How do aerosol histories affect solar “dimming” and “brightening” over Europe?: IPCC-AR4 models versus observations



[1] A multidecadal decrease in downward surface solar radiation (solar “dimming”) followed by a multidecadal increase in surface radiation (solar “brightening”) have been reported over Europe. The trends mainly occur under cloud-free skies, and they are primarily caused by the direct aerosol radiative effect. The present study compares observed cloud-free solar “dimming” and “brightening” trends with corresponding output from IPCC-AR4 20th century simulations and furthermore examines how sulfate and black carbon aerosol histories, used as model input, affect simulated surface radiation trends. Outputs from 14 models are compared to observed cloud-free surface radiation fluxes derived from a combination of (1) satellite cloud observations, synoptic cloud reports, and surface solar irradiance measurements and (2) sunshine duration measurements and variability of the atmospheric transmittance derived from solar irradiance measurements. Most models display a transition from decreasing to increasing solar irradiance, but the timing of the reversal varies by about 25 years. Consequently, large discrepancies in sign and magnitude occur between modeled and observed “dimming” and “brightening” trends (up to 4.5 Wm−2 per decade for Europe). Considering all models with identical aerosol histories, differences in cloud-free radiation trends are in all but one case less than 0.7 Wm−2 per decade. Thirteen of the fourteen models produce a transition from “dimming” to “brightening” that is consistent with the timing of the reversal from increasing to decreasing aerosol emissions in the input aerosol history. Consequently, the poor agreement between modeled and observed “dimming” and “brightening” is due to incorrect aerosol emission histories rather than other factors.

1. Introduction

[2] Solar radiation at the Earth's surface is a driving factor for the Earth's temperature and hydrological cycle, thus influencing the entire climate and ecosystem [e.g., Ramanathan et al., 2001; Wild et al., 2004]. Studies based on surface radiation measurements have shown that solar irradiance declined from about 1950 to the mid 1980s [e.g., Gilgen et al., 1998; Stanhill and Cohen, 2001], popularly called solar “dimming.” Recent publications report that solar radiation began to recover over large parts of the world [Wild et al., 2005; Ohmura, 2006], known as solar “brightening.” The increase in solar irradiance in Europe largely occurs under cloud-free skies [Ruckstuhl et al., 2008] or under conditions where the radiative effects of cloud cover anomalies have been removed [Norris and Wild, 2007]. This suggests that the recent “brightening” in Europe is related to decreasing aerosol emissions reported by Streets et al. [2006] and others.

[3] Trends in aerosol emissions and associated “dimming” and “brightening” have likely influenced changes in the Earth's temperature [Wild et al., 2007]. Model simulations [e.g., Meehl et al., 2004] also show that future global warming depends on radiative forcings induced by natural and anthropogenic aerosols in addition to atmospheric greenhouse gas concentrations and internal feedback mechanisms. Large uncertainty in anthropogenic aerosol radiative forcings, however, is reported in the Intergovernmental Panel of Climate Change Fourth Assessment Report IPCC-AR4 [Solomon et al., 2007]. On the basis of an older set of models, Kiehl [2007] found that the wide range in anthropogenic forcings is caused by poor knowledge of past changes in aerosol emissions and by different treatment of aerosol processes in the models. In particular, there was a negative correlation between total anthropogenic forcing and climate sensitivity and a positive correlation between the total anthropogenic forcing and the aerosol forcing. Knutti [2008] confirmed the correlation between total anthropogenic forcing and climate sensitivity with IPCC-AR4 models, officially called the World Climate Research Programme's (WCRP's) Coupled Model Intercomparison Project phase 3 (CMIP3) multimodel data set. These correlations imply that models with high climate sensitivity simulate the observed 20th century warming by partially compensating positive greenhouse gas radiative forcing with large negative aerosol radiative forcing and models with low climate sensitivity simulate the observed warming by compensating with small negative aerosol forcing. Consequently, we can reduce uncertainty in climate sensitivity and future global warming by narrowing uncertainty in past anthropogenic aerosol radiative forcing.

[4] Since current temperature projections for the 21st century are mostly based on coupled atmosphere-ocean general circulation models (CGCMs), it is of particular importance to determine how well these models simulate solar irradiance at the Earth's surface and its decadal variability due to changes in aerosol content by comparing them with surface observations. It has been shown recently that the ability of general circulation models (GCMs) to realistically simulate surface solar irradiance has generally improved [Wild, 2005; Wild et al., 2006], although relatively large discrepancies between models still exist in certain regions [e.g., Sorteberg et al., 2007]. In the work of Romanou et al. [2007], all-sky surface solar radiation trends from the Baseline Surface Radiation Budget Network (BSRN) are compared with nine IPCC-AR4 models. The average trend of the nine models is considerably lower than the observed BSRN trend at each of the eight sites. Since such all-sky short-term trends are largely influenced by weather related cloud cover variations and CGCMs do not simulate the same weather as occurred historically, the discrepancy between models and observations is not unexpected.

[5] Aerosol emission histories and scenarios used as model input are another large source of model uncertainty. Substantial efforts have been made with the Special Report on Emission Scenarios (SRES) [Nakićenović et al., 2000] to have consistent emission scenarios for the IPCC-AR4 model runs, but many different emission histories were nonetheless used in the IPCC-AR4 20th century experiments (20C3M) [e.g., Smith et al., 2001; Boucher and Pham, 2002]. Moreover, not all IPCC-AR4 models consider the same aerosol types.

[6] The present study compares two independent data sets of observed solar irradiance under cloud-free skies with cloud-free solar irradiance from IPCC-AR4 20C3M simulations over the region of Europe. One data set is based on cloud-free irradiance measurements [Ruckstuhl and Philipona, 2008] (RP08), and the other is estimated from all-sky measurements from which the radiative effects of monthly cloud cover anomalies have been removed [Norris and Wild, 2007] (NW07). We focus on cloud-free solar irradiance trends because (1) the direct aerosol radiative effect (scattering and absorption of solar radiation by aerosols) is included in all IPCC-AR4 models, (2) indirect aerosol radiative effects (modification of clouds by aerosols) are included in fewer than one third of the IPCC-AR4 models, and (3) regional all-sky solar irradiance on short timescales depends mainly on cloud cover variations produced by weather that differ among the models and observations. Although all-sky fluxes ultimately affect Earth's climate and hydrological cycle, our approach provides insight into how well IPCC-AR4 models simulate the direct aerosol radiative effect and how well they represent the solar “dimming” and “brightening” observed under cloud-free conditions over Europe since the 1970s. We will furthermore show how the simulated cloud-free irradiance trends are closely related to the aerosol emission histories used by the models and demonstrate how better emission histories may help reduce model discrepancies and enable more accurate projections of future climate change.

2. Data and Models

2.1. Solar Irradiance Observational Data

[7] Cloud-free shortwave downward radiation (SDRcf) from IPCC-AR4 models is compared with measured SDRcf (RP08) and with estimated SDRcf independently derived from monthly all-sky measurements from which the effects of cloud cover anomalies have been removed (NW07). The latter data set is not truly SDRcf since it may include radiative effects of cloud albedo variations that are uncorrelated with cloud cover variations, but we nonetheless use the term SDRcf to describe it because it closely resembles measured SDRcf (shown later on).

[8] The method to detect cloud-free skies from RP08 is based on the observed sunshine duration (SSD) and the variability of the atmospheric transmission derived from shortwave downward radiation (SDR) measurements. Both measurements, SDR and SSD, require a time resolution of 1 hour. An hour is defined as cloud-free, if SSD is 60 min and the variability of the atmospheric transmission is smaller than an empirically pre-defined value. This method has been applied to data from the German Weather Service (DWD) in northern Germany (8 sites) and from the MeteoSwiss automatic meteorological network (ANETZ) at 25 lowland sites in Switzerland [Ruckstuhl et al., 2008]. In the work of Ruckstuhl et al. [2008], cloud-free solar irradiance is scaled by observed SSD to obtain results weighted by the frequency of cloud-free conditions, whereas in this study results are provided for only 100% cloud-free skies. Hence, the reported anomalies and trends in this study are larger than those reported by Ruckstuhl et al. [2008]. Between 1981 and 2005, there was a SDRcf increase of +3.5 [+2.2 to +4.8] Wm−2 per decade (dec) in northern Germany and a smaller +1.8 [+0.9 to +2.8] Wm−2 dec−1 increase in Switzerland. In the following we refer to RP08 SDRcf data as “surface observations.”

[9] NW07 investigated trends in surface solar radiation and cloud cover using data from the Global Energy Balance Archive (GEBA) [Gilgen and Ohmura, 1999], synoptic cloud reports [Hahn and Warren, 2003], and the International Satellite Cloud Climatology Project (ISCCP) [Rossow and Schiffer, 1999]. The effect of cloud cover anomalies on solar irradiance (called cloud cover radiative effect (CCRE) anomalies by NW07) was calculated by multiplying monthly cloud cover anomalies by the ratio of climatological cloud radiative effect divided by climatological cloud cover. NW07 then removed CCRE anomalies from GEBA solar irradiance anomalies via linear regression to produce residual anomalies, hereafter referred to as “residual flux.” Residual fluxes include contributions from radiation anomalies under cloud-free conditions and the effects of anomalies in cloud albedo that are linearly uncorrelated to cloud cover anomalies. Multidecadal variations in residual flux are likely primarily controlled by changes in the direct aerosol radiative effect and, if it is substantial, the first indirect aerosol radiative effect (enhancement of cloud albedo due to more cloud droplets by an increasing number of aerosols acting as cloud condensation nuclei [Twomey, 1974]). Pan-European residual flux decreased between −2.7 and −3.5 Wm−2 dec−1 from 1971 to 1986 and started to recover between 1987 and 2002 with a rate of +2.0 to +2.3 Wm−2 dec−1, depending on the trend calculation method.

[10] We tested whether NW07 residual flux is an appropriate measure to evaluate modeled SDRcf trends by comparing it to measured SDRcf, available for northern Germany and Switzerland. For Germany, the residual flux time series from five ISCCP grid boxes were averaged together with weighting according to the number of surface radiation sites located therein and used by Ruckstuhl et al. [2008]. Because no Swiss measurements entered the GEBA archive and thus were unavailable to NW07, the NW07 procedure was applied to average of the 25 Swiss ANETZ sites, the average of 15 MeteoSwiss stations providing synoptic cloud reports, and ISCCP data for the grid box centered on Switzerland.

[11] Figures 1a and 1b display the time series of residual flux anomalies and SDRcf anomalies from surface observations in northern Germany (Figure 1a) and in Switzerland (Figure 1b). In Figure 1c, the correlation between residual flux and surface observation SDRcf anomalies is shown. Root mean square (RMS) difference and correlation coefficient (R) are 1.86 Wm−2 and 0.89 in northern Germany and 1.63 Wm−2 and 0.73 in Switzerland. The independent methods reveal good agreement of SDRcf anomalies and trends even if only a few (northern Germany) or a single ISCCP grid box is considered (Switzerland). Since residual fluxes include direct and first indirect aerosol radiative effects, and SDRcf from surface observations includes only the direct effect, an assessment of the magnitude of the first indirect aerosol effect is possible. Although the aerosol optical depth (AOD) strongly declined in this region [Ruckstuhl et al., 2008] no significant difference between the SDRcf trends from these two methods is apparent. This implies that the first indirect aerosol effect is much smaller than the direct aerosol effect and hence the NW07 residual flux data can justifiably be used to evaluate SDRcf from the IPCC-AR4 models.

Figure 1.

Time series of SDRcf anomalies in (a) northern Germany and (b) Switzerland. The solid line stands for observational fluxes according to RP08. The dashed line is derived from the NW07 method, referred to as residual flux. (c) Scatterplot of the SDRcf anomalies with linear regression lines. Dotted line is the 1:1 line. Root mean square (RMS) differences and correlation coefficients (R) are also given.

2.2. IPCC-AR4 Models

[12] The 14 models used in this study are listed in Table 1 with the corresponding abbreviation, center of origin, reference, number of runs performed for the IPCC-AR4 20C3M experiments which provide SDRcf, and references for the sulfate (SO4) or sulfur dioxide (SO2) (precursor of sulfate aerosol) and black carbon (BC) input data sets. Furthermore, Table 1 indicates which models include natural forcings such as solar variability and volcanic forcing. Output data from the model runs are available from the WCRP CMIP3 multimodel data set, which is maintained by the Program for Climate Model Diagnosis and Intercomparison (PCMDI).

Table 1. List of the 14 IPCC-AR4 Models Used in This Study
AbbreviationModel CenterReferenceNumber of RunsSulfate/Sulfur DioxideBlack CarbonSolar ForcingVolcanic Forcing
CNRM_CM3Météo-France/Centre National de Recherches Meteorologiques, FranceSalas-Mélia et al. [2005]1Boucher and Pham [2002]Tanré et al. [1984] scaled by Novakov et al. [2003]nono
GFDL_CM2.0/2.1Geophysical Fluid Dynamics Laboratory, New Jersey, United StatesDelworth et al. [2006]3/3Horowitz [2006]Horowitz [2006], based on the work of Cooke et al. [1999]Lean et al. [1995]Sato et al. [1993] and Ramachandran et al. [2000]
IAP_FGOALS-g1.0Institute of Atmospheric Physics, Beijing, ChinaYu et al. [2004]3Boucher and Pham [2002]noLean et al. [1995]no
INM_CM3.0Institute for Numerical Mathematics, Moscow, RussiaGalin et al. [2003]1Smith et al. [2001, 2004]noLean et al. [1995]Ammann et al. [2003]
IPSL_CM4Institut Pierre Simon Laplace, FranceMarti et al. [2005]1Boucher and Pham [2002]nonono
MIROC3.2hires/medresCenter for Climate System Research, JapanK-1 Developers [2004]1/3Nozawa et al. [2007]Nozawa et al. [2007]Lean et al. [1995]Sato et al. [1993]
MIUB_ECHO-GMeteorological Institute of the University of Bonn, Meteorological Research Institute of KMA, and Model and Data group, Germany/KoreaLegutke and Voss [1999]5Roeckner et al. [1999]noCrowley [2000] based on the work of Lean et al. [1995]Crowley [2000]
MPI_ECHAM5Max-Planck-Institute for Meteorology, GermanyRoeckner et al. [2003]3Boucher and Pham [2002]nonono
MRI_CGCM2.3.2Meteorological Research Institute, JapanYukimoto et al. [2006]1Mitchell and Johns [1997]noLean et al. [1995]Sato et al. [1993]
NCAR_CCSM3.0National Center for Atmospheric Research, Colorado, United StatesCollins et al. [2006]8Smith et al. [2001, 2004]Collins et al. [2002] scaled by global populationLean et al. [1995]Ammann et al. [2003]
UKMO_HadCM3Hadley Centre for Climate Prediction and Research/Met Office, United KingdomJohns et al. [2003]2Smith et al. [2001, 2004]noLean et al. [1995]Sato et al. [1993]
UKMO_HadGEM1Hadley Centre for Climate Prediction and Research/Met Office, United KingdomMartin et al. [2006]2Smith et al. [2001, 2004]Nozawa et al. [2007]Solanki and Krivova [2003]Sato et al. [1993]

[13] To compare SDRcf from observations with model outputs, the model data were linearly interpolated to the exact location of each surface radiation site. The averages of the 8 locations in northern Germany and the 25 locations in Switzerland were then calculated. For comparison to residual flux, the model data were linearly interpolated to the centers of the 44 ISCCP grid boxes displayed in Figure 1 of NW07 and then averaged to what we call a pan-European mean.

2.3. Aerosol Emission/Burden Data

[14] Sulfate and BC are the most important anthropogenic aerosols affecting the climate system [e.g., Ramanathan et al., 2001]. Although sulfate mainly scatters solar radiation and BC mainly absorbs solar radiation, the effect of these two aerosol types on the surface radiation budget is thought to be on the same order of magnitude on a global scale. According to the aerosol transport-radiation model study of Takemura et al. [2005], clear-sky total (shortwave plus longwave) surface forcing (pre-industrial to present-day) is about −0.94 Wm−2 and −0.41 Wm−2 for BC and sulfate, respectively. Because of the positive longwave forcing of BC, shortwave BC forcing will be slightly larger (more negative) than the total forcing. Consequently, the fact that only 7 out of the 14 analyzed IPCC-AR4 models use BC emissions is likely to have a significant impact on the magnitude of the modeled SDRcf changes during the 20th century. Moreover, the crude assumptions made by some of the modeling groups to generate BC fields in the absence of accurate data become important. Brief descriptions of the spatial and temporal distributions of sulfate and BC aerosols over Europe used in the IPCC-AR4 models follow, with pan-European time series displayed in Figure 2.

Figure 2.

(a) SO2 emission (dashed lines) and SO4 burden (solid lines) from different data sources used in IPCC-AR4 simulations. Data represent pan-European averages (44 equal-area ISCCP grid boxes used in NW07) and are normalized to the 1960–1990 long-term mean for better comparability. (b) same as Figure 2a, but for BC.

2.3.1. Sulfur Dioxide/Sulfate

[15] In the GFDL_CM2.0/2.1 runs, SO2 emission data from the EDGAR V2.0 [Olivier et al., 1996] and EDGAR-HYDE V1.3 [Van Aardenne et al., 2001] are used according to the assimilation described by Horowitz [2006]. Sulfur dioxide emissions are most pronounced in the United Kingdom and eastern Germany. The U.K. emissions show decreasing trends since the 1970s, whereas pan-European emissions peak in 1990.

[16] The SO2 emissions provided by Nozawa et al. [2007] show decreasing tendencies in the United Kingdom since 1960, whereas pan-European emissions are increasing until 1980. Since then, pan-European SO2 emissions also decline, most distinctly in northwestern Germany.

[17] The Climate and Global Dynamics Division of the National Center for Atmospheric Research (CGD/NCAR) provides a frequently used data set of SO2 emission data constructed from country-level emissions inventories and regional fossil fuel sulfur content information by Smith, et al. [2001] and Smith et al. [2004] (available at These data show an emission increase from 1950 to 1970/1980, depending on the region. From 1980 to 2000 the emission decrease is most pronounced in the United Kingdom, northwestern and eastern Germany, and Poland.

[18] Four of the fourteen models make use of the sulfate burden data set provided by Boucher and Pham [2002], which generally shows an increase from the 1950s to 1980 in Europe. The magnitude of the trend gradually increases from northern Europe (United Kingdom and Scandinavian countries) to southeastern Europe (Bulgaria). A decrease of sulfate burden occurs from 1980 to 2000.

[19] The MRI_CGCM2.3.2 uses sulfate burden data from Mitchell and Johns [1997], which exhibits a larger increase over eastern Europe than western Europe and a maximum in eastern Poland and Romania. According to this data set, sulfate burden reaches its pan-European maximum in 1990.

[20] The sulfate burden data set used in MIUB_ECHO-G simulations, described by Roeckner et al. [1999], rises in Europe until the 1970s and lessens afterward. Sulfate burden is largest in eastern Europe (Czech Republic, Poland).

[21] Sulfur dioxide emission (dashed lines) and sulfate burden (solid lines) data used in the IPCC-AR4 20C3M experiments are summarized in Figure 2a. Time series are pan-European averages normalized by the 1960–1990 long-term mean of the corresponding data set to overcome the problem of differing input parameters used in the models. Although the lack of direct comparability between SO2 emission and SO4 aerosol burden due to chemistry and transport processes will introduce some uncertainty into the regional pattern and timing of the reversal from increasing to decreasing aerosols, we believe the short lifetime of the aerosols (around 1 week) will ensure that this uncertainty is small. It is apparent from Figure 2 that the differences between data sets providing either emissions or burdens are as large as the differences between burden and emission data.

2.3.2. Black Carbon

[22] Horowitz [2006] describes the BC aerosol assimilation for the GFDL_CM2.0/2.1 model runs. BC emissions decrease in the United Kingdom since the 1960s, whereas the pan-European decline does not start until after 1990. The United Kingdom and eastern Germany/Poland are the two main emission centers in the Horowitz [2006] assimilation.

[23] Nozawa et al. [2007] BC emissions are based on energy consumption. Pan-European emissions peak around 1970 and show a continuous decrease since then. Emissions are most pronounced in eastern Germany and Poland.

[24] The BC spatial distribution in the NCAR_CCSM3.0 model runs is described by Collins et al. [2002], and the temporal evolution of BC burden follows global average population and consequently does not show the transition from increasing to decreasing BC content as do the other data sets.

[25] In the CNRM_CM3 simulation [Salas-Mélia et al., 2005], BC concentration from Tanré et al. [1984] is scaled by the total fossil fuel BC emissions estimated by Novakov et al. [2003]. These data show highest concentrations at the end of the 1980s.

[26] Pan-European time series for BC emission (dashed line) and BC burden (solid line) are displayed in Figure 2b. The time series are normalized to the average from 1960 to 1990. Large differences occur between the data sets, which reveals the urgent need for more accurate BC emission histories for further studies.

3. Results

3.1. Comparison Between Models and Observations in the 20th Century

[27] Figure 3 depicts pan-European SDRcf time series from individual models and the average over all models (black line). If a model center provides several runs (see Table 1), the displayed time series is the average of these runs. Estimated SDRcf anomalies from NW07 (“residual flux”) are also shown (red line). The multimodel average exhibits a decline of about −1.5 Wm−2 dec−1 from the 1950s to the mid 1970s, then a leveling off, and finally a slight recovery after about 1980 (+0.8 Wm−2 dec−1). The leveling off, however, is actually an artifact of averaging rather than the behavior of individual models. The timing of the transition from “dimming” to “brightening” shows a wide range between the models, with the earliest reversal occurring in the early 1960s (MIROC3.2 models) and the latest around 1990 (MRI_CGCM2.3.2). According to the multimodel average, SDRcf did not reach the level it had in the middle of the 20th century until the end of the century.

Figure 3.

Pan-European SDRcf time series from observations (red), fourteen IPCC-AR4 20C3M simulations, and as average over the fourteen models (black).

[28] We now compare modeled trends with observational trends during the periods when solar “dimming” (1971–1983) and “brightening” (1984–1999) occurred according to the observations. Although solar “dimming” began before 1971, the lack of complete observational data prevents comparison prior to that time. Output is also not available from all models after 1999. Figure 4 presents trends for the observations (orange bars) as well as the individual models (red bars) during the “dimming” (Figures 4a, 4c, and 4e) and “brightening” (Figures 4b, 4d, and 4f) periods. The blue error bars indicate the 95% confidence interval of the linear trend for the average of all available runs from each model, and the green error bars indicate ± one standard deviation of the trends of the different runs performed by each model. Pan-European trends show a wide spread between models, ranging from SDRcf decrease of −2.3 Wm−2 dec−1 to an increase of +1.2 Wm−2 dec−1 for the period from 1971 to 1983, when solar “dimming” was observed (Figure 4a). The multimodel average underestimates the observed SDRcf decline.

Figure 4.

Bar plot of SDRcf trends for the (a, c, and e) solar “dimming” period (1971–1983) and (b, d, and f) “brightening” period (1984–1998) for observations (orange bars) and IPCC-AR4 20C3M simulations (red bars). The blue error bars indicate the 95% confidence interval of the linear trend of the average from each model; the green error bars indicate ± one standard deviation of the trend from the different model runs. The models with statistically significant differences from the observed residual flux are marked with an asterisk.

[29] The models show an even larger spread on sub-European scales. Trends for northern Germany range from −2.2 to +3.1 Wm−2 dec−1 (Figure 4c), and for the multimodel average, nearly no trend is found. It is not clear, however, which models are more realistic since the residual flux trend is also highly uncertain for this region and time period. Over Switzerland (Figure 4e), the models present a similarly inconsistent pattern, ranging from −1.6 to +2.1 Wm−2 dec−1. Lack of observational data does not allow us to compare the models with observations for this period and location. During the “brightening” period (Figures 4b, 4d, and 4f) observations and models exhibit better agreement. On average, the models reproduce the “brightening,” although the magnitude is smaller than is shown by the observations. Trends for the surface observations and residual flux are both more positive for northern Germany than for Switzerland, and only the two MIROC3.2 models capture this regional difference in trend magnitude. Although the observed trends in Switzerland are not significant for 1984–1999, they are significant and show similar magnitude if measurements until 2005 are considered.

3.2. Relation Between SDRcf Trends and Emission Histories

[30] To investigate how input emission histories affect the modeled solar “dimming” and “brightening,” the radiation trends from 1984 to 1999 are plotted versus the trends from 1971 to 1983 for each model for pan-Europe (Figure 5a) and from 1984 to 1999 versus 1960 to 1983 for northern Germany and Switzerland (Figures 5b and 5c). The different time spans have been chosen as a compromise between availability of observational data and more robust model trends due to longer time periods. Pan-European residual fluxes are available for both periods; the box shows the 95% confidence area of the trends. In northern Germany and Switzerland, observations are only available for the “brightening” period; their range is represented with the gray bars. Marker colors and marker types stand for the sulfate and BC data sets used as model input, respectively. The circle marker indicates models without BC aerosol. Error bars show the 95% confidence interval of the linear trends.

Figure 5.

Scatterplots of SDRcf trends (a) in pan-Europe from 1971 to 1983 versus 1984–1999, (b) in northern Germany, and (c) in Switzerland from 1960 to 1983 versus 1984–1999. Gray box indicates 95% confidence interval of residual flux tends in Figure 5a, whereas in Figures 5b and 5c, observations are only available for the solar “brightening” period. The marker colors indicate the SO2 or SO4 data set used in the simulations, and the marker form stands for the BC data set. Error bars denote the 95% confidence interval of the linear trends for the respective period.

[31] Aside from the GFDL_CM2.0/2.1 models, pan-European model trends are very significantly different from the residual flux trends, and solar “dimming” is especially poorly represented. Even the GFDL models have imperfect agreement, since the SO2 and BC input data used in GFDLCM2.0/2.1 models [Horowitz, 2006] show steadily increasing pan-European emission until 1990 (see Figure 2), which causes the largest SDRcf “dimming” trends but also the underestimation of the “brightening” trend. Contrastingly, the observations show that the reversal clearly occurs before 1990. Even when model output is processed according to the method of NW07 rather than using actual SDRcf, no better agreement with the observed residual flux trends exists (not shown).

[32] Large spatial variability in aerosol over Europe and the occurrence of nation-scale emission changes (e.g., United Kingdom and eastern Germany) may blur direct aerosol effects in the pan-European average. Focusing on smaller areas allows better comparison between modeled SDRcf trends and aerosol input data. In northern Germany (Figure 5b), only three models (MIROC3.2hires/medres and UKMO_HadGEM1) show trustworthy agreement with the observations for the “brightening” period. These three models use SO2 emission data from either Nozawa et al. [2007] or Smith et al. [2001, 2004] and are the only ones using BC emission data from Nozawa et al. [2007]. This emission data set has the most pronounced BC decrease for pan-Europe (see Figure 2b) and northern Germany (not shown). On the other hand, these three models incorrectly exhibit SDRcf increases for the expected “dimming” period, which is likely driven by the Nozawa et al. [2007] transition from increasing to decreasing BC emission in the early 1960s for northern Germany. All other models show negative trends for this period. In Switzerland (Figure 5c), where the observed trends are smaller, the pattern of model trends is less clear, but the MIROC3.2hires/medres and UKMO_HadGEM1 models again have large positive or the smallest negative trends for the “dimming” period.

[33] Would better emission data sets help the models reproduce solar “dimming” and “brightening” as seen in the observations? Do the models represent solar radiation trends according to their input aerosol emission histories? To answer these questions, we calculated model SDRcf trends for the time periods when their respective aerosol emissions / burdens were increasing and when they were decreasing. This takes into account the timing of the transition from “dimming” to “brightening” that differs from model to model, according on the emission / burden data set used. The 2 years following the eruptions of El Chichón in 1981 and Mount Pinatubo in 1991 have been omitted from linear trend analysis for the models using volcanic forcings (see Table 1), since these years can bias the trends in different ways depending on the onset of the “brightening”. Table 2 gives the number of models (out of all models) showing decreasing (Solar Dimming) and increasing solar irradiance (Solar Brightening) for the time periods during which the aerosol histories show rising and declining trends. Numbers in brackets indicate the models with statistical significant trends (95% confidence interval). Two of the fourteen models do not display an increase in SDRcf in northern Germany and in Switzerland for the period during which SO2 emission or SO4 burden shows a decline. These are CNRM_CM3 and IAP_FGOALS-g1.0 for northern Germany and GFDL_CM2.1 and IAP_FGOALS-g1.0 for Switzerland. CNRM_CM3 and GFDL_CM2.1 consider also BC aerosols, but with a later transition timing from increasing to decreasing emissions than sulfate. This caused the disagreement between the sign of the SDRcf trend and the reversal of sulfate burden. Hence, only the IAP_FGOALS-g1.0 model does not represent SDRcf according to the input sulfate burden data set. Considering the reversal from increasing to decreasing BC aerosols, all models using BC as forcing agent show signs of SDRcf trends that are consistent with their input BC data set.

Table 2. Number Out of the Total Models Showing Solar “Dimming” and “Brightening” Consistent With the Timing of Rising and Declining Sulfur Dioxide or Sulfate Aerosols and BC Aerosolsa
 Sulfur Dioxide/SulfateBC
Pan-EuropeNorthern GermanySwitzerlandPan-EuropeNorthern GermanySwitzerland
  • a

    With numbers of models having statistical significant (95% confidence interval) solar radiation trends in brackets.

Solar dimming14 (13)/1414 (13)/1414 (13)/147 (7)/77 (7)/77 (7)/7
Solar bightening13 (12)/1412 (10)/1412 (10)/146 (6)/66 (4)/66 (5)/6

[34] The negative correlation between aerosol emission/burden and SDRcf (Figure 6) strengthens the hypothesis that better aerosol data sets would improve the models' ability to represent solar irradiance trends. Correlation coefficients (R) are given for all data points and for “dimming” (data points in the upper half of the diagram) and “brightening” (data points in the lower half of the diagram) periods separately. For “dimming” and “brightening” periods the negative correlation is weaker than the overall correlation, but Figure 6 still indicates that higher (lower) percentage changes in aerosol are associated with higher (lower) trends in SDRcf. This suggests that, in addition to the model response to aerosol, the magnitude of the change in the input aerosol history is a major contributor to the change in SDRcf. Models with the similar radiation codes but different aerosol histories produce different solar radiation trends (not shown), and most models with similar aerosol histories produce similar solar radiation trends.

Figure 6.

(a) Negative correlation between SO2 emission/SO4 burden trends and SDRcf trends and (b) between BC emission/burden trends and SDRcf trends. SDRcf trends are calculated for the time periods when aerosol emissions/burden show increasing and decreasing tendencies; therefore they have different periods. Correlation coefficient (R) is given for all data points, for data points belonging the “dimming” (data points in the upper half of the graphs), and “brightening” period (data points in the lower half of the graphs).

[35] An investigation of seasonal trends in aerosol emission/burden and simulated SDRcf produced no additional information. This is because input aerosol histories have no or only negligible differences in the seasonal timing from increasing to decreasing emissions/burden. Modeled SDRcf trends consequently exhibit no notable seasonal differences apart from greater trend magnitude in seasons when insolation is larger.

4. Summary and Conclusion

[36] The large spread in the timing of the transition from solar “dimming” to “brightening” under cloud-free skies between the IPCC-AR4 models in 20C3M simulations reflects the variability of aerosol histories used in these simulations. For instance, sulfate burden or sulfur dioxide emission peaks anytime between the mid 1960s and 1990, and BC starts to decline in the early 1960s or shows a gradual increase through the second half of the 20th century. These facts reveal that the uncertainty of the aerosol forcing reported by Solomon et al. [2007] is not only related to the aerosol treatment in the models, but also to the aerosol histories used. Differences in SDRcf trends among models using the same aerosol histories are generally small (less than 0.7 Wm−2 dec−1). The maximum SDRcf trend difference among models using the same aerosol history is 2.5 Wm−2 dec−1 for the IAP_GGOALS-g1.0 model compared to the MPI_ECHAM5 and IPS_CM4 models. This is a further indication that the IAP_FGOALS-g1.0 model has difficulty representing the direct aerosol effect correctly.

[37] If the model trends are calculated for time periods when the respective aerosol history shows an increasing or decreasing trend, virtually all models reproduce the transition form solar “dimming” to “brightening” correctly. This leads to the conclusion that the models would much better reproduce observed solar “dimming” and “brightening” if the input data (i.e., aerosol history) were more realistic. At present, we cannot determine which emission or burden data set is most reliable for the European region, as some better represent the “dimming” period and some the “brightening” period. Nevertheless, aerosol histories should have their maximum emissions or burdens for Europe as a whole before the 1990s, in contrast to some of the data sets that have a maximum around 1990. Additionally, our analysis emphasizes the necessity to consider sulfate and BC aerosols to obtain the magnitude of the observed solar irradiance trends. This is consistent with the modeling results of Takemura et al. [2005], who show that BC has a slightly larger impact on the surface forcing than sulfate has under clear-sky conditions over land as well as on a global scale. The incorporation of newly available BC emission data sets [Bond et al., 2007; Junker and Liousse, 2008] will be a important step toward producing more realistic historical climate change simulations.

[38] Although we investigated trends only over Europe, we expect that input aerosol histories also exert a dominant influence on simulated trends in SDRcf on a global scale. Hence, considering the large differences between BC aerosol histories in the data sets used in the 20C3M simulations and the facts that fewer than half of the IPCC-AR4 models use BC as a forcing agent and fewer than one third of the models include indirect aerosol effects, a final question remains: Why do the models reconstruct the 20th century global surface temperature so well and so consistently [e.g., Hegerl et al., 2007]? This is especially the case if BC is thought to play such a crucial role on the surface temperature [Ramanathan and Carmichael, 2008]. One possible reason for the apparent good agreement may be compensation between high climate sensitivity and large negative aerosol radiative forcing and vice versa, as discussed by Kiehl [2007] and Knutti [2008]. Another possible reason may be compensation between errors in the magnitude of sulfate and BC aerosol forcing. A third possible reason is compensation between errors in the magnitude of aerosol direct and indirect radiative forcing. We did not examine trends in aerosol indirect forcing, but [e.g., Quaas et al., 2008] suggest that indirect aerosol effects might be generally overestimated, and fewer than a third of the IPCC-AR4 models include indirect aerosol effects. More research is needed, and in other regions of the world. We believe observed multidecadal variations in downward solar irradiance at the surface can provide a useful constraint on the time history of past aerosol forcing and thereby model climate sensitivity and future projects of climate change.


[39] The work of Christian Ruckstuhl is supported by the Swiss National Science Foundation (SNSF), Grant no. 119674. An NSF CAREER award, ATM02–38527, supported the work by Joel R. Norris. GEBA database is maintained at ETH Zurich. ISCCP cloud and radiation flux data were obtained from the NASA Goddard Institute for Space Studies. The authors thank MeteoSwiss for providing synoptic cloud observations and solar radiation data. We also acknowledge the modeling groups, the PCMDI and the WCRP's Working Group on Coupled Modeling (WGCM) for their roles in making available the WCRP CMIP3 multimodel data set. Support of this data set is provided by the Office of Science, U.S. Department of Energy. The authors are grateful for emission data and comments from Olivier Boucher, William Collins, Larry W. Horowitz, Seung-Ki Min, Tihomir Novakov, Jean-Francois Royer, David Salas, Gary Strand, Evgeny Volodin, Martin Wild, and Seiji Yukimoto.