Estimations of anthropogenic dust emissions at global scale 1 from 2007 to 2010 2

Abstract. Dust emissions refer to the spatial displacement of dust particles from wind forcing, which is a key component of dust circulation. It plays an important role in the energy, hydrological, and carbon cycles of the Earth's systems. However, most dust emission schemes only consider natural dust, neglecting anthropogenic dust induced by human activities, which led to large uncertainties in quantitative estimations of dust emissions in numerical modeling. To fully consider the mechanisms of anthropogenic dust emissions, both indirect and direct anthropogenic dust emission schemes were constructed and developed in the study. Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) retrievals were used to constrain the simulations at global scale. The results showed that the schemes reasonably reproduced the spatio-temporal distributions of anthropogenic dust from 2007 to 2010. The high centers of anthropogenic dust emission flux appeared in India, eastern China, North America, and Africa range from 0.9 to 11 μg m−2 s−1. Compared with natural dust emissions, indirect anthropogenic dust emissions have indistinctive seasonal variation, with differences less than 3.2 μg m−2 s−1. Pasturelands contribute higher anthropogenic dust emissions than croplands, with emissions of approximately 6.8 μg m−2 s−1, accounting for 60 % of indirect anthropogenic dust emissions. Moreover, average anthropogenic dust emissions in urban areas have a value of 13.5 μg m−2 s−1, which is higher than those in rural areas (7.9 μg m−2 s−1). This study demonstrates that the environmental problems caused by anthropogenic dust in urban areas cannot be ignored.


Introduction
Dust emissions are the result of a key process in the dust cycle that determines long-term transport, dry/wet deposition, radiation forcing, and other dust-related processes at both regional and global scales (Tegen and Fung, 1994;Gong et al., 2003Gong et al., , 2004;;Han et al., 2004Han et al., , 2010Han et al., , 2012Han et al., , and 2013;;Shao et al., 2004Shao et al., , 2011;;Huang et al., 2006a, b, c;Chen et al., 2013Chen et al., , 2014a, b;, b;Li et al., 2016).Scientists have been constructing and developing dust emission schemes for simulation studies since the 1980s.Based on different assumptions and simplifications, Zender et al. (2003), Han et al. (2004), and Shao et al. (2006Shao et al. ( , 2011) ) divided dust schemes into three categories: empirical schemes, schemes based on simplified physical processes, and schemes based on detailed microphysical processes.The development of dust schemes has deepened our understanding of dust-related processes and dust's influences on the environment and the climate at the regional and global scales (Gong et al., 2006;Zhao et al., 2010Zhao et al., , 2013;;Huang et al., 2007Huang et al., , 2014;;Chen et al., 2014Chen et al., , 2017a, b;, b;Liu et al., 2016).
Current studies on simulated dust emissions have mostly focused on natural dust.There remains a gap with regard to simulating dust emissions induced by human disturbances.In the early 1990s, Penner et al. (1994) and Tegen and Fung (1995) suggested that it was inaccurate to classify dust aerosols as natural aerosols.Dust aerosols should be classified as either natural dust or anthropogenic dust according to its source region.Natural dust emissions essentially originate from natural dust source regions.Sufficiently strong winds occur over bare soil surfaces and dust particles are lifted and emitted into the atmosphere (Shao et al., 2004;Ginoux et al., 2012).Anthropogenic dust can be interpreted as dust emitted through modifying or disturbing soil particles through direct and indirect human activity (Penner et al., 1994;Tegen and Fung, 1995;Huang et al., 2015).Furthermore, anthropogenic dust emissions are mainly derived from wind erosion in anthropogenic dust land use due to the dryland vulnerability (i.e., "indirect anthropogenic dust emissions") (Xi et al., 2015) and human endeavors directly including urban activities, industrial activities (e.g., construction, cement production, and transportation), and farming (e.g., harvesting, plowing, and overgrazing) (i.e., "direct anthropogenic dust emissions") (Moulin and Chiapello, 2006;Munkhtsetseg et al., 2017).
The contributions of anthropogenic dust to the total dust mass cannot be ignored (Ginoux et al., 2012;Huang et al., 2006aHuang et al., , b, 2015Huang et al., , 2016;;Xi et al., 2016;Guan et al., 2015, Chen et al., 2014b, 2017c;Kang et al., 2015;Luan et al., 2017).Compared with naturally occurring dust, anthropogenic dust particles can more easily be emitted continuously from anthropogenic dust source regions, mostly because these areas contain freshly exposed soil with more fine materials (Tegen and Fung, 1995;Zheng et al., 2016) which makes the soil more susceptible to erosion due to the lower threshold friction velocity (Tegen et al., 2004).Such large amounts of anthropogenic dust particles are likely to have a considerable impact on regional climate variations.Previous studies have pointed out that the radiative forcing induced by anthropogenic dust is likely to be equivalent to other anthropogenic aerosols, although these simulations had a large degree of uncertainty (Sokolik and Toon, 1996).Other studies have found that anthropogenic dust can have an impact on nutrient deposition (Mahowald et al., 2008) and regional snowpack (Ye et al., 2012;Semborski, 2013;Zhao et al., 2013).Therefore, quantitative estimation of anthropogenic dust emission is crucial to reinforce the understanding of the dust cycle and its climate effects at the regional and global scale, and decrease the uncertainties of dust emission flux in the numerical modellings.
At present, the uncertainty of determining anthropogenic dust sources and constructing dust emission schemes has led to larger biases in the estimation of anthropogenic dust emissions.This research has noted that simulated anthropogenic dust contributions to the total dust loading mass have ranged from 10% to 60% (see Table 1).Below, we have summarized several key reasons for such large uncertainties.
First, lacking of observation constraints on estimations of anthropogenic dust.Due to the difficulties in detecting and discriminating of anthropogenic dust, simulated anthropogenic dust has always been limited by a lack of observational constraints.Ground observations can not capture the anthropogenic dust emission well because observed dust loading is a mixture of natural dust and anthropogenic dust.With the development of remote sensing and inversion algorithms, Ginoux et al. (2010) identified anthropogenic and natural dust sources in western Africa based on Moderate Resolution Imaging Spectroradiometer (MODIS) Deep Blue aerosol products in combination with land use data.This approach indicated that anthropogenic dust accounts for 25% of all mineral aerosols (Ginoux et al., 2012).However, their retrieval method was only applicable over bright surfaces in the visible wavelength and was unable to properly exclude natural dust aerosols due to the lack of vertical information.Huang et al. (2015) proposed a new technique to identify anthropogenic dust using Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO).Their estimated anthropogenic dust contribution was approximately 25% of global dust loading and 52% of it in semi-humid and semi-arid areas.These studies provide valuable observations that could be used to constrain simulated anthropogenic dust in numerical modeling.
Second, neglecting the influence of dynamic land surface in the anthropogenic dust emission.Most dust emission schemes have employed "climatological" land cover to identify dust source distributions but have neglected temporal variations linked to surface bareness (Kim et al., 2013(Kim et al., , 2017)).Compared with natural dust sources, anthropogenic dust sources have diverse feature types, scattered distribution, and high spatiotemporal variability.
Therefore, anthropogenic dust source regions have more significant seasonal and inter-annual variations (Huang et al., 2015).These dynamic land cover changes should be considered when estimating anthropogenic dust sources.
There is a statistical relationship between the normalized difference vegetation index (NDVI) and dust concentrations in dust source regions (Zender and Kwon, 2005).Thus, Kim et al. (2013) 2016) further used the dynamic dust source functions from Kim et al. (2013) to quantify anthropogenic dust emissions from agricultural land use.
Third, neglecting the direct anthropogenic dust emissions induced by human activities.Previous studies have commonly employed indirect anthropogenic dust emissions in agricultural land use (e.g., Xi et al., 2016).However, rapid urbanization and increasingly frequent human activity are likely to produce large amounts of anthropogenic dust particles in urban areas.
Observations have shown that anthropogenic dust mass loading is stronger than natural dust loading in densely populated regions with a high level of human activity.For example, anthropogenic dusts accounts for more than 91.8% and 76.1% of the total dust loading in east China and India, respectively (Huang et.al., 2015).Guan et al. (2016) further pointed out that direct anthropogenic dust loading in congested areas where the population density is more than 90 people per square kilometer (people km −2 ) is much larger than the indirect anthropogenic dust from croplands, pasturelands, and grasslands.Taking East Asia as an example, the population density has been growing significantly over the past half century.The urban population in East Asia is approximately 60.1% of the entire population of East Asia, which was more than half the global urban population until 2015 (Mitchell et al., 2016).Thus, direct anthropogenic dust emission scheme should be considered in dust modeling.
In this study, we estimated the spatial distribution of anthropogenic dust emissions at global scale from 2007 to 2010.This paper is organized as follows.

Methods
Previous research has found that human endeavors can directly and indirectly contribute to anthropogenic dust uplift (Zender et al., 2004;Xi et al., 2016).According to differences in the mechanisms of anthropogenic dust emissions, we divided these dust emissions into direct and indirect anthropogenic dust uplift, respectively.We used different methods to simulate these two anthropogenic dust emission sources using observations and reanalysis data, respectively.

(1) Indirect anthropogenic dust emission
To isolate the role of meteorology from the land surface effects, Marsham et al. (2011) simplified the dust emission scheme developed by Marticorena and Bergametti (1995).The scheme neglected differences from using wind speed at 10 m rather than at threshold velocity (Marsham et al., 2011).Instead, they substituted the threshold wind velocity by a constant of 7 m s −1 .Although this approach neglected the second-order effects of stability and roughness, it is a simple and easy method to better quantify the effects of meteorology on dust emissions at global scale over long time periods (Cakmur et al., 2004).Cakmur et al. (2004), Marsham et al. (2011), andEvan et al. (2016) pointed out that this dust emission scheme could reflect potential dust emissions, which is closely related to real-world dust emissions.
Indirect anthropogenic dust emissions are commonly caused by the erosive force of wind over anthropogenic land surfaces.Therefore, we used the simplified dust emission scheme by Marticorena and Bergametti (1995) to estimate the indirect anthropogenic dust emission.The influence of dynamic land surface in the indirect anthropogenic dust emission was also considered.Indirect anthropogenic dust emission flux G1 (μg m −2 s −1 ) was calculated as follows: where C is an empirical proportionality constant (units: μg s 2 m −5 ), Sad is the anthropogenic dust source function, u is the wind speed at 10 m, and ut is the threshold velocity depending on surface characteristics (when u > ut, soil particles are possibly being uplifting).Here, we chose ut = 6.5 m s −1 according to Tegen et al. (2004) because human disturbances make the soil more susceptible to erosion.
The anthropogenic dust source function Sad represents the probability of indirect anthropogenic dust uplifting with a range between 0 and 100.Sad is calculated by multiplying the accumulated sediments H, with the surface bareness percent B (Kim et al., 2013).Notable, Sad is only calculated for anthropogenic land covers (i.e., C4 croplands and C4 pasturelands; for detailed information please see Section 2.2.2).Furthermore, high values of snow cover and soil moisture are excluded in this Sad calculation.Surface bareness B is a "static" function that did not reflect the seasonal and inter-annual variations of land cover and soil bareness in the most research of dust emission schemes (e.g., Ginoux et al., 2001;Chin et al., 2002).Chen et al., (2014bChen et al., ( , 2017a) ) pointed out that the "static" dust source function could lead to uncertainty in the simulated seasonal dust emission flux.Kim et al. (2013Kim et al. ( , 2017) ) used the NDVI to obtain a dynamic dust source function in Sahel, choosing 0.15 as the threshold to discriminate the bare ground.Hence, the NDVI of 0.15 was chosen to define the threshold surface bareness in this study.Next, B was calculated as the ratio of the number of NDVI pixels below 0.1 (i.e., N<0.15) to the total number of NDVI pixels within the 1°× 1°grid cell (i.e., Ntotal) as follows: Additionally, topographical depression features H, is defined as the probability of having accumulated sediments in grid cell i of altitude zi (the local averaged surface elevation).A map of H was constructed from the altitude in the grid cell i relative to the altitude of the surrounding areas within a 5°× 5°235 grid in this study (Ginoux et al., 2001;Kim et al., 2013).
where zmax and zmin are the maximum and minimum elevations within the ×5˚grid, respectively, and zi is the altitude in the grid cell i.
(2) Direct anthropogenic dust emissions Direct anthropogenic dust emissions primarily originate from human activities and urbanization processes, such as city construction, cement production, traffic, and transportation.Population density, urbanization, and the levels of regional economic development, as important driving factors, should be contained in the calculating direct anthropogenic dust.The STIRPAT (the stochastic impacts by regression on population (P), affluence (A), and technology (T)) model is widely used to analyze the effects of driving forces on a variety of environmental impacts (Dietz and Rosa, 1997;Soulé and DeHart, 1998;Shi, 2003;and York et al., 2003).Here, we employed the STIRPAT model to calculate direct anthropogenic dust emissions based on the population density, compounded nighttime light index (CNLI), and Engel coefficient.The direct anthropogenic dust emissions, G2 (μg m −2 s −1 ), was calculated using the following equation: (5) into a linear form, we have ln( ) ln ln( ) ln( ) ln( ) To determine the coefficients for a, b, c, and d, we used the least squares method based on the anthropogenic dust column from CALIPSO, together with the independent variables of P, CNLI, and EC to determine that b=0.1, c=0.1, and d=1.6.In the regression calculation using the least squares method, we validated the well-parameterization for equation with a reasonable error (root mean square error=1.21)and a good fit (coefficient of determination=0.26).At a significance level of α =0.005, all of the independent variables passed an F-test, indicating that direct anthropogenic dust emissions showed good agreement with the population density, CNLI, and EC.Notably, high vales of soil moisture were excluded, because they reinforced inter-particle cohesion forces thus limiting the probability of dust emissions (Fécan et al., 1998).

ERA-Interim
We used the reanalysis product ERA-   (Redelsperger et al., 2006).Therefore, wind speed at 10 m from the ERA-Interim datasets was chosen to calculate indirect anthropogenic dust emission flux in this study.

Land cover datasets
Land cover datasets from Meiyappan and Jain (2012) incorporate 28 types of land cover, including 16 types of natural land cover (e.g., forests, grasslands, shrubs, etc.) and 12 types of land cover disturbed by human activities (e.g., secondary forests, croplands, pasturelands, and urban environments) (Table 2).
This land cover dataset, combined with the Historical Database of the Global Environment (HYDE 3.1) (Klein et al., 2010(Klein et al., , 2011)), wood harvest data, and urban land data, was used to construct the anthropogenic land cover map including cropland, pastureland, and urban regions, which provide dynamic land cover variations at 0.5°×0.5°resolutionfrom 1770 to 2010.Friedl et al. (2010) pointed out that the datasets effectively captured the spatial and temporal distributions of land cover at global scale compared with MODIS retrievals from 2000 to 2005.
Huang et al. ( 2015) decided to limit the mapping to three surface types (croplands, grasslands, and cropland mosaics) to define anthropogenic dust source regions based on the land cover products from MODIS Collection 5.1.
Here, we chose both cropland and pastureland as indirect anthropogenic dust source types.In addition, according to differences in photosynthesis dark croplands and C4 pasturelands as indirect anthropogenic dust source regions according to the land cover data provided by Meiyappan and Jain (2012).

Compounded night light index (CNLI)
Researchers have always utilized nighttime light, based on the Defense Meteorological Satellite Program/The Operational Linescan System (DMSP/OLS), to extract spatial distribution characteristics of urban areas (Elvidge et al., 1997).The DMSP satellite carries the OLS, which has a visible channel and an infrared channel with gray levels ranging from 0 to 63 and 0 to 255, respectively.The OLS has a strong photoelectric amplification capacity be recorded in the nighttime imagery.This has been widely used abroad to detect urban areas, supervise fires, etc. (Sutton, 1997;Elvidege et al., 1997Elvidege et al., , 2013;;Lo et al., 2001).
To simulate direct anthropogenic dust emissions caused by human activities, we estimated the CNLI from 2007 to 2010 based on nighttime light datasets that recognize urbanization levels at global scale.The CNLI proposed by Zhuo et al. (2003) represents urbanization level with a value between 0 and 1.
CNLI is defined as the ratio of lit urban areas to the whole region R and average night light brightness I within a 1°× 1°grid cell as follows: R is computed using the following formula (8): where AreaN is the area of lit urban areas in a region and Areatotal is the 1°362 × 1°grid in the calculation.Night light brightness I is presented below: where the digital number (DNi) is the ith gray level of the DN value within the 1°×1°grid; ni is the total number of lit pixels belonging to the ith gray level; and P is the threshold value representing the beginning of an increasing urbanization trend.DNM is the maximum potential DN value within the 1°× 1°368 grid, and NL is the number of lit pixels whose DN value is between P and DNM (i.e., the number of lit pixels).(Winker et al., 2007;Hu et al., 2007aHu et al., , 2007bHu et al., , 2009)).This study used CALIPSO retrievals to calculate the anthropogenic dust optical depth (ADOD) and the anthropogenic dust loading, following the approach of Huang et al. (2015).The first step was to detect the total dust column from CALIPSO retrievals, and the second step was to select the potential dust source regions based on the datasets from Meiyappan and Jain (2012).Then we calculated the height of the planetary boundary layer (PBL) because most anthropogenic dust particles accumulate under the PBL (Jordan et al., 2010;Yu et al., 2012).Finally, we calculated the anthropogenic dust optical depth and the anthropogenic dust column.A detailed description of this anthropogenic dust detecting procedure based on CALIPSO retrievals can be found in Huang et al. (2015).

Indirect anthropogenic dust emission
The land cover datasets used in this study reproduced the spatial distributions of anthropogenic land cover over the past 100 years.The dominant 17 types of land cover from Meiyappan and Jain (2012) are shown in Figure 1a.
Land cover types can be divided into anthropogenic land use and natural land use (Table 1).Croplands were mainly distributed in the eastern and central North America, east and central Asia, as well as throughout Europe.
Pasturelands dominate central North America, eastern South America, central Asia, and the southern Sahara.
In this study, we only included C4 croplands and C4 pasturelands as potential indirect anthropogenic dust sources (Figure 1b and 1c), demonstrating the wide spread of indirect anthropogenic dust.C4 croplands have common crops, such as corn and sorghum (Ehleringer and Cerling, 2002), that are distributed extensively throughout the tropics and subtropics in regions such as central and eastern North America, the southern Sahara, southern Europe, eastern China, and western India.C4 pasturelands are also mainly distributed in the tropics and subtropics, in regions such as central North America, the southern Sahara Desert, the northern region of South America, and the southern region of the Yangtze River Basin in China.Its turf grass is mainly comprised of poapratensis and fescue grasses (Ehleringer and Cerling, 2002).Although C4 pasturelands are comparatively less extensive than C4 croplands in Europe and the central and east of Asia, the proportion of C4 pasturelands is significantly higher in the east of South America, the southern region of the Sahara Desert and Africa, as well as western India.For example, the percentage of C4 pasturelands can reaches up to 50% in South Africa and South America, while C4 croplands rarely occupy more than 20% of the total area.
The NDVI values indicate that there are significant seasonal variations in vegetation cover and surface bareness, especially in anthropogenic land areas (Figure 2).On the whole, NDVI in July is generally higher than that in January in the Northern Hemisphere, where the difference can reach up to 0.3.The variations in NDVI are comparatively slight in deserts like the Sahara, western regions of Asia, and the Taklimakan Desert in Australia, all with differences of approximately 0.1.Because the two hemispheres have opposite seasons and that special climate characteristics were measured at regional scale, NDVI values decreased from January to July in a few regions like southern Africa (Figures 2a     and b).This is consistent with previous results from Kim et al. (2013).
The global surface bareness map was constructed using NDVI data (Figures 2c and d).The surface bareness in cold seasons is more extensive and intensive than in warm seasons, especially in the south Sahara Desert and in central and east Asia.Interesting, the bareness in Australia is the opposite due to the unique climate and vegetation characteristics at the regional scale.This is likely because evergreen trees dominate the northern part of Australia and tend to be denser in July.Moreover, some of the regions, like southwest Australia, experience a Mediterranean climate in which vegetation grows thicker in July (Scott et al., 1993;Bowman et al., 2005).
The indirect anthropogenic dust source function Sad reflects the probability of indirect anthropogenic dust uplifting, which is constructed using soil bareness (Figures 2c and d) and topographic features (Figure 3b) in C4 croplands and C4 pasturelands.The higher topographic depression H reflects flatter regions, which is more likely to have accumulated sediment (Figure 3b).
In Figures 3c and 3d, we can see that Sad experienced significant variation, and was distributed in central and east Asia, the southern Sahara, and western North America.It was more widespread in January than in July in the Northern Hemisphere, especially in western regions of North America, the southern Sahara, eastern China, and central Asia.The Southern Hemisphere tends to experience the opposite of these variations, although Australia is an exception, as discussed earlier.
The global distribution of seasonal indirect anthropogenic dust emission flux from 2007 to 2010 is shown in Figure 4.The highest centers of indirect anthropogenic dust emission flux occurred in North America (1.80 μg m −2 s −1 ), India (3.39 μg m −2 s −1 ), and eastern China (2.60 μg m −2 s −1 ) due to the wide range of C4 croplands and C4 pasturelands (Figure 1).As human disturbances can make soils more susceptible to erosion, anthropogenic land cover types contributed a considerable proportion of indirect anthropogenic dust to total anthropogenic dust (Justice et al., 1996;Huang et al., 2015).Indeed, indirect anthropogenic dust emission flux consistently showed indistinctive seasonal variation compared with natural dust emission flux.The variations in anthropogenic dust emission flux over four seasons were no more than 0.64 μg m −2 s −1 .CNLI is the light index reflecting the development of urbanization (Figure 5b).

Direct anthropogenic dust emission
The high centers of CNLI appeared in India, the east of China, Europe, and the east of North America, indicating a higher urbanization level in these regions, which showed good agreement with the population density.
Moreover, direct anthropogenic dust emissions also depend on economic development.Huang et al. (2015) found that direct anthropogenic dust was negatively correlated with regional economic progress.Hence, EC indicates the proportion of the total food expenditure to the total amount of consumer spending, as this plays an essential role in evaluating the living standard of residents and the region's stage of economic development (Zhang et al., 2010).
Previous studies have showed that a region with an EC value higher than 0.6 can be defined as poverty stricken.When the value falls between 0.5 and 0.6, the population is barely meeting its daily needs.If the value falls between 0.4 and 0.5, there is a moderately well-off standard of living.If the value falls between 0.3 and 0.4, there is well-to-do standard of living.Finally, if the value falls below 0.3, the population is generally quite wealthy (Zhang et al., 2010).
As shown in Figure 5a, the economic development of a country is negatively correlated with EC.The United States, England, and France are the most advanced developed countries in the world with EC values as low as 0.08, 0.13, and 0.17, respectively.China and India are both developing countries with an EC value of 0.22 and 0.26, respectively.Direct human activities dominate anthropogenic dust emissions.
Magnitudes of direct anthropogenic dust emission flux are nearly three to four times higher than that of indirect anthropogenic dust emission flux (Figure 6).
Direct anthropogenic dust emission flux shows clear regional heterogeneity.
Developing countries, such as India and China, contribute the greatest amount of direct anthropogenic dust (up to 4.3 μg m −2 s −1 and 3.0 μg m −2 s −1 , respectively) due to their incomplete industrial structures, city construction, and less restrictive environmental regulations.Outside of these two countries, in densely populated regions of developed countries, the average direct anthropogenic dust emission flux is comparatively less at approximately 1.6 μg m −2 s −1 because city development and environmental policies are more mature.

Total anthropogenic dust emission
Total anthropogenic dust emissions are overlaid by both direct and indirect anthropogenic dust.Figure 7 shows the simulated seasonal variations of anthropogenic dust emission flux at the global scale.The global annual mean anthropogenic dust column was approximately 0.11 g m −2 , and in regions like India, it could reach up to 0.87 g m −2 .It is evident that the simulated spatial distribution of anthropogenic dust emissions is highly consistent with that produced by CALIPSO retrievals (Figure 8).Anthropogenic dust emission flux has indistinctive seasonal variation compared with natural dust emissions due to the main contributions of human activities in urban regions (Huang et al., 2015).
The high centers of anthropogenic dust emission flux appear in eastern China, India, North Africa and North America, which is highly consistent with that of the anthropogenic dust column calculated by Huang et al. (2015) and Guan et al. (2016).However, the simulations underestimated the anthropogenic dust emissions in North America compared with CALIPSO retrievals due to the bias of estimating urbanization.Furthermore, Figure 9 shows the normalization anthropogenic dust from CALIPSO retrievals and our simulations in China, India, and North and South America.The simulations capture the differences of Compared with natural dust source regions, anthropogenic dust source regions are more complicated due to diverse types, scattered distribution, and high spatio-temporal variations.Divergences in anthropogenic land cover types are induced by the concept of "people managed" areas (Meiyappan et al., 2014).
Cropland, pastureland and urban belong to human managed land area.
Quantitative estimates of anthropogenic dust emissions in different land cover types are crucial to reinforce the understanding of dust emissions in anthropogenic land cover.For three major anthropogenic land covers, the contribution of anthropogenic dust emissions from croplands, pasturelands, and urban areas to the total anthropogenic dust column is 20.76%, 28.38%, and 50.86%, respectively (Figure 10), indicating that direct anthropogenic dust emissions from urban areas play a dominant role in anthropogenic dust.
Pasturelands includes pastures and artificially sparse grasslands, which contribute higher anthropogenic dust emissions than croplands due to the more intense distributions and higher anthropogenic dust source functions in C4 pastureland compared with those of C4 croplands (Figure 1, Figure 3).Further, although rural areas is larger than urban, anthropogenic dust emissions (13.54 μg m −2 s −1 ) in urban areas is higher than that in rural areas (7.89 μg m −2 s −1 ), suggesting that anthropogenic dust is more likely produced in urban areas than that in remote and rural areas (Figure 11).Therefore, policymakers should be paying much more attention to the control of air pollutions in urban areas.Further, anthropogenic dust emissions (13.54 μg m −2 s −1 ) in urban areas is higher than that in rural areas (7.89 μg m −2 s −1 ), suggesting that anthropogenic dust is more likely produced in urban areas than that in remote and rural areas (Figure 11).It is because that the larger bareness, more intensive population and high value of EC as well as CNLI in urban areas which result in both greater direct anthropogenic dust emission and indirect anthropogenic dust emission in urban than these in rural areas.Recently year, with increasing numbers of people who have migrated from rural areas to urban areas, the imbalanced distribution of anthropogenic dust emissions will be intensified, causing more ecological pressure for urban areas.Moreover, there exist much more human activities in urban areas, such as urban construction, cement production, transportation, energy consumption, etc, which cause large direct anthropogenic dust emissions.Therefore, policymakers should be paying much more attention to the control of air pollutions in urban areas.

Discussions and conclusions
The contribution of anthropogenic dust to the total atmospheric dust column is significant and should not be ignored (Huang et al., 2015).Previous dust emission modellings merely focused on the natural dust emissions and there is a great knowledge gap of the investigation of the anthropogenic dust emissions.There are more difficulties and uncertainties in anthropogenic dust emissions simulations compared with those in natural dust emissions, due to the diverse types, scattered distribution, and high spatio-temporal variability of anthropogenic dust sources (Xi et al., 2016).Thus, quantitative estimations of anthropogenic dust emissions are crucial to reinforce the understanding of the dust cycle and its climate effects, and decrease the uncertainties of dust emission fluxes in the numerical modellings.According to different anthropogenic dust emission mechanisms, both "indirect" and "direct" anthropogenic dust emission schemes were constructed and developed in the study, respectively.Indirect anthropogenic dust emissions are caused by the erosive force of wind on anthropogenic land surfaces as croplands and pasturelands.The simplified dust emission scheme proposed by Marsham et al. (2011) was used to simulate seasonal variations of indirect anthropogenic dust emissions in this study.In addition, previous studies focused on dust emissions in identifying dust sources only retained the static land cover types (Ginoux et al., 2001;Kumar et al., 2014;Nabavi et al., 2017).However, the static land cover types cannot reflect the dynamic change of dust sources well owing to seasonal variations of sparse vegetation.Therefore, dynamic land cover changes were considered by the anthropogenic dust source function based on the NDVI datasets in the study.Generally, indirect anthropogenic dust emission fluxes consistently showed indistinctive seasonal variation compared with natural dust emission fluxes.The highest centers of indirect anthropogenic dust emission flux occurred in North America (1.80 μg m −2 s −1 ), India (3.39 μg m −2 s −1 ), and eastern China (2.60 μg m −2 s −1 ).Notably, pasturelands (including pastures and artificially sparse grasslands) contribute higher anthropogenic dust emissions than croplands, with emissions of approximately 6.8 μg m −2 s −1 , accounting for 60% of indirect anthropogenic dust emissions.
Direct anthropogenic dust emissions primarily originate from direct human activities and urbanization processes.The mechanism of direct anthropogenic dust emission is quite different from that of indirect anthropogenic dust emissions.Population density and urbanization dominate the magnitude of direct anthropogenic dust emission fluxes (Guan et al., 2016).We utilized the STIRPAT model to simulate the spatial distribution of direct anthropogenic dust emissions from 2007 to 2010, taking the impacts of population, urbanization, and the economic development of a region into consideration.Results showed that direct anthropogenic dust emissions reflect regional heterogeneity.
Developing countries such as India and China act as dominant direct anthropogenic dust source regions (up to 4.3 μg m −2 s −1 and 3.0 μg m −2 s −1 , respectively) owing to their incomplete industrial structure, ongoing city construction, and less restrictive environmental regulations.For developed countries, lots of regions are a large urban agglomeration with dense population such as England in Europe and New York in North America, the magnitude of direct anthropogenic dust emission fluxes is comparatively less.The more Total anthropogenic dust emissions consist of both direct and indirect anthropogenic dust emissions.Total anthropogenic dust has a wide spread and the high value centers are concentrated in areas with a large population density and intense human activities in developing countries, such as India and eastern China.It indicates that direct anthropogenic dust plays an important role in total anthropogenic dust that cannot be ignored.Especially, anthropogenic dust emissions in urban areas (13.5 μg m −2 s −1 ) are higher than those in rural areas (7.9 μg m −2 s −1 ), suggesting that there is a greater potential for higher anthropogenic dust emissions in urban areas than in rural or remote areas.
In addition, due to the difficulties in detecting and discriminating of anthropogenic dust, the previous studies about simulations of anthropogenic dust emission are poorly evaluated by observations.As a unique observational constrain of anthropogenic dust, Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) retrievals were used to evaluate the simulations at global scale in the study.Huang et al., (2015) pointed out that anthropogenic dust is hard to lift up to the planet boundary layer for having a long-range transport, the anthropogenic dust column is generally contributed by the anthropogenic dust emissions in local regions.Therefore, we have compared the simulated anthropogenic dust emissions with the CALIPSO anthropogenic dust columns as noted in Huang et al. (2015) and Guan et al. (2016), without even having other observations.The comparisons indicated that the simulations captured the spatial distributions of CALIPSO anthropogenic dust well.In the future, the development of integrated systematic observations of anthropogenic dust in different land cover types is necessary for improving anthropogenic dust emission schemes and simulations.
developed a time dependency feature for their dust source function in the Goddard Chemistry Aerosol Radiation and Transport (GOCART) model simulations using NDVI from the advanced very highresolution radiometer (AVHRR) from 2002 to 2007.Xi et al. ( Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-890Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 1 December 2017 c Author(s) 2017.CC BY 4.0 License.Methods and datasets are described in Section 2. Model evaluation and discussion of anthropogenic dust emissions from 2007 to 2010 are presented in Sections 3. A broader discussion and conclusions are presented in Section 4.
an empirical proportionality constant (units: μg s 2 m −5 ); and b, c, and d represent the driving force indices.P represents the population density, CNLI (see Section 2.2.4) represents the urbanization level, and EC is the Engel coefficient, indicating the proportion of total food expenditure to total amount Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-890Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 1 December 2017 c Author(s) 2017.CC BY 4.0 License. of consumer spending, which represents economic development.Converting Eq.
Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-890Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 1 December 2017 c Author(s) 2017.CC BY 4.0 License.bias of ERA-Interim is 0.27 m s −1 .MERRA and NCEP-CFSR have values of 0.70 m s −1 and −0.62 m s −1 , respectively, compared with observations from the network of African Monsoon Multidisciplinary Analyses (AMMA) Population data was obtained from the Gridded Population of the World dataset, version 3 (GPWv3, http://sedac.ciesin.columbia.edu/gpw),which is supported by the Center for the International Earth Science Information Network and the Centro Internacional de Agricultural Tropical.GPWv3 provides spatial distributions of population density at global scale.The spatial resolution is 0.5°×0.5°and the time resolution is every 5 years from 1990 to 2010 (i.e., the population data are for theyears 1990, 1995, 2000, 2005, and     2010).
Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-890Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 1 December 2017 c Author(s) 2017.CC BY 4.0 License.and can obtain fire data and upper atmospheric source data, such as the appearance of the northern lights.Nighttime light products are derived from the average visible band digital number (DN) of cloud-free light detections multiplied by the percent frequency of light detection.Therefore, city lights can Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-890Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 1 December 2017 c Author(s) 2017.CC BY 4.0 License.CALIPSO launched on 28 April 2006 and combines an active Lidar instrument with passive infrared and visible images to delineate the vertical profiles and properties of aerosols and clouds at a global scale.This tool provides new insights into the influence of clouds and aerosols on Earth's weather, climate, and air quality Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-890Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 1 December 2017 c Author(s) 2017.CC BY 4.0 License.
Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-890Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 1 December 2017 c Author(s) 2017.CC BY 4.0 License.Huang et al. (2015) and Guan et al. (2016) suggested that anthropogenic dust has a close relationship with population density and the level of urbanization.Therefore, to calculate direct anthropogenic dust emissions, we used population density values, the CNLI, and the Engel Coefficient (EC), which are shown in Figure 5.The values for human population density are as high as 400 people km −2 in eastern China.Mumbai, located in northern India, has the largest population density, which were 29,650 people km −2 in 2010.
Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-890Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 1 December 2017 c Author(s) 2017.CC BY 4.0 License.these three high anthropogenic dust regions.The magnitude of anthropogenic dust in India is highest due to the main contributions of direct anthropogenic dust emissions; it is second highest in China and lowest in the United States.
Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-890Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 1 December 2017 c Author(s) 2017.CC BY 4.0 License.thing, various important factors are not considered in the direct anthropogenic dust scheme, such as the influence of city traffic, areas of urban roads, urban construction, urban development, and environmental policies.These factors will be considered in by developing more detailed direct anthropogenic dust emission schemes and constructing fugitive road dust emission inventories in our future study.For another, the indirect dust emission scheme only considered a few key factors that contribute to anthropogenic dust emissions in the paper.More complicated anthropogenic dust emission schemes, taking anthropogenic dust size distributions, soil moisture, chemical composition, etc into consideration, will be coupled with the Weather Research and Forecasting model with chemistry (WRF-Chem) model under constraints of satellite retrievals and ground observations.

Figure 1 .
Figure 1.Spatial distribution of land cover from Meiyappan and Jain (2012) and the percentages of C4 croplands and C4 pasturelands from 2007 to 2010.

Figure 10 .
Figure 10.Percentages of anthropogenic dust emission flux in croplands, pasturelands, and urban areas.