1. Introduction
The total net surface radiation is the sum of the net shortwave (SW) and longwave (LW) fluxes into the earth’s surface. Under steady-state conditions, this radiative input is balanced locally by the latent and sensible heat fluxes and the horizontal heat flux divergence below the surface. A good representation of the surface radiation budget is then important for both land surface processes and ocean heat transport (e.g., Essery et al. 2003; Gleckler 2005). Several recent studies have examined the representation of the surface radiation budget in climate models (Wild 2005; Bodas-Salcedo et al. 2008a; Wild 2008). These studies show that, globally, current models generally show an excess of surface downwelling shortwave radiation (SDSR) and a deficit of surface downwelling longwave radiation (SDLR). Webb et al. (2001) and Williams and Webb (2009) show that a lack of bright midlevel-top cloud in the midlatitude oceans contributes to a weak shortwave cloud radiative effect (CRE) in several models, which leads to an excess in surface downwelling solar radiation (SDSR). More recently, Trenberth and Fasullo (2010) show that the third Coupled Model Intercomparison Project (CMIP3; Meehl et al. 2005) models show a consistent positive bias in the absorbed shortwave radiation (ASR) over the southern oceans (in the belt between 45° and 60°S). This region also shows a strong negative trend in the ASR in the projections for the end of the twenty-first century, and they argue that this trend is only tenable because of the large cloud biases in the present-day climate simulations. However, changes in cloud optical depth can also dominate the changes in high-latitude clouds in future climate projections (Senior and Mitchell 1993; Zelinka et al. 2012). Changes in high-latitude clouds can also affect meridional energy transport (Zelinka and Hartmann 2012). Because these biases in the surface radiation budget will introduce biases in the coupled SSTs or impact the ocean heat transport in coupled models (Gleckler 2005), it is important to minimize them in the AGCMs to facilitate their coupling with ocean models.
The atmosphere-only configuration of the Met Office Unified Model (MetUM) shows a significant surplus of SDSR in the Southern Ocean when compared to satellite-derived estimates (Fig. 1b). Recent changes introduced to the model have reduced the bias (Fig. 1c), although a large proportion of the bias still remains. However, one cannot judge from the changes in the SDSR only whether the reduction in the bias is due to a reduction of model errors or to the introduction of compensating errors. In this study, we apply clustering and compositing techniques to identify the cloud regimes that are responsible for the bias and assess whether the recent changes in the model have targeted the right cloud regimes. We also propose and test changes to the boundary layer parameterization scheme that may help reduce the bias in future versions of the MetUM.
The structure of the paper is as follows: Section 2 describes the MetUM and the observations. Section 3 presents the December–February (DJF) climatological comparison of the MetUM simulations versus a wide range of observational datasets. Then, section 4 discusses the results of a clustering algorithm applied to cloud and top-of-atmosphere (TOA) fluxes data. This serves to identify those cloud regimes that are responsible for the SW bias in the Southern Ocean. The clustering results are composited around cyclone centers in section 5. This allows us to link the cloud and radiation biases with the large-scale dynamics and to develop parameterization changes to reduce the biases. Following this, section 6 presents a test of the impact of a change to the boundary layer scheme implemented. We discuss the results and draw conclusions in section 7.
2. Experimental design and observational data
a. The Met Office Unified Model
We use two standard configurations of the MetUM: Global Atmosphere 2.0 (GA2.0) and 3.0 (GA3.0). The global atmosphere is a single choice of dynamical core and atmospheric parameterizations. GA3.0 has been formulated by converging the development paths of the Met Office’s weather and climate global atmospheric components. Walters et al. (2011) describe GA3.0 in detail and present initial results from assessments of this configuration. For the sake of completeness, Table 1 describes the main details of GA3.0 that are relevant for the analysis presented in this paper. The developments of GA3.0 with respect to GA2.0 are also described in detail in Walters et al. (2011), so we only list the main scientific changes here:
Radiation: upgraded treatment of ozone UV absorption, upgraded solar spectrum, included treatment of cloud inhomogeneity, and exponential–random cloud overlap.
Large-scale precipitation: modified rainfall speed and prognostic rain formulation with a sub–time step of 2 min.
Convection: relaxed conditions for the diagnosis of shallow convection, which makes the conditions for the diagnosis of deep convection more stringent.
Summary of physical parameterizations for GA3.0. A more detailed description is given by Walters et al. (2011).
We run the MetUM in atmosphere-only mode forced with prescribed SSTs, from September 1981 to December 2008. The analysis presented in sections 3–5 is carried out with the climate horizontal resolution N96 (1.875° longitude by 1.25° latitude) and with 85 levels in the vertical. The case study presented in section 6b is run with a 1.5-km horizontal resolution and local area configuration of the MetUM (Roberts et al. 2009; Wilkinson et al. 2012), whereas the global test presented in section 6c is run for 10 yr.
b. Observations
We use data from the D series of the International Satellite Cloud Climatology Project (ISCCP; Rossow and Schiffer 1999) and radiative fluxes at TOA and at the surface from the same database (Zhang et al. 2004). We use daily cloud optical depth τ versus cloud-top pressure (CTP) histograms and TOA and surface radiative fluxes, labeled as ISCCP-D1 and ISCCP-FD products, respectively. We also use daily surface pressure from the 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40; Uppala et al. 2005). As independent databases for cloud properties, we use monthly data from the Multiangle Imaging Spectroradiometer (MISR; Diner et al. 2005) and the Moderate Resolution Imaging Spectroradiometer (MODIS; King et al. 2003). MODIS produces CTP–τ histograms like ISCCP but uses different retrieval algorithms. MISR determines cloud-top height (CTH) using a stereo-imaging technique (Moroney et al. 2002; Muller et al. 2002). MISR also retrieves cloud optical depth from the visible radiances, although only over ocean. These retrievals allow the computation of joint CTH–τ histograms.
The Cloud Profiling Radar (CPR) is onboard CloudSat (Stephens et al. 2008), and the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) is onboard the Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO; Winker et al. 2010). CloudSat and CALIPSO are active instruments capable of providing vertical information on clouds and aerosols. CloudSat CPR operates at a frequency of 94 GHz (Im et al. 2005), and its pulses sample a volume of 480 m in the vertical, with an approximate horizontal resolution of 1.4 km. We use the CloudSat geometric profile (2B-GEOPROF) dataset, which provides the radar reflectivity and identifies where hydrometeors occur (Marchand et al. 2008). CALIOP operates at 532 and 1064 nm. It is nadir pointing with a beam diameter of 70 m at the earth’s surface and produces footprints every 333 m in the along-track direction. We use the monthly GCM-Oriented CALIPSO Cloud Product (GOCCP; Chepfer et al. 2010).
To account for the characteristics of the different observational datasets, we simulate diagnostics comparable with the ones provided by the observational datasets from the model fields. We achieve this by using the Cloud Feedback Model Intercomparison Project (CFMIP) Observation Simulator Package (COSP; Bodas-Salcedo et al. 2011). COSP is a flexible software tool that enables the simulation of data from several satellite-borne active and passive sensors taking model variables as inputs. We use the following simulators within COSP: ISCCP (Klein and Jakob 1999; Webb et al. 2001), MISR (Marchand and Ackerman 2010), MODIS (Pincus et al. 2012), CloudSat (Haynes et al. 2007), and CALIPSO (Chiriaco et al. 2006; Chepfer et al. 2008). ISCCP diagnostics are output for the entire length of the runs. The other COSP diagnostics are produced for years 2006–08 in the GA2.0 and GA3.0 experiments.
The Clouds and the Earth’s Radiant Energy System (CERES; Wielicki et al. 1996) provides high-quality TOA radiative fluxes and cloud products from MODIS in a similar format to the ISCCP products. Currently, COSP does not have a module capable of producing CERES cloud diagnostics, which is a limitation for using the CERES ISCCP-like products. We have tested the sensitivity of the analysis presented in sections 4 and 5 with other datasets. We have used MODIS cloud data and Earth Radiation Budget Experiment (ERBE) radiative fluxes as alternative datasets and the results are very similar, suggesting that the results are robust.
3. Southern Ocean climatology
Figure 1a shows the DJF climatology of the SDSR in the Southern Hemisphere from 20 yr of ISCCP-FD data (1984–2003). The pattern is quite zonal, with large amounts of solar radiation equatorward of 45°S and a local minimum in the midlatitudes between 50° and 70°S. GA2.0 shows a strong positive bias across the entire Southern Ocean belt (Fig. 1b), ranging between 5 and more than 40 W m−2 locally. The largest biases are found in the south Atlantic. Figure 1c shows the impact of the changes introduced in GA3.0. These changes reduce the bias in the Southern Ocean, with a smaller impact in the regions equatorward of 40°S and in the Antarctic continent (Fig. 1c). Zhang et al. (2004) estimate the overall uncertainty of the surface fluxes to be between 10 and 15 W m−2. This uncertainty estimate is represented by the white areas in Figs. 1b,c. The mean bias in the region between 40° and 70°S is reduced from 20 to 14 W m−2.
The average bias over the Southern Ocean (70° to 40°S) has a strong annual cycle (Fig. 2), because it scales with the available solar radiation. It ranges from less than 5 W m−2 in winter to more than 20 W m−2 during the summer months. GA3.0 reduces the bias throughout the year, with the reduction being also dependent on the insolation at TOA, from little more than 1 W m−2 in winter to around 6 W m−2 in summer.
Figures 1 and 2 show evidence that the simulation of the shortwave radiation budget in the Southern Ocean has been improved with the changes introduced at GA3.0, reducing the mean annual bias by 25%. However, a good simulation of the radiation budget does not necessarily mean that the cloud radiative properties are well represented as there may be compensating errors that combine in a way that reduce the mean bias (e.g., Webb et al. 2001). We construct monthly-mean joint ISCCP CTP–τ histograms at 2.5° × 2.5° longitude–latitude resolution, and then calculate the DJF climatology over the (70°S, 40°S) region (Fig. 3a). ISCCP observes most of the clouds in the middle and low levels, with optical depths mainly between 1.3 and 60. ISCCP simulator outputs from GA2.0 (Fig. 3c) show that the main biases occur in clouds identified as having tops at midlevels (between 440 and 680 hPa) with a wide range of optical depths between 1.3 and 60 and low clouds (CTP > 680 hPa) with optical depths between 3.6 and 9.4. Figure 3e shows that the biases in the cloud properties have been reduced in GA3.0, in the midlevel clouds with optical depths greater than 9.4.
In the solar portion of the spectrum, the radiative effect of a cloud can be characterized by the area fraction that it occupies and its optical depth, so the information available in the ISCCP histograms can be used to estimate the radiative impact that each of the CTP–τ bins has on the TOA shortwave radiation budget over a dark surface like the ocean surface. Also, the fact that there is a robust relationship between the TOA albedo and the surface shortwave radiation budget (Li and Leighton 1993) makes it possible to infer changes in the SDSR by analyzing changes in the outgoing shortwave radiation (OSR). We estimate the contribution of each CTP–τ bin to the OSR by estimating the mean cloud albedo in each τ bin and multiplying that by the average insolation at TOA and the cloud fraction in each CTP–τ. These estimates can be computed both from the observational and model CTP–τ histograms (Figs. 3b,d,f). Figure 3d shows that the main contributors to the lack of OSR (excess of SDSR) are clouds with tops between 440 and 800 hPa and optical depths between 9.4 and 60. This bias is partially compensated by excess OSR coming from high-top thick cloud and low-level cloud with moderate optical depth, between 3.6 and 23. The improvements in the simulation of mean cloud properties in GA3.0 (Fig. 3e) has reduced the biases of the radiative impact of clouds with optical depths between 9.4 and 23, but there is still a substantial negative radiative bias for optical depths between 23 and 60, partially compensated for by positive biases in high, thick cloud and low-level cloud with moderate optical depth. These results suggest that there is a need to increase the optical depth of the low-level cloud with moderate optical depth and increase the population of midlevel cloud with moderate to high optical depths.
Figure 4 shows the comparison for DJF over (70°S, 40°S) for five different diagnostics, from COSP and observations. They all are histograms of relative frequency of occurrence (RFO), although the normalization method is not the same for all of them (see figure caption for details). The left-hand column shows the observational results, and the middle and right columns are the COSP diagnostics from GA2.0 and GA3.0. The ISCCP, MISR, and MODIS histograms (first three rows in Fig. 4) are conceptually similar, but the different sensors, retrieval algorithms, and spatiotemporal sampling introduce differences among them (e.g., Marchand et al. 2010; Bodas-Salcedo et al. 2011). The differences are mainly driven by responses of the retrievals to low-level clouds under temperature inversions, subpixel-scale low-level clouds, and multilayer clouds (Marchand et al. 2010). MISR and MODIS GA2.0 diagnostics (Figs. 4d–i) show a lack of cloud with tops at midlevels (CTP from 440 to 680 hPa or CTH between ≈3 and ≈7 km) and also not enough low-level cloud (CTP greater than 680 hPa or CTH below ≈3 km) with optical depths greater than 9.4, consistent with the comparison against ISCCP. GA3.0 slightly reduces these biases, although the main signature of the differences with the observational datasets still remains. The fact that comparisons with MISR and MODIS are consistent with the interpretation of the ISCCP histograms confirms that the differences are robust.
The diagnostics from the active instruments (CALIPSO and CloudSat; Figs. 4j–o) provide complementary information. Although both model configurations show realistic simulations, they show the same type of biases that are common across models (Bodas-Salcedo et al. 2011). Models usually explore a smaller region of the histogram than the observations. Figures 4j–l show that the MetUM does not simulate clouds between 2 and 7 km with large scattering ratios (>60), typically associated with liquid clouds. This may be an indication of the presence of supercooled clouds in reality, which are not simulated by the models. We estimate the optical depth of a single 480-m-thick cloud with scattering ratio of 60 to be around ≈1. Assuming that these supercooled clouds are only a few hundred meters thick (Hogan et al. 2003), they will appear in the MISR histograms in the intervals with optical depth between 0.3 and 3.6. The MISR simulator outputs show less cloud between 2 and 5 km high in that optical depth interval, which is consistent with these estimates. GA3.0 seems to improve the simulation of low cloud as seen by CALIPSO, although the spread in scattering ratios in the clouds above 5 km has reduced, making the comparisons against the observations worse.
The CloudSat histograms also show biases that are common across models. A lack of bimodality in the range of reflectivities at low levels, associated with a lack of nonprecipitating low-level clouds (Bodas-Salcedo et al. 2008b; Stephens et al. 2010). Both GA2.0 and 3.0 also show a lack of spread in the reflectivities for any given level for clouds above 3 km high, although this is less apparent in GA3.0. In the MetUM, part of this lack of spread is explained by the monotonic increase of ice cloud fraction with gridbox-mean ice water content (Bodas-Salcedo et al. 2008b; Delanoë et al. 2011). The model has little cloud in the area of the CloudSat histogram with low reflectivities (<−20 dB) and heights between 2 and 5 km. This corresponds to the area occupied by the “cumulus congestus” regime as identified by Zhang et al. (2007) in the CloudSat observations.
4. Cloud regime analysis over the Southern Ocean
The Southern Ocean is a region of transient weather systems, so, in order to understand which cloud types are contributing to the bias, it is necessary to go beyond a climatological-mean analysis and explore in which regimes the bias occurs. In this section, we follow the methodology of Williams and Webb (2009) to assign model grid points at daily temporal resolution to one of seven midlatitude cloud regimes obtained by spatiotemporal clustering of daily ISCCP histograms. The assignment of the grid point is based upon the gridpoint-mean cloud-top pressure, cloud albedo, and cloud cover from the ISCCP simulator. For ease of reference, these regimes are named shallow cumulus (cu), cumulus–stratocumulus transition (transition), stratocumulus (sc), midtop, thick frontal (frontal), cirrus (ci), and thin cirrus (thin ci). These names are intended to indicate the typical characteristics of the majority of cloud that makes up the regime, rather than suggesting that all of the cloud within the regime would be identified as those morphological types. A detailed description of the methodology and the radiative properties and geographical distribution of the cloud regimes is discussed in Williams and Tselioudis (2007) and Williams and Webb (2009). Gordon and Norris (2010) and Haynes et al. (2011) also study cloud regimes in the southern oceans. Although all these studies use clustering techniques to define the cloud regimes, the methodologies and geographical domains are slightly different, so a direct quantitative comparison is not possible. However, the results are qualitatively very similar (Gordon and Norris 2010; Haynes et al. 2011).
We use the TOA CRE to look at the impact of clouds in the radiation budget, which is the difference between clear-sky and all-sky fluxes. Each regime has an effect on the radiation balance through the CRE when the cloud regime is present and through the relative frequency with which the regime occurs. Hence, the climatological SW bias in the model can be through the model having the incorrect RFO of the regimes and/or the radiative properties of the regimes being incorrect when they are simulated. The relative contribution from these two terms, together with a cross term, which is usually small (Williams and Webb 2009), is shown for each cloud regime in Fig. 5.
SW CRE is negative so the large positive bias in the midtop and stratocumulus regimes indicates that errors in these cloud regimes largely account for the excess SW reaching the surface. The bias occurs through both the regimes being simulated too infrequently and the regimes not being sufficiently reflective. GA2.0 tends to largely simulate shallow cumulus in place of the midtop and stratocumulus regimes. However, the overestimation of the RFO of the shallow cumulus regime does not compensate for the positive CRE bias in the other two regimes. We have also compared the model with observational data from MODIS, processed into the ISCCP CTP–τ bins, and used ERBE flux data rather than ISCCP-FD (not shown). The midtop and stratocumulus regimes being simulated too infrequently and not being reflective enough is also evident from this second set of independent observational data and so may be considered a robust error.
The model improvements made between GA2.0 and GA3.0 (specifically the introduction of prognostic rain and altered fall speeds) have increased the frequency with which the midtop and stratocumulus regimes are simulated at the expense of shallow cumulus. This confirms that the reduced climatological bias in GA3.0 noted in the previous section has occurred due to a more realistic balance between different cloud types being achieved rather than, say, through the introduction of a compensating error. However, the radiative bias when these regimes are simulated remains.
5. Cyclone compositing
Some of the cloud regimes identified above can occur in a number of places around synoptic weather systems. We now use the cyclone compositing methodology of Field and Wood (2007) to identify the typical synoptic conditions in which the cloud regime biases occur.
Minima in daily-mean sea level pressure are identified following Field and Wood (2007) over the latitudes 40°–70°S. A box covering 60° in longitude and 30° in latitude is centered on the cyclone. By averaging across all the cyclones in a 2-yr period, the mean frequency with which each grid point within the box is assigned to each cloud regime is obtained. This box is large enough that mature cyclones and to some extent transient ridges ahead or behind the cyclone can be included but not so large to be seriously affected by a following large cyclone. Cyclones with centers over sea ice and points over sea ice are not included in the compositing, in order to reduce the impact of scene misidentification over bright surfaces. Two years has been found sufficient to give robust results. The numbers of cyclones identified are 3192 and 3488 for ERA-40 and GA3.0, respectively. The numbers for GA2.0 are similar to GA3.0. This implies that we are sampling a fraction of the Southern Ocean smaller than ≈70%. From these, we only retain 1152 and 2221 over ice-free ocean. The compositing does not change significantly if all the cyclones are included, and it is also stable to changes in the length of the time series used, which gives robustness to the results. No attempt is made to link the identified cyclones in time; therefore, individual cyclones at different times in their life will be identified as separate entities.
The results for the cloud regimes derived from ISCCP observations, composited using ERA-40 surface pressure to identify the cyclone, are in Fig. 6. The ERA-40 surface pressure contours are overlaid in Fig. 6g. As noted by Field and Wood (2007), on any one cyclone at one time the associated fronts can be in different locations relative to the center; however, on average for Southern Hemisphere cyclones they occupy the northeast quadrant of the composite box (Govekar et al. 2011). This can be seen to be the case for the frontal cloud regime in Fig. 6. Figure 6h shows a schematic of the typical position of the fronts in the cyclone composite and is included to help understand the description of the model results presented below. Midtop cloud occurs across the composite box and is likely to cover a variety of cloud types; however, there is a maximum on the cold-air side of the cyclone just behind the cold front. The stratocumulus regime also frequently occurs on the cold-air side of the cyclone. We interpret this as the northward moving cold air frequently containing fields of closely spaced or closed-cell cumulus, growing into congestus cloud as they continue to move north over relatively warm water. This cold-air region is a region of subsidence (Bauer and Del Genio 2006), where supercooled clouds often occur (Naud et al. 2006).
The frontal cloud regime in GA2.0 appears to be well simulated in terms of its frequency and mean location relative to the cyclone (Fig. 7). Also consistent with observations, there is midtop cloud simulated in the frontal sector, which is probably stratiform in nature. However, the model lacks midtop cloud elsewhere, particularly on the cold-air side of the cyclone. The majority of the GA2.0 cloud in the cold air is assigned as shallow cumulus, which indicates that the gridbox cloud cover is too low (it is simulating more open-cell style cumulus) and the convection is not deep enough. GA3.0 is an improvement with more midtop cloud across the composite box (although still less than observed by ISCCP in the cold air but more comparable to MODIS–not shown) (Fig. 8). Cloud in the cold air is also largely assigned as transition with some stratocumulus in GA3.0 rather than shallow cumulus. This regime shift indicates that more of the grid box is now being covered and the albedo of the cloud has increased. Although this is a significant improvement, a further brightening and increase in cloud cover in these cold-air situations is required for the stratocumulus regime to be diagnosed frequently enough to match observations. Although there are some differences in the simulation of the Cirrus regime, the impact of these differences in the radiative fluxes is small (Fig. 5).
The bias in the top-of-atmosphere radiation budget, which is dominated by the shortwave, is mainly present on the cold-air side of the composite cyclone (Fig. 9). This is consistent with the global results from an older version of the Met Office global forecast model (Field et al. 2011). This suggests that the bias is not specific to Southern Hemisphere cyclones. The increased frequency of the transition, stratocumulus, and midtop cloud regimes at the expense of shallow cumulus have had the effect of making this cold-air side more reflective and so reducing the net radiative bias. This confirms that the improvement in the climatological radiative balance in GA3.0 is through an improved simulation in those meteorological situations that were mainly contributing to the bias. However, this analysis shows that clouds are still not reflective enough in the cold-air side of the cyclone, mostly dominated by low- and midlevel cloud.
We now work on the hypothesis that model changes that target the meteorological conditions that prevail in the cold-air side may help improve the cloud regimes in those areas, without having a detrimental impact in other cloud regimes.
6. Representation of vertical mixing
As with many GCMs, the MetUM makes a decision on whether vertical mixing in clouds is driven primarily by narrow penetrative cumulus clouds, in which case either the mass-flux scheme or the resolved dynamics (in high-resolution, kilometer-scale simulations of the MetUM) are used to represent the vertical mixing or through more homogeneous turbulent mixing, as in stratocumulus clouds, for which the boundary layer scheme was specifically designed. This does not preclude the two operating together, such as where cumulus clouds rise into a layer of stratocumulus, but in the MetUM the boundary layer scheme is stopped from mixing across cloud base when cumulus is diagnosed. This diagnosis has important implications for the cloud layer evolution, with both the mass-flux scheme and kilometer-scale dynamics tending to lead to much smaller cloud amounts.
As described in Lock et al. (2000), an initial diagnosis of planetary boundary layer (PBL) type is made based on the surface buoyancy flux and, for unstable boundary layers, the thermodynamic characteristics of the layer identified by lifting a moist parcel adiabatically to find its level of neutral buoyancy, denoted zpar (Fig. 10). This process also triggers the cumulus convection scheme if a cloud-capped boundary layer has a significant total moisture gradient in the cloud layer; the property used to identify that the layer is not well-mixed stratocumulus. If this occurs, transport into the cloud layer is assumed to come solely from the cumulus elements and so both the surface-based nonlocal mixing and the local Richardson number-based mixing within the boundary layer scheme are stopped from mixing across cloud base (at zlcl). To allow boundary layers initially diagnosed as well mixed (rather than cumulus capped) to decouple, an additional constraint is placed on the buoyancy consumption of turbulent kinetic energy (TKE) arising from boundary layer mixing.
The above constraints consider only the buoyancy and moisture gradients in the boundary layer. To allow for the effects of shear generation of turbulence, which can extend mixing into regions of weak static stability, two additional dynamical constraints are included that can suppress the triggering of the convection scheme (and therefore maintain PBL mixing across the cloud base): 1) if the level where the local gradient Richardson number Ri = N2/S2 (where N2 is the static stability and S the vertical shear of the horizontal wind) first becomes supercritical zh(Ri) is above zpar (i.e., there is sufficient wind shear to generate turbulent mixing throughout the cloud layer) and 2) if the bulk measure of stability −zpar/L is small (i.e., the PBL is close to neutral), where L is the surface Obukhov length.
a. Changes to boundary layer scheme
We modify the first of these dynamic constraints to be more aggressive, requiring that zh(Ri) only extend some distance into the cloud layer. Physically it is hypothesized that this indicates sufficient turbulent mixing driven by wind shear to disrupt the formation of cumulus elements. The revised process is illustrated in Fig. 10. After an initial cumulus diagnosis (Fig. 10, left; because of a significant moisture gradient in the cloud layer: not shown), the Ri profile is examined. Either there is sufficient wind shear in the lower part of the (statically stable) cloud layer such that zh(Ri) > zthres = zlcl + f(zpar − zlcl), where f is a tunable parameter (here shown with f ~ 0.2), in which case a shear-dominated PBL is diagnosed (Fig. 10, bottom right)—the initial cumulus diagnosis will be overruled and the top of the nonlocal surface-based mixing coefficient KPBL is kept at zpar—or zh(Ri) < zthres, in which case the cumulus diagnosis remains and PBL mixing across the cloud base is suppressed (Fig. 10, top right).
b. Case study test
We apply this modification to the high-resolution local area configuration of the MetUM to test its performance in a cold-air outbreak case study. Despite some of the model physics being different to that of GA3.0 (e.g., different cloud scheme, convection parameterization switched off), the boundary layer scheme is the same. This gives us confidence to use this configuration to test the changes in a cold-air case study, as a representative situation of the cold-air sector of an extratropical cyclone. A global model run (N512; ≈25-km resolution at midlatitudes) was started from a 1200 UTC 30 January 2010 analysis and run for 36 h, generating hourly boundary conditions to drive 12-, 4-, and 1.5-km nested models. The 4- and 1.5-km model runs began at 1800 UTC 30 January 2010 and output (OSR) is shown from 1230 UTC 31 January 2010 for comparison with a satellite image (Fig. 11). The MODIS image (channel-4 radiance) shows a closed-cell cloud region between 60° and 64°N and between 8° and 12°W. South of this region, the cloud breaks up into open-cell convection.
The region of stratiform cloud present in the MODIS image is not present in the control model. Instead, the control model has cloud similar in appearance to those found to the south. In contrast, the run using the modified boundary layer diagnosis (with f = 0.5) was able to produce an extensive stratiform cloud layer in the region indicated by the MODIS image. This qualitative agreement suggests that the modified boundary layer parameterization is a more appropriate way of handling the mixing of shallow boundary layers in cold-air outbreaks (≈1.5 km deep) than allowing the model to attempt to do so explicitly. These results motivate us to test the changes in the global configuration.
c. Global test
These changes were implemented on top of GA3.0, and a 10-yr test was run setting f = 0.2. A more aggressive setting for f was needed to make a noticeable impact in the global model configuration. These changes reduce the biases in the SW CRE for the midtop and stratocumulus regimes (Fig. 5). This is mainly due to a better simulation of the frequency of occurrence. The impacts of these changes on the cyclone compositing are shown in Fig. 12. The changes are mainly concentrated in the cold-air side, where cloud is shifted from the transition regime into the stratocumulus regime, making it more reflective, and there is also more midtop cloud. They lead to a stronger (more negative) SW CRE on the cold-air side (Fig. 9). These changes are improvements with respect to ISCCP (Fig. 6). The higher cloud top gives more LW CRE across much of the cyclone, which partially offsets the shortwave effect (Fig. 9). These changes reduce the bias in the Southern Ocean all year-round, the reduction being more noticeable during winter and spring (Fig. 2). The SDSR bias annually averaged over the (70°S, 40°S) latitude band is reduced from 13 W m−2 in GA2.0 to 9 W m−2 in GA3.0 and 7 W m−2 in the shear-driven test. Altogether, the results from Figs. 5 and 12 suggest that an increase in optical thickness of the clouds in the stratocumulus regime is still required, but the frequency of occurrence and distribution are well simulated. However, the analysis of the midlevel regime is a bit more complex. Although the overall frequency of occurrence is well simulated, its distribution within the cyclone composite is not correct. The maximum in occurrence is located very close to the frontal region, whereas in the observations it is far behind. This contributes to the large SW CRE bias of more than 30 W m−2 that still exists in part of the cold-air region, which is dominated by the midlevel cloud regime in the observations. Because the frequency of occurrence of the transition regime is well simulated in that region, this suggests that the lack of midlevel in the northwest quadrant is not associated with cloud not penetrating deep enough and being classified as transition. The determination of the causes for this requires further investigation.
Trenberth and Fasullo (2010) suggest that the excess of SDSR is more acute in the Southern Hemisphere (SH) than in the Northern Hemisphere (NH). The MetUM is consistent with this result. Although we focus here on the Southern Ocean, and have not carried out the compositing for the NH, the case studies analyzed and the changes in the seasonal-mean shortwave fluxes suggest that the impact of the shear-driven changes is very similar in the NH.
7. Conclusions
We study the role of clouds in the shortwave radiation bias in the Southern Ocean in the atmosphere-only configuration of the Met Office climate model. First, we present a climatological comparison of the two most recent versions of the MetUM versus a wide range of observational datasets. Then, we discuss the results of a clustering algorithm applied to cloud and TOA fluxes data. This serves to identify those cloud regimes that are responsible for the SDSR bias over the Southern Ocean.
We develop a novel methodology that composites the clustering results around cyclone centers, which allows us to study the role of each cloud regime in a mean composite cyclone. We apply this methodology to model outputs and observations and identify the cloud regimes that are responsible for the bias and assess whether the recent changes in the model have targeted the right cloud regimes. The bias is mainly present on the cold-air side of the composite cyclone. The changes introduced in the most recent configuration of the model increase the frequency of the transition, stratocumulus, and midtop cloud regimes at the expense of shallow cumulus. This makes the cold-air side of the cyclone more reflective and reduces the radiative bias, although the transition regime is now too frequent and the stratocumulus regime is still too infrequent. This confirms that the improvement in the climatological radiative balance is through an improved simulation in the meteorological situations that were mainly contributing to the bias.
This new methodology allows us to propose model developments that target the simulation of the cloud regimes that are mainly responsible for the bias based on the meteorological situations in which they occur (cold-air side of cyclones). Based on this analysis, we develop and test a new diagnosis of shear-dominated boundary layers. This change improves the simulation of the cloud radiative effect; although the results suggest that there is still a need to increase the optical depth of the low-level cloud with moderate optical depth and cloud with tops at midlevels.
The boundary layer diagnosis changes are also implemented in a multiscale study, using the global model and nested models at 12-, 4-, and 1.5-km horizontal resolution. The results from a case study of a cold-air outbreak northwest of the British Isles suggest that the modified boundary layer parameterization provides a better cloud simulation. We generalize this result by running a global model experiment with these changes implemented and show that the frequency of occurrence of the cloud regimes in the cyclone composite are clearly improved. However, the radiative properties of two regimes (midtop and stratocumulus) are still biased, not being reflective enough. Overall, with the changes introduced in GA3.0 plus the boundary layer changes presented in this study, the SDSR bias over the Southern Ocean has been reduced from 20 to 13 W m−2 for the DJF season and from 13 to 7 W m−2 annually.
Although the boundary layer changes have helped reduce the SDSR bias in the Southern Ocean, their main impact is seen in the changes in the frequency of occurrence of the cloud regimes. A substantial bias in the radiative properties of the midtop and stratocumulus cloud regimes is still present, not being bright enough. This suggests a lack of cloud water path, too large cloud droplet sizes, or a combination of the two. Also, the midtop level cloud regime seems to be simulated in the wrong position in the cyclone composite. Future work will concentrate on looking at the causes of the biases in the radiative properties of these regimes and the wrong position of the midtop cloud regime.
Some of the differences in the cloud simulation may be due to differences in the large-scale thermodynamics between the observational composites and the models. Since we are using atmosphere-only models, the differences in SSTs are not significant. The fact that case studies like the one presented in section 6b, where humidity biases will be minimized by the data assimilation system, show biases consistent with the cyclone compositing suggests that the biases in the thermodynamic variables are not playing a leading role. However, we cannot rule out the possibility that they still play a significant role, and we will look into this in future studies.
We have focused our analysis on cyclonic conditions. The fact that the Southern Ocean sustains an average cloud cover that exceeds 80% suggests that anticyclonic conditions may also play a significant role in explaining the radiative biases shown by climate models (Trenberth and Fasullo 2010). This should be a focus of future studies.
This work demonstrates the benefits of the use of process-based diagnostics as tools to understand the causes of model errors in the simulation of clouds and radiation and how this can be used to steer the direction of new model developments. It also demonstrates the benefits of a multiscale and unified approach in model evaluation and development by sharing diagnostic techniques across time scales and applying the same changes to models at very different resolutions (Senior et al. 2010; Brown et al. 2012).
Acknowledgments
A. Bodas-Salcedo and K. D. Williams were supported by the Joint DECC/Defra Met Office Hadley Centre Climate Programme (GA01101). CloudSat data were obtained from the CloudSat Data Processing Center (http://cloudsat.cira.colostate.edu). CALIPSO-GOCCP data were obtained from the IPSL ClimServ data center (http://climserv.ipsl.polytechnique.fr/cfmip-obs.html). ISCCP data were obtained from the NASA Langley Research Center Atmospheric Sciences Data Center (http://eosweb.larc.nasa.gov). Thanks to those members of the NASA Langley Research Center and NASA Jet Propulsion Laboratory who are generating the MISR CTH-OD. MODIS data were obtained from the Level 1 and Atmosphere Archive and Distribution System (LAADS), NASA Goddard Space Flight Center. ERA-40 was obtained from the ECMWF archive (http://data-portal.ecmwf.int/data/d/license/era40/). Mark Webb provided constructive comments on the paper. We thank J. Fasullo and two anonymous reviewers for their constructive comments.
REFERENCES
Abel, S. J., D. N. Walters, and G. Allen, 2010: Evaluation of stratocumulus cloud prediction in the Met Office forecast model during VOCALS-REx. Atmos. Chem. Phys., 10, 10 541–10 559, doi:10.5194/acp-10-10541-2010.
Barker, H. W., and Z. Li, 1995: Improved simulation of clear-sky shortwave radiative transfer in the CCC-GCM. J. Climate, 8, 2213–2223.
Bauer, M., and A. D. Del Genio, 2006: Composite analysis of winter cyclones in a GCM: Influence on climatological humidity. J. Climate, 19, 1652–1672.
Bodas-Salcedo, A., M. A. Ringer, and A. Jones, 2008a: Evaluation of the surface radiation budget in the atmospheric component of the Hadley Centre Global Environmental Model (HadGEM1). J. Climate, 21, 4723–4748.
Bodas-Salcedo, A., M. J. Webb, M. E. Brooks, M. A. Ringer, K. D. Williams, S. F. Milton, and D. R. Wilson, 2008b: Evaluating cloud systems in the Met Office global forecast model using simulated CloudSat radar reflectivities. J. Geophys. Res., 113, D00A13, doi:10.1029/2007JD009620.
Bodas-Salcedo, A., and Coauthors, 2011: COSP: Satellite simulation software for model assessment. Bull. Amer. Meteor. Soc., 92, 1023–1043.
Brown, A. R., R. J. Beare, J. M. Edwards, A. P. Lock, S. J. Keogh, S. F. Milton, and D. N. Walters, 2008: Upgrades to the boundary-layer scheme in the Met Office numerical weather prediction model. Bound.-Layer Meteor., 128, 117–132, doi:10.1007/s10546-008-9275-0.
Brown, A. R., S. Milton, M. Cullen, B. Golding, J. Mitchell, and A. Shelly, 2012: Unified modelling and prediction of weather and climate: A 25-year journey. Bull. Amer. Meteor. Soc., in press.
Chepfer, H., S. Bony, D. Winker, M. Chiriaco, J.-L. Dufresne, and G. Sèze, 2008: Use of CALIPSO lidar observations to evaluate the cloudiness simulated by a climate model. Geophys. Res. Lett., 35, L15704, doi:10.1029/2008GL034207.
Chepfer, H., S. Bony, D. Winker, G. Cesana, J.-L. Dufresne, P. Minnis, C. J. Stubenrauch, and S. Zeng, 2010: The GCM-Oriented CALIPSO Cloud Product (CALIPSO-GOCCP). J. Geophys. Res., 115, D00H16, doi:10.1029/2009JD012251.
Chiriaco, M., R. Vautard, H. Chepfer, M. Haeffelin, J. Dudhia, Y. Wanherdrick, Y. Morille, and A. Protat, 2006: Ability of MM5 to simulate ice clouds: Systematic comparison between simulated and measured fluxes and lidar/radar profiles at the SIRTA atmospheric observatory. Mon. Wea. Rev., 134, 897–918.
Cusack, S., J. M. Edwards, and J. M. Crowther, 1999: Investigating k distribution methods for parameterizing gaseous absorption in the Hadley Centre climate model. J. Geophys. Res., 104 (D2), 2051–2057.
Davies, T., M. J. P. Cullen, A. J. Malcolm, M. H. Mawson, A. Stainforth, A. A. White, and N. Wood, 2005: A new dynamical core for the Met Office’s global and regional modelling of the atmosphere. Quart. J. Roy. Meteor. Soc., 131, 1759–1782, doi:10.1256/qj.04.101.
Delanoë, J., R. J. Hogan, R. M. Forbes, A. Bodas-Salcedo, and T. H. M. Stein, 2011: Evaluation of ice cloud representation in the ECMWF and UK Met Office models using CloudSat and CALIPSO data. Quart. J. Roy. Meteor. Soc., 137, 2064–2078, doi:10.1002/qj.882.
Diner, D. J., and Coauthors, 2005: The value of multiangle measurements for retrieving structurally and radiatively consistent properties of clouds, aerosols, and surfaces. Remote Sens. Environ., 97, 495–518, doi:10.1016/j.rse.2005.06.006.
Edwards, J. M., and A. Slingo, 1996: Studies with a flexible new radiation code. I: Choosing a configuration for a large-scale model. Quart. J. Roy. Meteor. Soc., 122, 689–720.
Essery, R. L. H., M. J. Best, R. A. Betts, P. M. Cox, and C. M. Taylor, 2003: Explicit representation of subgrid heterogeneity in a GCM land-surface scheme. J. Hydrometeor., 4, 530–543.
Field, P. R., and R. Wood, 2007: Precipitation and cloud structure in midlatitude cyclones. J. Climate, 20, 233–254.
Field, P. R., A. Bodas-Salcedo, and M. E. Brooks, 2011: Using model analysis and satellite data to assess cloud and precipitation in midlatitude cyclones. Quart. J. Roy. Meteor. Soc., 137, 1501–1515, doi:10.1002/qj.858.
Fritsch, J., and C. Chappell, 1980: Numerical prediction of convectively driven mesoscale pressure systems. Part I: Convective parameterization. J. Atmos. Sci., 37, 1722–1733.
Gleckler, P. J., 2005: Surface energy balance errors in AGCMs: Implications for ocean-atmosphere model coupling. Geophys. Res. Lett., 32, L15708, doi:10.1029/2005GL023061.
Gordon, N. D., and J. R. Norris, 2010: Cluster analysis of midlatitude oceanic cloud regimes: Mean properties and temperature sensitivity. Atmos. Chem. Phys., 10, 6435–6459, doi:10.5194/acp-10-6435-2010.
Govekar, P. D., C. Jakob, M. J. Reeder, and J. Haynes, 2011: The three-dimensional distribution of clouds around Southern Hemisphere extratropical cyclones. Geophys. Res. Lett., 38, L21805, doi:10.1029/2011GL049091.
Grant, A. L. M., 2001: Cloud-base fluxes in the cumulus-capped boundary layer. Quart. J. Roy. Meteor. Soc., 127, 407–421.
Gregory, D., and P. R. Rowntree, 1990: A mass-flux convection scheme with representation of cloud ensemble characteristics and stability dependent closure. Mon. Wea. Rev., 118, 1483–1506.
Gregory, D., and S. Allen, 1991: The effect of convective downdraughts upon NWP and climate simulations. Proc. Ninth Conf. on Numerical Weather Prediction, Denver, CO, Amer. Meteor. Soc., 122–123.
Haynes, J. M., R. T. Marchand, Z. Luo, A. Bodas-Salcedo, and G. L. Stephens, 2007: A multi-purpose radar simulation package: Quickbeam. Bull. Amer. Meteor. Soc., 88, 1723–1727.
Haynes, J. M., C. Jakob, W. B. Rossow, G. Tselioudis, and J. Brown, 2011: Major characteristics of Southern Ocean cloud regimes and their effects on the energy budget. J. Climate, 24, 5061–5080.
Hogan, R. J., A. J. Illingworth, E. J. O’Connor, and J. P. V. Poiares Baptista, 2003: Characteristics of mixed-phase clouds. II: A climatology from ground-based lidar. Quart. J. Roy. Meteor. Soc., 129, 2117–2134, doi:10.1256/qj.01.209.
Im, E., C. Wu, and S. L. Durden, 2005: Cloud profiling radar for the CloudSat mission. IEEE Aerosp. Electron. Syst. Mag., 20, 15–18, doi:10.1109/MAES.2005.1581095.
King, M. D., and Coauthors, 2003: Cloud and aerosol properties, precipitable water, and profiles of temperature and humidity from MODIS. IEEE Trans. Geosci. Remote Sens., 41, 442–458, doi:10.1109/TGRS.2002.808226.
Klein, S. A., and C. Jakob, 1999: Validation and sensitivities of frontal clouds simulated by the ECMWF model. Mon. Wea. Rev., 127, 2514–2531.
Kristjánsson, J. E., J. M. Edwards, and D. L. Mitchell, 2000: Impact of a new scheme for optical properties of ice crystals on climates of two GCMs. J. Geophys. Res., 105 (D8), 10 063–10 079.
Li, Z., and H. G. Leighton, 1993: Global climatologies of solar radiation budgets at the surface and in the atmosphere from 5 years of ERBE data. J. Geophys. Res., 98 (D3), 4919–4930.
Lock, A. P., 2001: The numerical representation of entrainment in parameterizations of boundary layer turbulent mixing. Mon. Wea. Rev., 129, 1148–1163.
Lock, A. P., A. R. Brown, M. R. Bush, G. M. Martin, and R. N. B. Smith, 2000: A new boundary layer mixing scheme. Part I: Scheme description and single-column model tests. Mon. Wea. Rev., 128, 3187–3199.
Marchand, R., and T. Ackerman, 2010: An analysis of cloud cover in multiscale modeling framework global climate model simulations using 4 and 1 km horizontal grids. J. Geophys. Res., 115, D16207, doi:10.1029/2009JD013423.
Marchand, R., G. G. Mace, T. Ackerman, and G. L. Stephens, 2008: Hydrometeor detection using Cloudsat—An Earth-orbiting 94-GHz cloud radar. J. Atmos. Oceanic Technol., 25, 519–533.
Marchand, R., T. Ackerman, M. Smyth, and W. B. Rossow, 2010: A review of cloud top height and optical depth histograms from MISR, ISCCP, and MODIS. J. Geophys. Res., 115, D16206, doi:10.1029/2009JD013422.
Martin, G. M., M. R. Bush, A. R. Brown, A. P. Lock, and R. N. B. Smith, 2000: A new boundary layer mixing scheme. Part II: Tests in climate and mesoscale models. Mon. Wea. Rev., 128, 3200–3217.
Meehl, G. A., C. Covey, B. McAvaney, M. Latif, and R. J. Stouffer, 2005: Overview of the Coupled Model Intercomparison Project. Bull. Amer. Meteor. Soc., 86, 89–93.
Moroney, C., R. Davies, and J.-P. Muller, 2002: Operational retrieval of cloud-top heights using MISR data. IEEE Trans. Geosci. Remote Sens., 40, 1532–1540, doi:10.1109/TGRS.2002.801150.
Muller, J.-P., A. Mandanayake, C. Moroney, R. Davies, D. J. Diner, and S. Paradise, 2002: MISR stereoscopic image matchers: Techniques and results. IEEE Trans. Geosci. Remote Sens., 40, 1547–1559, doi:10.1109/TGRS.2002.801160.
Naud, C. M., A. D. Del Genio, and M. Bauer, 2006: Observational constraints on the cloud thermodynamic phase in midlatitude storms. J. Climate, 19, 5273–5288.
Pincus, R., S. Platnick, S. A. Ackerman, R. S. Hemler, and R. J. P. Hofmann, 2012: Reconciling GCM-simulated and satellite-observed views of clouds: MODIS, ISCCP, and the limits of instrument simulators. J. Climate, 25, 4699–4720.
Roberts, N. M., S. Cole, R. M. Forbes, R. Moore, and D. Boswell, 2009: Use of high-resolution NWP rainfall and river flow forecasts for advance warning of the Carlisle flood, north-west England. Meteor. Appl., 16, 23–34, doi:10.1002/met.94.
Rossow, W. B., and R. A. Schiffer, 1999: Advances in understanding clouds from ISCCP. Bull. Amer. Meteor. Soc., 80, 2261–2287.
Senior, C. A., and J. F. B. Mitchell, 1993: Carbon dioxide and climate: The impact of cloud parameterization. J. Climate, 6, 393–418.
Senior, C. A., and Coauthors, 2010: Synergies between numerical weather prediction and general circulation climate models. The Development of Atmospheric General Circulation Models: Complexity, Synthesis, and Computation, L. Donner, R. Somerville, and W. Schubert, Eds., Cambridge University Press, 76–116.
Stephens, G. L., and Coauthors, 2008: CloudSat mission: Performance and early science after the first year of operation. J. Geophys. Res., 113, D00A18, doi:10.1029/2008JD009982.
Stephens, G. L., and Coauthors, 2010: Dreary state of precipitation in global models. J. Geophys. Res., 115, D24211, doi:10.1029/2010JD014532.
Trenberth, K. E., and J. T. Fasullo, 2010: Simulation of present-day and twenty-first-century energy budgets of the southern oceans. J. Climate, 23, 440–454.
Uppala, S. M., and Coauthors, 2005: The ERA-40 Re-Analysis. Quart. J. Roy. Meteor. Soc., 131, 2961–3012.
Walters, D. N., and Coauthors, 2011: The Met Office Unified Model Global Atmosphere 3.0/3.1 and JULES Global Land 3.0/3.1 configurations. Geosci. Model Dev., 4, 1213–1271, doi:10.5194/gmd-4-919-2011.
Webb, M., C. Senior, S. Bony, and J. J. Morcrette, 2001: Combining ERBE and ISCCP data to assess clouds in the Hadley Centre, ECMWF and LMD atmospheric climate models. Climate Dyn., 17, 905–922.
Wielicki, B. A., B. R. Barkstrom, E. F. Harrison, R. B. Lee III, G. L. Smith, and J. E. Cooper, 1996: Clouds and the Earth’s Radiant Energy System (CERES): An Earth observing system experiment. Bull. Amer. Meteor. Soc., 77, 853–868.
Wild, M., 2005: Solar radiation budgets in atmospheric model intercomparisons from a surface perspective. Geophys. Res. Lett., 32, L07704, doi:10.1029/2005GL022421.
Wild, M., 2008: Short-wave and long-wave surface radiation budgets in GCMs: A review based on the IPCC-AR4/CMIP3 models. Tellus, 60A, 932–945, doi:10.1111/j.1600-0870.2008.00342.x.
Wilkinson, J. M., A. N. F. Porson, F. J. Bornemann, M. Weeks, P. R. Field, and A. P. Lock, 2012: Improved microphysical parametrization of drizzle and fog for operational forecasting using the Met Office Unified Model. Quart. J. Roy. Meteor. Soc., doi:10.1002/qj.1975, in press.
Williams, K. D., and G. Tselioudis, 2007: GCM intercomparison of global cloud regimes: Present-day evaluation and climate change response. Climate Dyn., 29, 231–250, doi:10.1007/s00382-007-0232-2.
Williams, K. D., and M. J. Webb, 2009: A quantitative performance assessment of cloud regimes in climate models. Climate Dyn., 33, 141–157, doi:10.1007/s00382-008-0443-1.
Wilson, D. R., and S. P. Ballard, 1999: A microphysically based precipitation scheme for the UK Meteorological Office Unified Model. Quart. J. Roy. Meteor. Soc., 125, 1607–1636.
Wilson, D. R., A. C. Bushell, A. M. Kerr-Munslow, J. D. Price, and C. J. Morcrette, 2008a: PC2: A prognostic cloud fraction and condensation scheme. I: Scheme description. Quart. J. Roy. Meteor. Soc., 134, 2093–2107, doi:10.1002/qj.333.
Wilson, D. R., A. C. Bushell, A. M. Kerr-Munslow, J. D. Price, C. J. Morcrette, and A. Bodas-Salcedo, 2008b: PC2: A prognostic cloud fraction and condensation scheme. II: Climate model simulations. Quart. J. Roy. Meteor. Soc., 134, 2109–2125, doi:10.1002/qj.332.
Winker, D. M., and Coauthors, 2010: The CALIPSO mission: A global 3D view of aerosols and clouds. Bull. Amer. Meteor. Soc., 91, 1211–1229.
Zelinka, M. D., and D. L. Hartmann, 2012: Climate feedbacks and their implications for poleward energy flux changes in a warming climate. J. Climate, 25, 608–624.
Zelinka, M. D., S. A. Klein, and D. L. Hartmann, 2012: Computing and partitioning cloud feedbacks using cloud property histograms. Part II: Attribution to changes in cloud amount, altitude, and optical depth. J. Climate, 25, 3736–3754.
Zhang, Y., W. B. Rossow, A. A. Lacis, V. Oinas, and M. I. Mishchenko, 2004: Calculation of radiative fluxes from the surface to top of atmosphere based on ISCCP and other global data sets: Refinements of the radiative transfer model and input data. J. Geophys. Res., 109, D19105, doi:10.1029/2003JD004457.
Zhang, Y., S. Klein, G. G. Mace, and J. Boyle, 2007: Cluster analysis of tropical clouds using CloudSat data. Geophys. Res. Lett., 34, L12813, doi:10.1029/2007GL029336.