The contribution of different aerosol types to direct radiative forcing over distinct environments of Pakistan inferred from the AERONET data

To quantitatively estimate and analyze the contribution of different aerosol types to radiative forcing, we thoroughly investigated their optical and radiative properties using the Aerosol Robotic Network (AERONET) data (2007–2018) over an urban-industrial (Lahore) and coastal (Karachi) cities located in Pakistan. The contribution of inferred aerosol types following the threshold applied for FMF500 versus SSA440 and EANG440−870 versus AANG440−870 were found the highest for pure dust (PUD, 31.90%) followed by polluted continental (POC, 24.77%) types of aerosols, with moderate contribution was recorded for polluted dust (POD, 20.92%), organic carbon dominating (OCD, 11.85%), black carbon dominating (BCD, 8.77%) and the lowest for the non-absorbing (NOA, 1.79%) aerosol type. Seasonally, the mean (±SD) aerosol optical thickness at 440 nm (AOT440) was found maximum (0.73 ± 0.36) for PUD type in summer and minimum for BCD (0.25 ± 0.04) during spring at Karachi. However, the mean (±SD) AOT440 varied from 0.85 ± 0.25 during summer to 0.57 ± 0.30 in winter at Lahore, with the highest contributions for POC (29.91%) and BCD (22.58%) and the lowest for NOA (5.85%) type of aerosols. Further, the intensive optical properties showed significant temporal and spectral changes and the complexity of inferred aerosol types over the study sites. The results are well substantiated with the air mass analysis obtained from the concentration weighted trajectory (CWT) model for different aerosol types. The Santa Barbara DISORT Atmospheric Radiative Transfer (SBDART) model revealed the strong presence of BCD aerosol type led to a surface (BOA) and top of atmosphere (TOA) forcing of −70.12, −99.78 Wm−2 and −9.60, −19.74 Wm−2, with an annual heating rate of 2.10 and 2.54 Kday−1, respectively, at Karachi and Lahore sites.


Introduction
Rapid urbanization and industrialization have led to an increase in air pollution over several cities in Asia during recent years. The South Asian region is one of the most densely populated and distinct geographical domains with multiple emission sources, where the aerosols have not only affect the regional and local climate, apart from the hydrological scale but, indirectly alter the Earth's radiation budget (IPCC 2017). However, two megacities in Pakistan of the Asian continent, the coastal location of Karachi and landlocked urban-industrial environment of Lahore, play a dominant role in the distribution of air pollutants. Since a long time, the global monitoring of aerosols through the ground-based measurements from several networks (e.g. the AErosol RObotic NETwork (AER-ONET)) provides a unique chance to characterize them by varying types (Choi et al 2016), and assess their impacts on radiative forcing (Tiwari et al 2015, Kumar et al 2020. In the recent decades, numerous authors (e.g. Srivastava et al 2012, Kaskaoutis et al 2012, Alam et al 2012, 2014, Kumar et al 2013, 2020, Bibi et al 2016, Boiyo et al 2019, Khan et al 2019 had benefitted from the AERONET retrieved direct and inversion products for the characterization of aerosol optical (e.g. aerosol optical thickness (AOT); Ångström exponent (ANG), single scattering albedo (SSA)) and microphysical (volume size distribution (VSD), effective radius (R eff )) properties over different places in the world. In this study, a classification technique utilized from the works of Lee et al (2010) was deployed to characterize different absorbing and non-absorbing aerosol types using the aerosol optical parameters (such as the fine-mode fraction of AOT measured at 500 nm (FMF 550 ) and SSA 440 . A brief explanation of aerosol optical properties and their discrimination of aerosol types can affect aerosols on the simulation of radiative forcing. Recent studies (Choi et al 2016, Kumar et al 2018, 2020, Boiyo et al 2019, Rupakheti et al 2019, Shin et al 2019 have shown a statistically significant and robust association between the aerosol optical properties and radiative forcing inferred from various aerosols. A few studies (Choi et al 2016, Che et al 2018 have discussed the effect of different aerosol types on radiative forcing and heating rate, with the long-term ground-based data. A recent study by Khan et al (2019) had identified key aerosol types from six AERONET sites over Southeast Asia, where they found dominant contributions of biomass burning and urban-industrial aerosol-types followed by the mixed nature of aerosols. An earlier report by Bibi et al (2017) used the AERONET data to characterize the aerosol optical properties by implementing multiple clustering techniques for the seasonal classification of aerosol-types during 2007-2013 over the Indo-Gangetic Plain (IGP), which include Lahore and Karachi apart from the other sites. Along these lines, Shaheen et al (2019) adopted a clustering technique by using the long-term (2005-2017) AER-ONET retrieved parameters over Beijing, China.
The current work has been unique of its kind in re-iterating the thresholds to discriminate the aerosol types following the clustering techniques given in the previous works (table S1 of supplementary material (SM) (available online at stacks.iop.org/ERL/15/114062/mmedia)). This study has been conducted following the long-term measured ground-based AERONET data (2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018) to understand the contribution of major absorbing aerosol types, and their associated impact on radiative forcing in a 3D-way over two megacities (Karachi and Lahore) prevailed with distinct environments in Pakistan. The dominant aerosol types and their contribution to the atmospheric column abundance of aerosols are performed in terms of the temporal and spectral variations of AOT, SSA, ANG, and VSD following the clustering framed of FMF versus SSA. We compared our results observed to those from previous studies for the same sites to ensure the validity of the present approach. Further, we identified the potential source contribution of inferred aerosol types at both the locations using the concentrationweighted trajectory (CWT) method following the meteorological data provided by the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model. The ARF and associated heating/cooling rates in the atmosphere and its efficiency inferred for the aerosol types have been quantified on temporal scales using the Santa Barbara DISORT Atmospheric Radiative Transfer (SBDART) model over the two sites.

Site description
The ground-based aerosol climatology is derived over the two AERONET sites located in Pakistan to account for the heterogeneity in aerosol optical, microphysical, and radiative properties inferred from various aerosol types in the most recent years (2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018). The selected regions of interest comprise two important high aerosol laden sites: Karachi (24.87 • N, 67.03 • E, 26 amsl) and Lahore (31.54 • N, 74.32 • E, 712 amsl) where they were identified on the spatial map (with red color solid circles) representing the topography of the country (figure 1). Being proximity to an immense the Thar Desert (Lahore) and the Arabian Sea (Karachi), both the study sites experience the influence of dust and marine aerosols mostly during the spring and summer seasons (Alam et al 2012, 2014, Bibi et al 2016, Khan et al 2019. The pollutants at these two stations are, thus, a complex mixture of natural and anthropogenic (densely populated and urban-industrialized) aerosols that cause variability in the characterization of aerosol optical properties and radiative effects (Bibi et al 2017).
The monthly mean variations of major meteorological parameters (such as air temperature (AT), wind speed (WS), relative humidity (RH), and rainfall (RF)) obtained from the Pakistan Metrological Department (PMD) measured during 2007-2018 over the two sites is shown in figures 1(a)-(b). The results revealed that both the places are generally influenced by enhanced precipitation during the humid and hot summers, characterized by relatively high temperatures and RH. The WS remains moderately varied in the range 0.71-3.9 ms −1 and 0.3-1.22 ms −1 over Karachi (figure 1(a)) and Lahore ( figure 1(b)), respectively. However, the minimum (maximum) ATs occurred in January (June) with 19.65 • C (32.04 • C) and 12.95 • C (33.21 • C) for Karachi and Lahore, respectively. Besides, due to its proximity to the Arabian Sea, the RH at Karachi remains consistently high (∼81%, figure 1(a)), with a significant influence on the number of aerosol processes (Alam et al 2012). For the annual assessment, we considered four different seasons mentioned as follows: spring (from March to May), summer (June to August), autumn (September to November), and winter (December to February).

Instrument and data
The primary data set used constituted the remotelysensed ground-based measurements conducted by the AERONET's CE-318 Sun photometer (Cimel Electronique, France), and the retrieved data for all the stations given by the AERONET is wide open and downloaded at https://aeronet.gsfc.nasa.gov/. In the present work, we used the Version 3.0 and Level 1.5 (cloud screened) spectral AOT data in the wavelength range of 440-1020 nm, and ANG estimated for the wavelengths between 440 and 870 nm. Moreover, the other inversion products include absorption AOT and ANG (AAOT, AANG), asymmetry parameter (ASP), extinction AOT and ANG (EAOT, EANG) used in work are obtained in the spectral range of 440-1020 nm, and VSD. The uncertainty in AOT under cloud-free conditions is ±0.01 and ±0.02 for higher and lower wavelengths, respectively (Eck et al 1999). Whereas, the uncertainty involved in the retrieved inversion products of SSA 440 and ASP 440 (with AOT 440 > 0.40) were found ±0.03 and ±0.02, respectively (Dubovik et al 2006). While analyzing, we noticed gaps in the data are due to thick cloud cover, more precipitation days, calibration procedure, and system malfunction. It is mentioned that the recent collaboration (AERONET) provides AOT data, in three different quality levels: Level 1.0 (unscreened), Level 1.5 (cloud-screened and quality controlled), and Level 2.0 (quality-assured) (Holben et al 1998). The reason behind using level 1.5 instead of level 2.0 data is that level 2.0 datasets are not provided by the AERONET since late 2017, at the observation locations. Also, the recently released version 3.0 data products are capable in advanced cloud screening, and automated data quality assurance with higher air masses up to 7 (in level 1.5) in contrast with version 2.0 (which is up to 5) (Kokkalis et al 2018, Khan et al 2019, Kumar et al 2020. Hence, owing to the excellent availability of data count provided with the real-time and continuous long-term duration of data, level 1.5 was utilized for analyzing the aerosol characteristics in its different types in this study (figure S1 of SM).

Cluster technique for aerosol typing
A study in the classification of absorbing and nonabsorbing aerosol types over different locations was carried out based on the approach proposed by Lee et al (2010) via., the SSA 440 versus FMF 500 relationship previously conducted by several authors (table S1 of SM) (e.g. Srivastava et al 2012, Tiwari et al 2015, Choi et al 2016, Chen et al 2016, Moreno et al 2019. Lee et al (2010) characterized aerosol types based on dominant size mode (FMF 500 ) and radiation absorptivity (SSA 440 ), over North Africa and the Arabian Peninsula by adopting a safety margin of 0.2, by FMF to be higher than 0.6 (less than 0.4) as are defined fine-mode (coarsemode) aerosols. Aerosols in the safety margin of the thresholds (that is, between 0.4 and 0.6) are classified as a 'mixture' of coarse-and fine-mode aerosols (figure S2 and table S1 of SM).
The main types of aerosols observed from the clustering of SSA 440 versus FMF 500 were divided into six categories, namely, Pure Dust (PUD; only dust dominating aerosols), Polluted Dust (POD; dust dominating relative to anthropogenic aerosols), polluted continental (POC; anthropogenic dominating relative to dust aerosols), highly absorbing (BCD; only anthropogenic dominating due to black carbon aerosols), low absorbing (OCD; only anthropogenic dominating due to organic carbon aerosols), and non-absorbing (NOA; aerosols with no absorption, i.e. scattering nature). Values of FMF < 0.4 and SSA > 0.9 corresponds to PUD type, and FMF < 0.4 with SSA ⩽ 0.9 represents POD type; whereas, the thresholds with 0.4 ⩽ FMF ⩽ 0.6 and 0.4 ⩽ SSA ⩽ 0.9 is associated with POC type of aerosols. However, for all the values of FMF > 0.6 with SSA < 0.9, 0.9 ⩽ SSA ⩽ 0.95, and SSA > 0.95 indicates BCD, OCD, and NOA types of aerosols, respectively. Besides, the other aerosol optical parameters such as AANG 440−870 and EANG 440−870 are also clustered together to investigate the dominant aerosols representing their annual and seasonal variations over the sites; however, both the methods are found to be well associated with each other presenting the similar dominant type of aerosols. To supplement this method, the threshold applied on AANG 440−870 (EANG 440−870 ) for BB (biomass burning type of aerosols which are comparable with BCD type), Dust (corresponds to PUD + POD), and UI (urban-industrial aerosol type similar to NOA) follow 1.1-2.3 (0.8-1.7), 1.0-3.0 (0-0.4) and 0.6-1.2 (0.8-1.6) respectively, by utilizing the approach adopted from Giles et al (2012) and Rupakheti et al (2019).

Source receptor model
The CWT method is the most widely used technique to identify the distant emission sources of air masses for the different aerosol types (Hsu et al 2003). Here, we used the same model provided the AOT 440 input data related to the following four aerosol types (Dust (PUD + POD), POC, BCD, NOA) about the height of 500 m above ground level and for the time duration of −72 h (i.e. three days backward) to capture the potential source areas affecting the receptor sites (Karachi and Lahore). This approach is based on the TrajStat plug-in used in the GIS-receptor based model software and is found elsewhere (Wang et al 2009; www.meteothinker.com/Documents/Wang TrajStat Manuscript.pdf). Additionally, each CWT value reflects a conditional probability describing the potential contribution of a grid cell of the high pollutant loadings at the receptor site; however, this approach is capable of distinguishing primary sources from moderate ones (Hsu et al 2003). In the CWT method, each grid cell is assigned a weighted concentration by averaging the sample concentrations, which have associated trajectories that crossed the grid cell as follows: where C ij is the mean weight concentration of the back trajectory 'h' in the ij cell; C h represents inferred aerosol type concentration (here AOT) in the trajectory 'h' through ij cell; τ ijh represents the time that trajectory 'h' resides in the ij cell. W (n ij ) used in the CWT is to reduce the uncertainty in cells.

Radiative transfer model
The daily net fluxes obtained for the inferred aerosol types at the BOA and TOA of the atmospheres were calculated for the clear-sky shortwave (0.3-4.0 µm) direct ARF using the sun/sky radiometer measured spectral values of AOT, ANG, SSA, and ASP as inputs into the SBDART model (Ricchiazzi et al 1998).
Besides these optical parameters, the surface albedo, which plays a vital role in the forcing calculation was obtained over the study sites from the Aura Ozone Monitoring Instrument (OMI-3) reflectivity data set as well as retrieved from the AERONET inversion products. Besides, the diurnally averaged radiative forcing at the TOA and BOA was obtained from the AERONET at every one-hour interval for a 24 h period, with an accuracy of ±2 Wm −2 (Kumar et al 2017). Moreover, the model is capable of estimating the radiative flux within 2% of direct and diffuse irradiance measurements , and has been  Figure 2 presents the scatter plots between SSA 440 and FMF 500 as a function of AOT 440 observed at Karachi and Lahore to infer different aerosol types. A wide range of SSA (0.86-0.92) was found at both the sites for every kind of aerosol. However, the deviation in FMF at Lahore was evident only for BCD, OCD, and NOA types of aerosols, attributed to changes in emission sources. It is evident from the earlier studies that the pollutants from the burning of agricultural residue, especially in the harvesting season (autumn), contribute significantly to organic (OCD) aerosols. Besides, secondary aerosols from high traffic flow (fossil fuel burning) and a large amount of coal consumption for household heating and domestic cooking at these locations is the second reason for the observed variations in these types of aerosols. Figures 2(c)-(d) shows the monthly fluctuations in the occurrences of dominant aerosol types. Statistically, the occurrence represents the number of times dominant aerosol types occur during the valid operating hours of the instrument in the almucantar geometry, being found the highest for POD and BCD aerosol types during April and December at Karachi and Lahore, respectively. Further, the statistics illustrate that there was the more frequent occurrence found for PUD types over Karachi between April and June, which is mainly due to a large amount of windblown desert and mineral dust particles, considering that the site (Karachi) is located on the coast of Arabian Sea. The OCD (BCD) type is also found high in November and December months over Karachi (Lahore). However, the occurrence range is low for NOA comparatively, throughout the period, especially at Karachi, attributed to the difference in population, geography, elevation (figure 1), and use of anthropogenic sources at the study sites. Another reason behind may be the availability of fewer data points attributed to some specific flaws, including cloud covers, technical errors, and instrument sent for calibration, etc. The occurrence of a more substantial contribution by POC and BCD types at Lahore and PUD over Karachi is related to anthropogenic particles (Tiwari et al 2015), and sea salt aerosols from the marine environment (Bibi et al 2016). The carbonaceous (absorbing) aerosol types such as BCD and OCD revealed a sharp winter peak over both the sites. In particular, the contribution of BCD type increased from October to December, while the OCD type increased during October-February. The PUD (31.90%) followed by POC (24.77%) types of aerosols were recorded among the highest contribution relative to the rest of the aerosol types, being dominant during the spring and autumn seasons, respectively (figure 2(e)). Whereas, the NOA (1.79%) followed by the BCD (8.77%) aerosol types were found minimum in all seasons, with the lowest in spring at Karachi (figure 2(e)). However, the OCD (11.85%) type of aerosols contributed moderately to the aerosol load at both the sites.

Classification of aerosol types
The season-wise distribution of scatter plots between AANG 440−870 and EANG 440−870 to classify different aerosol types is presented in figure 3. Three prominent classes of aerosols were reported over the sites such as dust, BB, and UI. Absorption and extinction thresholds revealed coarse-mode dominating particles at both the sites especially, during the spring and summer seasons, while fine-mode anthropogenic aerosols were found dominating during winter and autumn. However, both the clustering techniques  (FMF versus SSA and AANG versus EANG) showed similar results and revealed prominent consistency between them. The results obtained in this study are identical to that of Bibi et al (2016) and Alam et al (2016) utilizing clustering analysis between AANG and EANG to classify various types of aerosol over the two sites.

Variations in aerosol optical properties
The annual contribution representing different seasons of varied aerosol types for AOT 440 observed at Karachi and Lahore is shown in figures 4(a)-(b). The maximum occurrences were found for NOA type of aerosols in all seasons, with a maximum AOT NOA ranging from 0.87 ± 0.51 (Lahore) to 0.72 ± 0.13 (Karachi) in summer attributed to hygroscopic growth of aerosols. Whereas, both the sites have observed the highest AOT for OCD during spring and autumn seasons, with the mean (±SD), ranged between 1.27 ± 0.63 (Lahore) and 0.63 ± 0.26 (Karachi) (table 1), strongly related to the seasonal pollution (e.g. dust, haze, and smog) events observed frequently over the IGP region from past few years (Alam et al 2018). Bibi et al (2016) and Tiwari et al (2015) have shown a similar distribution of absorbing aerosols for the same sites, which peaks during the autumn attributed to long-range transport and locally generated pollutants. Besides, the dominant components at the coastal site are sea salt and inorganic aerosol particles, which are an active agent for highly scattering particles. The seasonal mean  from the Middle East, Arabian Sea, and the Thar and Cholistan Deserts of Pakistan. Moreover, the comparatively high AOT (0.63 ± 0.33) and ANG (1.21 ± 0.13) at Lahore is due to the occurrence of more BCD type of aerosols suggests the dominant presence of fine-mode aerosols mainly from biomass burning and industrial-vehicular emissions transported from the most prominent economic zone of IGP. Figures 4(g)-(h) shows the seasonal variations of SSA 440 with more scattering (>0.93) found during the summer for NOA, and POC aerosol types, are likely due to the influence of dust particles transported from the desert regions. However, the lower (strong absorption) (<0.81) in winter for BCD type could be due to the effect of biomass burning activities at both the sites. Another reason is, both places are under the influence of heavy air pollution due to haze and smog caused by the high aerosol loading due to more anthropogenic activities and dense population. Moreover, the seasonality in aerosol properties over the regions is also due to local emissions, apart from the change in the meteorological phenomenon (Khan et al 2019).
The 3D scatter plot (figure 5) shows that sliding at the shorter scale range, of EANG and SSA, resulted in Dust (PUD + POD) type aerosols at both the stations with slightly high data points for Karachi site mainly associated with natural desert surfaces and locally produced dust and marine particles with high winds. However, going towards a higher scale range of SSA and EANG combined with the maximum EAOT results BCD (POC) in predominance for Karachi (Lahore) followed by OCD and NOA aerosol type, being consistent with the prevalence of regional anthropogenic activities and biomass burning mainly in the spring season.
Furthermore, the CWT model analysis revealed that the air masses traveled predominantly from the marine environments (the Arabian Sea) located in the south, and arid regions in the east (from the Thar Desert in India) and west (Afghanistan) that regulate the columnar aerosol loading at both the sites (figure 6). Owing to the annual dust episodes, the CWT revealed significant influence (CWT Dust > 0.9) of potential sources from the dust areas over the coastal site (Karachi) (figure 6), followed by CWT POC (~0.8) and CWT BCD (<0.7) over Lahore, moderately affected due to CWT POC (<0.6), and low CWT Dust (<0.4) over Karachi and Lahore sites, respectively, indicates significant diverse contributions from the regional aerosol sources to AOT 440 primarily, governed by the distant and localized sources. Figure 7 gives the seasonal mean changes observed in aerosol VSD patterns for inferred aerosol types, whereas the annual mean changes shown in figures S5(a)-(b) and S6(a)-(h) of SM are noted at 22 size bins with different radii between 0.05 and 15.0 µm during the entire study period. The gray shaded area indicates the standard deviation of the mean for a given aerosol-type. The VSDs exhibited almost bimodal structure in all seasons, with the secondary peak (coarse-mode) relatively higher than the primary (fine-mode), indicating the varied contribution of particles. There are several possible explanations for the observed differences between the two modes; however, the variation of VSD significantly influences the radiative properties of aerosols. The VSD curves for the PUD and POD aerosol-types were more easily distinguishable than the rest of all types. The volume particle concentration of coarsemode was found higher in all seasons for most of the aerosol-types, except NOA. The fine-mode peak was prominent throughout the study period at both the study sites attributed to the inferred aerosol-type, environment, and meteorological conditions (Alam et al 2012). As expected, the PUD type was found relatively higher over Lahore during spring (0.80 ± 0.03) and summer (0.35 ± 0.02) seasons than at Karachi, which is (0.28 ± 0.01) and (0.24 ± 0.04), respectively. This is evident and consistent with the previous studies (e.g. Bibi et al 2016) that the geographical location of Lahore city with its higher proximity to the dust source regions (e.g. Thal, Thar, and Cholistan Deserts) compared to Karachi, apart from the long-distant transport of dust particles from the deserts of Arabian Peninsula. The annual mean (±SD) particle volume concentration (and R eff ) in the fine-mode was noticed high for PUD Karachi which is 0.37 ± 0.01 µm 3 µm −2 (0.12 ± 0.01 µm) followed by NOA Lahore and OCD Lahore with 0.15 ± 0.07 µm 3 µm −2 (0.20 ± 0.04 µm) and 0.12 ± 0.07 µm 3 µm −2 (0.18 ± 0.03 µm), respectively. The low was observed for POD Karachi with 0.02 ± 0.01 µm 3 µm −2 (0.11 ± 0.01 µm) (table 2). Moreover, a noticeable increase in coarse-mode volume concentration was observed for the OCD types during the winter and autumn season at both the sites, attributed to the seasonal inputs of plant residue, soil and desert dust, and biomass burning components.

Implications to radiative forcing
The monthly (figures S7 and S8 of SM) and seasonal (figure 8) mean changes of radiative forcing was estimated for different aerosol types at the TOA and BOA using the SBDART model for the two sites during the entire study period. The radiative forcing at the BOA revealed a strong influence during the study period. However, the variability in TOA forcing is weaker, presenting maximum values for PUD in June (−93 W m −2 ) and July (−86 W m −2 ) and minimum in October (−31 W m −2 ) for OCD at Karachi (figure S7 of SM). It is observed that amongst all the types, the BCD aerosol-type has registered the highest positive ATM forcing in January (65 W m −2 ) and February (63 W m −2 ), with an annual mean of 56 ± 6.8 Wm −2 and corresponding heating rate of 2.31 Kday −1 (figure 8). Additionally, the forcing obtained at both the sites was highly dependent on the aerosol column load, which increases with the AOT and its type, especially for black carbon and organic carbon types during the winter and spring seasons. Likewise, the decreasing tendency in RF seen manifest, affiliated with the reduced AOT values for POD and PUD aerosol types, during winter and summer, respectively (figures 8 and S7, S8 of SM), which ultimately affect the range of HR in the upper atmosphere over the study regions (García et al 2008).
It is observed that the atmospheric HR of finemode aerosol increases with the AOT at both the sites, with the mean values larger than~2.1 Kday −1 . The highest HR associated with the presence of BCD aerosol-type was found in Karachi varied between 1.95 Kday −1 and 2.31 Kday −1 during winter and autumn seasons, respectively, with the corresponding AOT 440, found to be >0.5 (figures 8(a)-(d)). The seasonality in TOA forcing is due to BCD type of aerosols significantly associated with various assumptions such as low precipitation rate in winter, dry and cold weather conditions resulted in increased household burning, biomass/agriculture residue burning (Bibi et al 2017) and so on. However, the more amount of rainfall during June and July affects the growth of vegetation, indirectly changes the surface cover, and hence, the surface albedo in summer , Boiyo et al 2019, is another crucial reason for the fluctuation in ARF. As such, it can be inferred that aerosol and other atmospheric constituents, in addition to clouds, play a significant role in the attenuation of solar radiation over the sites.
In contrast, Lahore showed the highest BOA (−158.21 W m −2 ) and ATM forcing (+99.7 W m −2 ) for PUD and BCD aerosol types, respectively, during the spring, with HR of 2.31 Kday −1 (figures 8(f)-(i)). However, the lowest radiative forcing was observed for NOA with the BOA and ATM forcing values of −80.03 and +60.7 W m −2 , respectively, along with the HR of 1.4 Kday −1 , which is slightly more than the value (~1.02 Kday −1 ) reported by Kumar et al (2018) over Kanpur in IGP, andBibi et al (2017) and Alam et al (2014) for the same study sites of Pakistan. Numerous studies have emphasized the significant role of radiative forcing by absorbing aerosols over  . Nevertheless, it is seen that the BOA forcing for PUD/POD aerosol-type becomes a minimum than that of the TOA forcing due to less scattering by different aerosol types at both the sites. Apart from this, the radiative forcing is highly dependent on SSA (Srivastava et al 2012, Boiyo et al 2019, Kumar et al 2020. For lower SSA, the ARF found highly negative. Further, the ARFEs (the rate of forcing per unit AOT) (figures 8 and S9, S10 of SM) showed a strong influence during winter/autumn with maximum values for BCD BOA in January (−230 W m −2 τ −1 ) and February (−239 W m −2 τ −1 ), and minimum for NOA BOA during autumn in October (−105 W m −2 τ −1 ) and September (−100 W m −2 τ −1 ) at Karachi and Lahore, respectively. A distinct seasonal variability can be found for ARFEs at TOA. It is observed that amongst all the types, the PUD aerosol-type has registered the highest TOA efficiency in December (−80 W m −2 τ −1 ) at Karachi followed by Lahore (−78 W m −2 τ −1 ) and lowest for BCD type during August (−20 W m −2 τ −1 ) and January (−30 W m −2 τ −1 ) for Lahore and Karachi, respectively suggested, more absorbing aerosols produce a lower forcing efficiency at the TOA.

Summary of conclusions
In the present work, we investigated and compared the characteristics of aerosol types over Karachi and Lahore via the cluster analysis of optical, microphysical, and radiative properties obtained from 12 years of ground-based AERONET data set. The primary purpose of the present work is to investigate and understand the heterogeneity in different aerosol species, its origin, and the associated direct radiative impacts on regional climate and their heating effect. The implementation of two different clustering techniques together with a regional and local meteorological database is the scientific novelty of this work. It allows us to better understand the aerosol optical and microphysical properties and improving our understanding of the uncertainties involved in the radiative forcing via modifications in atmospheric warming or cooling by inferred aerosol types. Six different aerosol types dominated with coarse-mode (fine-) sizes of PUD, POD, and POC (BCD, OCD, and NOA) were identified based on their size distributions and absorption capabilities following the two different approaches (SSA 440 versus FMF 500 and AANG 440−870 versus EANG 440−870 ) over the sites. However, both methods are found to be well asso-ciated with each other and revealed a similar type of aerosols.
The distinct seasonal discrepancies between both the sites highlight the dominance of PUD (POC) type of aerosols, with the highest contribution of 71.63% (28.24%) during summer in Karachi (Lahore) site, respectively. The mean SSA and ASY at Karachi and Lahore both were highest in the NOA and PUD type, respectively, considered to be more irregular size than other classes. Seasonally, high AOT 440 (1.43 ± 0.51) with corresponding high SSA 440 (0.97 ± 0.01) was noticed during autumn for NOA type at Lahore. At the same time, the lowest was found for BCD aerosol type with mean AOT 440 (0.25 ± 0.04) and corresponding low SSA 440 (0.83 ± 0.03) at Karachi site, suggesting the dominance of significant anthropogenic emissions (absorbing fine mode) aerosols, while the former is associated with scattering-type (dust particles).
The CWT analysis revealed that potential sources of Dust at Karachi and POC over Lahore were found dominating among inferred aerosol types, with a considerable influence of CWT Dust > 0.9 followed by CWT POC (~0.8) suggesting the air masses are local as well as transported from distinct sources (from natural, and anthropogenic origin) to the study areas. The highest atmospheric forcing of +60.01 W m −2 (+80.07 W m −2 ) was observed for Karachi (Lahore), with the corresponding heating rate of 2.48 Kday −1 (2.05 Kday −1 ) for BCD type. While, the lowest value of +30.51 W m −2 (+43.07 W m −2 ), with the corresponding heating rate of 1.48 Kday −1 (1.53 Kday −1 ), was observed for NOA type of aerosol during the study period. However, the ARFEs at the TOP was found maximum (minimum) as −80 W m −2 τ −1 (−20 W m −2 τ −1 ) for PUD (BCD) type during winter (summer), respectively, at both the sites suggesting, absorbing aerosols produce a lower ARFEs at the top of the atmosphere.

Supporting material
The supplement material about the text, figures, and tables related to this article is available online at https://iopscience.iop.org/journal/1748-9326.