Thirteen Years of Aerosol Radiative Forcing in Southwestern Iberian Peninsula

Notable uncertainties in the aerosol impact on the Earth's radiative budget still remain, mainly at local scales. This study aims to accurately quantify the shortwave aerosol radiative forcing (ARF) and its efficiency (ARFE) at Évora (Portugal, Southwestern Europe). This is an interesting region affected by different aerosol types and sources. Towards this goal, ARF and ARFE are calculated for Évora during thirteen years (2003–2015), which is the longest period ever analyzed for this region. For this calculation, the most updated AERONET data have been used to simulate irradiance with the libRadtran code. Thus, 1676 daily values were analyzed, obtaining a mean ARF of –7.4 W m at the Earth's surface (SURF) and –1.1 W m at the top of the atmosphere (TOA), indicating a cooling of the Earth–atmosphere system. The difference in ARF between TOA and SURF indicates a mean net gain of 6.3 W m for the atmosphere. While no ARF seasonal pattern is observed at the TOA, a clear seasonal cycle is detected at the SURF with the most negative values occurring in summer. On the other hand, ARFE shows mean values of –62.7 W m τ at the SURF and –14.3 W m τ at the TOA. Results show that ARFE highly depends on the aerosol single scattering albedo, showing opposite behavior at the SURF and at the TOA. The large temporal variability detected in the aerosol properties in the region of study confirms the general need for long time series of measurements to achieve an accurate estimation of the ARF and ARFE.


INTRODUCTION
Atmospheric aerosol particles, of natural or anthropogenic origin, interact directly with solar and terrestrial radiation through scattering and absorption as well as through emission processes, modifying the Earth's radiative balance.This effect is quantified by the aerosol radiative forcing (ARF) calculated as the change in the radiative balance due exclusively to the existence of aerosols.According to the Fifth Assessment Report delivered in 2013 by the Intergovernmental Panel on Climate Change, direct and indirect effects of aerosols in the atmosphere result in an overall radiative forcing of -0.9 (-1.9 to -0.1) W m -2 at the TOA (Boucher et al., 2013).Although this global cooling effect due to aerosols is well established, significant uncertainties still remain, mainly referred to their spatial and temporal variability at regional and local scales.Thus, due to nearby point sources, the aerosol radiative forcing at the local scale can exceed the global values by an order of magnitude.Additionally, most climate models underestimate the negative radiative forcing caused by anthropogenic aerosols (Hansen et al., 2011).Hence, this short-scale information is essential to achieve an accurate knowledge of the impact of aerosols on climate.It is, therefore, important to quantify the long-term aerosol radiative forcing in different areas worldwide, paying special attention to regions of climatic significance.
Within this framework, the Iberian Peninsula is a particularly interesting region affected by occasional intrusions coming from the largest source of dust worldwide: the Sahara Desert.The frequency of these intrusions is modulated by the annual latitudinal displacement of major semi-permanent pressure systems, showing a typical seasonal pattern (Cachorro et al., 2005;Toledano, 2005;Salgueiro et al., 2014) associated to specific synoptic situations (Rodriguez et al., 2001;Querol et al., 2002;Escudero et al., 2005).In addition to the desert dust intrusions, the Iberian Peninsula is notably affected by other sources of aerosols.In particular, the southwestern interior part of the Iberian Peninsula is also exposed to local continental aerosols and marine aerosols coming from the Atlantic Ocean (Pereira et al., 2005;Elias et al., 2006;Silva et al., 2006;Pereira et al., 2008;Pereira et al., 2011;Obregón et al., 2012).Additionally, wildfires are an occasional but intense source of aerosols for the area in summer.Due to its exposure to a diversity of local and remote aerosol sources, the Iberian Peninsula is a privileged location for studying the aerosol impact on the radiation balance.
A major research site for aerosols in this area is located in Évora (Portugal), deployed by the Institute of Earth Sciences at the University of Évora, where the AErosol Robotic NETwork (AERONET, http://aeronet.gsfc.nasa.gov)has a monitoring station.This station started to operate in 2003 and, since then, provides radiometric measurements of aerosol properties averaged in the atmospheric column according to AERONET standards and protocols.Notable research has been conducted based on these measurements focusing mainly on obtaining typical values for the physical and optical properties of the aerosols during long periods and specific interesting events (Silva et al., 2003;Obregón et al., 2012), identifying most frequent aerosol types in the area (Elias et al., 2006), and studying their radiative effect (Santos et al., 2008;Mateos et al. 2014;Obregón et al., 2015a).However, these latter studies are based on a limited period of time (Santos et al., 2008;Mateos et al., 2014) or focus on specific short events (Obregón et al., 2015a), failing to provide an accurate and complete estimation of the aerosol radiative forcing.Therefore, this study gathers the longest series of aerosol measurements ever used for this area in order to accurately estimate the direct aerosol radiative forcing and its efficiency in the shortwave spectral range.In order to thoroughly analyze the impact of aerosols in the total atmosphere, the aerosol radiative forcing and its efficiency will be calculated at the Earth's surface (SURF) and at the top of the atmosphere (TOA).
The paper is structured as follows: first, a brief description of the study region, instrumentation, measurements and methodology is provided (Section 2), followed by the description of the main results (Section 3), and, finally, the main conclusions (Section 4).

Site Description
The current study is based on aerosol experimental measurements registered at the AERONET site located in the city of Évora (Portugal), in southwestern Iberian Peninsula.Fig. 1 shows the location of this station (38.6°N, 7.9°W, 293 m a.s.l), approximately 100 km eastwards from the Atlantic coast.
Évora shows a mainly continental climate attenuated by the maritime influence of the Atlantic Ocean.This climate is classified as Csa (Mediterranean climates with hot and dry summers) according to Köppen-Geiger types (Köppen, 1936).It features very hot and dry summers with prevailing cloud-free conditions.Very high irradiance values, among

Instrumentation
This ground-based station is equipped with a CIMEL CE-318 sunphotometer integrated in the NASA AERONET (Aerosol Robotic NETwork) network (Holben et al., 1998) since 2003.This network provides aerosol information with the aim to establish a ground-based climatology and to identify and characterize the most influential aerosols on the climatic system.The CIMEL CE-318 sunphotometer performs direct sun measurements at 340,380,440,500,675,870,940 and 1020 nm, and measurements of sky radiances in the almucantar and principal plane geometries at 440, 675, 870 and 1020 nm.These measurements are processed following AERONET protocols (Holben et al., 1998) to obtain a variety of aerosol products at different quality levels (1.0, 1.5 and 2.0).The sunphotometer is calibrated annually following AERONET protocols by the AERONET-NASA, RIMA and PHOTONS networks (Holben et al., 1998;Toledano et al., 2006).

Method
It is important to note that the radiative forcing estimates obtained in this study differ from the IPCC definition of radiative forcing.While the IPCC calculates the radiative forcing as the difference between irradiances in the preindustrial and present day atmosphere due to changes in the concentrations of a certain atmosphere component, in this study, radiative forcing stands for the instantaneous contribution of a particular component (in this case aerosols) to the radiation budget.Thus, it is calculated as the difference in the radiative energy at a given level between a situation where aerosols are present, F, and a situation where these atmospheric particles are absent (background/baseline conditions), F 0 : where F is the flux (shortwave irradiance), the subscript "dn" indicates the downward direction and the subscript "up" the upward direction.In this study, the ARF values calculated at the SURF and at the TOA are analyzed.Thus, radiative forcing can be defined at these two levels as: where A s is the surface albedo, assumed to be the same for both aerosol situations (background/baseline and presence of aerosols).
To estimate the amount of solar radiations trapped in the atmosphere due to aerosols, the difference in ARF between TOA and SURF is computed.Then, the heating rate (HR) due to aerosol absorption is calculated according to the first law of thermodynamics and assuming hydrostatic equilibrium, as suggested by Liou (2002): where ∂T/∂t is the HR in K day -1 , g is the acceleration due to gravity (9.8 m s -2 ), Cp is the specific heat capacity of air at constant pressure (i.e., 1006 J kg -1 K -1 ), and ΔP is the atmospheric pressure difference between surface and 3 km altitude, where most aerosols are present.According to several authors, in this study, ΔP was taken as equal to 300 hPa (Kaskaoutis et al., 2013;Singh et al., 2016).
From the ARF instantaneous data, the daily average value (DARF) can be calculated as follows: where ARF hourly are hourly averaged values and the sum extends over the whole day (24 h).
ARF is very sensitive to the amount and type of aerosols existing in the atmosphere, increasing with the AOD.In order to quantify the rate at which the forcing varies per unit optical depth, the aerosol radiative forcing efficiency (ARFE) is defined.The mean ARFE was calculated as the slope of the regression line between ARF and AOD as suggested by Bush and Valero (2003).In this study AOD at 550 nm has been used.The conversion of AOD at 500 nm, which is the AERONET direct product, to AOD at 550 nm has been performed using the Ångstrom power law: where α is Ångstrom Exponent in the wavelength range of 440-870 nm (Alam et al., 2011).
The dataset used in this study is composed of cloud-free shortwave irradiance simulated values integrated over the spectral interval 285-2800 nm.The simulations were performed during those days in the period 2003-2015 when level 2.0 AERONET data are available for this station.Thus, 1676 days were used, which is a much higher number than, for example, the 200 days used by Esteve et al. (2014).It is important to note that level 2.0 data correspond exclusively to cloud-free conditions.
The shortwave irradiances were simulated using the libRadtran version 1.7 radiative transfer code (Mayer and Kylling, 2005;Emde et al., 2016).The uncertainties of libRadtran model were assessed by Emde et al. (2016), who showed that the estimated radiance and irradiance results agree very well with a reference validated model, presenting differences lower than 0.05%, with a few exceptions for cases that are not treated in this work (for example circum-solar radiation).References therein (Emde et al., 2016) to a number of model intercomparison studies demonstrate the performance of libRadtran in comparison to a variety of other models.In addition, Obregón et al. (2015b) analyzed the performance of libRadtran to simulate shortwave solar irradiances under different aerosol conditions, through comparison with corresponding calibrated pyranometer measurements.The authors obtained relative differences between irradiance simulations and measurements lower than 0.2%.Additionally, Román et al. (2014) analyzed the sensitivity of SW irradiances simulated with libRadtran model to the input uncertainties, obtaining a global sensitivity usually below 10%.
The main inputs for simulations were aerosol optical properties, precipitable water vapor column, surface albedo, and total ozone column.The inputs to describe the aerosols were provided by AERONET and consist of direct products as aerosol optical depth (AOD) and Ångström parameter, and inversion products as asymmetry parameter (ASY), and single scattering albedo (SSA).Ångström parameters were derived from level 2.0 AERONET aerosol optical depth (AOD) in the range 440-870 nm.Level 2.0 AERONET dataset also provided ASY values.Level 2.0 SSA data are very scarce, with only 226 values available in the period of study.The main reason is the requirement to only consider cases with AOD at 440 nm higher than 0.4.This AOD value is very unusual in Évora (the typical median value is 0.1) and corresponds almost exclusively to desert dust intrusions, which is a very specific situation.Therefore, in order to include different aerosol types and to have enough data available, level 1.5 SSA had to be used.Additionally, data were inspected to search for anomalous values.No negative AOD cases and only eight (0.06% of a total number of cases of 13920) negative Ångström exponent cases were found.The latter were filtered.SSA and ASY values for 675 nm have been used as suggested by some authors (Di Biagio et al., 2010;Foyo-Moreno et al., 2014).This wavelength has been chosen among the wavelengths available at AERONET data as it is close to the maximum of the solar irradiance per unit wavelength.Precipitable water vapor data were also obtained from level 2.0 AERONET products.Daily total ozone column was provided by the AURA Validation Data Center (AVDC) via its website (http://avdc.gsfc.nasa.gov/index.php?site=1593048672&id=28). Surface albedo was obtained from the Surface and Atmospheric Radiation Budget (SARB) working group, part of NASA Langley Research Center's Clouds and the Earth's Radiant Energy System (CERES) mission (http://ceres.larc.nasa.gov/).In order to take into account the time of the year, midlatitude summer/winter standard vertical profiles of thermodynamic variables, main gases and aerosols were considered depending on the season (Shettle et al., 1989;Mayer and Kylling, 2005).The extraterrestrial irradiance was considered according to Gueymard (2004).The radiative transfer equation solver DISORT2, which is an improved version of the multistream discrete ordinate method of Stamnes et al. (2000), was used with 16 streams.
In order to more accurately model the instantaneous irradiance, 10-minute irradiances were simulated.For this, available AERONET values (α and β Ångström coefficients, ASY, SSA and PWC) were interpolated on a 10-minute basis.The dataset used was composed of 108843 10-minutes values of ARF corresponding to the 1676 days with available AERONET data.Fig. 2 shows the number of days analyzed in this study for each month during the period 2003-2015.Summer is the season with more days with available data (668 days, 40%), due to the prevailing cloud-free conditions during this period of the year.The relative frequency of days with available data for the other seasons was 21% (352 days) in spring, 20% (343 days) in autumn, and 19% (314 days) in winter.
To evaluate the aerosol radiative effects, simulations with and without aerosols were computed for the 108843 cases.Finally, ARF was calculated at the TOA and at the SURF levels, substituting the simulated irradiance in equations 2 and 3, respectively.The reliability of the libRadtran model for simulating SW irradiance in Évora was previously confirmed by comparing simulated values with irradiance measurements registered by an Eppley Black & White pyranometer (Obregón et al., 2015a).This comparison included clean and turbid cases corresponding to periods before or after (clean), and during (turbid) two particular desert dust intrusion events occurred in summer 2012 (Obregón et al., 2015a).Additionally, in this study the reliability of the libRadtran simulations will be tested by comparing our estimated instantaneous ARF with values provided by AERONET.

Spectral Variation of AOD, SSA and ASY
As mentioned in the methodology, aerosol optical properties such as AOD, SSA and ASY along with their spectral variability are required for calculating ARF and ARFE.These properties have different spectral variations revealing information about the particle size and their absorptivity.Fig. 3 illustrates the AOD, SSA and ASY mean spectral variation for the different seasons.
The spectral variation of AOD provides information about the particle size distribution (Eck et al., 1999).AOD seems to be more wavelength dependent during summer, when it decreases as wavelength increases, indicating the presence of fine particles.Conversely, the smaller spectral dependence is observed in winter, indicating the prevalence of coarser particles.Mean AOD in summer is 0.17 ± 0.12 at 440 nm wavelength and 0.07 ± 0.08 at 1020 nm, whereas, in winter, the mean value is 0.08 ± 0.05 at 440 nm and 0.03 ± 0.02 at 1020 nm.
In regard to SSA, Fig. 3 shows the mean values for each season of the year.As can be seen, mean SSA values for the four seasons decrease as wavelength increases.This SSA spectral behavior indicates the presence of absorbing aerosols (Dubovik et al., 2002).This wavelength dependence is more intense in winter than during the other seasons of the year, with mean values of 0.88 ± 0.08 at 440 nm and 0.80 ± 0.14 at 1020 nm.This dependence is also reported for Lahore and Karachi by Alam et al. (2012), Delhi by Singh et al. (2010), and four locations in the Indo-Gangetic plains by Bibi et al. (2016), but only during winter, when dust is not a major contributor to the aerosol load.
Additionally, the mean value corresponding to AERONET level 2.0 SSA data has been added to Fig. 3(b).In contrast to the behavior found for the SSA data used in this work, level 2.0 SSA presents higher values and increases or is neutral with wavelength, which is a clear signature of the predominance of mineral dust (Dubovik et al., 2002).This fact is due to the extremely restrictive threshold used for level 2.0 data, which requires AOD higher than 0.4.In the case of Évora, this threshold mostly retains those cases with desert dust aerosols originating from the Sahara region.Since this study attempts to analyze the mean aerosol radiative forcing in southwestern Iberian Peninsula during thirteen years, a variety of aerosol conditions needs to be sampled.Therefore, for this study, the level 2.0 AERONET data is too restrictive regarding the SSA.
Similarly to AOD and SSA, ASY shows spectral dependence (Fig. 3(c)).In general ASY decreases with increasing wavelength, except for high wavelengths, where a marginal increase can be observed for spring, summer and autumn seasons.Mean ASY (standard deviation) is 0.70 (0.03) at 440 nm and 0.66 (0.04) at 1020 nm in spring, 0.70 (0.03) at 440 nm and 0.65 (0.05) at 1020 nm in summer, 0.70 (0.03) at 440 nm and 0.64 (0.05) at 1020 nm in autumn, and 0.70 (0.03) at 440 nm and 0.62 (0.06) at 1020 nm in winter.The largest ASY decrease is thus observed in winter, in line with the results reported by Alam et al. (2012) in Lahore and Karachi.

Aerosol Radiative Forcing
In this section, the values of ARF at the SURF and at the TOA over Évora during the period 2003-2015 are analyzed.Fig. 4 shows the values of DARF at the SURF and at the TOA obtained in Évora during the period 2003-2015 in a box and whisker plot.As can be seen, the DARF at the SURF shows some variability during the year, with more negative values in warm months and less negative values in cold months.DARF ranged from -0.1 to -54.9 W m -2 , with a mean (and standard deviation) of -7.4 (5.5) W m -2 ,  at the SURF, and from -15.6 to +18.9 W m -2 , with a mean value of -1.1 (2.5) W m -2 , at the TOA.The negative values indicate a decrease in the solar radiation reaching the SURF and an increase in the light scattered back to space at the TOA.This loss in net radiation at the TOA leads to cooling of the Earth-atmosphere system.It is worth mentioning that 491 DARF values are positive (29% of the total number of cases).These values are near zero, with 75% of cases below 1.3 W m -2 .
It is interesting to note that the absolute magnitude of the mean DARF at the SURF is higher than at the TOA, producing a mean net gain of energy by the atmosphere estimated in 6.3 (with a standard deviation of 4.8) W m -2 .This gain is due to the absorption of radiation by atmospheric components.
The monthly mean of this absorption by the atmosphere is represented by the difference between the two red lines in Fig. 4.These values ranged from 2.6 W m -2 in December to 7.1 W m -2 in August.These absorbing properties of aerosols are also particularly important for energy balance and cloud-aerosol interactions, as shown in the work of Su et al. (2008).
While no seasonal DARF pattern is observed at the TOA, a clear seasonal cycle is detected at the SURF (Fig. 4).This seasonality is not only due to the variation in the radiation along the year, but also to the seasonal pattern shown by AOD (Fig. 5(a)).This figure presents a box and whisker plot of daily AOD (at 550 nm) values over Évora for each calendar month, showing its annual variation.The lowest optical depths occur in November and December, contrasting with the highest optical depths found in August and September, when AOD is higher than 0.18 in more than 25% of the cases.These high AOD values during the summer months can be explained by a combination of several factors such as Saharan dust intrusions and forest fires, being the latter associated with high temperatures, very low atmospheric relative humidity and lack of precipitation.The arrival of air masses from the Saharan region transporting desert-dust aerosols constitutes a substantial source of aerosols, affecting the southwestern Iberian Peninsula generally in summer (Toledano et al., 2007).This fact is confirmed by the low α values observed in August (Fig. 5(b)).Summer forest fires, particularly those extensive and long lasting that occurred in Portugal and Western Spain in 2003 and 2005 (Elias et al., 2006), could mean also a substantial contribution to high optical depth values during these months.Both conditions (low AOD in winter months and high AOD values in summer months) are consistent with the ARF values shown in Fig. 4 (less negative in winter and more negative in summer).Also, a secondary maximum is observed in spring, being consistent with ARF values and observations reported for other stations of the Iberian Peninsula (Estellés et al., 2007;Toledano et al., 2007;Obregón et al., 2012;Esteve et al., 2014).
Fig. 5(b) depicts the annual variation of daily α, showing the highest mean values in January, December, February, and November.Conversely, α is low in April, May and August, being associated with large particles, typical of desert dust episodes or/and intrusions of marine aerosol in spring and summer (associated with sea breeze).In August, the low α values together with high AOD load, indicate the intrusion of dust from North Africa.Other studies have reported a similar seasonal pattern of AOD and α at other locations within the Iberian Peninsula, such as El Arenosillo (Cachorro et al., 2005;Toledano et al., 2007), Boecillo (Cachorro et al., 2000), and Cáceres (Obregón et al., 2015c).As it can be seen in Fig. 5(a), although AOD is high and similar in August and September, α differs between these two months.In August, α is lower than in September, indicating a stronger influence of desert dust aerosol in August than in September, and a higher cooling at the SURF.This tendency of cooling the Earth-atmosphere system due to the effect of desert dust aerosols agrees with results reported for other regions (e.g., Costa et al., 2006;Derimian et al., 2006;  Lyamani et al., 2006;Cachorro et al., 2008;Guerrero-Rascado et al., 2009;Huang et al., 2009;García et al., 2012;Santos et al., 2013;Valenzuela et al., 2012;Obregón et al., 2015a).
It is worth mentioning that our findings are not in line with results for the same station reported by Mateos et al. (2014) within a general study where several stations were analyzed.The monthly ARF values obtained here are generally more negative than theirs.In particular, the magnitude and the time of the year with the most negative ARF differ between both studies; while this study finds -9.5 W m -2 in August, they find around -8 W m -2 in April.One reason for this discrepancy could be related with the limited period of study considered by Mateos et al. (2014), which, for example, excluded the year 2003.During this year, as already mentioned above, there were numerous intense intrusions of desert dust and several forest fires all over Iberian Peninsula and south of France.The difference in ARF values emphasizes the need for relying on long term representative series of measurements in order to achieve an accurate estimation of the average regional aerosol radiative effect, mainly in areas with high variability such as the Iberian Peninsula.
An example of the spatial variability is the difference in ARF values reported for the Atlantic and the Mediterranean coastal areas within the Iberian Peninsula.Thus, the magnitude of the ARF obtained in this study at an Atlanticinfluenced station such as Évora is around half the values measured at a Mediterranean station such as Burjassot (Valencia, Spain) (Esteve et al., 2014).This difference is due to several factors such as the influence of the oceans, the prevailing circulation patterns and local aerosol sources.The results obtained in this study also show great differences with respect to those obtained in large polluted cities such as Lahore and Karachi in Pakistan.At these stations the ARF values are around -70 and -19 W m -2 at the SURF and at the TOA, respectively (Alam et al., 2011;Alam et al., 2014).
Fig. 6 shows comparisons between ARF at both levels SURF (a) and TOA (b), as provided by AERONET vs values estimated in this study using libRadtran.The comparisons are performed for the period of study.Values at the SURF show a good general agreement, with a high correlation coefficient of 0.97.This value is similar to those obtained by Alam et al. (2012) for Lahore and Karachi using SBDART model (0.99 and 0.98, respectively), and notably higher than those obtained by Bibi et al. (2017) over Karachi (0.51), Lahore (0.81), Jaipur (0.66) and Kanpur (0.69).This good correlation between AERONET and libRadtran surface forcing demonstrates that the input parameters, particularly the atmospheric profiles, constitute an appropriate representation of the prevailing atmospheric conditions in our location of study.On the other hand, the comparison between ARF at the TOA presents also a very high correlation coefficient (0.91) at the TOA (Fig. 6(b)).This value is notably higher than the values obtained by Bibi et al. (2017) over Karachi (0.46), Lahore (0.73), Jaipur (0.57) and Kanpur (0.55).
The libRadtran code allows to select two standard aerosol profiles, one typical of spring-summer and another of autumn-winter conditions.These profiles (Shettle, 1989) describe the change in the aerosol profile throughout the year.In order to analyze the sensitivity of the radiative forcing estimates to the vertical distribution of aerosols, these two profiles have been used.Thus, all the calculations have been repeated for both profiles separately regardless of the true season of the year and the daily aerosol radiative forcing has been obtained at the SURF and at the TOA.For the comparison analysis, the whole year 2012 has been used in order to account for the annual variability of the frequency of the aerosol conditions.The DARF estimated using both profiles show a very good agreement with a correlation coefficient near one.Differences in the DARF are very low at both levels, with mean differences of -0.001 Wm -2 (standard deviation of 0.009 Wm -2 ) at the SURF, and of 0.012 Wm -2 (standard deviation of 0.050 Wm -2 ) at the TOA.These results are in line with the sensitivity study performed by Mishra et al. (2015), who concluded that aerosol forcing in the visible-near infrared range is less sensitive to changes in the aerosol layer height as compared to thermal-infrared and ultraviolet ranges.
As mentioned before, since the ARF depends on the particle load in the atmosphere, the forcing efficiency ARFE was calculated and analyzed at the SURF and at the TOA levels.The mean ARFE has been calculated as the slope of the regression line between ARF and AOD values.Since the ARF is expected to be zero when AOD is zero, a zero-intercept linear regression has been used to estimate the ARFE.Fig. 7 shows DARF at the SURF and at the TOA versus daily mean AOD (550 nm).The uncertainty in ARFE is quantified by the error associated to the regression slope.The ARFE obtained were -62.7 ± 0.4 W m -2 τ -1 and -14.3 ± 0.3 W m -2 τ -1 (per unit AOD or τ at 550 nm) for SURF and TOA datasets, respectively.In order to analyze the effect of SSA, in Fig. 7 the points have been colored according to their SSA value.As can be seen, for a given AOD, the most negative ARF values at the SURF correspond to low SSA values, i.e., to the most absorbing aerosols.Conversely, the opposite behavior is found at the TOA (Fig. 7(b)).Thus, the increase in the absorption by aerosols reduces the scattered radiation also towards TOA, thereby contributing to less negative (or even positive) ARF values at the TOA.These results are in line with those found during the CLEARCOLUMN/ACE2 campaign conducted at Sagres, Portugal (Silva et al., 2002).They are also confirmed by the ARFE values calculated for different SSA intervals.Thus, the ARFE obtained for SURF were -140.2, -104.3, -81.4 and -55.48 W m -2 τ -1 for the SSA ranges [< 0.7], [0.7-0.8],[0.8-0.9],[>0 .9],respectively.Similarly, the following ARFE values: +44.5, +15.2, -3.4,-19.1 W m -2 τ -1, were obtained for the TOA for the same SSA intervals mentioned above.While the aerosols at the SURF with low SSA result in highly negative ARFE values, at the TOA those aerosols are associated with less negative, or even positive, ARFE values.
The ARFE was also calculated at 500 nm in order to compare with the study performed by Esteve et al. (2014) for a Mediterranean station during the period 2003-2011.As mentioned before, notable differences exist between the Atlantic and the Mediterranean areas.Thus, while Évora shows -55.1 W m -2 τ -1 at the SURF and -15.1 W m -2 τ -1 at the TOA, Burjassot shows more negative values (-139 W m -2 τ -1 at the SURF and -19.2 W m -2 τ -1 at the TOA).
Fig. 8 shows the mean ARFE calculated for each season of the year at the SURF and at the TOA.Points have been colored according to their SSA value.No seasonal effect can be perceived at any level, with ARFE around -62 W m -2 τ -1 and -14 W m -2 τ -1 at the SURF and at the TOA, respectively.However, the aerosol tends to have a slightly stronger impact at the TOA in spring (-16.1 ± 0.7 W m -2 τ -1 ) than in the remaining seasons: winter (-14.2 ± 0.9 W m -2 τ -1 ), autumn (-13.9 ± 0.6 W m -2 τ -1 ) and summer (-13.8 ± 0.4 W m -2 τ -1 ).

Heating Rate
The amount of solar radiations trapped in the atmosphere by the aerosols as quantified by the atmospheric heating rate has been analyzed.The average heating rate (K day -1 ) estimated from DARF during the study period was 0.18 (with a standard deviation of 0.15).Percentiles 10 and 90 indicate that in 80% of the days, HR falls between 0.04 and 0.35 K day -1 .These values are in line with the value of 0.11 K day -1 obtained by Valenzuela et al. (2017) for the same location (Évora) during one specific day.The heating rate in Évora is generally lower than values reported for stations with high atmospheric absorption such as, for example, two polluted mega-cities in Pakistan, with values higher than 1 K day -1 (Alam et al., 2012).It is also lower than the values obtained for two desert dust events analyzed by Gharibzadeh et al. (2017), which showed mean values of 2.16 and 0.89 K day -1 .In order to analyze the annual variation of the HR in Évora, the mean HR value was calculated for each calendar month (Fig. 9).The higher HR values occur mainly during summer (mean of 0.22 and standard deviation of 0.15), when ARF is also high.It is followed by spring (mean of 0.18 and standard deviation of 0.18), autumn (mean of 0.15 and standard deviation of 0.12) and winter (mean of 0.10 and standard deviation of 0.10).Spring is the season that shows the higher variability, with many high HR values during April.The annual variation in the amount of solar energy trapped by the atmosphere has strong implications for temperature changes and thermal structure of the atmosphere as well as important consequences for many processes, like cloud formation, precipitation, and synoptic circulation.

CONCLUSIONS
The aerosol radiative forcing (ARF) at Évora station has been studied with the goal of reducing the uncertainty of the aerosol impact at local scales.This station provides information about Southwestern Iberian Peninsula, a region of climatic significance largely affected by different aerosol types such as local continental particles, Atlantic marine aerosols, and desert dust coming from the largest source of dust worldwide: the Sahara Desert.Towards this aim, the aerosol radiative forcing and its efficiency (ARFE) in the shortwave solar spectrum have been calculated and analyzed.The study period extends from 2003 to 2015, being the longest period ever used in studies of aerosol radiative forcing in this region.
ARF and ARFE have been calculated at both the Earth's surface (SURF) and top of the atmosphere (TOA) levels.For this purpose, a high temporal resolution dataset, composed of 10-minute irradiance values simulated using the libRadtran code and previously validated, has been used.The ARF values calculated on a 10 minute basis have been integrated over 24 h to obtain daily mean values, DARF, for 1676 days.While no seasonal DARF pattern is observed at the TOA, a clear seasonal pattern can be distinguished at the SURF, with more negative values in summer, likely associated not only to the variation in the radiation along the year but also to the annual variability of AOD.The DARF values obtained for SURF ranged from -0.1 to -54.9 W m -2 , with a mean (and standard deviation) of -7.4 (5.5) W m -2 , and from -15.6 to +18.9 W m -2 at the TOA with a mean value of -1.1 (2.5) W m -2 .The negative values indicate a decrease in the solar radiation that reaches the Earth's surface and an increase in the light that is scattered back to space, cooling the Earthatmosphere system.The difference in DARF magnitude between SURF and TOA indicates a mean net gain of energy by the atmosphere estimated in 6.3 (with a standard deviation of 4.8) W m -2 , due to absorption of radiation by atmospheric components.
The values of ARF at the SURF calculated in this study were compared with those reported by other authors for different regions, showing significant differences and indicating the importance to perform accurate local studies to achieve a thorough description of the impact of aerosols.The comparison with the study of Mateos et al. (2014) for the same station revealed notable differences, mainly in the ARF seasonal pattern.This discrepancy is likely associated to the use of a limited period of measurements in that study, emphasizing the need to use long high quality time series of data in order to guarantee an accurate description of the aerosol radiative forcing.
Instantaneous ARF values simulated with libRadtran have been compared with values provided by AERONET.The comparison showed a very good agreement at both levels: SURF and TOA, confirming the reliability of the calculations performed in this study with the libRadtran model.The ARFE (ARF per unit optical depth) was also calculated, and found to be -62.7 ± 0.4 W m -2 τ -1 and -14.3 ± 0.3 W m -2 τ -1 (per unit AOD or τ at 550 nm), at the SURF and at the TOA, respectively, showing no seasonality.These ARFE values are directly related to the SSA, with opposite behavior at the SURF and at the TOA.Thus, while at the SURF the aerosols with low SSA result in highly negative ARFE values, at the TOA those aerosols are associated with less negative or even positive ARFE values.This fact is due to the higher absorption by these aerosols, which results in less scattered radiation also towards the TOA and, therefore, in less negative ARFE at the TOA.
The values of ARF and ARFE reported for the region of study positively contributes to improve the understanding of the impact of aerosols in this area, mainly referred to its magnitude and temporal variability.In particular, the high temporal variability found in ARF and ARFE suggests the need to use long time series to guarantee the temporal representativeness in this interesting region.

Fig. 3 .
Fig. 3. (a), (b) and (c) show AOD, SSA and ASY mean values at wavelengths of 440 nm, 675 nm, 870 nm, and 1020 nm for different seasons.AERONET level 2.0 SSA has also been plotted in (b) for comparison purposes.

Fig. 4 .
Fig. 4. Box and whisker plot of daily ARF values, DARF, at the surface and the top of the atmosphere for each calendar month over the 2003-2015 period.The mean (filled circles), median (horizontal segment within the box), 25 th and 75 th percentiles (top/bottom box limits) and data within 1.5 times the interquartile range (whiskers) are shown.

Fig. 5 .
Fig. 5. Box and whisker plot of daily AOD (a) and α (b) values for each calendar month over the 2003-2015 period.The mean (filled circles), median (horizontal segment within the box), 25 th and 75 th percentiles (top/bottom box limits) and data within 1.5 times the interquartile range (whiskers) are shown.

Fig. 6 .
Fig. 6.Comparisons between ARF provided by AERONET and ARF estimated in this study using libRadtran, (a) for surface level and (b) for the top of the atmosphere level.

Fig. 7 .
Fig. 7. Aerosol radiative forcing as a function of the AOD at 550 nm during the study period.(a) for surface level and (b) for the top of the atmosphere level.The regression equations are included and the points have been colored according to their SSA value.

Fig. 8 .
Fig. 8. Aerosol radiative forcing as a function of the AOD at 550 nm for each season of the year.The regression equations are included and the points have been colored according to their SSA value.

Fig. 9 .
Fig. 9. Box and whisker plot of daily HR values for each calendar month over the 2003-2015 period.The mean (filled circles), median (horizontal segment within the box), 25 th and 75 th percentiles (top/bottom box limits) and data within 1.5 times the interquartile range (whiskers) are shown.