The Climate Impact of Aerosols on Lightning : Is it Detectable from 2 Long-term Aerosol and Meteorological Data ?

7 Qianqian Wang1, Zhanqing Li*1,2, Jianping Guo*3, Chuanfeng Zhao1, Maureen Cribb2 8 9 10 11 12 13 1State Laboratory of Earth Surface Process and Resource Ecology, College of Global Change and 14 Earth System Science, Beijing Normal University, Beijing, China. 15 2Earth System Science Interdisciplinary Center, University of Maryland, College Park, MD, 16 USA. 17 3State Key Laboratory of Severe Weather, Chinese Academy of Meteorological Sciences, 18 Beijing, China 19 20 21 22 23 24 25 26 27

with a turning point at around AOD=0.3, below which lightning flashes increase monotonously with increasing AOD in both ROIs.As AOD approaches the optimal value, lightning activity seems to be saturated under smoky condition, likely due to the tradeoff between the aerosol invigoration effect and the radiative effect that tends to enhance and suppress lightning, respectively.In contrast, lightning activity in ROI_1 is suppressed by dust aerosol presumably due to the more dominant radiative heating effect of dust aerosol under dry environment.
Since the pioneering work in 1990's by Westcott (1995) who attempted to link the summertime cloud-to-ground lightning activity to anthropogenic activities, the roles of aerosols in lightning have been increasingly recognized, as comprehensively reviewed on the topic associated with aerosol-cloud-precipitation interactions (e.g.Tao et al., 2012, Fan et al., 2016;Li et al., 2016;2017a).The aerosol effect can be generally divided into radiative and microphysical effects (Boucher et al., 2013;Li et al., 2017b).The radiative effect suggests aerosols can heat the atmospheric layer and cool the surface by absorbing and scattering solar radiation, thereby reducing the latent heat flux and stabilizing the atmosphere (Kaufman et al., 2002;Koren et al., 2004;Koren et al., 2008;Li et al., 2017a).As a consequence, convection and electrical activities are likely inhibited (Koren et al., 2004).By acting as cloud condensation nuclei (CCN) with the constrained water, increasing aerosols loading tends to reduce the mean size of cloud droplets, suppress coalescence and delay the onset of warm-rain processes (Rosenfeld and Lensky, 1998), which permits more liquid water to ascend higher into mixed phased region where it fuels lightning.
A conspicuous enhancement of lightning activity was found to be tightly connected to volcanic ashes over the western Pacific Ocean (Yuan et al., 2011) and aerosol emissions from ships over the equatorial Indian Ocean (Thornton et al., 2017).In terms of the response of clouds to aerosols, an optimal aerosol concentration was found to exist, based on observation analysis (Koren et al., 2008;Wang et al., 2015) and a simple parcel model calculation (Rosenfeld et al., 2008).Biomass burning activities, anthropogenic emission and desert dust are the three major atmospheric aerosol sources (Rosenfeld et al., 2001;Li et al. 2011;Fan et al. 2018), which are recognized to have different climate effects.The increased rainfall in the south China and drought in the north China are thought to be related with increased black carbon aerosols (Menon et al., 2002) while the effect of dust on cloud properties tends to decrease precipitation from a feedback loop (Rosenfeld et al., 2001).
However, most studies on aerosol-convection interactions account for aerosol burden (i.e.AOD, the number concentration of aerosol, PM2.5, or CCN) rather than aerosol species.It was not until recently that the effect of aerosol species in modulating lightning activities was investigated (e.g.Stolz et al., 2015;2017), which prompts us to perform more detailed analyses in this study.
Depending on aerosol properties and atmospheric conditions, aerosols may enhance (Fan et Khain et al., 2005;2008) or suppress convection (Khain et al., 2004;Rosenfeld, 2000;Rosenfeld et al., 2001;Zhao et al., 2006).In general, aerosols tend to suppress convection for isolated clouds formed in relatively dry condition but to invigorate convection for convective systems inside a moist environment.Under conditions of strong vertical wind shear, aerosols tend to reduce the strength of single deep convection clouds due to higher detrainment and larger evaporation of cloud hydrometeors (Richardson et al., 2007;Fan et al., 2009), thereby altering the lightning activities (Altaratz et al., 2010;Farias et al., 2014).Apart from invigoration effect induced by aerosol, lightning activities are enhanced by increases in NCAPE and SHEAR, but inhibited by increasing RH and warm cloud depth (Stolz et al., 2015).

2007;
Most previous studies were based on short-term data, while we try to employ long-term (11 years) lightning and meteorological data to investigate and quantify the relative roles of aerosol and meteorology on lightning.Section 2 describes the dataset and method used in this study, Section 3 shows regions of interest, and Section 4 examines (1) long-term climatological features of lightning and aerosol, (2) differences of lightning, AOD, meteorological variables under relatively clean and polluted conditions, (3) response of lightning to meteorology, (4) contrast in the response of lighting to dust and smoke aerosol, and (5) relative roles of meteorology and AOD in lightning activity, finally followed by a summary of the key findings in section 5. We employ lightning data from the Lightning Imaging Sensor (LIS) onboard Tropical Rainfall Measuring Mission (TRMM), which is designed to acquire and investigate the distribution and variability of total lightning (i.e.intra-cloud and cloud-to-ground) on a global basis, span all longitudes between 38°N-38°S, during day and night (Boccippio, 2002;Christian et al., 2003).

Aerosol data
Aerosol loading is characterized by aerosol optical depth (AOD), which is obtained from observations collected by the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard Aqua satellite that passes at ~ 13:30 local time (LT).Here, the level 3 atmosphere monthly global product (MYD08_M3) on 1°×1° grid is used.This AOD is retrieved at 0.55 µm, based on dark target-deep blue combined algorithm, which is particular suitable over desert regions (Hubanks et al., 2015;Levy et al., 2013).The Modern Era Retrospective-analysis for Research and Application (MERRAero) provides dust, black carbon, organic carbon, and total extinction AOD, and the total Angstrom Exponent (AE) at a spatial resolution of 0.625°×0.5°(da Silva, et al., 2015).These data characterize aerosol species and particle size.

Meteorological data
Meteorological data utilized are from the Medium-Range Weather Forecasting (ECMWF) ERA-Interim reanalysis product (Dee et al., 2011), including the surface upward sensible heat flux (SHF), surface upward latent heat flux (LHF), sea level pressure (SLP), 2m air temperature (T), convective available potential energy (CAPE), relative humidity (RH), wind field at 925 hPa and 500 hPa (U, V), divergence at 200 hPa (Div) with a spatial resolution of 1°×1°.With reference to the findings from previous studies, we choose the following factors to characterize meteorology: 1) Convective available potential energy (CAPE).CAPE is the most commonly used thermodynamic parameter, which describes the potential buoyancy available to idealized rising air parcels and thus denotes the instability of atmosphere (Riemann-Campe et al., 2009;Williams, 1992).The bigger it is, the atmosphere is more unstable, and more favorable to the occurrence and development of convection.Lightning activity increases with the CAPE (Williams, 1992).The conversion efficiency of CAPE to updraft kinetic energy (Williams et al., 2005) depends on the strength and width of updraft.Unfortunately, we don't have any reliable updraft measurements to tackle with its role in this study.
2) Sea level pressure (SLP).It's widely known that pressure is a key meteorological factor affecting the weather, for it defines basic weather regimes.Low pressure systems are usually associated with strong winds, warm air, and atmospheric lifting, thus normally producing clouds, precipitation, and severe weather events such as storms and cyclones.The implication of summertime SLP anomalies in the tropical Atlantic region shows the inverse relationship between SLP and tropical cyclones (Knaff, 1997).
3) Potential temperature (θ).Many researchers have studied the role of air temperature in influencing lightning activities (Markson, 2003;Williams, 1992;1994;1999;Williams et al., 2005).However, the direct comparison of temperature for different regions is problematic because air temperature is systematic decline with altitude.We choose potential temperature instead, which corrects for the altitude dependence and provides a more meaningful comparison.Here, potential temperature θ is calculated from 2 m air temperature T (K): T .
(1) 4) Relative humidity (RH).As one of the important thermodynamic factors, RH has been identified to affect the relationship between aerosols and convection (Fan et al., 2007;Fan et al., 2009;Khain and Lynn, 2009).It was found that for isolated clouds formed in a relative dry condition convection is suppressed by aerosols, whereas for convective systems inside a moist environment convection is invigorated (Khain and Lynn, 2009;Khain et al., 2008).The mean RH values at 700 and 500 hPa levels (mid-level RH) are employed in this study.(Coniglio et al., 2006;Rotunno et al., 1988;Weisman and Rotunno, 2004), but also qualitatively determines whether aerosols suppress or enhance convective strength (Fan et al., 2009).It is calculated from daily wind field at 925hPa and 500hPa as follows:

SHEAR
(2) 6) Divergence (Div).Air divergence is especially useful, because it can be linked to diabatic heating processes, of which, the non-uniformity gives rise to atmospheric motion (Homeyer et al., 2014;Mapes and Houze, 1995).Fully developed clouds are usually accompanied by upperlevel divergence especially in raining regions (Mapes and Houze, 1993).A pronounced divergence maximum exists between 300 and 150 hPa due to deep convective outflow (Mitovski et al., 2010).
Besides, bowen ratio (BR) is calculated from SHF and LHF to describe surface property:

Data collocation
A roughly three-month smooth average is chosen to allow LIS to progress through the diurnal cycle at a given location twice (Cecil et al., 2014), and show the normal annual variation of lightning activity due to the seasonal meridional migration of the intertropical convergence zone (Thornton et al., 2017;Waliser and Gautier, 1993).In order to match lightning data, all AOD and

Statistical analysis method
Correlation coefficients are used in statistics to measure how strong a relationship between lightning flashes and one predictor (SLP, θ, RH, CAPE, SHEAR, Div, AOD).Pearson correlation is commonly used in measuring linear correlation.And partial correlation is performed to control other predictors and make it possible to study the effect of each predictor separately.The correlation is considered to be significant when it passes the significant test at 0.05 level.
In order to explore the relative roles of meteorological variables and AOD in lightning activities, we use multiple-linear regression method follow previous studies (e.g.Igel and van den Heever, 2015;Stolz et al., 2017) and establish standardized regression equation for AOD greater and less than 0.3 respectively.This subsection regression is performed to reduce nonlinear effect of AOD.Note that, all data used here are processed by averaging 10 samples sorted by AOD from small to large to mitigate data uncertainties.The standardized regression equation with seven predictor variables x , x ,…, x (SLP, θ, RH, CAPE, SHEAR, Div, AOD) and a response y (lightnng flashes), can be written as: Here, y and x are standardized variables, which are derived from raw variables Y and X by subtracting the sample mean (Y , X ) and divided by the sample standard deviation (δ , δ ): The sample mean is calculated of N valid samples: The sample standard deviation is: Standardized regression coefficients ignore the independent variables' scale of units, which make slop estimate comparable and show the relative weights to the changes of lightning flashes.

Region of Interest (ROI)
The northern and southern Africa has high aerosol loadings of dust and smoke aerosols respectively, as seen in Fig. 1.Northern Africa is the world's largest source of mineral dust (Lemaître et al., 2010), with the most widespread, persistent dust aerosol plumes and the most dense particulate contribution found on Earth (Prospero et al., 2002).It has been estimated about 2-4 billion tons of blown dust globally comes from the Saharan desert (Goudie and Middleton, 2001).Dust of relevance to atmospheric processes are the minerals that can be readily suspended by wind (Shao, 2008) with particle size up to 70 m.Besides, Africa is also the single largest source of smoke emissions due to widespread biomass burning, accounting roughly for 30 to 50% of the total amount of vegetation burned globally each year (Andreae, 1991;Roberts et al., 2009;van der Werf et al., 2003;2006).In Central and Southern Africa, biomass burning due to wildfires and human-set fires has strong diurnal and seasonal variability (Ichoku et al., 2016;Roberts et al., 2009).
The upper left panel of Figure 1 shows the global distribution of mean aerosol optical depth  and black, respectively.Note that AOD used hereinafter is derived from MODIS, and the lowest (highest) third of AOD is labeled as clean (polluted) condition.The diurnal curves in Fig. 2b show the same afternoon peak in lightning activities, which is in accordance with the strong convection in terms of both the number and intensity of convective clouds in afternoon over land (Nesbitt and Zipser, 2003).Over both ROI_1 and ROI_2, the lightning peak times under polluted (dusty/smoky) conditions are delayed by about 1h compared with that under clean condition.This result is consistent with the obsevation-based finding in southern China (Guo et al., 2016) and model simulation results by Lee et al. (2016).Numerious studies have noted that aerosols modulate convection and lightning activities through both radiative and microphysical processes, as reviewed extensively by Li et al. (2016) in Asia and around the world (Li et al., 2017b).When we look into the monthly variation, dust loading changes little through the whole year (Fig. 2c), while smoke shows pronounced seasonal variation of a huge contrast between dry and wet seasons (Fig. 2d).Lightning in both regions is most active in summer and rarely occurs in winter, which implies that lightning activity is mainly controlled by thermodynamic conditions.Besides, also shown in Fig. 2 is the apparent enhancement of lightning activity under smoky conditions superimposed on both the diurnal and seasonal cycles.Under dusty conditions, however, the impact is much weaker.

Long-term climatological features of lightning and aerosol
Apart from different aerosol effects, different climate conditions between dust and smoke aerosol dominant region may also contribute.A key factor is relative humidity which is much lower over the desert region of ROI_1 (BR>10) than over the moist region of ROI_2 covered with rain forest (BR<0.4).It's widely known that high values of relative humidity are more favorable for the invigoration effect and vice versa (Fan et al., 2008;2009;Khain, 2009;Khain et al., 2008), which is likely a major cause for the difference.Besides, ROI_1 locates in the vicinity of the African Easterly Jet (AEJ) (Burpee, 1972) while ROI_2 locates in the ITCZ (Waliser and Gautier, 1993) that lead to differences in wind shear and instability.

Response of lightning to meteorology
As thermodynamic condition is considered to play the main role in lightning diurnal and seasonal variation.Firstly, we have a look at the response of lightning to thermodynamic condition which is characterized by six meteorological variables (SLP, SHEAR, θ, CAPE, RH and Div).
Violin plot is an effective and attractive way to visualize the distribution of the data and the shape of distribution that allows quick and insightful comparison of multiple distributions across several levels of categorical variables, and is used in the analysis between lightning and meteorological variables.It synergistically combines the box plot and the density trace into a single display (Hintze and Nelson, 1998).associated high level divergence.One thing to notice is the variable density shape in Fig. 3f.The bimodal distribution indicates that small to moderate high level divergence may due to atmosphere movement in clear sky or in-cloud with smaller updraft velocity that doesn't produce lightning, while large divergence usually characterize strong upward movement which is closely associated with lightning activity.Figures 3a, e show inverse correlations of lightning flashes to SLP and SHEAR and Figure 3b shows weak correlation to θ, all with quite small correlation coefficients that can not be considered as correlated.
Figure 4 shows the linear correlations between lightning flashes and six meteorological variables associated with strong convection for ROI_2.It is evident that RH, CAPE, and Div are positively associated with the occurrence of lightning, in sharp contrast to SLP, θ, and SHEAR that are negatively connected with lightning.In particular, Figures 4a, c, d, e show that CAPE, RH, Div, and SLP are significantly correlated with lightning (│R│>0.75,p<0.05), indicating that these four meteorological variables could be the major factors modulating the changes in lightning.By comparison, the weak linear relationship exists between lightning and θ (R=0.47), which is case for the relationship between lighting and SHEAR (R=-0.08),indicative of their minor effect on lightning (Figs.4b, 4g).Note that the correlation coefficients obtained here can only describe the possible dependencies between lightning and meteorological variables, which cannot imply the causal relationship.
In order to give a visual comparison of ROI_1 and ROI_2, we show the spatial distribution of correlation coefficients between lightning and meteorological variables.Spatially, Figure 5 shows that lightning flashes are well correlated with RH, CAPE, and Div throughout ROI_1 and ROI_2.

Contrast in the response of lighting to dust and smoke aerosol
Aerosols can modulate lightning activities through participating in radiative and microphysical processes.The roles of dust and smoke aerosols in modulating lightning activity are examined in Fig. 2 and Fig. 7.The diurnal cycle of lightning flashes in Fig. 2 shows that, for both dust and smoke aerosols, the lightning peak times under polluted conditions are delayed by about 1h.
Besides, smoke aerosols tend to apparently enhance lightning activity on both diurnal and seasonal cycles, while the impact of dust is much weaker.
Figure 7 shows that from clean to polluted condition, lightning response to AOD in an apparent boomerang shape (Koren et al., 2008) in both dust and smoke aerosol dominant regions, with a turning point around AOD=0.3.In order to reduce the effect of non-linearity, data are divided into two subsets before and after the turning point when perform correlation and regression.For AOD less than 0.3, lightning flashes increase monotonously with increasing aerosol loading in both regions.Clusters of the data are aligned around the regression lines rather tightly, implying that lightning flashes are strongly and positively correlated (R>0.75) with AOD.However, for AOD exceeding 0.3, the data points are more scattered and the trend is reversed, which imply that under large aerosol loadings, lightning is mainly influenced by other factors, such as meteorology.
Specifically, under smoky conditions, the correlation coefficient becomes negative (-0.15) and fails the significant significance test at 0.01 level, suggesting a suppressing effect is yielding to the invigoration effect.Under dusty conditions, the negative correlation is more significant with a correlation coefficient of -0.41, suggesting a stronger suppression under heavy dust conditions (Farias et al., 2014).Besides, we can easily find that lightning activity is much more intense in smoke aerosol dominant region that located in ITCZ where air is hot and humid, regardless of aerosol loading.In contrast, dust dominant region is much drier that is not so easy to produce intense convection and lightning.

Environmental dependence of aerosol effect
To further clarify the joint influences of meteorology and aerosols on lightning activity, the along the diagonal which shows that RH is highly correlated with CAPE, and they affect lightning activity in the same direction.Generally, intense lightning activity occurs under the condition of both high RH (>40%) and high CAPE (>100 J kg -1 ) in ROI_1.In ROI_2, high CAPE and high RH are still conductive to lightning production, but data disperse in a larger range, which suggests that the correlation of RH and CAPE are not as high as in ROI_1, and the restriction of RH is reduced.
As shown in Fig. 2 and Figs. 7, 8, the differences of lightning response to aerosols in ROI_1 and ROI_2 may also contribute to different climate conditions.In order to isolate the signal attributed to aerosol loadings from that attributed to environmental forcing, lightning flashes are separated by six meteorological variables (SLP, SHEAR, θ, CAPE, RH and Div) respectively.Besides, to note that, when SLP gets lower and RH gets larger, the differences of lightning flashes (polluted minus clean datasets) becomes larger.This suggests that under conductive condition, aerosols are more likely to participate in cloud process, convection development and thus modulate lightning activity.

Relative roles of meteorology and AOD in lightning
The response of lightning to changes in AOD may indicate an aerosol effect on lightning activity, but can also be the result of meteorology impacting on aerosol loadings and cloud microphysical process that is closely associated with lightning production.To further explore this complex process, the correlation of aerosollightning, meteorological variableslightning, aerosol-meteorological variables were examined before and after the turning point (AOD=0.3)respectively, results are shown in Fig. 10.
For clean condition (AOD<0.3) in the ROI_1, all meteorological variables and AOD show good correlations with lightning (Pearson1: │ R │ >0.5).Considering the interaction between aerosol and meterology, the correlation coefficients between AOD and six meteorological variables are calculated.Results show strong and positive correlation between AOD and RH, CAPE, Div, θ (Pearson2: R>0.6), negative correlation between AOD and SLP, SHEAR (in order of correlation strength, so as the following).To investigate the relative roles of these variables (AOD and six meteorological variables), we carry out partial correaltion analyses between lightning and any influencial factor while constraining all the others, and establish the standardized smaller role in the humid environment.Besides, AOD becomes more important.In both regions, aerosols correlate well with meterology, which could be the reason for the tight cluster distribution for AOD less than 0.3 in Fig. 9. CAPE measures the amount of moist static energy that is available to drive the convection (Rosenfeld et al., 2008).The high correlation between AOD and CAPE in both regions suggests that aerosol might modulate environment variables and participate in cloud microphysical process.A possible explanation is that more aerosols acting as CCN, lead to narrower cloud droplet size spectrum, delay warm-rain process, allow more liquid water to ascend higher into mixed-phase cloud, thus release more laten heat, produce more unstable atmosphere Under polluted conditions, CAPE and RH are still of paramount importance for lightning activity (Pearson1: R>0.8; Partial: R>0.35), but the correlation between aerosol and meterology are weakened.This weak connection between aerosol and meterology results in a large dispersion of lighting flashes under polluted conditions in both regions.The most important finding appears to be that the negative correlation of AOD with CAPE (Pearson2: R=-0.51), and RH (Pearson2: R=-0.33) in ROI_1 suggests that increasing dust aerosols may make environment drier, and atmosphere more stable through aerosol radiative effect which leads to the suppression of convection and lightning.While in ROI_2, AOD is negatively correlated with RH (Pearson2: R=-0.24) and θ (Pearson2: R=-0.74) which indicate the similar role of smoke aerosols in making environment drier and surface cooling through radiative effect.But, there's no significant correlation between AOD and CAPE (Pearson2: R=0, p>0.05), which may imply the radiative effect and invigoration effect are comparable under heavy loading codition.

Conclusions
Depending on specific environmental condition, aerosols are able to invigorate or suppress convection induced lightning activity in previous studies, the majority being case based.The Despite complex and diverse climatic conditions, the response of lightning to dust and smoke aerosols show in a boomerang shape with an optimum value of AOD around 0.3.We performed correlation analysis and standardized multiple regression analysis in attempt to constrain the effects of meteorology, and quantify the relative roles of these factors in modulating lightning.products on a 2.5°×2.5°grid (Ceil et al., 2001;2006;2014).
5) Wind Shear.The vertical shear of horizontal wind (U, V), hereafter simply referred as wind Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-251Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 April 2018 c Author(s) 2018.CC BY 4.0 License.shear or SHEAR, has been shown not only to affect the dynamical flow structures around and within a deep convective cloud Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-251Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 April 2018 c Author(s) 2018.CC BY 4.0 License.meteorological data are calculated by taking 3-month running mean and resampled to 2.5°×2.5°resolution grids in the climatic analysis.In order to make the comparison within the same AOD range and increase the number of data samples, climatological features of lightning, AOD and meteorology under polluted and clean conditions are limited to the cases with AOD<1.0 over both ROIs.The top third of AOD is labeled as polluted case and the bottom lowest third is labeled as clean case.All data are sorted by AOD, and divided into three equal-sample subsets to create sufficient contrast between polluted and clean conditions.

(
AOD) from the MODIS onboard Aqua satellite during the period from 2003 to 2013.The lower left panel is the Angstrom Exponent (AE) obtained from the Modern-Era Retrospective analysis for Research and Application (MERRA) Aerosol Reanalysis (MERRAero) at a 0.625°×0.5°spatial resolution for analysis of contributions from different aerosol species, chiefly, dust, black carbon (BC), organic carbon (OC) and total extinction AOD.Note that the satellite retrievals of the AE have too large uncertainties over land.The Africa stands out very distinctly with very large AOD in two regions: Sahara covered by dust (the upper right panel) and central to south (lower right panel) Africa by smoke, characterized by small and large values of angstrom index (lower left panel).Due to their distinct differences in aerosol species, they are selected as the regions of our interest ROI_1 and ROI_2 (Fig.2a).The ratios of dust (ROI_1) or (BC+OC) (ROI_2) extinction AOD to total extinction AOD are larger than 50% averaged over the period2003-2013, which      enables us to study any different effects on lightning activities.

Figure 3
Figure3shows the linear relationships between lightning flashes and six meteorological variables in ROI_1.CAPE, RH and Div are the top three meteorological variables strongly and positively correlated with lightning flashes (R>0.7), which suggests that high RH and CAPE are conductive to the development of intense convection and lightning occurrence that is always Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-251Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 April 2018 c Author(s) 2018.CC BY 4.0 License.After we constrain the variations of AOD and other five meteorological variables, partial correlation analyses show that only CAPE and RH are the top two factors affecting lightning activities (Fig. 6).

Figure 9
Figure9shows the differences of lightning flashes between polluted and clean conditions (polluted minus clean datasets) as a function of SLP, SHEAR, θ, CAPE, RH and Div.Generally, there are more lightning flashes for all these meteorological variables under polluted conditions when compared with that under clean conditions in both ROI_1 and ROI_2.All the lighting enhancement under polluted conditions are highly significant (>99%), based on Student's test.The differences Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-251Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 April 2018 c Author(s) 2018.CC BY 4.0 License.multiple regression equation, of which the coefficients represent the relative importance of each factor.After the common effects are constrained, partial correlation coefficients are much smaller compared to Pearson correlation coefficients, and the correlation between lightning and SLP, θ, AOD are not significant any more.It's envisioned that lightning doesn't respond much to dust aerosols directly, but dust can affect convection and lightning activities through modulating meterological variables.From these analyses, the top three factors are found to be RH, CAPE, and Div for dust dominant region under relatively clean condition.For clean ROI_2, the analyses show strong positive correlation between lightning and CAPE, AOD, Div (Pearson1:│R│>0.7),strong negative correlation between lightning and SLP (Pearson1: R=-0.94), weak negative correlation between lightning and θ, SHEAR (Pearson1:│R│<0.4).Main interplay exists between AOD and SLP, CAPE (Pearson2: │ R │ >0.75).The partial correlation coefficients and coefficients of standardized multiple regression equation reveal top three factors: CAPE, AOD, and RH (Partial: R>0.35).Different from RH as the top restraint facotr in dust dominant region, here it plays a Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-251Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 April 2018 c Author(s) 2018.CC BY 4.0 License.and larger CAPE which is conductive to convection development.The aerosol invigoration effect may play the key role during this stage (AOD<0.3).
controversial conclusions motivate us to study the long-term effects of aerosols and meteorology on lightning activity.Here, meteorology is characterized by six variables: sea level pressure (SLP), potential temperature (θ) at 2 m above ground level, mid-level relative humidity (RH), convective Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-251Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 April 2018 c Author(s) 2018.CC BY 4.0 License.availabe potential energy (CAPE), vertical wind shear (SHEAR), and 200 hPa divergence (Div).The current study investigates the response of lightning activity to meteorology and AOD over dust and smoke dominant regions from a climatological perspective.In particular, 11-year (2003In particular, 11-year ( -    2013) )  worth of coincident data are used, including lightning data from LIS/TRMM, AOD from MODIS/Aqua, and meteorological variables from ECMWF ERA-Interim reanalysis.Climatological features of diurnal and seasonal variations of lightning flashes show the peak in the afternoon and local summer respectively, which suggests a dominant role by thermodynamics, while the differences in lightning under relatively clean and polluted conditions signify the potential influences of aerosols.Specifically, under all stratified meteorological environments, lightning flashes are larger under polluted conditions than under clean condition.And the differences of lightning flashes in ROI_2 are larger than in ROI_1.Besides, lightning flashes increased much more when SLP gets lower and RH gets larger.

Figure 1 .
Figure 1.Global spatial distribution maps of (a) AOD (550 nm) as derived from MODIS at a spatial resolution of 1°×1°, and (b) Total aerosol angstrom parameter (470-870 nm) from MERRA dataset on a 0.625°×0.5°grid for the period 2003-2013.Also shown is the distribution of the ratio of dust AOD to total AOD over the region of interest outlined in red box (in panel a), and panel d is same as panel c, but for the ratio of (BC+OC) AOD to total AOD.Dust, black carbon, organic carbon and total extinction AOD used here are all derived from MERRAero dataset (da Silva et al., 2015).

Figure 2 .
Figure 2. (a) The distribution of 850 hPa mean wind field from ERA-Interim re-analysis with a spatial resolution of 1°×1° showing the prevailing wind direction over Africa and the neighbouring ocean over the region of interest (ROI) defined in Fig. 1.ROI_1 and ROI_2 (highlighted in black rectangles) are chosen with the ratio of dust or (BC+OC) extinction AOD to total extinction AOD being greater than 50% averaged over the period from 2003 to 2013, which enable us to better understand the potential effect of dust or smoke aerosols on lightning, respectively.Also shown are the diurnal cycle (b) and monthly variation (c, d) of AOD and lightning flashes as caculated under relatively clean and polluted (dusty/smoky) conditions in dust dominant region (ROI_1) and smoke dominant region (ROI_2), respectively.Unless noted otherwise, AOD used in this study is derived from MODIS, and the lowest (highest) third of AOD is labeled as clean (polluted) condition.Lightning flashes come from the low resolution monthly time series (LRMTS) and the low resolution diurnal climatology (LRDC)

Figure 3 .
Figure 3. Violin plots of lightning dispersion showing the relationship of lightning with six meteorological variables (SLP, θ, RH, CAPE, SHEAR, Div) in ROI_1.The five bins are equallyspaced.Box plots represent the interquartile range (the distance between the bottom and the top of the box), median (the band inside the box), 95% confidence interval (whiskers above and below the box)of the data, maximum (the end of the whisker above), minimum (the end of the whisker below), mean (orange dot) in each bin.Red '+' represent outliers.On each side of the black line is the kernel estimation to show the distribution shape of the data.The estimate is based on a normal kernel function, and is evaluated at 100 equally spaced points.Wider sections of the violin plot represent a high probability that members of the population will take on the given value; the skinnier sections represent a lower probability.The equations show the linear relationships (orange lines) between lightning and meteorological variables.Also given are Pearson correlation coefficients (R), p values and linear regression lines (in orange).

Figure 4 .Figure 5 .
Figure 4. Same as in Fig. 3, but for ROI_2.The mean values are represented by black dots, and the linear regression lines are shown in black.

Figure 6 .
Figure 6.Same as in Fig. 5, but for partial correlation coefficients between lightning flashes and any meteorological variable while controlling AOD and the other five meteorological variables.

Figure 7 .
Figure 7. Scatter plot showing the response of lightning to dust or smoke aerosols in a boomerange shape in both ROI_1 and ROI_2.Turning points are around AOD=0.3.Regressions are calculated before and after the turning points in these two regions of interest.Regression equations, correlation coefficients (R), p values, and linear fitting curves are in black (orange) for ROI_1 (ROI_2).All data used here are processed with taking an average of 10 samples after ordering AOD from small to large to reduce the uncertainty caused by large dispersion of data .

Figure 8 .
Figure 8. Joint dependence of lightning flashes on CAPE, RH, and AOD in both ROI_1 (a-c) and ROI_2 (d-f), respectively.Bold number in each cell indicate the number of samples in each cell.Colorbar denotes the number of lightning flashes averaged in each cell.

Figure 9 .
Figure 9. Differences (polluted minus clean subsets of data) of lightning flashes as a function of (a)sea level pressure (SLP), (b) potential temperature (θ), (c) mid-level relative humidity (RH) , (d) convective available potential energy (CAPE), (e) vertical wind shear (SHEAR), (f) 200 hPa divergence (Div) in ROI_1 (in orange) and ROI_2 (in black).Note that, the top third of AOD is labeled as polluted case and the bottom third is labeled as clean case.Vertical error bars represent one standard deviation.

Figure 10 .
Figure 10.Correlations between AOD and meteorology, and between lightning and meteorology, AOD.(a,d) Pearson correlation coefficients between lightning and six meteorological variables and AOD (Pearson1).(b,e) Partial correlation coefficients between lightning and any influential factor (AOD or meteorological variable) while controlling the other variables (Partial).(c,f) Pearson correlation coefficients between AOD and any given meteorological variable (Pearson2).Those bars with black dots denote that they fail the statistical significance test at 95% confidence level.Also shown are standardized multiple regression equations of lightning (y) onto six meterological variables (x -x )and AOD (x ), and standardized multiple correlation coefficients (R ).Six meterological variables are SLP (x ), θ (x ), RH (x ), CAPE (x ), SHEAR (x ), and Div (x ).
(Figs. 8b, 8d).In dust dominant region, lightning flashes monotonically increase as RH increases under any level of aerosol loadings, but changes a little as AOD increases in each RH bin.This suggests that, apart from CAPE, RH is another constraint of lightning activity in ROI_1.But for smoke aerosol dominant region, high lightning flashes appear in the environment of moderate RH and high aerosol loadings.When RH is fixed, the response of lightning flashes to AOD also shows a turning point in AOD around AOD=0.3, beyond 0.3, lightning flashes remain high value.When looking into the common role of RH and CAPE on lightning, the data distributes distribution of lightning flashes with AOD and the top two influential meteorological variables RH and CAPE (based on the results in Figs.3-6) are examined in Fig. 8. Lightning flashes are classified into 100 discrete cells by ten decile bins of AOD -CAPE, AOD -RH, and CAPE -RH respectively, which ensures approximately equal sample sizes among the cells.The mean values are calculated in each cell.Results show that, for each CAPE bin, lightning flashes increase with However, lightning flashes response to RH in different ways for ROI_1 and ROI_2 when AOD is Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-251Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 April 2018 c Author(s) 2018.CC BY 4.0 License.fixed Under relative clean conditions, standardized multiple regression coefficients of meteorology and AOD on lightning in both regions exhibit R  0.92, with RH and CAPE being the top two determinant factors.The contribution of RH is comparable with that of CAPE in dust dominant region, and less in smoke aerosols dominant region.The narrow distribution and high correlation between aerosol and meteorology imply that the impact of AOD on lighting activities is likely to Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-251Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 April 2018 c Author(s) 2018.CC BY 4.0 License.be exerted through microphysical effect which modulate meteorological variables.Beyond the optimum value, lightning shows more dispersed distribution and has much weaker dependence on AOD, which may be the consequence of competition between aerosol microphysical effect and radiative effect.Under smoky conditions, R for the standardized multiple regression equation diminishes to 0.77 with a strong negative correlation (r=-0.74)with θ, and weak negative correlation with mid-level RH and no correlation with CAPE.Note that aerosol cools surface and warms mid-level atmosphere through radiative effect, which may be less (for AOD<0.3),more Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-251Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 April 2018 c Author(s) 2018.CC BY 4.0 License.