Cloud properties from the Australian Community Climate and Earth System Simulator (ACCESS1.3) are evaluated using the Cloud Feedback Model Intercomparison Project (CFMIP) Observational Simulator Package (COSP). CloudSat, CALIPSO, and International Satellite Cloud Climatology Project (ISCCP) observations are used to evaluate the modeled cloud cover, condensate properties, and cloud optical depths for two seasons. The global distribution of cloud in the model is generally well represented with maximum high cloud in the tropics and low cloud over the eastern edges of the ocean basins. The model captures the observed position of the midlatitude storm track clouds and the modeled cloud top heights compare well with the observations in the upper troposphere. However, there is a lack of modeled midlevel cloud in the tropics and midlatitudes. The average high cloud cover in the Tropical Warm Pool region shows good agreement with CALIPSO. However, the modeled radar reflectivities and lidar scattering ratios are biased toward lower values, suggesting that the ice water contents and particles sizes of these clouds in the model are too small. Over the Southern Ocean the modeled cloud cover is underestimated due to a lack of mid- and low-level cloud. The low clouds over the Southern Ocean and the California stratocumulus clouds in the model have too little condensate and optical thickness and too much rain and drizzle. A sensitivity experiment showed that reducing the ice fall speeds improves aspects of the modeled cloud properties by increasing the frequency of occurrence of high clouds with large scattering ratios and optically thick low clouds. ACCESS1.3 has a reasonable representation of cloud. However, the underestimate of ice water content and particles sizes in high clouds and the too frequent occurrence of drizzle may impact the modeled cloud feedbacks and regional precipitation associated with current and perturbed climates.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Clouds are an integral part of the climate system, affecting the surface and top of atmosphere energy balance through their vertical redistribution of heat and moisture. To a large extent the global atmospheric energy balance is determined by the equilibrium between radiative cooling and latent heating associated with precipitation [Stephens, 2005] and the vertical structure of clouds contribute to determining these heating profiles. Wang and Rossow  showed through perturbing the cloud vertical structure in global climate model (GCM) simulations that clouds directly affect the atmospheric circulation. By keeping constant the cloud cover, condensate amount and particle sizes but modifying the vertical distribution of these cloud properties, they demonstrated direct changes to radiative cooling and static stability, which in turn induced changes in the Hadley circulation in their simulations.
 Cloud induced changes in the atmospheric energy balance and circulation come about due to direct and indirect effects because clouds tie together many of the components of the atmospheric energy and hydrological cycles. This coupling between clouds and dynamics, turbulence, radiative transfer, hydrological, chemical and surface processes, occurs across a range of spatial and temporal scales and has been recognized for some time to be a key aspect in the problems associated with modeling cloud processes [e.g., Arakawa, 1975] and their feedbacks in the climate system [e.g., Stephens, 2005]. Differences in cloud feedbacks between GCMs account for much of the uncertainty regarding climate sensitivity [Randall et al., 2007]. Senior  showed that in order to understand differences between cloud feedbacks in different models it is necessary to examine the horizontal and vertical cloud properties associated with climate changes and relate these to differences in physical parameterizations.
 The parameterization of clouds in GCMs has come a long way from the pioneering work of Sundqvist , and this progress has been intimately coupled to the advances in observing cloud properties. The new generation of satellite observations provides unparalleled opportunities to evaluate the global macro and microphysical properties of clouds simulated by GCMs. In this study we take advantage of these observations to evaluate the new Australian Community Climate and Earth System Simulator, ACCESS. The global distribution of simulated cloud cover, and cloud occurrence as a function of height is compared to observations, as well as the radar reflectivities, lidar scattering ratios and cloud optical depths, that give information about the amount and type of condensate within the clouds. Given that some key uncertainties in climate predictions are related to cloud feedbacks and regional precipitation, understanding how well the clouds in the current climate system are simulated is important for analyzing the ACCESS climate model output, such as that contributed to the Coupled Model Intercomparison Project 5 (CMIP5) [Taylor et al., 2012] and also helps guide physical parameterization development.
 A description of the model and the observations used in this study is given in section 2. Section 3 evaluates the global distribution and seasonal variation of clouds, and section 4 examines the detailed vertical structure of simulated clouds in three climatically important regions. The impact of reducing the fall speeds of ice aggregates is explored in section 5 and the conclusions of this study are summarized in section 6.
2 Description of Model and Observations
 ACCESS is a new coupled climate and earth system model that has been developed as a joint initiative between CSIRO and the Bureau of Meteorology in partnership with Australian universities. The model provides a framework for numerical weather prediction and coupled climate modeling, and enables research into processes occurring in the Earth system. The atmospheric component of ACCESS1.3 is the UK Met Office Unified Model (UM) with a horizontal resolution of 1.25° latitude and 1.875° longitude. The vertical coordinate is height based and terrain following and there are 38 vertical levels. The lowest 1.1 km has eight levels, with the lowest level at 10 m and the model top at 39 km. The model is based on the UM version GA1.0 described in Hewitt et al. . Modifications from this configuration are documented in detail in D. Bi et al. (The ACCESS coupled model: Description, control climate and evaluation, submitted to Australian Meteorological and Oceanographic Journal, 2012) and include changes to the representation of clouds and radiation as well as a different land surface scheme, the CSIRO Atmosphere Biosphere Land Exchange model (CABLE). These modifications are briefly summarized in Table 1. along with the standard GA1.0 model settings.
Table 1. Summary of Model Dynamics and Physical Parameterizations
Davies et al.  nonhydrostatic, two-time-level semi-Lagrangian advection scheme
Prognostic Cloud Prognostic Condensate (PC2) [Wilson et al., 2008], where each physical process in the model act as sources and sinks of cloud condensate and cloud fraction, includes the additional source of ice cloud fraction due to the process of depositional growth as described in Franklin et al. , cloud area parameterization from Boutle and Morcrette  used to account for the effects of coarse vertical resolution on liquid cloud cover
Wilson and Ballard  single-moment bulk microphysics scheme with ice content a prognostic variable
Modified mass flux scheme based on Gregory and Rowntree  with smoothed adaptive detrainment and CAPE closure based on relative humidity
Lock et al.  with explicit cloud top entrainment [Lock, 1998], air-sea flux parameterizations based on results from Fairall et al. 
Edwards and Slingo , two-stream equations, full treatment of scattering and aerosols, and the triple-cloud scheme of Shonk et al.  is used to account for cloud horizontal inhomogeneity
Gravity wave drag
Webster et al.  flow blocking scheme and spectral gravity wave scheme described in Warner et al. 
CSIRO Atmosphere Biosphere Land Exchange model (CABLE) [Wang et al., 2011]
 The Wilson and Ballard  single-moment microphysics scheme is used in the model, which includes a single prognostic ice water variable. For the microphysical process calculations this variable is separated into a large ice category (aggregates) and a small ice category (crystals) by a diagnostic function of temperature. The particle size distribution of these categories is exponential and takes the form
where D is the particle diameter, λ is the slope parameter and is the intercept parameter given by with TC the temperature in °C and the constant nai set to 4.0 × 106 m-4 for the aggregates and 80.0 × 106 m-4 for the ice crystals. The mass-diameter relationship is the same for both ice categories
where M is the particle mass (kg m-2), a is 0.0185 and b is 1.9 [Brown and Francis, 1995]. The fall speeds are determined by the mass- and area-dimensional power laws from Mitchell 
where vt is the fall speed (m s-1), ρ is the density, ρ0 is a reference density equal to 1 kg kg-1, η is the dynamic viscosity of air, g is the acceleration due to gravity and the values of the constants are given in Table 2.
Table 2. Values of the Constants Used in the Fall Speed Parameterizations of Ice
 The Cloud Feedback Model Intercomparison Project (CFMIP) Observational Simulator Package (COSP, version 1.3.1) [Bodas-Salcedo et al., 2011] has been implemented into ACCESS1.3 and the model outputs analyzed in this study are from the simulator running online in the model. COSP enables a more consistent evaluation of modeled cloud properties using satellite observations by following a model to satellite approach. In doing so the differences between variable definitions, sensor sensitivity limitations and vertical overlap assumptions between models and observations are minimized. COSP takes grid box mean profiles of modeled cloud, radiation and thermodynamic fields and breaks these into subcolumns using the Subgrid Cloud Overlap Profile Sampler (SCOPS) [Webb et al., 2001]. In this study model outputs are analyzed from the CloudSat, Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO), and International Satellite Cloud Climatology Project (ISCCP) simulators, the details of which are given in Bodas-Salcedo et al. .
 In order to compute the COSP radar reflectivity and scattering ratios the microphysical assumption made is that the ice water uses the particle size distribution for the large-ice category, as this will dominate the radar reflectivity which is sensitive to particle size. This is not the same treatment that is used by the radiation scheme in the model where the ice particle size is a function of temperature only [Edwards et al., 2007]. This methodology enables an evaluation of the cloud and microphysics schemes in the model with the CloudSat and CALIPSO observations. The effective radii of the liquid water components used in the radar and lidar simulators take the same values computed by the radiation scheme. The optical depths used by the passive instrument simulator is consistent with the radiation scheme. In the model the precipitating ice is treated as cloud and as such is included in the lidar simulator. However, liquid precipitation is not. It should also be noted that the radiation scheme uses the Tripleclouds scheme [Shonk et al., 2010] to account for the effect of cloud horizontal inhomogeneity. Tripleclouds divides the cloudy region of a grid box into two equal regions, one containing optically thicker cloud the other optically thinner cloud, and the radiative transfer calculations are performed for these regions. The COSP simulators do not take this inhomogeneity into consideration and instead use SCOPS to model the subgrid variability. The grid mean cloud water contents and cloud fractions are the same in both cases.
 As well as the assumptions described above that need to be made to use the simulators in the model, the simulators themselves use assumptions that should also be considered when comparing the simulator output with observations. The main assumptions for the radar simulator are that ice crystals are modeled as equal volume spheres, and multiple scattering is neglected [Haynes, 2007]. Multiple scattering is expected to increase the reflectivity below regions of large attenuation but is only appreciable for rain in excess of 3 mm h-1 [Marchand et al., 2009]. The lidar simulator also assumes that the particles are all spherical and the multiple scattering coefficient is set to a constant value of 0.7. Chepfer et al.  conducted sensitivity experiments and found that using a coefficient value of 0.3 made less than a 1% difference in the monthly mean cloud fractions derived from the lidar simulator. As described by Bodas-Salcedo et al.  the ISCCP simulator does not make full forward calculations of radiances and does not include biases due to the viewing angle or calibration that affect the observations. The effects of overlapping cloud are considered, and the maximum random overlap assumption used in the model is applied in COSP.
 The radar reflectivity observations used in this study are the 2B-GEOPROF data [Marchand et al., 2008] from the CloudSat cloud profiling radar. This near-nadir-looking radar has a frequency of 94 GHz, with a resolution of 480 m in the vertical and 1.4 km in the horizontal. The lidar scattering ratio observations are from the GCM-Oriented CALIPSO Cloud Product (GOCCP) [Chepfer et al., 2010]. These are level 1 data from Cloud-Aerosol Lidar with Orthogonal Polarsation (CALIOP) on board CALIPSO, which has a wavelength of 532 nm, a horizontal resolution of 333 m along track and 75 m across track, and a vertical resolution of 30 m below 8 km and 60 m above. The CloudSat and CALIPSO satellites are part of the A-Train constellation of satellites and are Sun synchronous, meaning that they pass over each location at approximately 1:30 local time in both the morning and afternoon. ISCCP uses observations of radiances from passive imagers onboard Sun-synchronous and geostationary satellites to produce an observational data product of cloud top pressure and cloud optical depth [Rossow and Schiffer, 1999].
 The model outputs analyzed in this study are from multiyear simulations using observed fields of sea surface temperatures and sea ice, as well as atmospheric gases and aerosol concentrations, i.e., an Atmospheric Model Intercomparison Project (AMIP) experiment [Taylor et al., 2012]. The monthly mean COSP outputs are produced from daily means of 3-hourly calculations and the number of subcolumns used in the simulators is 100. D. Konsta et al. (Evaluation of clouds simulated by the LMDZ5 GCM using A-Train satellite observations (CALIPSO-PARASOL-CERES), submitted to Climate Dynamics, 2012) showed that the CALIPSO simulator is insensitive to the frequency of the simulator calculations (whether it be 1.5-, 3-, or 6-hourly calculations) and the number of subcolumns provided there are at least 20. The results presented are for seasonal means and Bodas-Salcedo et al.  discussed their results for a season being insensitive to the spatial sampling, with their individual monthly results showing the same features as their seasonal results. While the Bodas-Salcedo et al.  study also used the UM, they used a global weather prediction version of the model to conduct short-range numerical weather prediction experiments with a radar simulator. In this study we use a coarser resolution climate version of the model with different parameterizations including a new prognostic cloud scheme PC2 [Wilson et al., 2008] and the CABLE land surface scheme, and multiple satellite simulators to evaluate long time integrations of the model. A thorough evaluation of the modeled cloud properties is undertaken in this study and we note that a detailed analysis of the model's radiation budget is presented in Z. Sun et al. (Modifications to atmospheric physical parameterizations aimed at improving the SST simulation in the ACCESS coupled model, submitted to Australian Meteorological and Oceanographic Journal, 2012).
3 Evaluation of Modeled Global Cloud Properties
 For both of the seasons December 2006 to February 2007 (DJF) and June–August 2007 (JJA) the GOCCP observation of the globally averaged total cloud fraction is 0.67. The value from the ACCESS1.3 lidar simulator in COSP is 0.61 for DJF and 0.60 for JJA. While the globally averaged cloud cover produced by the model is close to that observed, there are differences in the horizontal distributions of clouds as shown in Figure 1. The total cloud observations show maximum cloud cover for DJF over the tropical land regions, the Tropical Warm Pool (TWP) and Intertropical Convergence Zone (ITCZ), the North Atlantic and around the entire Southern Ocean. Apart from the ITCZ the model produces the tropical maxima. However, the maximum cloud cover regions are more localized than the observations and there is a tendency to produce too little cloud cover throughout the subtropical and trade wind regions. These cloud properties are explored in more detail in a companion paper that uses compositing techniques to evaluate the ACCESS1.3 tropical cloud properties (C. N. Franklin et al., Evaluation of clouds in ACCESS using the satellite simulator package COSP: Regime-sorted tropical cloud properties, submitted to Journal of Geophysical Research, 2012). The modeled total cloud cover in the northern midlatitudes compares reasonably well with the observations during DJF, except over Russia where the modeled cloud fraction is too low by up to 0.25. In the southern midlatitudes the cloud fraction is too low over the Southern Ocean, typically by 0.20. The bias in the model to produce lower cloud cover over the Southern Ocean is also apparent in JJA. The total cloud fractions greater than 0.9 observed over the North Pacific and North Atlantic are simulated very well by the model during JJA. However, there is about 0.3 too little cloud fraction produced by the model over much of the United States in this season.
 To examine in more detail the reasons for the model's successes and failures in producing the total cloud amount, Figure 2 shows the contributions from the high (50–440 hPa), midlevel (440–680 hPa) and low clouds (>680 hPa) from the lidar simulator in the model and the GOCCP observations for DJF. ACCESS1.3 has skill in modeling the high-cloud fractions with generally very good agreement between the model and the observations. The maxima in the model and observations occur in the regions of deep convection over land, the maritime continent and the TWP, with the model not producing as laterally extensive high cloud in the maritime continent region compared to the observations. The model well represents the high-cloud fractions in the South Pacific Convergence Zone (SPCZ) with underestimates typically less than 0.1.
 The more localized regions of maximum total cloud fraction over the tropical land areas are predominately due to the lack of midlevel cloud across the tropics (Figures 2c and 2d). Too little midlevel cloud is a well-known shortcoming of GCMs and in the tropics this has been attributed to being caused by a lack of detrainment from the cumulus parameterization [e.g., Bodas-Salcedo et al., 2008]. Much of the 0.4 underestimate of total cloud fraction over Eastern and Northern Australia in DJF is due to completely missing midlevel cloud.
 The GOCCP observations and the model simulator output both show low-cloud fractions greater than 0.6 at the eastern edges of the ocean basins in the subtropical oceanic regions during DJF (Figures 2e and 2f). The observed cloud cover in these regions tends to increase during JJA with some locations showing overcast conditions, which is also simulated by the model (Figure 3e and 3f). The minimum low cloud cover over the deep convective regions is due to the attenuation of the lidar signal that occurs where thick high clouds are present and prevents the lidar from determining whether lower-level clouds are underneath [Chepfer et al., 2010]. The model does not capture the GOCCP low-cloud fractions larger than 0.7 that are commonly observed in the Southern Ocean, particularly in the regions south of 60° during both DJF and JJA. The Southern Ocean total cloud fraction error in the model is due to too little cloud cover being produced in both the middle and lower levels. Taking an average of the DJF cloud fraction between 40°S and 60°S around the globe gives an underestimate of the low (middle) level cloud cover of 0.08 (0.12). An underestimate of middle- and low-level clouds in the model could result from upper level clouds that are optically thicker or occur more frequently in the model compared to the observed high clouds, obscuring the clouds below from being observed by the lidar simulator. However, this is not the case in the Southern Ocean, with the high-cloud fraction being overestimated by the model by less than 0.02, and as will be shown in section 4.2 the ice water content of the high clouds in this region tends to be underestimated.
 Table 3 shows the average root-mean-square errors (RMSE) and biases for the shortwave, longwave and net cloud radiative effect (CRE) at the top of the atmosphere for the DJF season. CRE is a measure of the effect of clouds on the downward radiative fluxes and the observations used to compute these statistics are from Clouds and the Earth's Radiant Energy System (CERES) Energy Balanced and Filled-Top of Atmosphere Ed6.2r [Loeb et al., 2008]. ACCESS1.3 underestimates the strength of the shortwave CRE with the average RMSE of about 23 W m-2 being very similar for the global, tropical (30°N–30°S) and Southern Ocean (40°S–60°S) regions (all longitudes). The lack of cloud over the Southern Ocean results in the largest bias occurring in this region, with an average underestimate of the strength of the shortwave CRE of 18 W m-2. The longwave CRE is underestimated on the average for the globe and the tropics, and is too strong over the Southern Ocean. The average net CRE globally is too low by 4.5 W m-2, with the best result in the tropical region with an average net CRE bias of 1.6 W m-2. The results in this table from the fall speed sensitivity test will be discussed in section 5.
Table 3. December 2006 to February 2007 Global, Tropical (30°N–30°S), and Southern Ocean (40°S–60°S) Cloud Radiative Effect (CRE, W m-2) Root-Mean Square Error (RMSE) and Bias Compared to CERES EBAF
Global fall speed
Tropics fall speed
Southern Ocean control
Southern Ocean fall speed
 Using the radar and lidar observations and simulators to evaluate the vertical structure of modeled cloud properties provides a wealth of information to understand the performance of current parameterizations and further model development. DJF zonal means from the CloudSat and GOCCP observations are shown in Figure 4 with the output from the COSP radar and lidar simulators. To reduce the effect of false detections that occur close to the limit of the CloudSat sensitivity, the CloudSat and radar imulator results in Figure 4 and those in section 4 use a threshold of −25 dBZ rather than −30 dBZ [Marchand et al., 2009]. In Figure 4 the large-scale circulation is evident and the model reproduces well the clear sky region of the descending branch of the Hadley circulation in the Northern Hemisphere. The cloud top heights follow the tropopause height, decreasing from the tropics to the poles and are well captured by the model. GOCCP has higher cloud tops than CloudSat due to the inability of the radar to detect small ice particles and this distinction is produced by the model. The model overestimates the maximum GOCCP cloud fraction of the high clouds in the tropics but has good agreement with the CloudSat observations, suggesting that the model has too many occurrences of small ice particles in the upper tropical troposphere. This illustrates the advantage of using both observational data sets that provide complementary information and this will be exploited in the next section.
 The CloudSat observations show maximum cloud or hydrometeor fraction in the low levels centered on 60° in both hemispheres (Figure 4a). Note that the levels below about 1 km in the CloudSat observations are not used due to ground clutter contamination. The radar and lidar observations show the cloud fraction in these frontal regions extending more vertically than what is shown in the model results, with the model tending to produce cloud in a thinner vertical layer with larger cloud fractions in this layer. The lack of modeled midlevel cloud is apparent in both data sets in the tropics and midlatitudes. Compared to CloudSat the model does not produce enough cloud between 2 and 7 km in the midlatitudes and between 2 and 10 km in the tropics. The result is similar for GOCCP. However, due to the different sensitivities of the lidar and radar, the cloud cover in the midlatitudes is overestimated by the model lidar simulator above 6 km between the latitude range of 40°–75° in both hemispheres by 0.05–0.1.
4 Regional Evaluation of ACCESS1.3 Clouds
 To examine the modeled cloud properties in more detail, normalized histograms as a function of height from the CloudSat, GOCCP and ISCCP simulators are compared with observations in three climatically important regions: the Tropical Warm Pool, the Southern Ocean and the California stratocumulus region. These regions differ in meteorological regime, with the clouds being produced by deep convection, synoptic storms and subtropical subsidence.
4.1 Tropical Warm Pool (5°S–20°N, 70°E–150°E)
 The TWP is characterized by deep convection that influences tropical variability and the global climate [e.g., Neale and Slingo, 2003]. The convection in this region is forced by a wide range of spatial and temporal scales including localized thunderstorms, the Madden-Julian oscillation and the monsoon, and it is challenging to represent these systems in GCMs. Given the importance of this region for the global climate, a detailed examination of the model's ability to represent the complex cloud structures observed in this region has been undertaken. Figure 5 shows the profiles of average DJF cloud fraction in this region from the A-Train satellite radar and lidar observations and model simulators. The observed and modeled lidar cloud fraction at each level is determined by dividing the frequency of occurrence of the lidar scattering ratios greater than 5 by the frequency of occurrence of all scattering ratios that are not fully attenuated. Comparing the GOCCP observations and the lidar simulator output shows that the cloud fraction in the upper troposphere associated with the outflow of deep convection in this region is represented very well in the model. As seen in the previous section the midlevel cloud in the TWP region is significantly underestimated in the model and Figure 5 shows that the modeled cloud cover is underestimated between the levels of 2–10 km. The average peak low-level cloud cover at 1 km is overestimated by 0.02 in the model.
 The CloudSat observations show that the mean cloud fraction profile in the TWP has more cloud cover in the levels below 10 km than the GOCCP observations. This is due to the large number of attenuated lidar profiles that occur in these regions where there are thick clouds [Chepfer et al., 2010]. The CloudSat radar reflectivity is sensitive to the cloud condensate and precipitation amounts, and in the Rayleigh scattering regime particle size has a large influence on the radar reflectivity due to the reflectivity of hydrometeors being proportional to their diameter to the sixth power [e.g., Doviak and Zrnić, 1984]. The cloud fraction profile from the model radar simulator is underestimated throughout the atmospheric column between 2 and 15 km. However, the GOCCP observations and the model lidar simulator agree well between 10 and 15 km with differences less than 0.02. This suggests that smaller particle sizes at these heights may be the leading cause of the underestimated radar detected cloud occurrence in the model over the TWP. Between 6 and 10 km there are not enough cloud/hydrometeor occurrences in the model and below this level the frequency of occurrence from the model closely resembles that shown in the observations.
 The CloudSat radar reflectivity–height histogram for the TWP region displays a boomerang type shape with maximum occurrences at 13 km in the low reflectivity range of −30 to −20 dBZ as shown in Figure 6a. The maximum occurrence reduces with height and moves to larger reflectivities so that at 5 km the reflectivity that occurs most often in the observations is between 5 and 10 dBZ. This structure reflects the aggregation process with particles growing in size as they collect other particles while descending [e.g., Bodas-Salcedo et al., 2011], as well as the larger water contents that occur more frequently as height reduces toward the freezing level [e.g., May and Rajopadhyaya, 1996]. At the melting level at about 5 km this structure reverses and the largest reflectivity occurrence moves to lower dBZ values due to attenuation. At the lowest levels detectable by CloudSat there is a bimodal structure in the observations with two regimes clustered around +5 and −30 dBZ, representing a raining and a nonprecipitating mode. The lower levels of the radar reflectivity from the model show a different behavior with one dominant regime being simulated that represents a wide spread light precipitation/drizzling regime near −10 dBZ. In the upper levels where only ice is present, the modeled reflectivity has the correct structure but is biased toward lower reflectivity values and shows a narrower distribution with no occurrence of clouds above 10 km with reflectivities greater than 0 dBZ. This indicates that either the ice water content in the model is too low and/or the ice particle sizes are too small in the clouds that are detected by the radar simulator.
 Complementary information can be gained by analyzing the lidar scattering ratio–height histogram in conjunction with the CloudSat radar reflectivity–height histogram due to the differences in measurements, with the radar being insensitive to thin ice cloud and the lidar unable to penetrate thick cloud and strongly attenuated by liquid. The lidar scattering ratio is measured relative to the backscatter a molecular atmosphere without clouds or aerosols would have produced. In the GOCCP data set clouds are defined as having scattering ratios (SR) larger than 5, with the values less than this due to aerosols, very thin cloud or instrument noise [Chepfer et al., 2010] and the histograms use only the nighttime observations to minimize these differences with the model simulator output. Deep convective cloud systems in the TWP produce many occurrences of SR values >10 at altitudes between 12 and 15 km (Figure 6c). The model captures the peak frequency of occurrence for observed SR values between 10 and 15. However these maxima occur at slightly lower heights in the model. SRs either side of this value though are different, with the model simulating too many smaller SR values and not enough larger SR>20. These results suggest that the model underestimates the ice water content in the upper troposphere and this is a common model deficiency [e.g., Su et al., 2011]. The cloud tops heights from the model compare well with the observations suggesting that the depth of convection is well modeled. The second region of the GOCCP histogram that shows a significant frequency of occurrence is between 5 and 9 km for SR>80. This area of the histogram corresponds to liquid cloud typical of congestus or mixed phase stratiform cloud and is almost completely missing in model. Running the lidar simulator without ice gives more occurrences of SR between 40 and 80 in the height range of 5–10 km (not shown). This demonstrates that the model does simulate thicker water clouds at these levels but these clouds are produced as part of deeper thick cloud systems that attenuate the lidar. Satoh et al.  showed that this cloud type was also underrepresented in their model evaluation using COSP. Given that they use a GCM of varying resolution in the TWP region of 3–15 km and their model does not use a cumulus convection scheme, this suggests that the lack of congestus-type cloud in the tropics is not purely due to horizontal resolution or convection scheme but is more likely due to the complex interplay of physical processes in models.
 In the lowest levels the model does not simulate the same variability in SR as the observations and tends to produce clusters within the SR range of 20–60, rather than a more uniform number of occurrences across all SR bins. This result is also apparent in the SR histogram figures for the other regions to be discussed in the following sections. A number of sensitivity tests were conducted to try to isolate the reasons for the model producing high occurrence rates in these SR bins. The lidar simulator was run once including only the convective cloud water and once with only the large-scale cloud water and these SR clusters were produced in each case, suggesting that it is not the thresholds that are imposed on the cloud water contents in the convection scheme or the autoconversion scheme that are exclusively responsible for this result. The effective radius used in the lidar simulator is calculated using the number concentration that is derived from the aerosol concentrations, testing using a constant cloud droplet number does not produce a more continuous distribution of low-level scattering ratios. Reducing the cloud water effective radius used in the lidar simulator by half smooths out the distribution for the range of SR between 15 and 40. However, the larger SR bins are simply shifted up to the next SR bin. For the highest SR this brings the model into better agreement with the observations suggesting that perhaps the cloud water effective radius being used in the radiation scheme is too large for the higher water contents. Further work is required to understand the reasons behind the SR histogram result in the low levels and we note that an earlier version of the UM, HadGEM1, produces the same result as does the Multiscale Modeling Framework 4 km model for the two lower SR bins [Bodas-Salcedo et al., 2011].
 The ISCCP histograms of cloud top pressure and cloud optical depth show a large occurrence of high optically thin cloud in the TWP region and the model tends to also produce these dominant cloud types in the levels above 180 hPa (Figures 6e and 6f). The most frequently simulated cloud by the model is very thin with an optical depth less than 0.3, which is below the detection limit of ISCCP. The number of clouds that occur above 180 hPa with optical depths greater than 10 are underestimated in the model, as are most clouds with tops between 680–180 hPa. The cloud type that occupies the upper right of the ISCCP histogram is characteristic of deep convection and the frequency of occurrence in the model for this cloud type is about half what is observed. However, there is a larger discrepancy for the midlevel topped clouds that are significantly underrepresented in the simulation. The frequency of occurrence of clouds with cloud tops below 680 hPa in the model is fairly similar to the ISCCP observations, although the frequencies tend to be biased low in agreement with the lack of thick low clouds in the model with lidar SR>60.
4.2 Southern Ocean (40°S–60°S, 130°W–175° W)
 The Southern Ocean plays an important role in the global climate system and its response to climate change due to ocean heat and carbon uptake. Trenberth and Fasullo  found that climate models simulate large biases in the atmospheric energy balance in this region due to the poor simulation of clouds. As discussed in section 3, ACCESS1.3 does not simulate enough lidar detected cloud cover over the Southern Ocean, predominately due to lack of midlevel cloud but also through too little low-level cloud cover (Figure 2). Comparing the average profile of cloud fraction from the lidar simulator in the model and the GOCCP observations in a region of the Southern Ocean shows that the model overestimates cloud fraction between 8 and 12 km by up to 0.1 (Figure 7). Between 2 and 8 km the modeled cloud fraction is underestimated by up to 0.07 and at about 1 km the cloud fraction is again overestimated by 0.05. The lack of midlevel cloud in the model compared to both observational data sets is consistent with the findings in the tropics discussed in the previous section and other studies of the UM [e.g., Bodas-Salcedo et al., 2008]. The convection scheme in the model detrains too little moisture in the midlevels and too much moisture in the upper levels [Derbyshire et al., 2011] and this contributes to the cloud errors seen in the Southern Ocean region. Comparing the average profile of cloud fraction from the radar simulator in the model and the CloudSat observations shows that the model does very well at matching the observed cloud/hydrometeor fraction profile, with a maximum difference of only 0.02–0.04. The reason for the better agreement with CloudSat in this region is due to the reflectivity being dominated by large particles and this will be discussed next.
 Figure 8 shows the histograms of radar reflectivity with height for the Southern Ocean region. Compared to the TWP, the maximum observed frequency of occurrence of ice cloud in the Southern Ocean region occurs at a lower altitude, 6–11 km, and across a broader range of reflectivities and the model simulates these differences in regional cloud properties. However, the model produces more occurrences than the observations of ice clouds above 5 km with reflectivities between −30 and −20 dBZ. The cloud top heights simulated by the model agree very well with the observations. Although the average cloud cover from the model compares well with CloudSat between 2 and 5 km, Figure 8 shows that this is due to the model simulating too many occurrences of reflectivities between −10 and 0 dBZ, and not enough between −30 and −20 dBZ. This suggests that the model is producing too many drizzle size raindrops and is the same error structure as was seen in the low levels of the TWP (Figure 6). The overproduction of drizzle has been observed in a global forecasting system using the UM [Bodas-Salcedo et al., 2008] and even though we use a different cloud scheme in ACCESS1.3 and, therefore, different cloud cover and in-cloud water contents, the use of the new prognostic cloud scheme makes little difference to this more microphysically driven error, as will be discussed in section 4.3.
 The GOCCP scattering ratio–height histogram for the Southern Ocean region (Figure 8c) shows a different structure to the TWP region. Cloud is observed to occur across all scattering ratios below 10 km. Between 10–14 km the observed lidar detected cloud occurs with SRs less than 30 and the maximum occurs within the SR range of 7–15. The histogram produced for this region from the model lidar simulator shows very few clouds occurring with scattering ratios larger than 25 in the levels above 2–3 km. The SR distribution from the model does not show the same amount of variability as that observed with essentially no SR greater than 60, which gives a narrower distribution. Instead the model produces more than double the occurrence of observed cloud with SRs between 5 and 15 in the levels between 8 and 11 km. This result shows that the model produces more occurrences of low ice water content and no occurrences of the high ice water contents that are observed in the Southern Ocean region. The most frequently observed scattering ratio in this region is the 60–80+ values that occur at heights between 1 and 2 km. The model produces close to the observed cloud cover in these levels but it does this by simulating more occurrences than observed for the lower SR between 15–60. The less frequently produced occurrence of SR>60 by the model leads to optically thinner clouds over the Southern Ocean as demonstrated by the weaker CRE from the model (Table 3).
 The observed ISCCP histogram (Figure 8e) shows that the cloud observed most frequently in the Southern Ocean region studied here, occurs with cloud tops between 680–800 hPa and optical depths between 3.6–9.4 typical of stratocumulus [e.g., Hahn et al., 2001]. Output from the model ISCCP simulator shows that the model produces cloud with the same cloud top pressure most often as the type that is observed. However, the modeled optical thickness is between 1.3 and 9.4 and the frequency of occurrence is lower. The model produces more frequencies of optically thin cloud with optical depths between 0.3 and 1.3 for all cloud top pressures than is observed, consistent with the lower ice water contents. Compared to the TWP region the model produces more cloud with tops in the midlevels in the Southern Ocean region, but again the frequency of clouds produced by the model is lower than the observations and also biased toward smaller optical depths.
4.3 California Stratocumulus (15°N–35°N, 110°W–140°W)
 Marine stratocumulus clouds are climatically important due to their large and persistent cloud fractions [Klein and Hartmann, 1993]. These clouds have a cooling effect due to the contrast in albedo that these clouds have compared to the underlying ocean surface. Marine boundary layer clouds are thought to be key in explaining the large model differences in cloud feedbacks [e.g., Bony and Dufresne, 2005], therefore, it is important to examine the detailed structure of these simulated clouds in ACCESS1.3 to better understand the model's ability to simulate present-day persistent stratocumulus clouds. For this region we use the JJA season as the California stratocumulus has larger cloud fraction in JJA compared to DJF, and there is also less cirrus in JJA to potentially obscure the satellite observations of the low cloud. Comparing the average profile of cloud fraction from the lidar simulator in the model to the GOCCP observations for JJA shows that the model agrees with the observations to within 0.02 in all levels above 1 km, however, below 1 km the cloud fraction is overestimated by the model by up to 0.1 (Figure 9). The comparison of the average profile from CloudSat and the radar simulator shows that the model produces the cloud frequency to within 0.05 of the observations, underestimating the cloud fraction between 5–15 km and overestimating below this level.
 The CloudSat observations in the California stratocumulus region in JJA (Figure 10) show an upper level cirrus cloud between 10–14 km and a low-level stratocumulus cloud between 1–2.5 km. The separation of these cloud layers in the observations is seen by low occurrences of all reflectivities with the minimum occurring between 2.5 and 5 km. The model, however, has more and/or larger hydrometeors present in these intermediate levels, suggesting that the model is not evaporating the hydrometeors as much as it should be above the low-level inversion. The model simulates fewer occurrences of cloud with reflectivities between −30 and −10 dBZ between 10 and 14 km, suggesting that the ice particle sizes and possibly the ice water content of the cirrus cloud are smaller than those observed. Between 1 and 2 km the observations show that most cloud occurs with reflectivities between −20 and −30 dBZ, representing nonprecipitating particles sizes. However, below these altitudes detectable by CloudSat extensive drizzle has been observed in this region [van Zanten et al., 2005]. The model produces cloud at this height predominately with reflectivities between 0 and −20 dBZ, which contains light rain and drizzle size particles. In the low levels the observations show a smooth transition across reflectivity, with the frequency of occurrence reducing as reflectivity increases. The model does not produce the same result and instead produces high occurrences across all reflectivity bins.
 The lidar scattering ratio–height histogram for the GOCCP observations shows that the model compares better with the GOCCP observations than with CloudSat with regard to the small cloud amount between 2 and 5 km (Figures 10c and 10d), suggesting that at these levels CloudSat and the radar simulator are detecting particles below higher clouds. The maximum frequency of occurrence for the cirrus clouds occurs in the model for the SRs between 10 and 15 at the heights between 12–15 km. This is also the SR where the observations show the highest frequency of occurrence. However, the observations extend over a deeper layer from 11–16 km for frequencies larger than 0.01. The scattering ratio–height histogram shows that the ice water contents simulated by the model in the cirrus cloud in the stratocumulus region are generally in good agreement with the observations, with a tendency to produce lower SRs. The stratocumulus cloud is predominately observed to have a scattering ratio greater than 80 and this large liquid water content is not simulated by the model at all. Instead the model produces low-level cloud in this region across the range of scattering ratios of 15–60. In this region the model and observations produce a very similar number of occurrences of full attenuation with frequencies of 0.03. The height of the low-level cloud is produced well by the model. However, the comparison with the Cloudsat and GOCCP observations suggest that the sizes and amounts of the cloud water in this cloud are different to those observed, with larger drizzle and raindrops in the model and lower liquid water paths. These results agree with those of Boutle and Abel  who studied the ability of the UM to simulate stratocumulus observed in the southeast Pacific. In their higher-resolution simulations they showed the same general features as those shown in Figure 10, with the model simulating too much drizzle and lower liquid water paths than the observations. Their study showed that these errors were not significantly reduced with increasing resolution and instead it was changes to the model microphysics parameterizations that resulted in a better simulation of stratocumulus.
 The ISCCP comparison in Figure 10 shows optically thinner clouds simulated by ACCESS1.3 in the low levels of the California stratocumulus region in agreement with the lidar simulator and GOCCP comparison. The stratocumulus clouds that occur most frequently in the observations occur with cloud tops between 680–800 hPa and optical depths between 3.6 and 23. In the model the cloud top height range for the most frequently occurring clouds is the same as the observations. However, the optical depths are lower with values between 1.3 and 9.4 occurring most often in the simulation. The model significantly underestimates the frequency of occurrence of shallow cloud with tops in the lowest levels between the surface and 800 hPa, and instead overestimates the number of clouds with tops between 560 and 680 hPa, in agreement with the CloudSat comparison. The optical thickness of the cirrus cloud from the model compares well with the ISCCP observations, with a tendency for the modeled cloud to be optically thinner, again in agreement with the lower reflectivities and ice water contents seen from the radar and lidar simulator comparisons with observations.
5 Impact of Reduced Ice Fall Speeds
 Using observations of ice fall speeds from a radar-lidar retrieval during the Tropical Warm Pool–International Cloud Experiment, Franklin et al.  showed that the PC2 cloud scheme produced particle fall speeds that were on average up to 25% faster than the observations in the levels between 4 and 8 km. Figure 5 shows that ACCESS1.3 does not produce enough midlevel cloud in the TWP region, particularly in the levels between 5 and 8 km, and a sensitivity experiment was performed to see what impact reducing ice fall speeds has on the modeled cloud properties. ACCESS1.3 has only one prognostic variable for ice water content. However, this is diagnostically separated by temperature into ice crystals and aggregates in the microphysical calculations to account for the different properties of these two types of ice hydrometeor. In the sensitivity experiment the fall speeds of the aggregates were reduced by changing the constant e (4) in Table 2 to 0.1381. In the calculation of the fall speeds this constant is used to determine the relationship between the Reynolds and Best numbers [Mitchell, 1996]. A reduction in this number without a corresponding change to the constant f in Table 2 equates to a reduction in the Reynolds number for a given Best number without a change in the slope of the relationship. Figure 1 in Mitchell  shows empirical Reynolds-Best number relationships from laboratory and field experiments. For the applicable range of Best numbers these measurements show a tendency for the slope to remain relatively constant and it is the intercept that shows the larger variability. In the sensitivity experiment the modification is made to the intercept parameter for the aggregates and there was no change made to the fall speeds of the ice crystals. This experiment is designed to test whether the impact of fall speed changes is significant on the modeled cloud properties and future work will explore the full sensitivity of the model to changes in the ice fall speeds through modifications to the particle sizes and mass-diameter relationships. The model was run for 6 months with the fall speed change before the DJF season to allow for any spin up effects. Figure 11 shows the results from this experiment for the TWP and Southern Ocean regions. The frequency of occurrence of midlevel cloud is increased when the fall speed of aggregates is reduced, as shown by comparing the radar simulator histogram in Figure 11a with that in Figure 6b. The high ice cloud is also affected in a positive way by increasing the frequency of occurrence of lidar scattering ratios larger than 20 (Figure 11c). The upper right corner of the ISCCP histogram has a small increase in the frequency of occurrence from the sensitivity experiment, which brings the modeled frequency of high cloud top altitudes and large optical depths closer to the observations. The frequency of occurrence of ISCCP midlevel topped clouds in the TWP region has also increased by a small amount, although it is still much less than the observations. Table 3 shows the results of the fall speed change on the CRE averaged over the tropics between 30°N and 30°S. The RMSE of the shortwave, longwave, and net CRE are improved when the fall speed of the aggregates is reduced. However, the bias of the shortwave and longwave CRE is increased. Together these results suggest that with this parameterization change the CRE in the tropics is modeled more accurately on a process level, but the integrated effect is not improved. There is no change in the tropical net CRE between the control and sensitivity experiments as there is the same cancellation of errors.
 The impact of the reduced aggregate fall speeds in the Southern Ocean region is to reduce the maximum frequency of occurrence of the ice cloud above 5 km, which was larger than the observations in the control run (Figure 11b). The lidar scattering ratio histogram shows that like in the TWP, the Southern Ocean high clouds occur more often with larger scattering ratios when the fall speeds of aggregates are reduced. The low clouds of the Southern Ocean region are impacted in a positive way, with significantly more low clouds occurring with scattering ratios between 60 and 80 in the sensitivity experiment. The ISCCP results reflect this by showing an increase in the frequency of optically thick cloud with tops between 800 and 680 hPa. Table 3 shows that these improvements in the cloud properties translate into improvements in the CRE over the Southern Ocean region averaged over all longitudes between 40°S and 60 °S. The shortwave, longwave and net CRE RMSE and biases are all reduced in the sensitivity experiment compared to the control. The largest improvement is in the SW CRE with a reduction of the RMSE and bias of 1.3 and 1.6 W m-2, respectively. The bias in the net CRE over the Southern Ocean has reduced by 1.6 W m-2. However, there still exists a significant average bias of 16.7 W m-2. The results from the sensitivity experiment in the California Stratocumulus region for the DJF season also show improvements with broader scattering ratio histograms for the high clouds and more optically thicker low clouds (not shown).
 The Cloud Feedback Model Intercomparison Project (CFMIP) observational simulator package COSP [Bodas-Salcedo et al., 2011] has been implemented in the ACCESS1.3 climate model and output from an atmosphere only simulation (i.e., an AMIP simulation) has been evaluated against observations. The simulator outputs evaluated are from the radar, lidar and ISCCP simulators within COSP. For the seasons December 2006–February 2007 and June–August 2007 the modeled total cloud cover from the lidar simulator compares reasonably well with the GOCCP observations and the errors in these seasons closely resemble the errors from 2008. The model simulates too little total cloud fraction over the Southern Ocean by up to 0.2, predominately due to an underestimate of midlevel cloud but there is also a contribution to the underestimate from the low-level clouds. Lack of midlevel cloud is also the cause for the underestimate of cloud fraction over Australia in DJF and the US in JJA by 0.3–0.4. The cloud cover in the SPCZ is modeled well. However, there is too little cloud in the ITCZ.
 In DJF the high cloud cover from the lidar simulator compares well with the GOCCP observations, with the regions of maximum high cloud over tropical continents tending to be more localized in the model. The tropical eastern Indian Ocean high cloud cover is underestimated in DJF and overestimated in JJA. The low-level cloud fraction from the model lidar simulator compares well with the GOCCP observations at the eastern edges of ocean basins with differences typically less than 0.05. However, over the Southern Ocean the low-cloud fraction is underestimated by 0.08 and is also underestimated in much of the trade wind regions. The ACCESS1.3 tropical cloud properties are examined in more detail in a companion paper (Franklin et al., submitted manuscript, 2012).
 Radar reflectivity–height, lidar scattering ratio–height and cloud top pressure–cloud optical depth histograms were examined in three regions: the Tropical Warm Pool, a region of the Southern Ocean and the California stratocumulus region. The comparison of the CloudSat, GOCCP and ISCCP histograms with those from the COSP outputs show that while the radar reflectivity, lidar scattering ratio and cloud optical depth that occurs most often in the upper levels of the modeled clouds in the TWP and Southern Ocean regions is the same as that observed, the number of modeled occurrences of large reflectivities, scattering ratios and optical depths are underestimated. This suggests that the ice water contents and particle sizes in the simulated clouds are smaller than the observations. The variability of simulated ice cloud properties is less than that observed, as indicated by the narrower radar reflectivity and lidar scattering ratio distributions for any given altitude in the upper levels. This result is likely due to the choice of prognostic microphysical variables in ACCESS1.3 and the more simplistic representation of microphysical processes in the bulk single-moment scheme used in the model. Given the warming effect of tropical cirrus and the impact that this could have on perturbed climate states, improving the ice water properties of these clouds in the model is important.
 The TWP and Southern Ocean regions show a lack of occurrence of midlevel cloud, with the average cloud fraction for the DJF season underestimated by 0.05 (0.1) in the TWP region for the GOCCP (CloudSat) comparison. In the Southern Ocean region the average midlevel cloud/hydrometeor frequency of occurrence from the radar simulator compares very well to CloudSat. However, this is due to too many occurrences of reflectivities between 0 and −10 dBZ characteristic of drizzle size drops and not enough occurrences of nonprecipitating cloud. The observations of the low-level clouds over the Southern Ocean show thick water clouds associated with large scattering ratios >80 and small drop sizes associated with reflectivities between −20 and −30 dBZ. The same cloud structure is observed in the California stratocumulus region, and the model exhibits the same biases for the Southern Ocean cloud in the low levels and the California stratocumulus, suggesting clouds with insufficient condensate and optical depths but also containing too much rain and drizzle.
 For the three regions analyzed the model shows a systematic bias to simulate too many occurrences of drizzle and light rain and not enough occurrences of nonprecipitating cloud. This occurred in all regions even though the meteorological regimes are very different: deep tropics, midlatitude frontal systems and subtropical stratocumulus. This finding agrees with that of Stephens et al.  who showed that models tend to produce precipitation twice as often as observed. Given the constraints on accumulated precipitation amounts due to the global energy balance, this implies that models produce precipitation with an intensity that is less than that observed. This has important implications for modeling the hydrological cycle in current and perturbed climates, as well as other physical processes that are sensitive to the frequency and intensity of precipitation such as land surface processes.
 A sensitivity experiment conducted, where the fall speed of ice aggregates in the model was reduced, showed improvements in many aspects of the cloud properties including: more occurrences of large lidar scattering ratios for the high ice clouds; modest increase in the tropical midlevel cloud; optically thicker clouds, both for high deep convection and low-level stratocumulus. These results highlight an avenue of model development, which has been shown to improve cloud properties in a number of ways including the shortwave cloud effects over the Southern Ocean in ACCESS1.3.
 The results of this study suggest a number of areas of parameterization development that might prove fruitful for improving the representation of clouds in the model. To address the lack of midlevel cloud a change to the convection scheme that increases the level of detrainment in the midlevels but not at the expense of the high levels would be helpful. Too little condensate may be related to resolution and later versions of the UM have increased resolution, therefore, investigating the changes in condensate amount due to resolution would be a useful future study. The ice water content is strongly influenced by the subgrid moisture distribution used in the PC2 cloud scheme [Wilson et al., 2008] and the representation and impact of this distribution could be explored. Incorporating a subgrid representation of moisture in the microphysical parameterizations is likely to improve the frequency and intensity of rainfall produced by the model and ideally this should be consistent with that used in the cloud scheme. Given the strong dependence of rain production on the autoconversion parameterization it would be useful to test the sensitivity of the model rain rates to this process. One possibly is to implement the scheme of Franklin  that has been shown to improve the simulation of stratocumulus compared to some other parameterizations (Y. Wang et al., Toward improving bulk microphysics parameterizations for regional and global simulations of aerosol indirect effects, submitted to Journal of Geophysical Research, 2012). Finally the cloud properties are highly likely to be improved by moving to a double-moment microphysics scheme. In particular, the sedimentation of particles is known to be significantly better represented with a double-moment scheme compared to a single-moment scheme [e.g., Milbrandt and Yau, 2005] and the prognostic number concentration should lead to a better representation of the effective radius and the optical properties of the modeled clouds.
 This work has been undertaken as part of the Australian Climate Change Science Program, funded jointly by the Department of Climate Change and Energy Efficiency, the Bureau of Meteorology, and CSIRO. This work was supported by the NCI National Facility at ANU. A. Bodas-Salcedo was supported by the Joint DECC/Defra Met Office Hadley Centre Climate Programme (GA01101). We thank and acknowledge the constructive comments from the reviewers.