AOD trends during 2001–2010 from observations and model simulations

. The aerosol Optical Depth (AOD) trend between 2001–2010 is estimated globally and regionally from observations and results from simulations with the EMAC (ECHAM5/MESSy Atmospheric Chemistry) model. Although interannual variability is apply only to the anthro- 5 pogenic and biomass burning emissions, the model is able to reproduce quantitatively the AOD trends as observed by MODIS satellite sensor, while some discrepancies are found when compared to MISR and SeaWIFS observations. Thanks to an additional simulation without any change in the emis- 10 sions, it is shown that decreasing AOD trends over the US and Europe are due to decrease in the emissions, while over the Sahara Desert and the Middle East region the meteorological changes do play a major role. Over South East Asia, both meteorology and emissions changes are equally impor- 15 tant in deﬁning AOD trends. Additionally, decomposing the regional AOD trends into individual aerosol components reveals that the soluble components are the most dominant con-tributers to the total AOD, as


Introduction
The atmosphere is a mixture of various gases and aerosols. The increase of greenhouse gases causing climate change is partially countered or enhanced by aerosol radiation modifications (the aerosol direct effect; Andreae et al., 2005). Ad-25 ditionally, aerosols can modify cloud properties (indirect effects; Ramanathan et al., 2001a, b;Kaufman et al., 2002).
Furthermore, depending on their composition aerosols affect the ecosystems, quality of life (cardiovascular and respiratory diseases; Lelieveld et al., 2013) and visibility. Since the 30 late 1980s the decline in solar radiation at the Earth's surface due to aerosol pollution (dimming) has reversed over the Northern Hemisphere (Wild et al., 2005;Wild, 2010). This change from dimming to brightening has important consequences for climate change, affecting the hydrological 35 cycle, cloud formation processes and surface temperatures, possibly intensifying the warming trend caused by carbon dioxide (CO 2 ) and other greenhouse gases. Remote sensing instruments and Atmospheric Chemistry Transport Models (ACTMs) provide possibilities for improved qualitative 40 and quantitative analysis of the global burden of atmospheric trace gases and aerosol particles. ACTMs and chemistry climate models (CCMs) are additionally used to assess the effects of future changes in aerosol (+ precursor) emissions on climate by making use of physical and chemical process 45 descriptions in dependency of emission inventories. These emission inventories are constructed from estimates for population and economic growth to determine present gas phase and aerosol emissions as well as future emission scenarios. The emission inventories include natural and anthropogenic 50 emissions relevant for aerosol formation, such as sulphur dioxide (SO 2 ), nitrogen oxides (NO x ), black (BC) and organic (OC) carbon, ammonia (NH 3 ) and many more compounds.
SO 2 and BC are mainly emitted from fossil fuel (coal and 55 petroleum) combustion. Primary sources for NO x are road transport and fossil fuel combustion for energy production, and NH 3 is mainly emitted from agricultural activities (waste burning and fertilizers) and to a small extent by the combustion of biofuels for energy use. The amount of NH 3 emitted by agricultural activities is related to the type of fertilizer, meteorological conditions (wet/dry) and soil properties. Since 1990 in Europe and in North America the emissions of aerosol precursors have dropped in response to the implementation of air quality legislation (Clean Air Act Amendpollution on the incoming solar radiation at the Earths surface (dimming and brightening phenomena), such as Wild et al. (2005), Wild (2010, special issue J. Geophys. Res., and references therein), Pinker et al. (2005); Hinkelman et al. (2009) ;Mishchenko and Geogdzhayev (2007); Remer et al. 95 (2008); Chylek et al. (2007); Lu et al. (2010); Zhang and Reid (2010) and Kishcha et al. (2009). These studies have shown elevated AODs (decreasing incoming solar radiation at Earths surface) over India, East Asia, Bay of Bengal, Arabian Sea and reduced AODs (increasing incoming solar radi-100 ation) over North America and Europe.
A recent work of Hsu et al. (2012) with the SeaWIFS instrument has shown high precision in trend derivation while reporting decreasing trends over the eastern USA and Europe and increasing trends over China and India. Hsu et al. 105 (2012) additionally have investigated the impacts of other uncertainty factors in trend estimates, e.g. retrieval algorithm deficiency and sampling bias, by comparing their results with AERONET and MODIS-Terra products. In particular the correlation analysis between large-scale meteorolog- 110 ical events (such as El Niño Southern Osciallation and North Atlantic Oscillation) and SeaWiFS-retrieved AOD indicated strong influences of the climatic indices on Saharan dust outflow and biomass-burning activity in the tropics.
In de Meij et al. (2012a)  The ECHAM/MESSy Atmospheric Chemistry (EMAC) model is a numerical chemistry and climate simulation system that includes sub-models describing tropospheric and middle atmosphere processes and their interaction with oceans, land and human influences . It uses the second version of the Modular Earth Submodel System (MESSy2) to link multi-institutional computer codes. The core atmospheric model is the 5th generation European Centre Hamburg general circulation model (ECHAM5, Roeckner et al., 2006). For the present study we use ECHAM5 version 5.3.02 and MESSy version 2.42. The EMAC model has been extensively evaluated for gas 180 tracers (e.g. Pozzer et al., 2007) and for aerosols (e.g. Pringle et al., 2010a;Pozzer et al., 2012;Astitha et al., 2012). The modeled AOD is calculated at 550 nm using concentrations of dust and sea salt particles, biomass burning products (black carbon and organic carbon) and anthropogenic 185 aerosols (sulphates, nitrates, etc). The aerosol optical properties are calculated with the EMAC submodel AEROPT, which is based on the scheme by Lauer et al. (2007) and makes use of predefined lognormal modes (i.e. the mode width σ and the mode mean radius have to be taken into 190 account). Lookup tables with the extinction coefficient, the single scattering albedo and the asymmetry factor for the shortwave and extinction coefficient for the longwave part of the spectrum are pre-calculated with explicit radiative transfer calculations (see Pozzer et al., 2012 (Fountoukis and Nenes, 2007), which is based on a ZSR relation (Stokes and Robinson, 1966) of the simulated aerosol compounds. Furthermore, additional compounds such as organic carbon further contribute to aerosol water following the κ-approach (Petters and Krei-205 denweis, 2007). Aerosol water uptake is limited by a relative humidty of 95% of grid box mean relative humidity. The extinction, single scattering albedo and asymmetry factor are determined with the help of the volume weighted complex refraction index of the particles assuming an internal mixture 210 of the individual components. This assumption partially influences the absolute values of the calculated AOD (see e.g. Klingmüller et al., 2014, and references therein). Potential impacts can be amplified absorption by soot due to multiple reflection as seen from a detailed core-shell treatment. How-215 ever, as all simulations use the same assumption of internal mixing potential trends in AOD should be affected only to a minor degree. Previous studies using the EMAC model have proven that the simulated AOD is able to capture the overall global pat-220 tern although a general underestimation is present (de Meij et al., 2012b;Pozzer et al., 2012). The seasonal cycle of AOD is in general well represented in the EMAC model simulations in most parts of the globe, especially over dust influenced regions. In heavily anthropogenic polluted regions, the 225 modeled AOD is slightly overestimating the observations.
In this study, the Chemistry Climate Model (CCM) EMAC has been used with a T63L31 resolution, corresponding to a horizontal resolution of ≈ 1.875 • × 1.875 • of the quadratic Gaussian grid, and with 31 level vertical levels up to 10 hPa 230 in the lower stratosphere. The model set-up used in this work was presented by Pozzer et al. (2012), and here only the differences and the central features are described.
The atmospheric chemistry is simulated with the MECCA (Module Efficiently Calculating the Chemistry of the At-235 mosphere) submodel by Sander et al. (2005Sander et al. ( , 2011, while the aerosol microphysics and gas-aerosol partitioning are calculated by the Global Modal-aerosol eXtension (GMXe) aerosol module (Pringle et al., 2010a, b). For descriptions of the emission and deposition routines we refer to Kerkweg  Pozzer et al. (2006) and Tost et al. (2007).
As in Pozzer et al. (2012) and Pringle et al. (2010a), both dust and sea-salt emissions are offline prescribed using offline monthly emission files based on AEROCOM and do not depend on the model meteorology, hence having no 245 multi-annual variability. However, the desert dust and sea salt aerosol concentrations that influence the AOD calculation exhibit multi-annual variability as the model description of the aerosol microphysical process (coagulation, condensation, ageing) as well as transport (advection, convection) and 250 deposition processes are subject to meteorological variability.
The biomass burning contribution was added using the Global Fire Emissions Database (GFED version 3, van der Werf et al., 2010) with a monthly temporal resolution.

255
In this work we used the emissions scenarios recently developed for the ACCMIP (Atmospheric Chemistry and Climate Model Intercomparison Project, www.giss.nasa.gov/ projects/accmip/) initiative, which focuses on emissions scenarios based on the Representative Concentration Pathways 260 (RCP) Meinshausen et al., 2011;van Vuuren et al., 2011b, a, and references therein). The RCPs consist of four emission scenarios, also called RCP 2.6, 4.5, 6.0, and 8.5 representing the radiative forcing of anthropogenic activity from 2.6 to 8.5 W m −2 in 2100, which de-265 pend on the mitigation or emission scenarios. Among them, the emission scenario RCP 8.5 is used in this study as Granier et al. (2011) showed that it is a "reasonable" choice for anthropogenic emissions after the year 2000 and for the recent past. Additionally, to compare the different simulations, the chemistry and the dynamics have been decoupled, such that there is no direct interaction and feedback between the atmospheric composition of the gas phase and aerosol particles with the dynamics of the atmosphere.

285
Accordingly, both simulations follow the same (i.e. binary identical) dynamics and meteorology, i.e. the CCM is used as a chemistry-transport model. The model dynamic has been weakly nudged (Jeuken et al., 1996;Jöckel et al., 2006) towards the analysis data of the European Centre for Medium-

290
Range Weather Forecasts (ECMWF) operational model to represent the actual day-to-day meteorology in the troposphere, which allows a direct comparison of the simulations results with observations.
Although the simulations cover the period 2000-2010, the 295 first year is used as spinup time, and the results of this work are based on 10 years of data (2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010). Additionally, the submodel SORBIT ) was used to sample the AOD at the correct local time of the satellite overpass. Hence, we can neglect any influence of the diurnal cycle in 300 the comparisons performed in the next sections.

Remote sensing data
In this work, observations from three different satellite sensors have been analyzed independently. Specifically, the following satellite datasets have been used:

305
-The Multi-angle Imaging SpectroRadiometer (MISR) instrument (Diner et al., 1998) is located onboard the Terra satellite and has been operational since February 2000. The instrument is designed to measure the solar radiation reflected by the Earth-Atmosphere sys- are averaged on a monthly basis and stored on a geographic grid of 0.5 • × 0.5 • . The validation of MISR AODs over land and ocean with AERONET (AErosol RObotic NETwork) data has shown that MISR retrievals are within ±0.05 or ±20% of that of AERONET (Kahn 330 et al., 2005(Kahn 330 et al., , 2010. -The MODerate resolution Imaging Spectroradiometer (MODIS) sensor is also located on the Terra satellite. In contrast to MISR, the MODIS instrument has only one NADIR looking camera which measures radiances in 36 335 spectral bands, from 0.4µm -14.5µm, with spatial resolutions of 250m (bands 1 -2), 500m (bands 3 -7) and 1000m (bands 8 -36). Daily level 2 (MOD04) aerosol optical thickness data are produced at the spatial resolution of 10 × 10km over land, aggregated from the orig-340 inal 1km × 1km pixel size. As the swath width is about 2330km, the instrument has almost a daily global coverage. MODIS aerosol products are provided over land  and water surfaces  with uncertainties being ±0.05 ± 0.15× AOD 345 (Chu et al., 2002;Remer et al., 2008;Levy et al., 2010) and ±0.03±0.05× AOD (Remer et al., 2002(Remer et al., , 2005, respectively. In this paper, AOD550 data from the MODIS Level 3 (Col. 051) gridded product are used at a spatial resolution of 1 • × 1 • . In this work the Deep Blue algo-350 rithm products was not used, as does not present any coverage after the year 2007 in this MODIS version. Nevertheless, this algorithm allows retrieval from bright surfaces and therefore could allow a full coverage also over desert area (Sayer et al., 2013).

355
-The Sea-WIFS (Sea-Viewing Wide Field-of-View Sensor) instrument operated on board GeoEye's OrbView-2 (AKA SeaStar) satellite providing data from September 1997 to December 2010. AODs at different wavelengths have been retrieved over land with the use of the 360 Deep Blue algorithm over land (Hsu et al., 2004(Hsu et al., , 2006 and ocean using the SeaWiFS Ocean Aerosol Retrieval (SOAR) algorithm (Sayer et al., 2012a) at a horizontal resolution of ∼ 13.5 km. Here, SeaWiFS v004 Level 3 AOD550 gridded data are used at a spatial resolution 365 of 1 • × 1 • following (Hsu et al., 2012). Validation studies of Sea-WIFS's AOD550 with ground-based observations from AERONET and data from other satellite sensors indicate an absolute expected error of ±0.03±15 % over ocean and ±0.05 + ±20 % over land at 550 nm 370 (Sayer et al., 2012a, b).
In addition to the satellite observations, AOD from station observations are also used in this work. The AOD observations have been obtained from the global AErosol RObotic NETwork ( AERONET Holben et al., 1998;Dubovik et al., 375 2000). The solar extinction measurements are used to calculate the aerosol optical depth with an accuracy of about ±0.01 AOD units (Eck et al., 1999). The cloud-screened quality-assured Level 2 AOD data used in this work were obtained from the website http://aeronet.gsfc.nasa.gov/cgi-bin/ 380 combined_data_access_new, and contain AOD daily averages. Furthermore, the AOD at 550 nm was calculated from the AOD values reported at 870 and 440 nm by using the information of the Ångström exponent (see Eqs. 1 and 2, de Meij et al., 2012b).

Model evaluation and AERONET observations comparison
The model AOD product was extensively evaluated already in a number of publications (Pringle et al., 2010a;de Meij et al., 2012b;Pozzer et al., 2012;Astitha et al., 2012). Impor-390 tant for this work is the study of Pringle et al. (2010b), which used the same water uptake coefficients used here. Pringle et al. (2010b) has shown that there is both qualitatively and quantitatively good agreement of aerosol hygroscopicity (and hence water uptake) with measurement campaigns all 395 over the world. Pringle et al. (2010b) included comparison of the modeled values with surface concentrations and vertical profiles (available from several campaigns). This agreement provides us with confidence that the representation of water uptake of the aerosol particles in the model is sufficiently 400 credible to draw conclusions from temporal trends.
In this section we therefore compare briefly the model with AERONET observations, to confirm the previous findings. Additionally, in this work the AERONET data is purely used for evaluation of the model performance, but no AOD trends 405 will be estimated and compared with the model as this was already performed in de Meij et al. (2012b).
As shown in Fig. 1, the model reproduces the AOD observed by AERONET within a factor of two in 73.5% and 72.9% of the cases for BASE and FIXEMI, respectively.

410
When the observed AODs are compared to the simulated AODs, it is interesting to notice that both simulations show similar linear fit: the correlation coefficients (0.57 and 0.52 for BASE and FIXEMI, respectively) and the bias (0.02 and 0.03 for BASE and FIXEMI, respectively) are very similar 415 while the slope is almost equal to 1 for both simulations. This implies that the average global observed and calculated AODs do not vary significantly during 2001 and 2010, which corroborates the study by Chin et al. (2014). It also indicates that both simulations are in general able to reproduce 420 the overall maxima/minima, although BASE shows slightly better agreement for correlation and slope of the linear fit. However, large differences are found in the AOD trends on a regional scale between BASE and FIXEMI and this will be further discussed in the next sections.  (Weatherhead et al., 1998) where Y t , µ, ω and X t denote the monthly time series, the offset, the trend (i.e. AOD yr −1 ), and the years of the time series (X t = t/12 with t as month), respectively. S t is a seasonal component representing the effect of the seasonal variations in the 435 trends estimates, while N t is the residual term of the interpolation. The seasonal component we have taken into account is based on the Fourier series as proposed by Weatherhead et al. (1998Weatherhead et al. ( , 2002 with: S t = 4 j=1 [β 1,j sin(2πjt/12) + β 2,j cos(2πjt/12)]. A statistically significant trend at 95 % 440 confidence level is defined by the absolute value of the ratio of the trend to its standard deviation which is larger than two (Tiao et al., 1990;Weatherhead et al., 1998), with the standard deviation of the trend estimated again with the approach of Weatherhead et al. (1998, Eq. 2). Finally, no trends 445 was calculated for location with less than 6 data points for each year between 2001 and 2010 (grey aread in Fig. 2).
It is clear that trends from monthly averages level 3 data should be taken with cautions, as the monthly averages are calculated independently of the representativeness of the data 450 sampled. The sampling of actual retrievals is highly nonuniform in space and time, even at the resolution of the products used in this work (Kahn et al., 2009) Therefore, especially in the case of location with very high frequency of cloud covered sky, the monthly averages could be calculated  Deep Blue dataset and this product is not used in this work.
In Fig. 3 the trends are estimated for both simulations (FIXEMI and BASE), using the same method as for the remote sensed observations. For both model simulations AOD data is sampled at the same co-located time of the observa-480 tions, thanks to the submodel SORBIT (Sample along satellite ORBIT, Jöckel et al., 2010), following the trajectories of the platform TERRA (on which MODIS and MISR instruments are on board) and OrbView-2 (with the SeaWIFS instrument on board).

485
The model present some discrepancies on developing region and over regions strongly influenced by biomass burning. On the other side, the sign of the trends is correctly reproduced over Norths America and Europe. The tendencies over North America and East Europe are consistent with the 490 results of various studies based on trend analysis of satelliteretrieved AOD (e.g. Zhang and Reid, 2010;Hsu et al., 2012;Yoon et al., 2014). Since the satellite-retrieved AOD data, which have been used in most of the trend analyses, are only available under cloud-free conditions, the trend esti-495 mates based on the cloud-free AOD products can be biased due to insufficient retrievals in cloudy seasons (Yoon et al., 2014). It must also be underlined that the simulation data is not screened for cloudy conditions, as in the satellite, but represents aerosol optial properties at all times. However, due to 500 the relative humidity limited aerosol water uptake, a large overestimation of AOD due to the high water available for aerosol growth in cloudy conditions is not expected. Nevertheless, the cloudy conditions contribute to AOD values at the upper edge of the extinction probability density function.

505
In the model results a significant increase of AOD both in simulations BASE and FIXEMI can be observed also over the Saudi Arabian peninsula, confirmed by the observations. In addition, a strong significant increase during the decade 2001-2010 is observed over East Asia, which is however not 510 fully corroborated by the observations. The strong decreasing trend over the Tropical/Southern Africa is not well reproduced by the model. The model results from simulation data FIXEMI (see Fig. 3, left plots) do not show significant trends both in Eu-515 rope and Northern USA, while they still do present significant trends both in Northern Africa (Sahara Desert) and South East Asia.
Discrepancies between simulation BASE and MODIS observations are clearly observed in Central Africa and South 520 America. Nevertheless this is not corroborated by other satellite observations. On contrary, both MISR and SeaWiFS present in this region not enough observations to estimate a trend. This points out the large effect that clouds have on these regions and the possible influence on the remote sensed 525 data. Therefore these regions will not be analyzed in detail, as observational datasets could be contaminated (Yoon et al., 2012).

AOD regional trends
The global analysis of the AOD trends discussed in the previ-  Fig. 4). In addition, three more regions have been included for completeness, i.e. the North Hemisphere (NH), South Hemisphere (SH) and the whole Globe (GL).
The regional trend analysis is illustrated with scatter 540 plots that show the comparison between the model and the satellite AOD trends (Fig. 5), each point representing the regional mean trend for the respective regions. Additionally Focusing on the target regions, over Western Europe (WE), the AOD trends are negative by the model simulations and the satellite retrievals, with BASE being closer to the 565 satellite-based trend. This is related to the influence of the  anthropogenic emissions regulation imposed in Western Europe that resulted in a decrease of the atmospheric aerosol load during the decade 2001-2010 (Vestreng et al., 2007(Vestreng et al., , 2009. The same applies to Eastern US (EUS) with results 570 by simulation BASE and observed AOD having a negative trend. Nevertheless, the model overestimates the negative trend by MISR and SeaWIFS and correlates better with the MODIS retrievals.
A common pattern appears in the AOD trends for the cli- simulations and satellite data. The positive trends calculated from the observations (the three satellite products agree) are somewhat higher than predicted by the model, which shows a slightly positive trends in both simulations (BASE and FIX-580 EMI). This means that the variation in the anthropogenic and biomass burning emissions did not affect the AOD trend in this region, (see FIXEMI results) which is mostly affected by desert dust emissions and the dynamic factors that con-trol dust transport and deposition. Moreover, the dust emis-585 sions were prescribed off-line in the model, i.e. independent on the wind fields and therefore independent on the meteorological conditions and without any interannual variability. Therefore, the positive trends in this region can be attributed to the decrease in the precipitation and the consequent reduc-  The latest is confirmed by the precipitation trends calculated from the model results, which is shown in Fig. 6.

595
Clearly the region over the North Africa present a negative precipitation trends, which influences the dust deposition and its transport. Further confirmation is by the Global Precipitation Climatology Project (GPCP), which shows a strong decrease in precipitation over the Middle East during the period The same behavior is found for the Sahara Desert (SD) region where the AOD trends remain unchanged between the two model simulations (i.e. implying a meteorological 610 driven trend), which can be attributed to the decrease precipitation over North Africa (see Fig.6). The trends observed by the satellites exhibit a small variation (for MODIS is slightly negative; for MISR and SeaWIFS is slightly positive) which can be attributed to the different algorithm assump-615 tions (Kahn et al., , 2009, calibration methods, and differences in the aerosol models used to construct the lookup tables in the retrieval algorithms (Abdou et al., 2005). Nevertheless, the difference in the trends is very small considering the high total AOD values in this region.

620
For South Asia (SA) and Eastern China (EC), the improved temporal correlation coefficients by the BASE simulations with the satellite trends shows the importance of the variability in the emissions compared to the constant emission assumption. The sign of the trend changes from neg-625 ative (FIXEMI) to positive (BASE) in the two simulations, which is due to the increase emissions of the fast developing countries like China and India. The high positive trends mod-eled by simulation BASE over EC deviate from the satellite trends and this can be explained by the influence of desert 630 dust aerosols that is not accurately represented by the model simulations. In fact, the MODIS trends for the entire EC region are slightly negative whereas the sign is positive by both MISR and SeaWIFS (see Fig. 5). This could be attributed to uncertainties in the AOD retrieval, e.g. sensor calibration 635 status, retrieval accuracy, and cloud contamination. In addition, insufficient sampling not reflecting actual data population due to different sampling times, limited orbital periods, and cloud occurrence, which can be serious over polluted and cloudy areas, could be another reason for differ-640 ences between the trends estimated from satellite observational datasets (see Yoon et al., 2012Yoon et al., , 2014. In South East Asia (SEA) the model (sampled on the TERRA platform overpass) calculated a statistically significant negative trend in FIXEMI (−1.77 ± 0.6 % yr −1 ) caused 645 by changes in the meteorological conditions. Interestingly, this trend is enhanced in BASE (−3.03 ± 1.65 % yr −1 , not significant) due the decrease of biomass burning emissions (Yoon and Pozzer, 2014) during this decade in the region. Therefore, AOD trend in this region seems to be caused both 650 by meteorological and emissions changes during the decade. Finally, the model present a significant negative trend in the in the North Hemisphere (NH) and globally (GL). These trends are also calculated for MODIS and MISR instruments for the same regions, although they results non significant.

655
Interesting in simulation FIXEMI, no significant trend is detected for the NH, which is emissions driven, while a significant decrease is estimated for the Souther Hemisphere (SH) and globally (GL).

Effects of aerosol components in modeled AOD trends 660
In this section the main causes of positive/negative trends in the regions of interest are analyzed. The trends for each of the regions defined in Sect. 5.2 are decomposed by estimating the AOD trends for the different aerosol components (see Fig. 7 and Table 1 Additionally, in this section, the full model output was used, so that the results are based on the full 10 years model results. In general, aerosol water content has the largest contribu-670 tion to the total AOD with the notable exception of desert area where dust dominates and therefore must be considered as the most effective extinction component in the aerosol polluted atmosphere (Gao, 1996). de Meij et al. (2012b found that over Europe and North America the associated aerosol 675 water contributes around 40-45% to the total AOD. In the Eastern part of the US and in the Western part of Europe, a decrease in the WASO components (i.e. ammonium, nitrate and sulphate) is found during the decade 2001-2010. Additionally, since the aerosol water uptake efficiency  and consequently the aerosol water content is decreased by the decrease presence of the WASO components, the negative AOD trends are further amplified. This highlights the fact that the observed AOD trends depend on the reduction in the emissions (van Vuuren et al., 2007;Vestreng et al., 2007, 685 2009), but also on the nature of the aerosols. The WASO-H 2 O relationship was analyzed by de Meij et al. (2012b), both for the EMAC model and the observations for different regions. They showed that, in Europe, the WASO aerosol contributes about 37% to the yearly mean and the associated 690 aerosol water 45%, respectively. In North America the inorganic part contributes 30 to 40% to the total AOD and the associated aerosol water 40%. Over SD and the ME the positive AOD trends are mainly caused by an increase of the natural dust. The WASO compo-695 nents reduce the total AOD trends in these regions due to the lower anthropogenic aerosol transport from Europe into this region. This results in a lower aerosol water content which also reduces the total AOD trends. The increase of the natural dust component to the total AOD does not only compensate 700 for this reduction but also exceeds it, causing, as mentioned before, an overall positive trend over the desert-covered regions. The decrease of WASO components in these regions, potentially increases the aerosol lifetime due to the missing coating effect (e.g. sulphate particles over dust) which makes the particles less susceptible for wet removal. This effect plays only a minor role here, as the AOD trends in the FIXEMI and BASE simulations are similar, implying that the trends are strongly dominated by meteorological factors, i.e. the decrease in precipitation as described in Sect.5.2 and 710 shown in Fig. 6.
The simulation with the BASE scenario for SA and EC (see Fig. 7) reveals a reverse situation. The highly soluble WASO components and the water content exhibit a positive AOD trend while the dust component exhibits a negative 715 trend with the overall AOD trend being positive in these two regions. The trends of the WASO components are very similar for the two regions while the aerosol water content trend over East China is nearly three times that of South Asia. This is due to the higher relative humidity over East China with 720 respect to South Asia (∼ 73% and ∼ 59%, respectively, for the 2001-2010 average) which causes a more effective water uptake due to the exponential relationship between water uptake and relative humidity.
Finally, over SEA, the negative AOD trends are due to 725 the decrease of all aerosol components (see Table 1). Dust and sea salt, however, show a significant decrease, implying trends driven by meteorological conditions. In particular, this trend is enhanced by the decrease of the aerosol water content (also significant), which is mostly due to the decrease of the highly soluble sea salt. Nevertheless, the BC/OC decrease (due to biomass burning decrease in the region during the 2001-2010 decade) enhances the negative trends by ≈ 0.4 × 10 −3 yr −1 . Therefore the total trend in this region (≈ 2.75 × 10 −3 yr −1 ) is a combination of a meteorological 735 effect and a decrease of the biomass burning emissions. However, it is unclear what meteorological parameter is driving such trend in the region. In fact, the only significant detectable trends in the simulation is a decrease of wind speed at the surface (see doi:10.5194/acp-0-1-2015-supplement), 740 which could possibly decreases the transport of Sea Salt from the open ocean to this region and consequently the water contribution to the AOD, being Sea Salt highly hydrophilic. Finally, the NH presents a decreasing significant trend (−0.798±0.247 10 −3 yr −1 ) 70% of which caused by the water uptake decrease. Interesting, on a global scale (GL, see Fig. 7 and Table 1) the same proportion is kept, with water contributing by 73% to the total (significant) AOD decrease.

Conclusions
In this work, the AOD was simulated for a period of 10 years 750 (2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010) with the EMAC Chemistry General Circulation Model. AODs from AERONET stations were used to evaluate the model results prior to the calculation of the global and regional AOD trends. Satellite retrievals from MODIS, Sea-WiFS and MISR have subsequently been used to estimate 755 the AOD trends and compare them with the simulated results. The trends in the aerosol extinction are qualitatively reproduced over North America, East Europe North Africa and Middle East, while some discrepancies are found over other regions. Seven regions of interest are selected to con-760 duct a regional analysis, based on the strength of the signal from both model and satellite AOD trends.
The main objective of this work is to identify the causes of the decadal AOD trends for the designated regions of interest by decomposing the AOD trends into its aerosol components 765 trends. Therefore, two simulations have been performed to address the research objective; one with changing and one with constant emissions, using identical same atmospheric dynamics. The differences between the two simulations show that the observed AOD increases in the model over Middle

770
East and North Africa are due to meteorological effects as a strong increase in the dust component of the AOD is found in both simulations for those regions. However, these results are obtained without any online dust emissions calculation, but rather with fixed prescribed monthly emissions. Therefore 775 this AOD increase could be enhance or dampened if interactive dust emissions were calculated. Over Eastern US and Western Europe, significant decreasing trends are deduced from the model simulation only when realistic, decreasing emissions in the decade 2001-2010 are included. This in-780 dicates the strong influence of anthropogenic emissions on aerosol load and the related AOD. Consequently, it can be confirmed that in Eastern US and Western Europe the AOD decrease is purely driven by emissions reduction policies.
For South Asia and East China the AOD trend is positive 785 both from model results and satellite observations. The differences between the two model simulations identify that the AOD is increasing due to a combination of changes in the anthropogenic emissions and the meteorological conditions in the denoted areas. Finally, for South-East Asia, the de-790 creasing trends are due to the decrease in the biomass burning emissions and the meteorological conditions, where the purely meteorological trend contributes approximately 50 % of the total AOD trend. The role of natural aerosols (i.e. desert dust) has proven to be significant for Middle East and 795 North Africa and non-negligible for China and South Asia. Future work in this area would include the online production of dust and sea salt emissions during the simulation time in order to identify the effects of meteorology in the dust component AOD by introducing emission variability for both sea 800 salt and dust in the model. Even though the existing parameterizations for emissions of natural species hide a number of uncertainties, the comparison with the AOD trends that results from off-line prescribed inventories will identify the relative importance of the two methods (e.g. if the AOD trend from dust changes significantly). Finally, the use of newly developed dataset of remote sensed observations (such as the MODIS collection 6), could improve the reliability and agreements of these dataset and decrease artificial trends due to calibration (Lyapustin,et al., 2014).