Exploiting the weekly cycle as observed over Europe to analyse aerosol indirect effects in two climate models

A weekly cycle in aerosol pollution and some meteorological quantities is observed over Europe. In the present study we exploit this effect to analyse aerosol-cloudradiation interactions. A weekly cycle is imposed on anthropogenic emissions in two general circulation models that include parameterizations of aerosol processes and cloud microphysics. It is found that the simulated weekly cycles in sulfur dioxide, sulfate, and aerosol optical depth in both models agree reasonably well with those observed indicating model skill in simulating the aerosol cycle. A distinct weekly cycle in cloud droplet number concentration is demonstrated in both observations and models. For other variables, such as cloud liquid water path, cloud cover, top-of-the-atmosphere radiation fluxes, precipitation, and surface temperature, large variability and contradictory results between observations, model simulations, and model control simulations without a weekly cycle in emissions prevent us from reaching any firm conclusions about the potential aerosol impact on meteorology or the realism of the modelled second aerosol indirect effects.


Introduction
In its latest report, the Intergovernmental Panel on Climate Change confirmed its previous conclusion that aerosol indirect effects constitute the most uncertain anthropogenic forcing of global climate change (IPCC, 2007).Anthropogenic pollutant aerosols modify cloud optical properties by acting as cloud condensation nuclei.Specifically, aerosols are thought to increase the cloud droplet number concentration Correspondence to: J. Quaas (johannes.quaas@zmaw.de)(CDNC), enhancing the cloud albedo (first aerosol indirect effect, Twomey, 1974).This may also lead to a decrease in the precipitation formation rate, increasing the cloud liquid water path (LWP), cloud lifetime, and subsequently total cloud cover (second aerosol indirect effect, Albrecht, 1989).Modelling indicates that the two effects are of about the same order of magnitude (Lohmann and Feichter, 2005) but considerable uncertainties remain.In light of these uncertainties IPCC (2007) only quantified the cloud albedo effect, with a range of −1.8 to −0.3 Wm −2 , and a best estimate of −0.7 Wm −2 .Recent studies constraining the aerosol indirect effect by satellite observations (Lohmann and Lesins, 2002;Quaas and Boucher, 2005;Quaas et al., 2006), or estimating it from satellite data (Quaas et al., 2008) suggest that the indirect effect indeed may be on the upper side of this range (i.e. of a small magnitude).
Measurements of anthropogenic pollution show a weekly cycle in many countries of Europe (Cleveland et al., 1974;Beirle et al., 2003;Shutters and Balling Jr., 2006;Bäumer et al., 2008;Stephens et al., 2008;Xia et al., 2008;Barmet et al., 2009), in China (Gong et al., 2007), and in the United States (Murphy et al., 2008).In this latter study the weekly pattern is more pronounced for black carbon than for other aerosols.This weekly cycle in aerosol concentration is related to a weekly cycle in emissions, with reduced emissions on weekends compared to weekdays due to decreased industrial activity, less commuter traffic, and, in some European countries, a driving ban for heavy duty vehicles on Sundays.
In the absence of a natural forcing with a seven-day period, one has to invoke a human-made cause.The weekly cycle in heat generation itself is too small to explain a weekly cycle in meteorology, so a different mechanism is required, e.g. through atmospheric chemistry.The cycle in aerosol effects has been proposed earlier as a potential cause for a weekly cycle in meteorological quantities (Cerveny and Balling Jr., 1998;Jin et al., 2005;Bäumer and Vogel, 2007;Gong et al., 2007;Bell et al., 2008;Sanchez-Lorenzo et al., 2008), but according to the literature, cycles in greenhousegases (Cerveny and Coakley, 2002) and large-scale dynamics (Sanchez-Lorenzo et al., 2008;Laux and Kunstmann, 2008) might also play a role.

Method
We use here a combination of various kinds of observations and global climate modelling to analyse the aerosol indirect effects by exploiting the observed weekly cycle.The region chosen is Europe (approx.35 • N-70 • N, 10 • W-30 • E) and we restrict the analysis to land areas.
Data from 177 ground-based stations from the European Monitoring and Evaluation Program (EMEP; Hjellbrekke, 2008) are analysed to investigate the weekly cycle of aerosols and aerosol precursors using measurements of near-surface sulfur dioxide (SO 2 ) and sulfate (SO 4 ) concentrations.We use here the 24-h accumulated air concentrations measured by filter methods.The time period 1 January 2000-31 December 2006 is analysed.A list of the stations and their locations can be found on http://tinyurl.com/emepdata 1round-based observations of various meteorological quantities from 41 stations of the German Meteorological Service (Deutscher Wetterdienst, DWD) are also analysed.Daily average data for the time period 1 April 2001-31 December 2006 are used.Locations of the data and time periods are detailed on http://tinyurl.com/dwddata2 .A subset of these data has also been analysed by Bäumer and Vogel (2007).
Satellite data from the MODerate Resolution Imaging Spectroradiometer (MODIS; Remer et al., 2005;Platnick et al., 2003) and the Clouds and the Earth's Radiant Energy System (CERES; Wielicki et al., 1996) are also used.The data from the Terra satellite with a sun-synchronous orbit overpassing a point at the Earth's surface at about 10.30 a.m.local time cover the March 2000-December 2006 period for CERES andMarch 2000-March 2008 for MODIS, and data from the Aqua satellite overpassing at about 1.30 p.m., the July 2002-December 2006 period for CERES and July 2002-July 2007 for MODIS.We use the aerosol optical depth (AOD) and cloud properties from the MOD08 D3 (Terra) and MYD08 D3 (Aqua) collection 5 datasets, and broadband short-wave planetary albedo and outgoing longwave radiation from the CERES Single Scanner Footprint (SSF) Edition 2 dataset applying the User Applied Revisions Rev1 (FM1 for Terra and FM3 for Aqua).We compute for each day of the time period an area average for the European continental area for each of the datasets, from which we then analyse the weekly cycle.
The global climate models we use are a developmental version of HadGEM2-AML (Jones et al., 2007) and the ECHAM5 (Roeckner et al., 2003).HadGEM2-AML (Atmosphere/Mixed-Layer ocean) includes interactive aerosol models (Jones et al., 2007;Bellouin et al., 2007;Rae et al., 2007) and a representation of aerosol indirect effects (Jones et al., 2001).Cloud cover and condensate are diagnosed from total water and liquid water potential temperature assuming a triangular probability distribution function.The width of the distribution is diagnosed from the variability of the moisture and temperature of the surrounding grid points.Transfers between water categories (ice, liquid water, vapor, and rain) are calculated based on physical process equations using particle size information (Martin et al., 2006 and references therein).The ECHAM5 model includes a comprehensive aerosol module considering the mass concentrations of sulfate, sea salt, dust, black carbon and organic carbon as soluble or insoluble internal mixtures resolved in seven log-normal modes (Stier et al., 2005).The aerosol module includes a simplified sulfur chemistry in which the oxidation of SO 2 precursor gas is computed in dry and aqueous chemistry using prescribed oxidant distributions (Feichter et al., 1997).Aerosol-cloud interactions are parameterized using a two-moment cloud microphysical scheme (Lohmann et al., 2007), where droplet activation is parameterized following Lin and Leaitch (1997), and autoconversion following Khairoutdinov and Kogan (2000).
In HadGEM2, a 10-year perturbed simulation has been performed where climatological emissions of anthropogenic aerosols and aerosol precursors have been increased by 10.5% during weekdays and decreased by 26.3% on Saturdays and Sundays (peak-to-trough difference in emissions of 38%).This follows the suggestion by Bäumer et al. (2008) that week-end emissions are one third lower than weekday emissions.The changes in emissions are such that the weekly average in emissions remains unchanged.The last five years of the perturbed simulation are analysed in order to use a similar period length as in the observations.A 5-year control simulation has also been performed but only a few model variables have been saved with a daily resolution.Although the weekly cycle in anthropogenic emissions may be different in different parts of the world, the same weekly pattern -mostly representative of Europe and North America -has been applied everywhere but results are only considered over Europe.
With ECHAM5, two 5-year simulations with a 3-month spin-up have been carried out using the AeroCOM year 2000 emissions (Dentener et al., 2005) for natural and anthropogenic aerosols and observed monthly-mean prescribed sea-surface temperature and sea-ice cover distributions as boundary conditions.In the control simulation monthly-or annual-mean emissions have been used depending on the aerosol type.In the experiment investigating the influence of a weekly cycle, anthropogenic emissions over European land areas are reduced by 33% on Saturdays and Sundays, and increased during the weekdays accordingly (peak-to-trough difference in emissions of 46%).
Note that there is no diurnal cycle in emissions in either model.This may perhaps shorten the week-end effect in the models as compared to the real world.Also no weekly cycle is applied to biofuel and biomass burning emissions.

Results
The weekly cycles found in the observations are compared against those diagnosed in the model experiments with a weekly cycle of emissions and also against the control simulations in Figs.1-4.Table 1 further summarises for each analysed quantity the mean weekly range (weekly maximum -weekly minimum) and the weekday occurrence of the extreme values.All the ranges quoted in this section are peakto-trough differences; percentage changes are relative to the weekly mean.Statistical significance of the seven-day period, for each time series, is tested using a red-noise (AR1) fit to the associated power spectra (Priestley, 1981;Weedon, 2005), and in Table 1 the statistical significance level of a peak at a 7-day frequency is listed, if it is above the 95% or 99% significance level.
In the surface concentrations measured by the EMEP network, a significant weekly cycle in aerosol precursor gas (SO 2 ) and aerosol (SO 4 ) is found.The range of the weekly cycle in SO 2 is about 0.07 µg m −3 (9%), and 0.03 µg m −3 (4%) for SO 4 .The weekly cycle imposed on aerosol emissions in the climate models (38% and 46% range in HadGEM2 and ECHAM5, respectively) results in a weekly cycle in SO 2 which is too strong (overestimated by about a factor of 2.5) compared with the EMEP observations.The  weekly cycle observed for the sulfate aerosol (SO 4 ) concentration is much smaller than the one in SO 2 .The likely reason is that SO 2 is directly emitted, and with a residence time of about 1 day, variability in its concentration is directly linked to variability in the emissions.SO 4 , on the other hand, is produced in the atmosphere through chemical processing of SO 2 and has a longer residence time thus changes are delayed and smoothed out.Also, oxidation of SO 2 into SO 4 might be limited by the availability in oxidants so that the weekly increase in SO 2 may not translate into a proportionate increase in SO 4 .Finally variability in environmental conditions (e.g., the presence of clouds allowing for the much more efficient aqueous chemistry) would increase the variability in SO 4 concentration, thus making the periodicity SO 4 less visible.Both models show a reduction in the weekly range of the same order of magnitude (by a factor of 2) as in the observations.This shows that the processes in the sulfate aerosol schemes in both models are qualitatively well simulated.The observations show the SO 2 and SO 4 minima both occur on Monday compared with the marked Sunday minimum in the models.
In satellite-retrieved AOD, the range is 0.010 (5.6%) and 0.008 (4.6%) for MODIS on Terra and Aqua, respectively.This value is similar to the 4.0% range found by Xia et al. (2008) from the ground-based sunphotometer network For SO 2 , SO 4 , and AOD, the control simulations do show a weak weekly cycle.However, the weekly cycle simulated in the experimental simulation is clearly distinguished from this noise.HadGEM2 captures better than the ECHAM5 the gradual increase in SO 4 and AOD from the Sunday/Monday minimum to the Friday/Saturday maximum as shown in the observations.In ECHAM5, the increase in aerosols at the beginning of the week, and the decrease at the end, are clearly too sharp; indicating that the chemical processing from aerosol precursors to aerosols might be too fast in this model.Quaas et al., 2006) for both datasets with a minimum on Monday (Sunday for Aqua) and a variation of 5 cm −3 (Terra) and 10 cm −3 (Aqua).There seems to be no clear reason why the weekly cycle in CDNC should show a much larger range in the one dataset (afternoon orbit) than in the other one (morning orbit).Thus, the difference of a factor of two should rather be considered as uncertainty in the observation-based result.Both models show a weekly cycle in CDNC similar in minimum (Sunday) and amplitude (though overestimated) to the observations.This weekly cycle is clearly a feature of the experiment simulations compared with the control runs.The overestimation is comparable to the one found for AOD.This may be interpreted as a suggestion that the parameterization of aerosol activation in the models, or the link between aerosol and cloud droplet concentrations, is reasonably well simulated.
The satellite retrievals show a consistent weekly cycle in the cloud liquid water path, with a clear minimum on Saturdays/Sundays, and larger values during weekdays (Fig. 2, middle row).The ECHAM5 model shows a similarly large variability.However, the modelled variability in the experiment and control simulations is equally large, so that no conclusion about an aerosol effect can be drawn (HadGEM2 data are not available for LWP).For cloud fraction, the observa-  tions from Terra and Aqua disagree on a weekly cycle.Both models show variability of the same order of magnitude for both the experiment and control simulations.This appears to indicate that no distinguishable signal of a second aerosol indirect effect can be found in cloud fraction, and perhaps not even in LWP.
Both direct and indirect aerosol effects influence the planetary albedo.Variability in planetary albedo for a given location is dominated by variability in cloud cover, cloud water path, CDNC, and aerosol concentration.As can be seen in Fig. 3 (top row), the observations from CERES on Terra show a weekly cycle in albedo which would be consistent with an influence of anthropogenic aerosols on albedo.However, a low value on Wednesdays, and the less clear cycle in the Aqua data indicate that this finding is not very robust.The ECHAM5 model shows a weekly cycle in albedo which is consistent with that found in the Terra observations, although the timing of the minimum (Monday vs. Saturday) is different.On the other hand, HadGEM2 shows a weekly cycle of the opposite sign.The likely reason is that in the simulation, the cloud cover happens to show a weekly cycle with a minimum during weekdays and a maximum during weekends, which is reflected in the albedo cycle.
The direct aerosol effect can be best seen in clear-sky conditions.For this, CERES footprints labeled as cloud-free have been selected to compute the weekly cycle in the clearsky albedo (Fig. 3, middle row).The weekly cycle found is not very robust (particularly in the Aqua observations), and the ECHAM5 model shows a cycle opposite to the expectation.Thus, the short available timescale is not sufficient to distinguish direct aerosol effects in the weekly cycle.
There have been speculations whether aerosols might invigorate convection, increasing cloud top height, which would lead to an enhanced cloud greenhouse effect (reduced outgoing longwave radiation; Devasthale et al., 2005;Koren et al., 2005).As shown in Fig. 3 (bottom row), such an effect cannot be confirmed by any discernible weekly cycle in the OLR data available for this study.
When analysing the five-year period comparable to the model simulations, the DWD station data show a weekly cycle which is not consistent with the expectation that the daily maximum (daytime) temperature would show a minimum on weekdays when aerosol effects are largest.A weekly cycle of different phase found in the ECHAM5 model simulation.Also, the control experiment with no weekly cycle in the aerosol emissions also happens to show a very similar cycle.While this is an uncanny coincidence, there is no reason why the control simulation could show a weekly cycle, and a fortuitous instance of natural variability is the only explanation.Similarly, variability of at least the same order of magnitude is found in the weekly cycle of the near-surface daily mean temperature in the observations and in both the experiment and control simulations from both models.The conclusion is that the available time series do not show a discernible aerosol-influenced weekly cycle in near-surface temperatures, despite the fact that the statistical analysis suggests a statistically significant seven-day period.. Compared to other variables the uncertainty in individual mean daily observed precipitation is unusually large.The apparent weekly cycle in this variable seen in Fig. 4 is not confirmed by the spectral analysis.However, the phase of this oscillation is consistent with a weekly cycle in the effect of aerosols on the cloud lifetimes.On the other hand, Also, such a weekly cycle in precipitation is not consistent with the cycles found for either LWP or cloud cover.Consequently, a weekly cycle in observed precipitation has yet to be confirmed.

Summary and conclusions
Weekly cycles have been analysed in surface observations of sulfur dioxide and sulfate concentrations, satellite observations of aerosol optical depth, cloud properties (cloud droplet number concentration, cloud liquid water path, total cloud cover), radiation (albedo, clear-sky albedo, and outgoing long-wave radiation), and surface observations of meteorological quantities (daily maximum and mean temperatures, and precipitation).The same quantities have been simulated in two different general circulation models (HadGEM2 and ECHAM5), where in an experiment simulation the aerosol emissions have been reduced during weekends (Saturday/Sunday) and increased during weekdays accordingly, yielding a range in emissions of about 40%.Control experiments without a weekly cycle in aerosol emissions have been performed for comparison.
A clear weekly cycle is observed in aerosol quantities (surface concentrations of SO 2 and SO 4 , and satellite-retrieved AOD),.The imposed weekly cycle in aerosol emissions in the models leads to an overestimation in the range of the weekly cycle of surface concentrations of SO 2 , SO 4 , and AOD.Both models and observations show a stronger cycle in SO 2 than in SO 4 and AOD, indicating that the models simulate the sulfate aerosol processes qualitatively well.Statistical analysis of the power spectra shows that the weekly cycles in aerosols in most of the datasets are significant (Table 1).In comparison to hypothetical 6-and 8-day weeks (Supplementary Fig. 1 and 3, http://www.atmos-chem-phys.net/9/8493/2009/acp-9-8493-2009-supplement.zip), the signal in the 7-day week is clearly distinguished.
The aerosol indirect effect is reflected in a clearly distinguishable weekly cycle in CDNC, which both models simulate with some skill.A weekly cycle consistent with a second indirect effect is found for LWP in the satellite observations, but not in total cloud cover.The variability in both quantities in both the experiment and control simulations suggests that no conclusion about a second indirect effect can be drawn from the data analysed here.Also, power spectra analysis and comparison to hypothetical 6-and 8-day weeks show that while there is some confidence in a weekly cycle in CDNC, this is not the case for either LWP or cloud cover.
For the planetary albedo, the observed weekly cycle is also uncertain.On the one hand ECHAM5 captures some consistent cycle in the experiment simulation, with less variability in the control simulation.On the other hand, the HadGEM2 shows a weekly cycle inconsistent with the observations, likely due to the variability in total cloud cover.The range in planetary albedo found in ECHAM5 is similar to the one found in the Terra observations despite the fact that the range in CDNC and AOD is overestimated.For the clear-sky planetary albedo, OLR, temperature (both daily maximum and mean), and precipitation, no clear weekly cycle was found.In the cases where the observations seem to show a weekly cycle, the models show variability of equal magnitude in both the experiment and control simulations.
In summary, clear weekly cycles have been found in aerosols and cloud droplet number concentration, and the models confirm that the contrast in emissions between weekdays and weekends leads to such a cycle.Cycles related to second aerosol indirect effects (LWP, cloud cover) or thermodynamic effects of aerosols on clouds (OLR) are not distinguishable in the datasets.Similarly, our results do not support the attribution of observed weekly cycles in temperatures (either maximum or daily-mean) and precipitation to aerosol effects.It could be that such observed weekly cycles are accidental.A mixed result has been found for planetary albedo, where one model does show a weekly cycle consistent with the observations, while the other one does not.
Our results suggest that weekly cycles in cloud liquid water path, cloud cover, surface temperature, precipitation and radiation quantities cannot readily be attributed to aerosol indirect effects, at least not given the current state of the art as implemented in global climate models.Observations of weekly cycles will become even more useful to evaluate climate model parameterizations as data records obtained from high-quality satellites, which are able to observe the Earth system on the large scale, cover even larger statistics.

Fig. 2 .
Fig. 2. As Fig. 1, but for cloud droplet number concentration (top row), liquid water path (middle row), and total cloud cover (bottom row).Observations are from MODIS satellite data on Terra (red) and Aqua (blue).

Fig. 4 .
Fig. 4. As Fig. 1, but for daily maximum near-surface temperature (top row), daily-mean near-surface temperature (middle row), and precipitation (bottom row).Observational data are from the DWD over Germany.
Su Mo Tu We Th Fr Sa Su Mo Tu We Th Fr Sa Fig. 1.Weekly cycle (percentage deviation from mean value) of surface SO 2 concentration (top row), surface SO 4 concentration (middle row), aerosol optical depth (bottom row).Left column: Observations (EMEP for SO 2 and SO 4 , and MODIS Terra (red) and MODIS Aqua (blue) for AOD), middle column: HadGEM2 model (experiment with weekly cycle in anthropogenic aerosol emissions in red, control run in dashed grey; no control run output available for SO 2 and SO 4 ), right column: ECHAM5 model (experiment in red, control in dashed grey).

Table 1 .
Ranges of the weekly cycles in observations/satellite retrievals and as simulated by the imposed emission cycle with a range of 46% in ECHAM5 and of 38% in HadGEM2, respectively.Please note that most of the EMEP stations are located in Central Europe, and that the models apply the same range in anthropogenic emissions throughout the entire area.
tent than that in the weekly cycle in SO 4 .This could indicate that natural aerosols -which should not exhibit any weekly cycle -are underestimated in the model or that the lifetime of anthropogenic aerosols is too short in the models.