Earth's radiative energy balance (solar radiative energy absorbed and terrestrial radiation emitted to space) determines current patterns of weather and climate, the complexity of which is illuminated by satellite observations of the evolving distribution and diversity of cloud structures. Representing clouds and the physical processes responsible for their formation and dissipation is vital in numerical weather and climate prediction, yet many approximations must be made in these detailed models of our atmosphere (e.g. Bony et al., 2006; Allan et al., 2007). Observations of cloud characteristics from satellite instruments and in situ or ground-based measurements are crucial for improving understanding of cloud processes and their impact on Earth's radiative energy balance (Sohn, 1999; Jensen et al., 2008; Su et al., 2010). The energy exchanges associated with cloud formation and precipitation are also a key component of the global water cycle, of importance for climate change (Trenberth, 2011).
In this paper, initially presented at a joint meeting of the Royal Meteorological Society and Institute of Physics on Clouds and Earth's Radiation Balance (Barber, 2011), the utility of combining weather forecast model output with satellite data in estimating the radiative effect of cloud is highlighted. Using a combination of models and satellite data a simple question is addressed: how do clouds influence the radiative energy balance of the atmosphere and the surface.
As an example of the radiative impact of cloud, Figure 1 displays thermal infra-red and visible channel narrow-band images of the European region from the Spinning Enhanced Visible and Infra-Red Imager (SEVIRI) on board the Meteosat-9 satellite (Schmetz et al., 2002). In both images clouds appear bright: this denotes relatively low infra-red emission to space and relatively high reflection of visible sunlight to space. The hot, generally clear regions of northern Africa are also noticeable in both images since they are associated with substantial thermal emission to space (dark regions in the infra-red image) and high surface reflection from the desert surface (bright in the visible image).
The brightest clouds in the thermal image correspond with (1) a trailing cold front extending from the coast of Norway, across Scotland and to the west of Ireland, (2) a developing low pressure system to the west of Iceland, and, (3) a low pressure system in the Mediterranean centred on Sardinia. These are regions of ascending air with relatively high altitude, low temperature cloud tops which depress the thermal emission to space compared with surrounding regions. These features are also present in the visible image. However, many more cloud structures are also present. There is a prevalence of low altitude cloud over the oceans: this cloud contains large amounts of water droplets which are highly reflective (e.g. Stephens et al., 1978). The imagery captures the complex cellular structure of this cloud (e.g. Jensen et al., 2008) over the region surrounding the Canary Islands. These cloud types are thought to contribute strongly toward uncertainty in climate projections (Bony et al., 2006). While these clouds also strongly attenuate infra-red radiation, their impact on the thermal radiation escaping to space is modest since cloud-top temperatures are not dissimilar to the surface at night and so they do not contribute significantly to the strong natural greenhouse effect of the clear-sky atmosphere.
The altitude and optical thickness of cloud determines the overall radiative impact of cloud, a combination of the warming greenhouse effect and the surface-cooling solar shading effect. Yet, probably an even stronger influence does not relate to the cloud itself. The time of day and time of year dictate the incident solar radiation and, therefore, modulates the strength of the short wave reflection: clearly at night the solar influence of cloud is absent.
2. Cloud radiative forcing
To quantify the influence of cloud on Earth's radiative energy balance it is informative to consider separately the broadband long wavelength and short wavelength portions of the electromagnetic spectrum. To infer the influence of cloud on the radiation balance, the radiative fluxes over cloudy scenes may be compared with corresponding fluxes where the influence of cloud has been removed. This may be computed trivially in models by repeating the radiative transfer calculations and neglecting the effects of cloud but this method cannot be applied observationally. However, compositing radiative fluxes for clear-sky conditions and comparing with radiative fluxes for all scenes is one way to infer cloud radiative effect from satellite measurements, although this presents a difficulty in comparing model simulations with satellite estimates of cloud radiative forcing, CRF (e.g. Cess and Potter, 1987; Allan and Ringer, 2003; Erlick and Ramaswamy, 2003).
A further method is to combine simulations of clear-sky radiation with satellite measurements. An example of this is illustrated in Figure 2 which shows long wave and short wave cloud radiative effect over the Africa–Atlantic hemisphere. Although this does not constitute a radiative forcing of climate (Forster et al., 2007), it is common practice to term the cloud radiative effect as a cloud forcing (Ramanathan et al., 1989). Here, long wave cloud radiative forcing (LWCF) is calculated as the difference between outgoing long wave radiation (OLR) and clear-sky OLR (OLRc):
In the example in Figure 2, OLRc is provided by simulations made using the Met Office operational global Numerical Weather Prediction (NWP) model (Allan et al., 2005) which uses the Edwards and Slingo (1996) radiative transfer scheme. The observationally-derived LWCF is presented in Figure 2(a) using OLR derived from radiances measured by the Geostationary Earth Radiation Budget (GERB) instrument (Harries et al., 2005) onboard the Meteosat-9 satellite (preliminary version ARG-V006). The GERB instrument measures total and short wave broadband radiances (long wave radiances are calculated by subtraction) every 17 min at a sub-satellite nominal resolution of around 50 km. The radiances were converted to radiative fluxes using angular dependence models (Clerbaux et al., 2008a, 2008b) that also rely on accurate information about the scene type (e.g. desert, ice cloud, etc), provided by SEVIRI also onboard Meteosat-9 (Schmetz et al., 2002). The model-simulated LWCF is shown in Figure 2(b). Further details of the model and GERB data are provided elsewhere (Allan et al., 2005, 2007; Harries et al., 2005).
Similarly, the short wave cloud radiative forcing (SWCF) is defined as:
where RSW is the reflected short wave radiation to space and clear-sky RSW (RSWc) is simulated by the NWP model. Figure 2(c) shows the GERB/model estimate of SWCF (multiplied by − 1, since SWCF is usually negative, a cooling of the climate system): GERB RSW is adjusted by a factor, f = Sm/Sg, to account for a slight time offset of the model data (1200–1215 UTC) and the GERB data (1150–1205 UTC), where S is the incoming solar radiation at the top of the atmosphere for the model time (m) and the GERB data time (g). The model − SWCF is displayed in Figure 2(d).
The strongest LWCF (>100 Wm−2) and SWCF (<− 400 Wm−2) occurs in the tropical rainy belt, a region of extensive convective cloud associated with high short wave reflectivity (high albedo) and low cloud top emission from the high altitude, cold cloud-tops. Mid-latitude systems are also visible: an intense system over the Argentine basin (around 40°W, 40°S) and a trailing cold front over the North Atlantic. There is also extensive Canarian stratocumulus visible in Figure 2(c) (SWCF ∼− 300 Wm−2) and extensive cloud over the Southern Ocean. It should be noted that the magnitude of SWCF is generally larger than the LWCF, one contributing factor being the proximity to local noon at which incident short wave radiation is strongest. This will be discussed further in Section 5.
While the model simulations capture the overall structure of CRF, there is a number of discrepancies which hint at the complexity in representing cloud systems, even for model analyses which are initialized from a previous model forecast using data assimilation to incorporate available observations. This is discussed elsewhere (Allan et al., 2007) and is beyond the scope of the current study. Nevertheless the example in Figure 2 demonstrates the utility of combining models and satellite data in quantifying cloud radiative effect.
3. Global estimates of cloud radiative forcing
By flying well-calibrated broadband radiation instruments onboard low-Earth orbit satellites, high resolution, global estimates of the Earth's radiation budget may be built up over a period of days and months. As for the geostationary GERB data, measured radiances must be converted to radiative fluxes using angular dependence models (Loeb et al., 2007) valid for distinct scene types such as thick water cloud or cloud-free ocean. Using these data, composites of clear-sky and all-sky fluxes are constructed from which cloud radiative forcing may be estimated using Equations (1) and (2). The current set of Earth radiation budget instruments are from the Clouds and the Earth's Radiant Energy System (CERES) sensors onboard the polar orbiting Terra and Aqua satellites (see Loeb et al., 2007). These have been providing global estimates of short wave, long wave and long wave window channel radiative fluxes since 2000.
Figure 3 shows the multi-annual mean top of atmosphere CRF from CERES on Terra (version 2.5Lite, a reduced set of products but with improved calibration compared to standard version 2 products) over the period 2001–2007. Clouds act to reduce the net incoming radiation through short wave reflection by up to about 100 Wm−2 (Figure 3(a)), in particular over the ocean storm-track regions, the tropical warm pool and Inter-Tropical Convergence Zone (ITCZ) and the extensive ocean stratocumulus decks off the coast of Peru, California and Namibia. The LWCF is strongest over the tropical warm pool, equatorial Africa and Columbia (up to 70 Wm−2) where deep convective clouds reach the coldest points of the tropical tropopause. The net cloud radiative forcing:
is generally one of cooling, with the SWCF dominating the LWCF. This is most pronounced over the marine stratocumulus regimes and the mid-latitude storm-track regions with NCF more negative than − 50 Wm−2. Despite the substantial LWCF and SWCF over the tropical western Pacific and Indian ocean, their effects cancel to a large degree with NCF close to zero (Kiehl, 1994). NCF is also small over the clear regions of northern Africa, Australia and also over polar regions where temperature and cloud greenhouse effect is small and S is small. The overall global net cloud radiative effect is one of cooling as documented previously (Ramanathan et al., 1989).
4. Cloud radiative forcing of the atmosphere and at the surface
It has been established that, globally, the cooling effect of clouds through enhanced scattering of sunlight dominates over the enhanced cloud greenhouse heating effect when considered as annual averages. Are the heating and cooling effects acting at the surface or in the atmosphere? This is fundamental in linking the surface-atmosphere fluxes of energy, knowledge of which is important in understanding the coupling between the global energy and water cycles (Trenberth, 2011). There is a net radiative cooling of the atmosphere of around 100 Wm−2 and this is balanced by latent and sensible heat transfer from the surface to the atmosphere (e.g. Trenberth et al., 2009). Figure 4 shows estimates of the cloud radiative effect on the top of atmosphere, the atmosphere and the surface using 1° resolution global, gridded data from the NASA Surface Radiation Budget dataset (SRB version 3; Stackhouse et al., 2011). SRB combines radiative transfer models (Pinker and Laszlo, 1992; Fu et al., 1998) with satellite measurements of cloud properties from the International Satellite Cloud Climatology Project (ISCCP; Rossow and Schiffer, 1999) and reanalysis data (a fixed NWP model system incorporating data assimilation of available observations to produce a realistic three-dimensional simulation of weather systems; Bloom et al., 2005).
The SRB dataset is constrained at the top of the atmosphere by satellite measurements explaining the similarity in top of atmosphere cloud radiative effect in Figures 3 and 4(a)–(c). Differences in LWCF also relate to the contrasting methodologies of estimating OLRc: CERES OLRc is composited from clear-sky scenes which tend to contain lower than average monthly humidity compared to all-sky conditions from which the SRB calculates OLRc. Therefore CERES OLRc is higher than SRB OLRc in cloudy regions and from Equation (1) it may be explained why CERES LWCF is larger than SRB by around 10 Wm−2 over the Pacific ITCZ. There are also similar inconsistencies between measurements and simulations of RSWc due to systematic changes in aerosol and humidity between cloudy and clear scenes; using model simulations, Erlick and Ramaswamy (2003) found the definition of clear sky to impact the RSW by 8–10% over marine stratocumulus cloud. These issues are discussed further in previous studies (Cess and Potter, 1987; Allan and Ringer, 2003; Erlick and Ramaswamy, 2003; Sohn and Bennartz, 2008; John et al., 2011; Sohn et al., 2010).
The advantage of the SRB methodology is that by combining reanalyses data with satellite retrievals and radiative transfer models, the entire month of clear-sky fluxes may be sampled, avoiding the more limited sampling of clear-sky fluxes using the CERES data alone and allowing direct comparison with climate model output. An additional advantage is that the SRB dataset allows an observational estimate of cloud radiative effect within the atmosphere (Figure 4(d)–(f)) and at the surface (Figure 4(g)–(i)). These are calculated as follows:
where SDL is the surface downwelling long wave radiation and SUL is the surface upwelling long wave radiation (c denotes clear-sky and sfc denotes surface). SUL is generally slightly larger than SULc (by 0.5 Wm−2 in the global annual mean) due to the additional long wave radiation emitted to the surface by clouds, a small proportion of which is reflected back by the surface, in particular for surfaces of low emissivity such as deserts where SUL can be up to 5 Wm−2 greater than SULc in the annual mean.
Similar to Equation (4), for the SWCF at the surface:
where SDS is the surface downward short wave radiation and SUS is the surface upward (reflected) short wave (SUS = SDSα, where α is the surface reflection coefficient, or albedo and is spectrally dependent). Similarly for the atmosphere (atm),
The NCFatm and NCFsfc are calculated as in Equation (3) for the atmosphere and the surface.
The negative SWCF is manifest primarily as a cooling of the surface with a residual heating effect applying in the atmosphere that is less than 10 Wm−2 in magnitude. This heating effect relates to enhanced absorption of short wave radiation by the atmosphere above (primarily low-altitude) cloud decks rather than a direct absorption of short wave radiation by the cloud itself.
Conversely, much of the LWCF in the moist tropics is manifest as a heating of the atmosphere, especially for high altitude cirrus cloud; cloud emission to the surface is small here since much of the surface downward long wave radiation originates from emission by the moist, near-surface layers of the atmosphere (e.g. Prata, 1996). In higher latitudes, and also for sub tropical stratocumulus regimes, where clouds are generally lower altitude, the atmosphere is more transparent to long wave radiation and where temperature inversions are common, there is a cooling effect of clouds on the atmosphere and a strong heating effect at the surface (additional emission to the surface is larger than reduced emission to space), explaining the modest heating effect calculated and measured at the top of the atmosphere (e.g. Sohn, 1999).
The small NCF in the tropical warm pool (due to compensation between LWCF and SWCF) is manifest as a strong cloud long wave heating effect of the atmosphere and a short wave cooling effect at the surface. As such, the presence of deep convective cloud is acting to stabilize the atmospheric profile radiatively, modifying the radiative convective balance of the clear-sky atmosphere (e.g. Manabe and Wetherald, 1967). At higher latitudes, clouds enhance cooling of the atmosphere to the surface through long wave radiative transfer.
Figure 5 summarizes the zonal mean effect of cloud on the top of atmosphere, atmosphere and surface radiation balances, highlighting that clouds heat the atmosphere and cool the surface in the tropics while at high latitudes they enhance long wave emission from cloud base to the surface. Overall, NCF is negative at all but the highest latitudes due to the dominance of negative SWCF over positive LWCF for the annual mean.
Global mean estimates of CRF are documented in Table I. Where CERES multi-annual mean data are missing, SRB values were inserted, thereby avoiding sampling differences (this altered the global mean values by less than 0.5 Wm−2, primarily affecting SWCF). The global mean NCF is approximately − 20 Wm−2 depending upon which dataset and method is used, and is comparable to estimates by Su et al. (2010). This cooling influence of clouds on the current climate system is primarily experienced at the surface; SWCF in the atmosphere is small and cloud long wave heating of the atmosphere in the tropics is counteracted by cloud radiative cooling of the atmosphere over higher latitudes (Figure 5).
Table I. Global multi-annual mean cloud radiative effect, 2001–2007
5. The influence of time of day on cloud radiative forcing
As discussed in Section 1, the dominance of SWCF over LWCF in Figure 5 is only applicable for annual averages, or more specifically for annual diurnal averages and is not applicable for night time where SWCF becomes zero due to zero incoming solar radiation. This is illustrated in Figure 6 using the combination of GERB Edition 1 and NWP model data described in Section 2. All GERB short wave radiative fluxes are scaled by a factor of 0.976 to account for updated calibration information (J. Russell, personal communication 2011) and also by the factor Sm/Sg as discussed in Section 2. Zonal mean CRF is calculated separately for 1200–1215 UTC and 0000–0015 UTC data over the period 2004–2007.
At 1200 UTC, the substantial incoming solar radiation leads to a large SWCF typically around − 100 Wm−2 which dominates over the modest LWCF of around 10–40 Wm−2 (see also Figure 2). The SWCF is more negative than − 300 Wm−2 at 45°S in November to January (Figure 6(b)) highlighting the importance of season on SWCF at higher latitudes with extensive cloud in the Southern Ocean reflecting back to space a significant proportion of the considerable solar radiation incident during the southern hemisphere summer. However, at 0000 UTC (black, dotted line), SWCF is zero and the positive NCF is explained exclusively by the LWCF which is of similar magnitude and structure to the 1200 UTC values. Thus, the radiative effect of changes in cloud cover or properties is highly sensitive not only to cloud type (height, optical thickness, extent) but also to the time of year and time of day at which the changes in cloud properties take place. This is of importance in assessing cloud climate feedbacks which contribute substantially to uncertainty in climate prediction (Bony et al., 2006).
6. Interannual changes in cloud radiative forcing
To assess cloud climate feedbacks, in addition to measuring cloud properties and assessing their representation by models, it is also informative to consider how cloud radiative effect changes with time (e.g. Wielicki et al., 2002; Clement et al., 2009; Dessler, 2010). The longest records of top of atmosphere radiation are from the CERES scanning instrument on the Terra satellite (2000-present) and from an earlier non-scanning radiation budget instrument (wide field of view, WFOV) onboard the Earth Radiation Budget Satellite (ERBS, 1985–2000), described in Wielicki et al. (2002). The WFOV instrument measures integrated radiative fluxes rather than scanning radiances and so does not need to rely on angular dependence models but samples at a low spatial resolution (around 5° in latitude and longitude) and requires 72 days to completely sample the diurnal cycle of radiative fluxes necessary to build up diurnal-mean fluxes (Wielicki et al., 2002; Wong et al., 2006). The near-global region 60°S to 60°N, the limit of the WFOV instrument, is considered in Figure 7 which shows net top of atmosphere radiation and NCF with the seasonal effects removed (de-seasonalized) for WFOV, CERES and also for a state-of-the-art reanalysis system, the European Centre for Medium-range Weather Forecasts (ECMWF) Interim reanalysis (ERA Interim; Dee et al., 2011). ERA Interim is based upon the ECMWF integrated forecast system and uses four-dimensional data assimilation of observational data to provide a six-hourly three dimensional representation of the atmosphere since 1989 (further details in Dee et al., 2011). The radiative transfer code used to simulate radiative fluxes is described by Morcrette et al. (2007).
Because the spatial and temporal resolution of the WFOV instrument is low it was not possible to construct clear-sky composites and to calculate NCF it was necessary to use simulations of clear-sky fluxes from ERA Interim. For consistency, and also to avoid the uneven spatial sampling of clear-sky regions, the same strategy in calculating NCF from CERES data is also adopted. The WFOV data provides flux estimates over 72 day averages which avoids aliasing in artefacts of orbital drift into the measurements of OLR and RSW (Wong et al., 2006). In calculating NCF, clear-sky radiative flux estimates from ERA interim were interpolated onto this 72 day temporal grid before estimating NCF anomalies. Anomalies were computed with respect to the monthly averages over the entire respective data records. The WFOV anomalies were adjusted by the mean 1989–1990 WFOV anomaly minus the mean 1989–1990 ERA Interim anomaly such that mean anomalies over the period 1989–1990 agreed with ERA Interim. Also shown in Figure 7 are clear-sky net radiative flux anomalies simulated by ERA Interim.
Changes in net radiation appear relatively stable over the period 1985–2010, fluctuating by ± 1 Wm−2 with a few notable exceptions. The large negative anomalies in 1991–1993 in the WFOV data relate to the explosive eruption of Mt. Pinatubo in June 1991 which caused an increase in Earth's albedo due to the injection of reflective sulphate aerosols into the stratosphere, resulting in negative instantaneous radiative forcing of around 3 Wm−2 (e.g. Soden et al., 2002). Since Pinatubo aerosols were not included in the ERA Interim simulations, the NCF calculated from WFOV-ERA Interim actually corresponds to a combination of cloud radiative effect and radiative forcing from Pinatubo over this period. Further inaccuracies in the simulation of clear-sky radiative fluxes due to the lack of representation of changes in aerosol radiative forcings will reduce the accuracy of NCF calculated in Figure 7. Nevertheless, with this caveat in mind, the comparison provides an interesting perspective on changes in NCF over recent decades.
For the remainder of the comparisons, there is remarkable agreement in changes in net radiative flux between CERES observations and ERA Interim (r = 0.83) and to a lesser extent WFOV and ERA Interim. Substantial negative anomalies in net radiative flux from ERA Interim are apparent in 1998 and 2010, both El Niño years, suggesting that the substantial re-organization of atmospheric and oceanic circulation systems act to remove energy from Earth during these periods. While a substantial proportion of the net flux change can be explained by the ERA Interim clear-sky fluxes in 1998, relating to an increase in OLRc due to substantial warming and drying of the subtropical descent regions (Chen et al., 2002), there is an overall negative NCF simulated by ERA Interim which appears to relate to a reorganization of clouds and associated SWCF. However, the negative anomalies in 2010 may be overestimated by ERA Interim compared to CERES estimates updated to December 2010 (N. Loeb, personal communication). Nevertheless, combining reanalyses datasets with satellite data potentially provides valuable information on Earth's energy flows and on how cloud radiative effects respond to and modify warming and cooling of the surface, critical in improving climate change projections.
Exploiting satellite measurements and combining them with NWP models initialized through assimilation of available observations enables the effect of clouds on the Earth's radiative energy balance at the surface and within the atmosphere to be quantified for the present day climate. Consistent with previous results (Ramanathan et al., 1989; Su et al., 2010), the cloud radiative cooling effect through reflection of short wave radiation is found to dominate over the long wave heating effect, resulting in a net cooling of the climate system of − 21 Wm−2. The short wave radiative effect of cloud is primarily manifest as a reduction in the solar radiation absorbed at the surface of − 53 Wm−2 for the global multi-annual mean. The magnitude of this effect is strongly modulated by the incoming solar radiation and the dominance of cloud short wave cooling over long wave greenhouse trapping is maximum around local noon (Nowicki and Merchant, 2004) while the cloud long wave heating effect dominates at night. The long wave greenhouse effect of cloud measured at the top of the atmosphere is manifest primarily as a heating of the atmosphere in the moist tropics, consistent with calculations by Sohn (1999). Over the marine stratocumulus regions and across higher latitudes the cloud-base emission to the surface becomes substantial and dominates over the reduced outgoing long wave radiation to space resulting in enhanced radiative cooling of the atmosphere and heating of the surface.
The cloud radiative influence on the exchange of radiative fluxes between the atmosphere and the surface are intimately linked with the water cycle through radiative-convective balance. While tropical, high-altitude clouds act to stabilize the atmospheric profile radiatively, clouds over polar regions tend to cool the atmosphere while heating the surface through enhanced atmospheric long wave radiative emission to the surface. In future work it would be informative to categorize these effects by cloud type further (e.g. Futyan et al., 2005) and compare with climate model simulations. These analyses are vital in constraining cloud feedback processes further and in linking to future changes in the water cycle (Stephens, 2005; Bony et al., 2006; John et al., 2009). A particular challenge is the accurate quantification of surface radiative fluxes due to the sparse ground-based observing network (Roesch et al., 2011) and also monitoring current changes in cloud radiative effect in satellite data, reanalyses and models (Wielicki et al., 2002); combining meteorological reanalyses with satellite data and surface observations provide a vital methodology for meeting these challenges.
The CERES, WFOV and SRB data were obtained from the NASA Langley Research Center Atmospheric Sciences Data Center. ERA interim data was obtained from the European Centre for Medium-range Weather Forecasts data server. The GERB data were provided by the Royal Meteorological Institute of Belgium (RMIB) and the GERB Ground Segment Processing System (GGSPS) at the Science and Technology Facilities Council Rutherford Appleton Laboratory, UK. Thanks to two anonymous reviewers who provided useful input. I also acknowledge the Royal Meteorological Society which hosted the meeting on ‘Clouds and Earth’s Radiation Balance—Observational Evidence' (http://www.rmets.org/events/abstract.php?ID=4496).