References

Cloud climate feedback constitutes the most important uncertainty in climate modelling, and currently even its sign is still unknown. In the recently published report of the intergovernmental panel on climate change (IPCC), from 20 climate models 6 showed a positive and 14 a negative cloud radiative feedback in a doubled CO2 scenario. The radiative budget of clouds has also been investigated by experimental methods, especially by studying the relation of satellite observed broad band shortwave and longwave radiation to sea surface temperature. Here we present a new method for the investigation of the dependence of cloud properties on temperature changes, derived from spectrally resolved UV/vis satellite observations. Our study differs from previous investigations in three important ways: first, we directly extract cloud properties (amount and altitude) and relate them to surface temperature. Second, we retrieve the cloud altitude from the atmospheric O2 absorption instead from thermal IR radiation. Third, our correlation analysis is performed using 7.5 years of global monthly anomalies (with respect to the average of the same month for all years). For most parts of the globe (except the tropics) we find a negative correlation of cloud fraction versus surface-near temperature. In contrast, for the cloud top height a positive correlation is found for almost the whole globe. Both findings might serve as an indicator for an overall positive cloud climate feedback. Another peculiarity of our study is that the cloud-temperature relationships are determined for fixed locations (instead to spatial variations over selected areas) and are based on the "natural" variability over several years (instead the anomaly for a strong El-Nino event). Thus our results might be especially representative for the extrapolation to long term climate changes. Climate models should aim to reproduce our findings: if substantial differences are found, this might indicate that important details are not yet well captured by these models. If good agreement is found, from the models reliable information on the magnitude and the detail mechanisms of cloud climate feedback could be gained.

vestigations in three important ways: first, we directly extract cloud properties (amount and altitude) and relate them to surface temperature. Second, we retrieve the cloud altitude from the atmospheric O 2 absorption instead from thermal IR radiation. Third, our correlation analysis is performed using 7.5 years of global monthly anomalies (with respect to the average of the same month for all years). For most parts of the globe 15 (except the tropics) we find a negative correlation of cloud fraction versus surface-near temperature. In contrast, for the cloud top height a positive correlation is found for almost the whole globe. Both findings might serve as an indicator for an overall positive cloud climate feedback. Another peculiarity of our study is that the cloud-temperature relationships are determined for fixed locations (instead to spatial variations over se- 20 lected areas) and are based on the "natural" variability over several years (instead the anomaly for a strong El-Nino event). Thus our results might be especially representative for the extrapolation to long term climate changes. Climate models should aim to reproduce our findings: if substantial differences are found, this might indicate that important details are not yet well captured by these models. If good agreement is found, from the models reliable information on the magnitude and the detail mechanisms of cloud climate feedback could be gained. ature (ST) with opposite signs, and their net effect depends on various cloud properties, especially on the cloud altitude (Stephens, 2005;Cess et al., 1992;Hartmann et al., 2001;Fu et al., 2002;Cess and Udelhofen, 2003;Webb et al., 2006;Kubar et al., 2007). Today, the magnitude, and even the sign of cloud feedback is still not known (see e.g. the IPCC 4th assessment report (Solomom et al., 2007) and references therein). 10 Besides modelling studies (Cess et al., 1990;Soden and Held, 2006;Webb et al., 2006;Cess et al., 1996;Larson and Hartmann, 2003;Bony and Dufresne, 2005;Williams et al., 2006), climate feedbacks have been investigated also from experimental data and their combination with model results. Most studies investigated the relation of broad band short wave and long wave radiation and cloud properties to 15 spatial variations of the surface temperature (Ramanathan et al., 1989;Harrison et al., 1990;Stephens and Greenwald, 1991). Other studies investigated seasonal variations, anomalies during strong El-Nino events or temporal trends (Cess et al., 1992(Cess et al., , 2003Bony et al., 1997a, b;Tian and Ramanathan, 2002). Most of these studies were restricted to specific areas, e.g. the continents or the tropics. 20 In this study we investigate the cloud response for fixed locations on a global scale using correlation analyses of monthly anomalies of cloud fraction and cloud top height (derived from the atmospheric O 2 absorption) with those of ST. The cloud fraction and cloud top height information is derived from 7.5 years of observations of the Global Ozone Monitoring Experiment (GOME) on board the European research satellite ERS-25 2. We relate the GOME results to ST observations for the same period (ST data are obtained from the Goddard Institute for Space Studies (Hansen et al., 2001;Reynolds et al., 2002)  The GOME instrument aboard the European research satellite ERS-2 measures sunlight reflected from the Earth's atmosphere and surface covering the wavelength range between 240 and 790 nm with moderate spectral resolution (Burrows et al., 1999) (0.2-0.4 nm FWHM). The satellite operates in a nearly polar, sun-synchronous orbit 5 at 780 km altitude with an equator crossing time of approximately 10:30 a.m. local time. This has to be taken into consideration for the interpretation of our results, which might be only representative for mid-morning because of the diurnal variation of clouds (Bergman and Salby, 1996). The ground pixels cover an area of 320 km east to west by 40 km north to south. Simultaneous to the spectral channels, also broad band intensi-10 ties are measured by the so called polarization monitoring devices (PMD). Compared to the spectral channels, they have a much finer spatial resolution of 20×40 km 2 . The Earth's surface is entirely covered within 3 days, and poleward from about 70 • latitude within 1 day.
3 Cloud products derived from GOME 15

Cloud fraction
An effective cloud fraction (CF) is derived using the Heidelberg Iterative Cloud Retrieval Utilities  (HICRU), which is based on the PMD observations. The CF is a product of the cloud top albedo (depending on the cloud optical depth) and the geometrical cloud fraction; it is almost independent from cloud height. From 20 sensitivity studies we estimate the absolute uncertainty of an individual measurement over surfaces free of snow and ice to range from 2% for almost clear pixels to 5% for CF>20% (expressed in absolute CF). Introduction The O 2 absorption at 630 nm is analysed using differential optical absorption spectroscopy (Platt, 1994) (DOAS); details of the retrieval can be found in Wagner et al. (2006a). One specific advantage of the DOAS method is that it is based on differential absorption structures and is therefore hardly affected by instrumental degra-5 dation. The statistical uncertainty of the retrieved O 2 absorption is negligible (<0.3%) for monthly mean values. The observed O 2 absorption depends on CF and the effective cloud top height (CTH) of the cloudy part of the observed scene.

Conclusions References
3.3 Retrieval of the cloud top height using radiative transfer modeling The calculation of the effective cloud top height (CTH) from the observed effective cloud 10 fraction (CF) and O 2 absorption is based on the results of atmospheric radiative transfer modelling of the O 2 absorption. While the results of such simulations provide the best means for the interpretation of the O 2 absorption measurements, it is also instructive to discuss the general dependencies of the observed O 2 absorption on cloud properties: 1) First observed O 2 absorption depends on the CTH: with increasing CTH an in- 15 creasing fraction of the atmospheric O 2 absorption is shielded. Thus high clouds strongly reduce the observed O 2 absorption (shielding effect).
2) Second, the observed O 2 absorption depends on the cloud top albedo: with increasing brightness of the cloud top (depending mainly on the cloud optical depth) an increasing number of the measured photons originates from atmospheric layers at the 20 cloud top (having "seen" a larger O 2 absorption). In addition, multiple scattering inside the clouds (also depending on the cloud optical depth) can further increase the atmospheric absorption paths. Both effects increase the observed O 2 absorption (the strongest increase is found for large cloud optical depth) and we will refer to them as albedo effect. For low cloud altitudes the albedo effect can even overcompensate the 25 shielding effect.
3) Third, the measured O 2 absorption depends on the CF. The higher the CF is, the Introduction EGU higher is the deviation of the measured O 2 absorption from the value for the cloud free case.
We investigated the dependence of the measured O 2 absorption on CF and CTH in a quantitative way using our full spherical Monte-Carlo radiative transfer model TRACY-2 (Deutschmann and Wagner, 2006;Wagner et al., 2007). For the modeling of the 5 O 2 absorption of an individual satellite ground pixel, we use the independent pixel approximation: the O 2 absorption of the total ground pixel is the average of the modeled O 2 absorptions for the clear and for the cloudy parts (weighted by their geometrical fractions and their respective top of atmosphere albedos). For the modeling of the cloudy part we assumed the following standardised cloud properties: optical depth: 10 50, asymmetry parameter: 0.85; single scattering albedo: 1, vertical thickness: 1 km. Note that the albedo of these clouds is about 80%. Finally, the O 2 absorptions are normalised with respect to the maximum O 2 absorptions found for the same solar zenith angle. This normalization has the advantage that the calculation of the CTH can be performed using a single inversion scheme for a wide range of solar zenith an-15 gles (SZA<80 • ). In the normalization procedure we also applied a correction for the effect of non-linearity between the measured O 2 absorption (for the resolution of the instrument) and the corresponding integrated O 2 concentration along the atmospheric absorption path (Wagner et al., 2003;. The calculations were performed for three different values of the surface albedo (2%, 20 10%, 30%). The modeled O 2 absorptions as a function of CF and CTH are shown in Fig. 1. The calculated relationships -in particular for low CF -are strongly dependent on the surface albedo. Based on the modeled dependence of the O 2 absorption we calculated the global distribution of the mean CTH from the average values of the CF and the O 2 absorption 25 for the period January 1996-July 2003 (Figs. 2 and 3). For these calculations we also made use of the average values of the surface albedo as derived by Koelemeijer et al. (2003) (see Fig. 4). For each 2×2 degree grid of the globe the CTH was determined by linear interpolation of the modeled dependencies ( Fig. 1)  EGU values of the CF, CTH and surface albedo. The determined global map of CTH is shown in Fig. 5. Note that in our current algorithm, we had to exclude continental areas with low CF (<7%) and high surface elevation (>1.2 km). For such conditions, the measured O 2 absorption is reduced, because the total atmospheric column is smaller than for zero elevation. In our current algorithm, this decreased O 2 absorption would be 5 apparently interpreted as shielding effect due to clouds. In connection with the strong dependence of the O 2 absorption on CF (see Fig. 1) this would lead to unrealistically high CTH. In future versions of our algorithm this limitation can be, however, generally corrected using information on the global surface elevation and respective adaptation of the radiative transfer modeling.
It should be noted that the derived global distributions of CF and CTH show overall good agreement with those of the ISCCP (Schiffer and Rossow, 1983;Rossow and Schiffer, 1999, http://isccp.giss.nasa.gov/climanal.html). It is important to keep in mind that no exact agreement can be expected because of the different sampling times, spectral ranges, spatial resolutions and methodologies.

Results
In this chapter we present correlation analyses of the cloud properties derived from GOME observations with ST. They are performed as follows: from the individual observations of the CF and O 2 absorption (and also ST) we calculated monthly average values for the period January 1996 to July 2003. From these monthly averages, anoma-20 lies for individual months are derived (deviation from the mean value of the respective month for all years). Using these monthly anomalies we performed correlation analyses of the CF versus ST and the O 2 absorption versus ST. Note that the amplitude of the ST anomalies for the selected period ranges between about 2 K and 15 K, depending on location on earth, with the highest amplitudes over the continents (see Fig. 6).
Global maps of the calculated correlation coefficients of the CF and the O 2 absorption versus ST are shown in Fig. 7. Over large parts of the globe negative correlations ACPD 7,2007 Dependence of cloud fraction and cloud height on temperature EGU are found for both quantities indicating that a positive (negative) anomaly of ST is associated with a negative (positive) anomaly in CF or O 2 absorption. One prominent exception is the pronounced positive correlation of the CF versus ST over the tropical oceans close to the equator. From the correlation analyses we also determined the slopes of the regression of 5 the monthly CF anomalies versus those of the ST (Fig. 8, top). The changes in CF with increasing ST are negative for most parts of the globe, except over the tropical oceans close to the equator. These findings are in good agreement with those of Bony et al. (1997a), who found a negative correlation of CF (and cloud optical depth) for ST<26 • C and a positive correlation for ST>26 • C. 10 We also determined the slopes of the regression for the O 2 absorption anomalies versus those of the ST (see Fig. 9). From these correlations and those of the CF versus ST (Fig. 8, top), we calculated the changes in the CTH per degree ST (Fig. 8, bottom) using radiative transfer modelling. Over almost the whole globe, positive temperature anomalies are correlated with increases of CTH. Very strong changes in CTH 15 are found over the tropical oceans close to the equator. Again our findings are in good agreement with those of Bony et al. (1997a), who found a weak positive correlation of CTH for ST<26 • C and strong positive correlation for ST>26 • C. They find a strong increase of cloud top height (from about 500 mb to 200 mb) for a temperature increase from about 26 • to 29 • . Also Larson and Hartmann (2003) found an increasing CTH of 20 tropical clouds for increasing ST. They found in particular that over tropical oceans, the frequency of situations with large scale uprising air and high clouds increases strongly for surface temperatures >26 • C.

Results for time series without the strong ENSO 1997/98
It is interesting to compare the results of the correlation analyses for periods including 25 or excluding the strong ENSO event 1997/98. In Fig. 10  EGU over most parts of the globe the results are almost identical. Only over parts of the tropical oceans substantial differences are found. These findings indicate that results of our correlation studies are representative also for the natural temperature variations beyond strong ENSO events.
4.2 Discussion of various error sources 5 As stated in Sects. 3.1 and 3.2, the errors of the data products (CF and O 2 absorption) are typically by far smaller (2% to 5% for the CF, and ∼0.3% for the O 2 absorption) than the observed dependencies per degree ST change. Nevertheless, there are a number of additional error sources (especially important for the correlation analyses), which are discussed in detail in this section.

10
One important question is whether the observed variation of the O 2 absorption might be simply caused by the changes of atmospheric pressure and temperature associated with variations of ST. In particular at mid latitudes cloud free situations are e.g. typically accompanied by high pressure systems. However, from a quantitative analysis it turns out that the magnitude (and for most regions also the sign) of the observed O 2 ab-15 sorption changes can not be explained by the direct influence of these meteorological parameters. It should be also noted that for large SZA (> about 80 • , corresponding to latitudes poleward of about 60 • ), several effects can lead to increasing uncertainties compared to observations at lower SZA. First, because of the long atmospheric light paths, the relative contribution of Rayleigh scattering to the total observed radi-20 ance increases. Second, because of the increased O 2 absorption, the non-linearity between the measured O 2 absorption and the integrated O 2 concentration along the atmospheric absorption path increases (see above). Third, and probably most important, the probability of interference with snow and ice-covered surfaces increases. The first two effects might lead to an error in the magnitude (but not the sign) of the 25 calculated CTH change. The third effect, however, might even affect the sign of the calculated change in CTH (and also CF) per degree ST for some regions of the world. We estimate this potential source of errors from individual correlation analyses for different 17125 Introduction In our radiative transfer calculations simplified homogenous cloud layers were assumed. In the presence of complex cloud formations, the photon paths between separated cloud fragments can become important because they can enhance the O 2 absorption. In the case of horizontally broken cloud fields, the additional horizontal photon 5 paths typically increase the O 2 absorption compared to a homogenous cloud layer with similar total optical depth. Taking this effects into account, the observed decrease of the O 2 absorption with increasing ST might be partly also caused by a change in the horizontal homogeneity of the cloud cover (towards a horizontally more homogenous distribution). However, from sensitivity studies using our radiative transfer model, we 10 find that the magnitude of the observed changes in the O 2 absorption can hardly be explained by this effect.
Also vertical inhomogeneities (cloud layers) can increase the O 2 absorption due to enhanced photon paths between separate cloud layers. From radiative transfer calculations we found that this effect is especially pronounced if rather thin cloud layers 15 (optical depth <5) are located above cloud layers with higher optical depth. For such cases, an increase of the height of the thin cloud layer can even increase the O 2 absorption (the shielding effect is then slightly overcompensated by the increased path lengths of the photons "reflected" between the cloud layers). Thus, in principle, an observed decrease of the O 2 absorption could even indicate a decrease of a thin cloud 20 layer above the thick cloud layer located below. However, again this effect is rather small, and can by far not explain the observed changes in the O 2 absorption. We conclude that only a change in the height of cloud layers with high optical depth (optical depth > about 10) can explain the observed large changes in the O 2 absorptions due to the shielding effect. Changes of the inhomogeneities in vertical and horizontal direction 25 might explain only part of the observed O 2 absorption changes.
In summary, we conclude that except in polar regions (see Sect. 4.3), the results of the correlation analyses can be interpreted in a quantitative way for the CF. For the CTH, the results of the correlation analyses can be only considered as qualitative EGU information. However, we believe that even qualitative information is very useful here, since currently even the sign of cloud feedback is still unknown.

Results derived for specific seasons
At mid and high latitudes the results of our correlation analyses can be potentially influenced by the variations of snow and ice cover during winter. This influence becomes 5 especially important, because it can be expected that for positive temperature anomalies the probability of snow and ice should be reduced compared to average conditions. As a consequence, the CF should be apparently decreased (because the HICRU algorithm can not discriminate between bright surfaces and clouds) leading to a negative correlation between CF and ST. Similarly, the O 2 absorption should be also reduced 10 for reduced snow and ice cover because of the albedo effect. Again this would lead to a negative correlation between the O 2 absorption and ST. Because of these potential interferences, the results of our correlation analyses at high latitudes must in principle be treated with care during winter months. One possibility to estimate the potential error caused by snow and ice is to perform the correlation 15 studies for separate seasons. The corresponding results for such seasonal correlation studies of CF with changing temperature are shown in Fig. 11. The results for most parts of the world are very similar (but with larger amplitudes and larger scatter) as analyzed for the whole time series (Fig. 8, top). However, especially in the northern hemispheric winter, also positive correlations can be found between the CF and 20 ST over parts of northern America and Siberia. Since the possible interference with snow and ice should lead to a negative feedback, we can conclude that the ice-and snow-induced interference should be rather small.
The corresponding results for the seasonal correlations of the O 2 absorption with changing temperature are shown in Fig. 12. Again, the results are very similar to those 25 obtained for the whole time series (Fig. 9). For all seasons a negative correlation between the O 2 absorption and ST is found for northern hemisphere high latitudes. From these results, we can not rule out that part of these negative correlations might 17127 Introduction EGU be caused by interference with snow and ice. However, from the seasonal correlation analyses for the CF and ST we conclude that the influence of ice and snow should be rather small. In Fig. 13 the changes in CTH associated with variations in ST are shown.

Conclusions
The observed dependencies of CF and CTH on surface-near temperature can provide 5 information on cloud climate feedback. However, since cloud climate feedback involves complex interactions, depending especially on cloud type, location and season, the interpretation of our results with respect to climate feedback is not straightforward. Therefore, the major conclusion of our study is that the results should aim to serve as a test case for climate models. Another limitation of our results is that they are only 10 representative for a fixed time of the day. Nevertheless, it is instructive to draw some general conclusions from the observed correlations. We therefore relate our results to the general dependencies of cloud forcing, which can be summarised as follows: a) in general, clouds tend to cool the Earth-atmosphere-system. However, this depends especially on season and latitude. The cooling effect of clouds is especially strong in the summer hemisphere. (Stephens, 2005;Ramanathan et al., 1989;Harrison et al., 1990). b) cloud heating increases with increasing cloud top height (Stephens, 2005;Cess et al., 1992;Kubar and Hartmann, 2007). 20 Based on the first dependency, no clear general conclusions can be drawn here. The only significant conclusion might be that especially in the summer hemisphere, the negative correlation between ST and CF indicates a positive cloud feedback (see results in Sect. 4.3). From the second dependency, however, we conclude that over almost the whole globe, our results indicate a positive cloud feedback. 25 One important general question is whether the observed dependencies might be extrapolated to a future temperature increase caused by climate change. While this ques-ACPD 7,2007 Dependence of cloud fraction and cloud height on temperature

EGU
tion is in general difficult to answer, our data sets might have some specific advantages to address this aspect. They describe cloud-temperature relationships for the "natural" variability at fixed locations over 7.5 years (while other studies investigate changes associated with strong changes associated with e.g. volcanic eruptions or ENSO events, or investigating the spatial variability). Thus our results might be especially well suited 5 for long term future predictions. Indeed, similar relationships between the CF, O 2 absorption and ST are found in the spatial distribution of trends (Wagner et al., 2006b) for the period January 1996-December 2002. However, the only means to provide reliable future predictions, are results of climate models. Our results should thus be used as a test case for such models. If these models are not able to reproduce the observed   Bony, S., Dufresne, J.-L., Emori, S., Gudgel, R., Knutson, T., Li, B., Lo, K., Musat, I., Wegner, J., Slingo, A., and Mitchell, J. F. B.: Evaluation of a component of the cloud response to climate change in an intercomparison of climate models, Clim. Dynam., 26, 145-165, doi:10.1007/s00382-005-0067-7, 2006 ACPD 7,2007 Dependence of cloud fraction and cloud height on temperature     ACPD 7,2007 Dependence of cloud fraction and cloud height on temperature   Fig. 8, but excluding the strong ENSO event (August 1997-July 1998). Even over the tropical oceans the differences compared to Fig. 8