compare to city center values (humidity affect), and (8) COVID-19 lockdown clearly exabit in the data blanch in both PM₁₀-AOD datasets. The spatial and temporal trend of PM₁₀ predictions matched the spatial and temporal trends of PM₁₀ measurements well in Karachi City.

Footnotes

Acknowledgments

The Environmental Research Center at Bahria University Karachi Campus in Pakistan.

Authors’ Contributions

A.D. and Y.N.: designed the experiment. A.D.: performed the field works. Y.N. and H.N.: carried out the data analyses. J.A.B. and K.G.: helped to gather data performed the artwork and wrote this article. K.G.: revised this article. All authors contributed to the discussion of the data and have approved for submission.

Author Disclosure Statement

No competing financial interests exist.

Funding Information

This research received no any external financial support.

References

Ahmed

, Shams

, Ahmed

, et al. Spatio-temporal variations of lower tropospheric pollutants and their relationship with meteorological factors in Karachi, Pakistan. Arab Gulf J of Sci Res, 2022; 40(2); doi: 10.21203/rs.3.rs-1077688/v1

Álvarez-Mendoza

, Lopez

, Vasquez

. Monitoring air quality using remote sensing based on a Google Earth engine application in countries with limited air quality data and control policies: A case study in Ecuador. In Remote Sensing Technologies and Applications in Urban Environments 2023;VIII(Vol. 12735, pp. 80–85). SPIE; 2023; doi: 10.1117/12.2672325

Bălă

, Râjnoveanu

, Tudorache

, et al. Air pollution exposure—the (in) visible risk factor for respiratory diseases. Environ Sci Pollut Res, 2021; 28(16):19615–19628; doi: 10.1007/s11356-021-13208-x

Bojjagani

, Mahanta

, Prasad

, et al. Source apportionment and analysis of spatial representativeness of fine particle pollution for an urban residential area in Lucknow, India. Environ Eng Sci, 2023; 40(12):678–688; doi: 10./1089/ees.0120

Caido

, Ong

, Rempillo

, et al. Spatiotemporal analysis of MODIS aerosol optical depth data in the Philippines from 2010 to 2020. Atmos, 2022; 13(6):939; doi: 10.3390/atmos13060939

Ceamanos

, Six

, Moparthy

, et al. Instantaneous aerosol and surface retrieval using satellites in geostationary orbit (iAERUS-GEO)–estimation of 15 min aerosol optical depth from MSG/SEVIRI and evaluation with reference data. Atmos Meas Tech, 2023; 16(10):2575–2599; doi: 10.5194/amt-2023-1

Das

, Nergis

, Butt

, et al. Spatial-temporal re-analysis of seasonal air quality data in karachi city for future prospects. IJEEG, 2022; 13(3):38–44; doi: 10.46660/ijeeg.v13i3.41

Edwards

, Wilkinson

, Rutter

, et al. Health effects in people relocating between environments of differing ambient air pollution concentrations: A literature review. Environ Pollut, 2022; 292(Pt A):118314; doi: 10.1016/j.envpol.2021.118314

Fan

, Ma

, Jin

, et al. Validation, analysis, and comparison of MISR V23 aerosol optical depth products with MODIS and AERONET observations. Sci Total Environ, 2023; 856(Pt 1):159117; doi: 10.1016/j.scitotenv.2022.159117

10.

Feng

, Wu

, Li

, et al. Effect of transportation operation on air quality in china based on MODIS AOD during the Epidemic. Sustain, 2023; 15(5):4064; doi: 10.3390/su15054064

11.

Gao

, Liu

, Xu

, et al. Environmental pollution analysis and impact study—A Case Study for the Salton Sea in California. Atmos, 2022; 13(6):914; doi: 10.3390/atmos13060914

12.

Gharibzadeh

, Abadi

ARS

. Estimation of surface particulate matter (PM2. 5 and PM10) mass concentration by multivariable linear and nonlinear models using remote sensing data and meteorological variables over Ahvaz, Iran. Atmos Environ, 2022; 14:100167; doi: 10.1016/j.aeaoa.2022.100167

13.

Hameed

, Khan

, Fatmi

, et al. Exploring the relationship between air quality and ischemic stroke admissions during the COVID-19 pandemic. J Stroke Cerebrovasc Dis, 2021; 30(8):105860; doi: 10.1016/j.jstrokecerebrovasdis.2021.105860

14.

Huang

. The impact of PM10 and other airborne particulate matter on the cardiopulmonary and respiratory systems of sports personnel under atmospheric exposure. Atmos, 2023; 14(11):1697; doi: 10.3390/atmos14111697

15.

Idrees

, Nergis

, Irfan

. Industrial emission monitoring and assessment of air quality in Karachi Coastal City, Pakistan. Atmos, 2023; 14(10):1515. https://www.worldometers.info/world-population/pakistan-population/

16.

Kalisa

, Zhang

, Igbawua

, et al. Spatial and temporal heterogeneity of air pollution in East Africa. Sci Total Environ, 2023; 886:163734; doi: 10.1016/j.scitotenv.2023.163734

17.

Kumar

, Sarma

, Pandey

, et al. COVID-19 Lock-down in Delhi: Understanding trends of particulate matter in context of land-use patterns, GIS mapping, and meteorological traits. Environ Eng Sci, 2023; 40(1):1–12; doi: 10.1089/ees.0020

18.

Liang

, Daniels

, Bailey

, et al. Integrating low-cost sensor monitoring, satellite mapping, and geospatial artificial intelligence for intra-urban air pollution predictions. Environ Pollut, 2023; 331(Pt 1):121832; doi: 10.1016/j.envpol.2023.121832

19.

Macías-Hernández

, Tello-Leal

, Barrios

, et al. Effect of environmental conditions on the performance of a low-cost atmospheric particulate matter sensor. Urban Clim, 2023; 52:101753; doi: 10.1016/j.uclim.2023.101753

20.

Mailepessov

, Ong

, Nasir

MZM

, et al. Association between exposure to ambient air pollution, meteorological factors and atopic dermatitis consultations in Singapore—a stratified nationwide time-series analysis. Sci Rep, 2024; 14(1):10320; doi: 10.1038/s41598-024-60712-4

21.

Maleki

, Sorooshian

, Alam

, et al. The impact of meteorological parameters on PM10 and visibility during the Middle Eastern dust storms. J Environ Health Sci Eng, 2022; 20(1):495–507; doi: 10.1007/s40201-022-00795-1

22.

Martins

, de Souza Eicardi

, Nogarotto

, et al. Health and economic benefits of lowering Particulate Matter (PM) levels: Scenarios for a Southern Brazilian Metropolis. Aerosol Sci Eng, 2024; 1–12; doi: 10.1007/s41810-024-00239-3

23.

Munir

, Mayfield

, Coca

. Understanding spatial variability of NO₂ in urban areas using spatial modelling and data fusion approaches. Atmos, 2021; 12(2):179; doi: 10.3390/atmos12020179

24.

Nakada

LYK

, Urban

. COVID-19 pandemic: Impacts on the air quality during the partial lockdown in São Paulo state, Brazil. Sci Total Environ, 2020; 730:139087; doi: 10.1016/j.scitotenv.2020.139087

25.

Nguyen

GTH

, La

, Hoang-Cong

, et al. An exploration of meteorological effects on PM2. 5 air quality in several provinces and cities in Vietnam. J Environ Sci (China), 2024; 145:139–151; doi: 10.1016/j.jes.2023.07.020

26.

Payra

, Sharma

, Mishra

, et al. Performance evaluation of MODIS and VIIRS satellite AOD products over the Indian subcontinent. Front Environ Sci, 2023; 11:1158641; doi: 10.3389/fenvs.2023.1158641

27.

Pedde

, Kloog

, Szpiro

, et al. Estimating long-term PM10-2.5 concentrations in six US cities using satellite-based aerosol optical depth data. Atmos Environ, 2022; 272:118945; doi: 10.1016/j.atmosenv.2022.118945

28.

Rahim

NAAA

, Noor

, Jafri

IAM

, et al. Prediction of PM10 level during high particulate event in Malaysia using modified model. In E3S Web of Conferences 2023(Vol. 437, p. e01002). EDP Sciences; 2023. 10.1051/e3sconf/202343701002

29.

Rahman

. Recommendations on the measurement techniques of atmospheric pollutants from in situ and satellite observations: A review. Arab J Geosci, 2023; 16(5):326; doi: 10.1007/s12517-023-11410-4

30.

Ranjan

, Patra

, Gorai

. A review on estimation of particulate matter from satellite-based aerosol optical depth: Data, methods, and challenges. Asia-Pacific J Atmos Sci, 2021; 57(3):679–699; doi: 10.1007/s13143-020-00215-0

31.

Sasmita

, Kumar

, Priyadharshini

. Assessment of sources and health impacts of PM10 in an urban environment over eastern coastal plain of India. Environ Chall, 2022; 7:100457; doi: 10.1016/j.envc.2022.100457

32.

Shareef

, Hashmi

. Impacts of COVID-19 pandemic on air quality index (AQI) during partial lockdown in Karachi Pakistan. JHER, 2020; 6(3):93–97; doi: 10.11648/j.jher.20200603.17

33.

She

, Zhang

, Bu

, et al. A deep-neural-network-based aerosol optical depth (AOD) retrieval from Landsat-8 Top of Atmosphere Data. Remote Sens, 2022; 14(6):1411; doi: 10.3390/rs14061411

34.

Shin

, Bae

, Park

, et al. Effects of air pollution on moderate and severe asthma exacerbations. J Asthma, 2020; 57(8):875–885; doi: 10.1080/02770903.2019.1611844

35.

Sipra

, Aslam

, Syed

, et al. Investigating the Implications of COVID-19 on PM2. 5 in Pakistan. Aerosol Air Qual Res, 2021; 21(2):200459; doi: 10.4209/aaqr.2020.07.0459

36.

Smotherman

, Sprague

, Datta

, et al. Association of air pollution with postmenopausal breast cancer risk in UK Biobank. Breast Cancer Res, 2023; 25(1):83; doi: 10.1186/s13058-023-01681-w

37.

Sohrab

, Csikos

, Szilassi

. Effects of land use patterns on PM10 concentrations in urban and suburban areas. A European scale analysis. Atmos Pollut Res, 2023; 14(12):101942; doi: 10.1016/j.apr.2023.101942

38.

Tella

, Balogun

. GIS-based air quality modelling: Spatial prediction of PM10 for Selangor State, Malaysia using machine learning algorithms. Environ Sci Pollut Res Int, 2022; 29(57):86109–86125; doi: 10.1007/s11356-021-16150-0

39.

Wei

, Li

, Xue

, et al. The ChinaHighPM10 dataset: Generation, validation, and spatiotemporal variations from 2015 to 2019 across China. Environ Int, 2021; 146:106290; doi: 10.1016/j.envint.2020.106290

40.

World Bank. The Global Health Cost of Ambient PM2. 5 Air Pollution. World Bank; 2020. Available from: http://hdl.handle.net/10986/35721

41.

, Wong

, Liu

, et al. Synergistic data fusion of satellite observations and in-situ measurements for hourly PM2. 5 estimation based on hierarchical geospatial long short-term memory. Atmos Environ, 2022; 286:119257; doi: 10.1016/j.atmosenv.2022.119257

42.

Zhang

, Meng

, Shi

, et al. Overview of particulate air pollution and human health in China: Evidence, challenges, and opportunities. Innovation (Camb), 2022a;3(6):100312; doi: 10.1016/j.xinn.2022.100312

43.

Zhang

, Zhou

, Zhao

, et al. Gap-filling MODIS daily aerosol optical depth products by developing a spatiotemporal fitting algorithm. GISci Remote Sens, 2022b;59(1):762–781; doi: 10.1080/15481603.2022.2060596

44.

Zoran

, Savastru

, et al. Peculiar weather patterns effects on air pollution and COVID-19 spread in Tokyo metropolis. Environ Res, 2023; 228:115907; doi: 10.1016/j.envres.2023.115907