Changes in atmospheric aerosol loading retrieved from space-based measurements during the past decade

The role and potential management of short-lived atmospheric pollutants such as aerosols are currently a topic of scientific and public debates. Our limited knowledge of atmospheric aerosol and its influence on the Earth’s radiation balance has a significant impact on the accuracy and error of current predictions of future climate change. In the last few years, there have been several accounts of the changes in atmospheric aerosol derived from satellite observations, but no study considering the uncertainty caused by different/limited temporal sampling of polar-orbiting satellites and cloud disturbance in the trend estimates of cloud-free aerosol optical thickness (AOT). This study presents an approach to minimize the uncertainties by use of weighted least-squares regression and multiple satellite-derived AOTs from the spaceborn instruments, MODIS (onboard Terra from 2000 to 2009 and Aqua form 2003 to 2008), MISR (Terra from 2000 to 2010), and SeaWiFS (OrbView-2 from 1998 to 2007) and thereby provides more convincing trend estimates for atmospheric aerosols during the past decade. The AOT decreases over western Europe (i.e., by up to about −40 % from 2003 to 2008). In contrast, a statistically significant increase (about +34 % in the same period) over eastern China is observed and can be attributed to the increase in both industrial output and Asian desert dust.


Introduction
Anthropogenic aerosol from both fossil fuel combustion and land use change is now well known to impact on human health and global climate change.Detailed knowledge of long-term temporal changes of local, regional, and global aerosols is needed to test our scientific understanding of its sources and sinks and to provide an evi- Health Organization (WHO) 2012; Solomon et al., 2007;Climate and Clean Air Coalition (CCAC), 2012;Richter et al., 2005).As a result of remarkable advances in technology over the last decades, the observations from remote sensing instrumentation on Earth orbiting satellite platforms now provide novel and unique global information about atmospheric aerosols (e.g., Advanced Very High Resolution Radiometer (AVHRR), Total Ozone Mapping Spectrometer (TOMS), Along Track Scanning Radiometer (ATSR) Multi-angle Imaging SpectroRadiometer (MISR), Moderate Resolution Imaging Spectroradiometer (MODIS), and Sea-viewing Wide Field-of-view Sensor (SeaWiFS)) (Li et al., 2009;Thomas et al., 2010;Yu et al., 2009;Zhang and Reid, 2010;Karnieli et al., 2009;Mishchenko et al., 2007;Mishchenko and Geogdzhayev, 2007;Zhao et al., 2008;Massie et al., 2004;Yoon et al., 2011;Hsu et al., 2012;de Meij et al., 2012).In particular the aerosol products from the sensors, MODIS, MISR and SeaWiFS are often used because of their relatively long-term observation periods, stable and accurate sensor calibration, and the validated, quantified and high accuracies of the aerosol retrieved data products.
The unrepresentative sampling of polar-orbiting platforms in time and space is a limitation for the change and trend analysis as discussed elsewhere (Li et al., 2009;Ignatov et al., 2005;Kahn et al., 2007;Levy et al., 2009;Yoon, et al., 2011Yoon, et al., , 2012)), and will remain partially unresolved until an adequate "fit for purpose" measurement system is established, replacing the experimental series of satellite instrumentation currently in space.The issue arises because of the different sampling times (e.g., local equatorial crossing times are 10:30 a.m. for Terra, 12:20 p.m. for OrbView-2, and 01:30 p.m. for Aqua), limited orbital period (i.e., roughly 100 min for each orbit), and frequent cloud contamination or disturbance i.e. from the small and wispy sub pixel clouds, which are challenging to separate from aerosol.
In this study, the changes and trends of atmospheric aerosol have been determined from the retrieved AOT data from MODIS (Terra and Aqua), MISR (Terra), and SeaW-iFS (OrbView-2) during the past 15 years.We firstly minimized the uncertainty caused by the unrepresentative sampling using the relevant polar orbiting satellites observa-Figures
The selection criteria for data from the different sensors in the trend analysis include the duration of the observation periods (National Aeronautics and Space Administration (NASA), MODIS Web, 2000; Jet Propulsion Laboratory -NASA, MISR Multiangle Imaging SpectroRadiometer, 2000; Goddard Space Flight Center -NASA, SeaWiFS Project, 1997; The International Ocean-Colour Coordinating Group (IOCCG), 2011) and the stability and accuracy of the sensor calibration (Kahn et al., 2005a;Bruegge et al., 2007;Li et al., 2009;Barnes et al., 2001;Gordon, 1998;Eplee et al., 2001).

Regions and datasets
The average values of the AOT derived using the retrieval algorithms, developed for the Recently changes in aerosol, resulting from both direct emissions from fossil fuel combustion and the secondary aerosol created by photochemical transformations of trace gases, have been reported in these regions (Streets et al., 2003;Zhao et al., 2008).In addition, the aerosol, downwind from deserts, is influenced by wind-blown mineral dust (Zhao et al., 2008;Yoon et al., 2012).Figures

Back Close
Full The main objective of these instruments is to improve our understanding of global dynamics and processes occurring on land, in the oceans, and in the lower atmosphere.They achieve global coverage every one to two days.The MODIS instruments have been well calibrated (∼ 2 % absolute and ∼ 1 % precision) (Li et al., 2009) using on-board, vicarious, and lunar targets.Their data yield aerosol products having high accuracy (±0.05 or ±15 % over land and ±5 % over ocean for AOT) (Kaufman et al., 1997;Remer et al., 2005Remer et al., , 2008;;Levy et al., 2010) suitable for trend analysis.Their monthly AOTs at 550 nm and cloud fraction (CF) from Level 3 Collection 5 global products (1 are used in this study., 2007) using on-board, vicarious, and lunar targets.In this study, monthly AOTs at 558 nm (±0.05 or ±20 % over land and ocean) (Kahn et al., 2005b(Kahn et al., , 2010) ) from Level 3 Component Global Aerosol Product version F15 (CGAS-F15) products (0.5 • × 0.5
In this study, SeaWiFS Level 1B Global-Area Coverage data (L1B GAC) are used for AOT retrieval over selected regions using the Bremen AErosol Retrieval (BAER) algorithm (von Hoyningen-Huene et al., 2003, 2006, 2011).It has been demonstrated that the AOT error of BAER retrievals ranges within ±0.05 or ±20-25 % over land and ocean (von Hoyningen-Huene et al., 2011;Yoon et al., 2011).The trends of atmospheric aerosols are analysed in this study using monthly AOTs at 510 nm from Level 3 Global Product with 1 • × 1 • spatial resolution.

AERONET (AErosol RObotic NETwork)
The AERONET is a global network of ground-based instruments (i.e.sun photometers) for monitoring aerosol optical properties and validating the aerosol products retrieved from satellite borne measurements.It provides long-term records of cloud-free AOT (Remer et al., 1997;Dubovik et al., 2002) with high temporal resolution as well as high retrieval accuracy (Holben et al., 1998(Holben et al., , 2001;;Eck et al., 1999).In this study, Level 2.0 AERONET AOTs are used not only for validation of AOT trends retrieved from the satellites but also to test uncertainty or error caused by temporal sampling limitations.The reduction of the error and uncertainty in the trend analysis of cloud-free AOT, retrieved from measurements of the upwelling solar and thermal infrared spectrum by instruments on polar-orbiting satellite, is achieved by a variety of approaches.This includes optimization of instrument calibration and the refinement of retrieval algorithms (Zhao et al., 2008;Karnieli et al., 2009;Mishchenko et al., 2007;Mishchenko and Geogdzhayev, 2007;Massie et al., 2004;Hsu et al., 2012;de Meij et al., 2012).Thus far, no study has fully addressed the issue of the unrepresentative sampling, induced by different/limited temporal sampling of the instruments (Li et al., 2009;Yoon et al., 2011;Ignatov et al., 2005;Kahn et al., 2007;Levy et al., 2009).Adequate sampling is a prerequisite for deriving reliable and representative trends.An advantage of the polar-orbiting satellite data used in this study is their global coverage, but they are limited in the continuous temporal observations at a given location.
The temporal pattern correlations between different samplings (at 10:30 a.m. ± 30 min for Terra, 12:20 p.m. ± 30 min for OrbView-2, and 01:30 p.m. ± 30 min for Aqua, and all available samplings) were investigated using monthly AERONET AOTs at 550 nm, which are determined from the retrieved AOTs at 440 nm by using knowledge of the Ångström exponent for the 440-675 nm region.Since there is no difference in retrieval accuracy, cloud filtering method, and spatial resolution for each station, the differences in the temporal pattern correlations from this investigation are caused only by the different and limited sampling.The AERONET stations were selected using the criterion that the minimum temporal length of the data set is five years (see Yoon et al., 2012).
Figure 2 shows the Taylor diagrams, which describe three statistical metrics in one plot, viz.temporal correlation, normalized standard deviation, and normalized centred root-mean-square (RMS) difference of the two time series (Taylor, 2001;Solomon et al., 2007;Meehl et al., 2007), for the temporal pattern correlation between the monthly AERONET AOTs resampled at the local equatorial crossing times As there is no difference in retrieval accuracy, cloud-filtering method, and spatial resolution as mentioned before, this difference is attributed only to the different and limited sampling times, which are related to diurnal variation of the aerosol sources (Smirnov et al., 2002).In particular, the temporal correlation coefficient at the station Beijing ranges from 0.72 to 0.83, and there is good chance of deriving different trends from the different/limited samplings over such a large urban agglomeration.This knowledge is needed to understand the difference in the AOT trends from the different/limited temporal sampling of the satellite data.
To minimize the impact of different/limited sampling on the determination of the trend, the present study analyses global AOT trends derived from multiple polar orbiting satellites observations: Terra (MODIS and MISR), OrbView-2 (SeaWiFS), and Aqua (MODIS).

Weighted trend model for considering cloud disturbance
To derive an accurate and reliable change/trend analysis of atmospheric AOTs, the impact of sub scene cloud needs to be considered.The insufficient number of AOT retrieval induced by thin or wispy clouds in the instrument field of view causes a bias in the trend analysis since it is a significant influence on the statistical representative-Figures

Back Close
Full ness of monthly AOT means (Yoon et al., 2011).Generally, cloud occurrence causes the decrease of observation number (n t ).A relatively large spatial standard deviation compared to the mean (σ t /y t ) is a good indicator for cloud contamination in cloud-free AOT retrieval (Yoon et al., 2011).Therefore, the combination of these parameters as a weighting factor yields a minimum impact of cloud contamination on the change/trend analysis of cloud-free global AOT (Yoon et al., 2012), and is used in this study.The monthly AOTs (y t ) are used for fitting the linear regression where R 2 is minimized by where, wt t is the monthly weighting factor ( √ n t / σ t /y t ), t is time in months (t = 1. ..T ), A is a constant term, B is the magnitude of the trend per year (x t = t/12), n t is the total number of observations per month, y t is the monthly AOT measurement, σ t is the standard deviation of the monthly AOTs.y m represents the total mean of y t (i.e. the climatological monthly varying pattern in AOT) for each month (m = 1. ..12) and accounts for the seasonal/natural AOT cycle in the trend estimation.
Figure 3 shows the simple linear and weighted changes/trends, as well as total means and standard deviations of cloud fraction (CF), derived from MODIS (Terra) products (March 2001 ∼ December 2009).The simple linear and weighted trends shown in Fig. 3a and b, are generally consistent.However, in terms of the intensity and tendency, significant differences are found predominantly over regions having large CF variability in Fig. 3c, e.g.: (i) A clear spatial division of the AOT trends in Fig. 3a is found near coastal locations, i.e. across the land sea boundary, even though they have a common dominant aerosol source from the land.The pattern of increasing trends over the Indian subcontinent is often different across its land and ocean boundaries; (ii) There is no positive signal over Brazil even though it is one of the BRICs (a group of the countries: Brazil, Russia, India, and China having advanced economic development (Goldman Sachs, 2003)).In spite of efforts to diminish the impact of Figures

Back Close
Full (iii) The simple linear AOT trends over South Africa and Southeast Asia show a clear discontinuity between land and surrounding ocean areas.
These findings are indicative of either instrumental (e.g.platform characteristics, sensor calibration etc.), or retrieval (AOT retrieval accuracy), or sampling or atmospheric (changes resulting from human activity or natural phenomena) issues.As mentioned above, one important systematic source of error for trend analysis of cloud-free AOT is attributed to cloud disturbance.Thus the trend analysis is expected to be less robust over regions, where frequent cloud occurrence persists throughout the year as shown in Fig. 3c and d e.g., most of the marine areas and tropical rain/cloud forests in the equatorial zone.
For the problematic regions, mentioned above, the differences in AOT at coasts are reduced in the weighted trend compared to that of the simple linear trend.The discontinuities of the MODIS-Terra AOT trend between land and ocean over South Asia, Southeast Asia, South Africa has disappeared, and a positive trend is now found over South America in Fig. 3b.Over the majority of the ocean, a continuous cloud disturbance results in unrepresentative sampling all year around as shown in Fig. 3c and d, and the derivation of statistically significant AOT trends over oceans is limited by this issue.Consequently, for the regional trend analysis of cloud-free AOT, the present study focuses on those regions, which are not significantly influenced by cloud disturbance as selected and shown in Fig. 1.

Outlier tests of the weighting factors
To remove outliers from the weighting factors we use the Grubbs test (Grubbs, 1969) and a Gaussian test within 95 % of confidence levels.Firstly, the Grubbs test is used to detect outliers in weighting factors using the assumption that an approximately normal Introduction

Conclusions References
Tables Figures

Back Close
Full distribution is the most probable distribution.
where, N, µ wt , and σ wt are the total number, total mean, and total standard deviation of wt i , respectively.t 2 (α/N,N−2) denotes the critical value of the t-distribution with (N − 2) degrees of freedom and a significance level of (α/N).If the weighting factor satisfies Eq. ( 2), it is rejected as an outlier.After removing outliers by using the Grubbs test, the remaining weighting factors follow a Gaussian distribution.In a second step, the hypothesis about no outliers within approximately 95 % of confidence level is discarded if wt t satisfies the following Eq.( 3): Using these statistical tests and assumptions, outliers are successfully removed in the weighted trend analysis.Figure 4 illustrates an example of the outlier tests for significant weighting factors.

Significance test of the weighting trends
For a meaningful analysis of regional AOT trends, the present study takes into account the weighted trend (B g ) for each grid cell ( 1• ×1 • for MODIS and SeaWiFS or 0.5 • ×0.5 • for MISR) with significance (|B g /σ B g |) larger than two, from which it can be concluded that the trend is significant within 95 % confidence level (Weatherhead et al., 1998;Zhao et al., 2008;Yoon et al., 2011).The standard deviation of the gridded trend (σ B g ) is estimated using the bootstrap method (Mudelsee, 2010) (aka, Monte Carlo error bars analysis), using 5000 resampling iterations of monthly AOT anomalies for each grid.Introduction

Conclusions References
Tables Figures

Back Close
Full The aerosol retrievals from different cloud masking approaches in the various algorithms can lead to different monthly, seasonal, and annual behaviours, and thereby influence the trend analysis.To minimize the cloud uncertainty, a weighted trend approach is introduced in this study.In addition there is another important source of difference in trends arising from the different temporal sampling, as discussed above and demonstrated in Fig. 2.
The trend for each of the satellite AOT data products has been validated by comparing it with that available from the AERONET stations, listed in Table 3. Figure 5 shows the scatterplots and correlation analysis.The AERONET stations are selected by the criteria of having three years or more of the continuous observation.The three-years time span may be insufficiently short for the trend estimation, but it is practical for the trend validation over global area.Since the AERONET monthly AOT is calculated from all available sampling with high temporal resolution and accuracy (Holben et al., 1998(Holben et al., , 2001;;Eck et al., 1999), the AOT trends derived from satellite observations are expected to be different against the AERONET AOT trends at least due to different and limited sampling times as discussed in Fig. 2. In summary, from this comparison we can show how much different the satellite-derived trends are against the AERONET AOT trends, "actual trends".
In Fig. 5a and b, the AERONET AOT trend shows better correlation with the MODIS (Terra) trend than with the MISR trend.This is partly explained by MISR having a smaller swath width and spatial coverage, even though MODIS and MISR are onboard the same space platform, Terra.However MISR provides the only valuable trends over desert regions in this study.This is because the MISR algorithm retrieves AOT over these highly reflecting surfaces using its multiple-viewing observations (Kahn et al., 2007(Kahn et al., , 2010)).The weighted trends of the SeaWiFS/BAER AOT and that of the AERONET AOT are generally in good agreement having high correlation (R = 0.8) in Fig. 5c.However, the low slope of the SeaWiFS trends compared to that of AERONET Introduction

Conclusions References
Tables Figures

Back Close
Full  et al., 2011;Yoon et al., 2011).In contrast, the weighted trends of MODIS (Aqua) and AERONET AOTs have a higher slope of 0.7, but somewhat poorer correlation (R = 0.6) with AEONET, as shown in Fig. 5d.Overall and taking into account any biases introduced by their different sampling times and limited orbital periods, it can be concluded that the AOT trends derived from satellite-and ground-based observations are correlated (0.4 ≤ R ≤ 0.9).

Regional trend analysis
Figure 6 shows the significances of weighted trends.In this study, statistically significant trends within 95 % confidence level are used for the regional trend analysis.The significant trends determined using a linear-weighted regression for each of the data sets, shown in Fig. 7, are different depending on the instruments.One significant reason for these differences is attributed to the difference in temporal sampling of the instruments as discussed.
In Fig. 8 the AOT trends shown for the regions 1 to 10 are estimated from the data products from each of the individual instruments.In Western Europe, region 1 in Fig. 1, AOT from industry and traffic sources decreases significantly (Marmer et al., 2007;Karnieli et al., 2009).This is attributed to the success of environmental regulation in the EU countries (Streets et al., 2003;Yoon et al., 2011Yoon et al., , 2012;;Hilboll et al., 2013).Over Eastern Europe, region 2, large emissions of smoke aerosols from summer peat and forest fires and industrial pollutants from urban areas have been reported during the measurement period (Richter et al., 2005;Hayn et al., 2009;Yoon et al., 2011).
However overall, AOT also decreases in this region.This more complex behaviour is a result of societal changes initiated within Perestroika in the 1980's and the subse- quent political and economic practices in Eastern Europe in the 1990's.These then in turn produced changes in the vehicle fleet and industrial practices.Overall emissions have been reduced.However, a significant temporal increase of MODIS (Aqua) AOT is observed in the afternoon.In this context, positive trends of NO 2 emissions (Zhou et al., 2012;Streets et al., 2006) over Eastern Europe in the afternoon have been identified and similarly burned areas from Global Fire Emissions Database (GFED) (Giglio et al., 2010) over Eastern Europe, as shown in Fig. 9 for the period 2004 to 2008 are increasing.
The cities of the Near/Middle East, such as Cairo, having a combination of increasing population, intense fossil fuel combustion and poorly regulated vehicle emissions, exhibit some of the highest local air pollution levels known.In addition mineral dust from the Saharan and Arabian deserts, dependent the prevailing wind direction, is an aerosol type, on which pollutants deposit, over the Near/Middle East (Sabbah et al., 2006;Zhao et al., 2008;Yoon et al., 2012), region 3 in Fig. 1.Significant amounts of fine-mode aerosols, produced by the petroleum industry and related shipping, are also observed (Basart et al., 2009;Yoon et al., 2012).MISR AOTs are the only available data and thereby they provide more representative mean of significant trends over the bright desert areas in this study as shown from blue bar chart of significant trend pixels in Fig. 8.The trend in the Near/Middle East shows a significant increase.This is explained by an increase of coarse-mode aerosols from deserts (Yoon et al., 2012) and of fine-mode aerosols from oil production, refining and other industry in and around the Red Sea and the Persian Gulf (Sadrinasab and Kämpf, 2004).Emissions from fossil fuel combustion by shipping passing through the Suez Canal and Red Sea (Franke et al., 2009;Richter et al., 2004;de Ruyter de Wildt et al., 2012) also play a role.
The aerosol over the Indian sub-continent, as shown in region 4 in Fig. 1, is increasing.AOT is influenced by a variety of emission sources (Dey et al., 2004;Ramanathan et al., 2007a,b): fossil fuel combustion, domestic burning of biofuels, biomass burning, forest fires, mineral dust, and maritime aerosol. in India, where GDP increased (The World Bank Group, 2012) by ∼ 7.5 % annually from 1998 to 2010, coupled with the second largest and growing population of the world, which is around 1.2 billion, with nearly ∼ 1 billion people living in and around the Ganges valley, are factors contributing to the significant enhancement in the release of aerosol and its precursors.
Regions 5-7 in China show markedly increasing AOT.The Chinese economy is the second largest in the world (about $7.3 trillion of GDP in 2011 (The World Bank Group, 2012) and had annual growth rates (The World Bank Group, 2012) of ∼ 10 % or more over the past decade.Moreover, China is the world's most populous country, having a growing population of more than 1.35 billion people.As a consequence of the growth of industry, related construction and changes of land usage, large amounts of aerosol and their precursors are emitted into the atmosphere (Zhao et al., 2008;Yoon et al., 2012) in the conurbations of China.Additionally, mineral dust from the Asian deserts is transported by the predominantly westerly winds in spring and summer into this region, prior to its transport into the Pacific and the amount is increasing (Zhang et al., 2003;Yoon et al., 2012).The visibility in China is well known to be poor and AOTs over China during the period of observation have significantly increased (Streets et al., 2003(Streets et al., , 2006;;Mishchenko and Geogdzhayev, 2007;Zhao et al., 2008;Yoon et al., 2012).The AOT trend over region 8 (Korea and Japan), located in the same belt of westerly winds, shows consistently an increase, but slightly less pronounced than over China.This behaviour is attributed to a combination of emissions associated with increasing urbanization, coupled with a growing industrial production, and the increasing desertification and the resulting increase in dust being transported over Asia.
In the Western USA (region 9), the trend observed by MODIS (Aqua) from 2003 to 2008 shows an increase in apparent disagreement with the trends derived from the other instruments.Lower rainfall and resultant enhanced fire activity (see Fig. might be much more significant in MODIS (Aqua) AOT, which flies in a sun synchronous orbit with an early afternoon equator crossing time.
Over the Eastern USA (region 10), a decreasing AOT change/trend is generally observed.This behaviour is attributed in large part to the results of legislation and the subsequent measures introduced to reduce pollutant emissions, as reported in previous studies (Streets et al., 2003;Zhao et al., 2008;Yoon et al., 2012;Hilboll et al., 2013).The increasing trend of MODIS (Aqua) AOT in the afternoon is different to the other derived trends.However, it should be noted that the MODIS (Aqua) trends over Eastern USA are not significant in most areas, but only contributed from the significant trends over central USA as shown in Figs.6d, 7d, and 8.
To investigate further the remarkable behaviour of AOTs over East China (region 6), where the largest aerosol loadings are now observed, the time series of atmospheric AOTs normalized to total means, tropospheric nitrogen dioxide (NO 2 ) and sulphur dioxide (SO 2 ) columns from SCIAMACHY (Burrows et al., 1995;Richter et al., 2005;Bovensmann et al., 1999;Hilboll et al., 2013), and Chinese GDP (The World Bank Group, 2012) are compared in Fig. 10  As is now well established the atmospheric oxidation of SO 2 produces sulphuric acid, H 2 SO 4 .This has low volatility but is highly hygroscopic, and thus an ideal cloud condensation nucleus (Wallace and Hobbs, 2006).The oxidation of NO 2 produces nitric acid, which has a high solubility, and is absorbed by tropospheric aerosol (Pozzoli et al., 2008).The SO 2 and NO 2 seasonal cycles reflect changes in their atmospheric lifetimes, the amount and type of fossil fuel combustion, and seasonal advection patterns.SO 2 has a strong temporal correlation with the fine-mode aerosol loading in winter (i.e.Introduction

Conclusions References
Tables Figures

Back Close
Full correlation coefficient R = 0.6 between SCIAMACHY SO 2 and AERONET fine-mode dominant AOT at Beijing (Yoon et al., 2012) The maximum aerosol loading over China occurs in spring and early summer.This is the dry and windy season when mineral dust is transported by westerly winds from the Asian deserts.The MISR AOTs over the Asian deserts are strongly correlated to the other three independent AOTs over East Asia (i.e. the correlation ranges from 0.8 to 0.9, as shown in Fig. 12).A significant increase of atmospheric aerosol in spring is observed in all datasets, which varies from +0.94 to +2.53 % yr −1 .This is attributed to the growth of the Asian desert (Jeong et al., 2011) and thus desert dust, accompanied by reduced precipitation, approximately −5.62 % yr −1 estimated from NCEP/NCAR reanalysis data (NOAA Earth System Research Laboratory, 2012).These processes, which may be both in part a natural phenomena and in part induced by anthropogenic activity, result in an increase in desert dust aerosol, acting as a surface for deposition of pollutants, over the populated regions of China, Korea and Japan as well as the amount transported to the Pacific and beyond.mented in the developed world (Zhao et al., 2008;Yoon et al., 2011Yoon et al., , 2012;;Hsu et al., 2012;de Meij et al., 2012).This is an optimistic sign that mitigation of anthropogenic activity, when introduced, is having the intended consequence of reducing a threat to human health.In contrast, regions with uncontrolled pollutant emissions, associated with rapid economic growth and impacted by desertification have increasing trends in AOT.These changes are certainly able to influence the global climate change, and therefore the further studies are needed to investigate the consequence.
The recent severe smog in China during the winter 2012/2013 is a dramatic consequence of the emissions from economic growth, as shown by the increases in gross domestic product shown in Fig. 10, coupled with minimal environmental legislation.This trend is only mildly influenced by the recent global economic crisis, which began in 2008, and the cleaning up operation for the 2008 Olympics including introduction of desulfurization in coal fired power plants.How the atmospheric aerosol loading will evolve in a polluted and warming climate is not yet clear.Measurements of the type described in this study and better are required to assess objectively the changes.The generation of pioneering space-based remote sensing instrumentation is near to, or at its end of life.New and improved systems are required urgently to provide continuity and improved knowledge in the next phase of the anthropocene, uniquely enabling early warning of the direction and magnitude of these changes, and testing our understanding of the sources, sinks and processing of aerosol in the troposphere.The data products form such a constellation of instruments and data products simultaneously provide an evidence base for international environmental policymaking, addressing the key short-lived climate substances (CCAC, 2012), which impact on human health, agriculture, food security, ecosystem services and climate change.Introduction

Conclusions References
Tables Figures

Back Close
Full "ALL" (i.e.shown as purple dashed circles).The AERONET stations are classified by 7 regional dominant aerosol types: industrial/biomass burning, free tropospheric, desert, and 8 rural aerosols (Yoon et al., 2012), and listed in Table 2. 9 10 Fig. 2. Taylor diagrams for different sampling times of the different satellite instruments used in this study, all available sampling, (a) 10.30 ± 30 a.m., (b) 12.20 ± 30 p.m., and (c) 001.30 ± 30 p.m. using AERONET AOT (550 nm) data.The normalized centred root-meansquare (RMS) difference is proportional to the distance to the point on the x-axis identified as "ALL" (i.e.shown as purple dashed circles).The AERONET stations are classified by regional dominant aerosol types: industrial/biomass burning, free tropospheric, desert, and rural aerosols (Yoon et al., 2012), and listed in Table 2. Introduction

Conclusions References
Tables Figures

Back Close
Full dence base for policymakers (World Meteorological Organization (WMO), 2011; World Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | independent instruments MODIS (Terra) from March 2000 to December 2009, MISR (Terra) from March 2000 to December 2010, SeaWiFS (OrbView-2) from January 1998 to December 2007, and MODIS (Aqua) from January 2003 to December 2008, are shown in Fig. 1.Ten regions are selected for more detailed investigation in Europe, Middle/Near East, Asia, and North America and include large urban agglomerations.
Imaging Spectroradiometers (MODIS)(NASA, MODIS Web,  2000;  The International Ocean-Colour Coordinating Group (IOCCG), 2011) on the Terra and Aqua spacecrafts have contributed uniquely to our knowledge of atmospheric aerosols over the last decade.The first MODIS instrument is mounted on NASA Terra, which was launched on 18 December 1999.The second instrument, mounted on NASA Aqua spacecraft, started to observe global aerosols on 4 May 2002.
Terra, March 2000-December 2010) The Multiangle Imaging SpectroRadiometer (MISR) (Jet Propulsion Laboratory -NASA, MISR Multiangle Imaging SpectroRadiometer, 2000; IOCCG, 2011) instrument, one of the sensors on-board the Terra spacecraft, provides Earth viewings in the visible wavelength range simultaneously at nine widely spaced angles.This unique feature enables different types of atmospheric aerosols, clouds, and land surface covers to be distinguished.In addition, the instrument provides global coverage at high spatial resolution (i.e.275 m × 275 m, 275 m × 1.1 km, 1.1 km × 1.1 km), but it takes around nine days to cover the entire Earth surface.The sensor is carefully calibrated (∼ 3 % absolute, 1-2 % channel-to-channel relative, 1 % precision) (Kahn et al., 2005a; Discussion Paper | Discussion Paper | Discussion Paper | et al.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | slash and burn deforestation since 2001 in Brazil, significant amounts of aerosol and NO x are produced by this activity; Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 5 Trend validation with AERONET observation Discussion Paper | Discussion Paper | Discussion Paper |might be attributed to either an underestimation of SeaWiFS AOT (up to 20 % near heavily polluted areas as a result of absorbing aerosols)(von Hoyningen-Huene et al., 2011), an OrbView-2 orbital drift (about two hours delay till end of 2007)(Yoon et al., 2011), or the strict cloud-filtering method applied in BAER (von Hoyningen-Huene Discussion Paper | Discussion Paper | Discussion Paper | Significant AOT increases are observed in all the data sets during the observation period.The recent rapid economic growth Discussion Paper | Discussion Paper | Discussion Paper | 9), which occurred over parts of the Western USA from 2003 to 2008 (La Niña phases) (Westerling et al., 2006), qualitatively provides an expansion of the difference.As wildfires typically ignite in the afternoon (Mu et al., 2011), the increase, resulting from the fires, Discussion Paper | Discussion Paper | Discussion Paper | from 2003 to 2008.This period is chosen because all the data sets are available.In spite of having different temporal samplings and retrieval algorithms, the relative behaviour of the AOTs retrieved from the set of satellite instruments are all in reasonable agreement with each other and the ground-based observations, e.g. the AERONET AOTs measured in Beijing.The AOT trends over East China in spring, summer, autumn, winter are up to +2.53 %, +3.25 %, +3.26 % and +3.58 % yr −1 from 2003 to 2008, respectively.
Discussion Paper | Discussion Paper | Discussion Paper |

7
Conclusions and outlook By using a new trend model (i.e.weighted least squares regression) and optimally different measurements (MODIS-Terra, MISR-Terra, SeaWiFS-OrbView-2, and MODIS-Aqua) from 1998 to 2010, we have established a benchmark for the rate of change of AOT in selected regions.The uncertainty in the trend analysis induced from the cloud disturbance and different/limited temporal sampling has been discussed for the first time, and successfully minimized in this study.The dramatic increases in AOT, associated with rapid industrial growth and desertification are clearly identified.The positive impact of legislation in reducing AOT and improving air quality is unambiguously docu-Discussion Paper | Discussion Paper | Discussion Paper |

Fig. 11 .
Fig. 11.Plot of the correlation between AERONET fine-mode dominant AOT at Beijing and SCIAMACHY tropospheric SO 2 in winter seasons from 2003 to 2008.Black and red error bars show natural variability and standard retrieval error within 95 % confidence level, respectively.
, see Fig.11, which shows how well they are correlated), when mineral dust from Asian deserts over East China is low.Winter trends of +22.41 % yr −1 for fine-mode dominant aerosols in Beijing and +9.32 % yr