J. Meteor. Res.  2019, Vol. 33 Issue (4): 651-665 PDF
http://dx.doi.org/10.1007/s13351-019-8206-y
The Chinese Meteorological Society
0

#### Article Information

ZHOU, Yihui, Yi ZHANG, Xinyao RONG, et al., 2019.
Performance of CAMS-CSM in Simulating the Shortwave Cloud Radiative Effect over Global Stratus Cloud Regions: Baseline Evaluation and Sensitivity Test. 2019.
J. Meteor. Res., 33(4): 651-665
http://dx.doi.org/10.1007/s13351-019-8206-y

### Article History

in final form March 6, 2019
Performance of CAMS-CSM in Simulating the Shortwave Cloud Radiative Effect over Global Stratus Cloud Regions: Baseline Evaluation and Sensitivity Test
Yihui ZHOU1,2, Yi ZHANG3, Xinyao RONG3, Jian LI3, Rucong YU3
1. State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing 100029;
2. University of Chinese Academy of Sciences, Beijing 100049;
3. State Key Laboratory of Severe Weather, Chinese Academy of Meteorological Sciences, China Meteorological Administration, Beijing 100081
ABSTRACT: The ability of climate models to correctly reproduce clouds and the radiative effects of clouds is vitally important in climate simulations and projections. In this study, simulations of the shortwave cloud radiative effect (SWCRE) using the Chinese Academy of Meteorological Sciences Climate System Model (CAMS-CSM) are evaluated. The relationships between SWCRE and dynamic–thermodynamic regimes are examined to understand whether the model can simulate realistic processes that are responsible for the generation and maintenance of stratus clouds. Over eastern China, CAMS-CSM well simulates the SWCRE climatological state and stratus cloud distribution. The model captures the strong dependence of SWCRE on the dynamic conditions. Over the marine boundary layer regions, the simulated SWCRE magnitude is weaker than that in the observations due to the lack of low-level stratus clouds in the model. The model fails to simulate the close relationship between SWCRE and local stability over these regions. A sensitivity numerical experiment using a specifically designed parameterization scheme for the stratocumulus cloud cover confirms this assertion. Parameterization schemes that directly depict the relationship between the stratus cloud amount and stability are beneficial for improving the model performance.
Key words: Chinese Academy of Meteorological Sciences Climate System Model (CAMS-CSM)     shortwave cloud radiative effect (SWCRE)     stratus cloud     model errors
1 Introduction

The performance of global models in simulating the cloud radiative effect is vitally important in representations of the present and future climates. Unfortunately, the poorly simulated clouds and associated radiative effects constitute one of the major reasons for the heavily discretized present-day simulations and future model projections of the climate (Cess et al., 1990; Dufresne and Bony, 2008; Webb et al., 2013).

The Chinese Academy of Meteorological Sciences Climate System Model (CAMS-CSM; Rong et al., 2018) is a newly developed global climate system model that will be a part of the Coupled Model Intercomparison Project Phase 6 (CMIP6). In this study, we evaluate the performance of this model in simulating cloud radiative effects over major global stratus cloud regions. By quantifying the model performance in this regard, we may have a better grasp of the model’s capabilities, which is helpful in understanding the effects of climate in the model.

The generation and maintenance of clouds and their radiative effects are related to multiscale atmospheric processes. Because the resolutions of global climate models are too coarse to explicitly resolve cloud dynamics, parameterization schemes based on statistical relationships and physical principles have been proposed to improve simulations of cloud radiative effects (Wang and Ding, 2005; Li et al., 2010; Wu et al., 2011; Guo and Zhou, 2014). Moreover, in the radiation transfer module of climate models, a poor representation of the subgrid cloud structures, such as vertical overlapping and horizontal variations in cloud condensates, also adversely impact the model’s performance (Shonk et al., 2010; Zhang and Jing, 2016).

Large inter-model differences in the cloud feedback are largely attributed to the shortwave component (Zelinka et al., 2013). Stratiform clouds are mainly responsible for the shortwave cloud radiative effect (SWCRE; Boucher et al., 2013), and lead to the largest annual mean changes in shortwave radiative fluxes (Zuidema and Hartmann, 1995; Chen et al., 2000). Additionally, the largest simulation uncertainty in tropical ocean regions is caused by low-level stratocumulus clouds (Bony and Dufresne, 2005). These marine boundary layer clouds typically occur under conditions of strong temperature inversion, low sea surface temperature, and moderate subsidence (Lin et al., 2009; Myers and Norris, 2013). In particular, the inversion strength and sea surface temperature are identified as the most important large-scale factors that govern the low cloud cover (Rieck et al., 2012; Qu et al., 2014; Brient and Schneider, 2016).

In contrast to marine boundary layer regions, stratus clouds over continental East Asia are more sensitive to the dynamic environment on the lee side of the Tibetan Plateau (Yu et al., 2001, 2004; Zhang et al., 2013). The cloud type and vertical distribution during the boreal cold season are different from the state during the boreal warm season (Luo et al., 2009; Zhang et al., 2014a; Yin et al., 2015). The deep stratiform clouds are largely related to the local dynamic structure and are mainly generated during the cold season (Li and Gu, 2006), in which the ambient environment is the most favorable.

Previous studies have reported the model error sources of these continental stratus clouds. Zhang and Li (2013) suggested that although most models in the CMIP3 and CMIP5 experiments do not well simulate SWCRE over both marine boundary layer cloud regions and continental East Asia, the underlying reasons seem to be different. Errors in marine low clouds are caused by complex small-scale processes that a coarse-resolution general circulation model (GCM) cannot well simulate, leading to failure in simulating the SWCRE dependence on the boundary layer stability. Over East Asia, most models cannot well simulate large-scale circulations. Based on a series of numerical experiments, Zhang et al. (2014b, 2015) suggested that altering the dynamic configurations (e.g., horizontal resolution, ambient environment, and dynamical core) of a GCM can lead to different model behaviors in this regard.

At present, the simulation difficulties and uncertainties related to low and stratus clouds are one of major issues for climate projections (Zelinka et al., 2017). Most climate models still show not good enough performance in the cloud radiation over stratus regions (e.g., Flato et al., 2013; Wang et al., 2014a). As a key aspect of understanding the model’s abilities in climate simulations, it is important to quantify the performance of CAMS-CSM in simulating SWCRE. Additionally, even if the model can reproduce the climatological mean state, there may be biases related to improper physical mechanisms. To address these issues, we first evaluate the cloud radiative effect simulations against the observational data, with an emphasis on the shortwave component over global stratus cloud regions. We further examine the relationships between SWCRE and dynamic–thermodynamic regimes, which provides a key performance metric to quantify the large-scale physical relations associated with the cloud radiative effect. The verified defects in representing the physical mechanism motivate us to explore a possible route for improving the model performance: the SWCRE simulation sensitivity over marine boundary layer regions is examined by implementing an additional parameterization scheme for the cloud fraction.

The rest of this paper is organized as follows. In Section 2, the model, reference data, and methods are briefly described. In Section 3, the global cloud radiative effect simulation is evaluated. The moisture conditions associated with stratus clouds are also examined. In Section 4, the relationships between SWCRE and dynamic–thermodynamic regimes over major stratus cloud regions are investigated. In Section 5, the SWCRE simulation sensitivity over marine boundary layer regions is explored. Finally, conclusions are provided in Section 6.

2 Model, data, and method 2.1 Model and data

We analyze the model results from Atmospheric Models Intercomparison Project (AMIP)-type experiments, which are driven by the observed sea surface temperature dataset. The outputs of the same AMIP experiments conducted by 24 CMIP5 models are compared with CAMS-CSM to locate the performance level of the model among the existing models. The ERA-Interim reanalysis (ERAIM) dataset (Dee et al., 2011) during 2004–10 as well as Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO), and GCM-Oriented CALIPSO Cloud Product (GOCCP) dataset (Chepfer et al., 2010; Cesana and Chepfer, 2012; Wang et al., 2014b) during 2006–10 are used as references. The observed SWCRE and longwave cloud radiative effect (LWCRE) are taken from the Clouds and the Earth’s Radiant Energy System (CERES) Energy Balanced and Filled (EBAF) dataset (Loeb et al., 2012) during 2001–10. All data are based on the monthly mean results.

2.2 Methods

The use of appropriate performance metrics in model evaluations is important for understanding the simulations. Because the definitions of clouds in models and satellite data are different, it is not straightforward to directly compare the simulated clouds with observational data. Instead, SWCRE is used as a proxy for stratus clouds to qualify the climatological distribution in the model. SWCRE is defined as the difference in upwelling the shortwave radiation between clear-sky and all-sky conditions at the top of atmosphere.

Five marine boundary layer cloud regions (Klein and Hartmann, 1993; Yue et al., 2011; Zhang and Li, 2013) and the eastern China region (Table 1) are selected as the major global stratus cloud regions. The moisture quantities over these regions are used to investigate the vertical distribution features of clouds.

Table 1 Symbol, latitude and longitude boundaries for each of the six stratus cloud regions
 Region Symbol Latitude/longitude boundary Peruvian coast P 10°–20°S, 80°–90°W Namibian coast N 10°–20°S, 0°–10°E Californian coast C 20°–30°N, 120°–130°W Australian coast A 25°–35°S, 95°–105°E Canarian coast CA 15°–25°N, 25°–35°W Eastern China EC 23°–33°N, 103°–118°E

We evaluate the SWCRE in the dynamic and thermodynamic regimes to identify the environmental conditions that are responsible for the stratus cloud formation. These relationships are important metrics that can be used to measure the model performance. The local stability and vertical velocity are used to construct dynamic and thermodynamic regimes. For eastern China, we use the vertical velocity at 700 hPa (ω700) and lower troposphere stability (LTS; difference in the potential temperature between 500 and 850 hPa) as two proxies. The upward motion between the mid- and low-level convergences is responsible for the stratus cloud formation. For marine boundary layer regions, we use the vertical velocity at 500 hPa (ω500) that represents the vertical motion in the free troposphere as a whole (Weaver and Ramanathan, 1997). The stratus clouds are generated under the descending motion, which reflects the large-scale subsidence environment. Additionally, the estimated inversion strength (EIS; Wood and Bretherton, 2006) is adopted to be the proxy for the thermodynamic condition. The EIS is recognized as a better predictor over these regions owing to the stronger relationship of EIS with stratus cloud amount. The EIS formula is:

 ${\rm{EIS}} = {\theta _{700}} - {\theta _0} - {{\varGamma }_{{\rm m},850}}\left({{Z_{700}} - Z} \right).$ (1)

Here, θ700 and θ0 are potential temperatures at 700 and 1000 hPa, respectively; Γm,850 is the moist-adiabatic potential temperature gradient at 850 hPa; and Z700 and Z are geopotential heights at the 700- and 1000-hPa based lifting condensation levels, respectively.

3 Climatological mean state 3.1 Cloud radiative effects

The global distribution of the annual mean cloud radiative effect in CAMS-CSM coincides well with that in the CERES dataset, although some minor differences are found in certain regions (Fig. 1). The spatial correlation coefficient of SWCRE between CERES and CAMS-CSM is about 0.89. In both CAMS-CSM and CERES, the maximum centers of SWCRE are mainly located over the North Pacific, North Atlantic, and oceans at the middle and high latitudes of the Southern Hemisphere. To the west of major continents at low latitudes, the maximum SWCRE centers over marine regions in CAMS-CSM are absent or weaker than those in the observations. The maximum center of SWCRE over eastern China occupies a smaller area in CAMS-CSM than that in CERES.

 Figure 1 Annual mean (2001–10) SWCRE, LWCRE, and NCRE (W m−2) in (a, d, g) CERES, (b, e, h) CAMS-CSM, and (c, f, i) their differences (CERES minus CAMS-CSM). The R value denotes the correlation coefficient between CERES and CAMS-CSM. The six red boxes in (a) locate the selected stratus regions.

The global distribution of LWCRE in CAMS-CSM is closer to that in CERES. The correlation coefficient is approximately 0.91. In CERES, high LWCRE magnitudes aggregate at low latitudes, and the LWCRE distribution is similar to that of SWCRE. The magnitude of LWCRE in CERES is slightly larger than that in CAMS-CSM, especially over the equatorial regions. The difference in LWCRE between CERES and CAMS-CSM is smaller overall than that of the global SWCRE.

The net cloud radiative effect (NCRE) distribution is dominated by SWCRE. The correlation coefficient is around 0.80. In the tropical regions, NCRE approaches zero because the SWCRE magnitude is nearly equivalent to that of LWCRE, which is found in both CERES and CAMS-CSM. To the west of several continents at low latitudes, NCRE is weaker in CAMS-CSM than that in CERES over the oceans. For these regions, the difference in NCRE between CERES and CAMS-CSM becomes larger, which is similar to that of SWCRE.

Errors in NCRE are largely attributed to SWCRE, which is mainly related to stratiform clouds. The seasonal variation in SWCRE is investigated over these selected stratus cloud regions. The regionally averaged SWCRE over each of the six stratus regions in CAMS-CSM is much weaker than that in CERES for all seasons (Fig. 2). For eastern China, the difference in SWCRE between CAMS-CSM and CERES is nearly constant during each season. Differences over the marine regions vary seasonally, which implies that the SWCRE bias may result from the deficient representation of certain conditions in eastern China, while sources of errors in the marine regions may be more complex.

 Figure 2 Regionally averaged SWCRE (W m−2) over the six stratus regions in CERES (blue bars) and CAMS-CSM (golden bars) during (a) December–February (DJF), (b) March–May (MAM), (c) June–August (JJA), and (d) September–November (SON).

To have a basic grasp of the performance of CAMS-CSM among state-of-the-art climate models, we compared the regionally averaged SWCRE of CAMS-CSM with those from the CMIP5 models during 2001–08. In eastern China, CAMS-CSM simulates SWCRE better than many of the other models (Fig. 3). Nearly all models reproduce a weaker SWCRE than the observation. The simulation error is slightly smaller in CAMS-CSM than the CMIP5 multi-model mean. Additionally, SWCRE in CAMS-CSM is close to those in six of the models whose simulation errors are smaller. Only quite a few models generate the SWCRE equivalent with the observation.

 Figure 3 The regional average of annual mean SWCRE (W m−2) over the six stratus regions. (a) SWCRE in CAMS-CSM and CMIP5 models, and (b) differences in SWCRE between the models and observation (CERES).

For the marine regions, SWCRE is weaker in most models than in the observation over all of the regions, except for the Canarian coast. The error in CAMS-CSM is larger or equivalent to the multi-model mean in each region. The performance level of CAMS-CSM varies in different marine stratus regions. CAMS-CSM slightly better simulates SWCRE in the Californian and Canarian coasts compared to the other models. However, the model has a poor performance in the Peruvian and Australian coasts.

3.2 Moisture conditions associated with stratiform clouds

SWCRE is mainly contributed by the stratiform clouds. Most clouds over the stratus regions are low stratus clouds with high liquid water content. To obtain a basic understanding of the vertical structure of simulated clouds, we compare the model results to CALIPSO-GOCCP results. Normally, comparing the simulated clouds with observations is not straightforward due to the inconsistent cloud definition criteria. The observed vertical structures and physical properties of cold-season stratus clouds on the lee side of the Tibetan Plateau were investigated by using the CloudSat/CALIPSO dataset (Zhang et al., 2014a). These observational results also provide a qualitative reference for the model evaluation. The results from CALIPSO-GOCCP are generally in agreement with the observational results in Zhang et al. (2014a).

Overall, the model reproduces the vertical distribution of clouds at low levels during November–March (NDJFM) over eastern China. In CALIPSO-GOCCP, most clouds are located on the lee side of the Tibetan Plateau (Fig. 4a). The geometric center of cloud fraction in CAMS-CSM is lower than that in CALIPSO-GOCCP (Fig. 4b). The model produces fewer clouds over eastern China. The climatological mean cloud water pattern is similar to that of the cloud cover in CAMS-CSM, with a large center located at 850 hPa over the eastern periphery of the Tibetan Plateau (Fig. 4c).

 Figure 4 The zonal cross-sections of the climatological mean (a, b) cloud cover (%), (c) cloud water (10–3 g kg–1), and (d, e) relative humidity (%) averaged from 23° to 33°N over eastern China during NDJFM in CALIPSO-GOCCP, ERAIM, or CAMS-CSM. Differences in (f, g) relative humidity and (h) cloud water between the composite and climate are also shown.

We examined the relative humidity distribution to identify moisture conditions for the stratus cloud formation. Large relative humidity values appear at low levels over eastern China in both ERAIM and CAMS-CSM (Figs. 4d, e). The relative humidity is quite large on the lee side of the Tibetan Plateau, corresponding to the maximum center of cloud cover in CALIPSO-GOCCP. The simulated high-value center of relative humidity appears at lower levels than in ERAIM. At 700 hPa, where the simulated cloud water content is small, the simulated relative humidity values are smaller than those in the observations. This lower relative humidity limits the SWCRE representation. The moisture content deficiency leads to the insufficient cloud fraction at low levels, which further results in a weakened SWCRE in the model.

To determine whether the model can capture the cloud property and moisture condition impacts on SWCRE, we investigated the anomalous fields of relative humidity and cloud water based on composite results. Here, the regionally averaged SWCRE over eastern China is used as an index to select months when the index values are higher than the climatological mean state. Then, the difference between the composite field and climatological mean field is examined.

In ERAIM, a larger relative humidity is at approximately 850 hPa over 108°–118°E, corresponding to more stratus clouds that intensify SWCRE (Figs. 4f, g). The vertical distribution and location of the high-value center of relative humidity in CAMS-CSM coincides with that in ERAIM. The model well simulates the variation in relative humidity and captures the observed feature. In addition, the higher cloud water content at 850 hPa over 108°–115°E is favorable for the strengthening of SWCRE (Fig. 4h).

Over the five marine boundary layer regions, the cloud fraction simulated by CAMS-CSM is considerably less than that in CALIPSO-GOCCP. Because of the similar distribution of cloud fractions in either the observations or model over each region, we present the results over the Namibian coast as an example. In the observations, the cloud fraction rapidly increases above the sea surface and reaches a maximum at approximately 1 km (Fig. 5a). The clouds aggregate beneath an altitude of 2 km. The model simulates much less cloud cover than the observations (Fig. 5b).

 Figure 5 The zonal cross-section of the climatological annual mean (a, b) cloud cover (%), (c) cloud water (10−3 g kg−1), and (e, f) relative humidity (%) averaged from 10° to 20°S over the Namibian coast in CALIPSO-GOCCP, ERAIM, or CAMS-CSM. Differences in (d) cloud water and (g, h) relative humidity between the composite and climate are also shown.

The cloud water simulated by CAMS-CSM aggregates over the eastern region of the Namibian coast, where the relative humidity vertical gradient is large (Fig. 5c). The cloud water generally increases near the sea surface when SWCRE intensifies (Fig. 5d). The cloud water increments are larger in areas where the climatological values are higher, which indicates the relatively centralized growth of the simulated stratus clouds. The model reproduces a coherent variation in cloud water as SWCRE changes, indicating that the model captures stratus cloud impacts on SWCRE in terms of the cloud properties.

CAMS-CSM fails to capture the relative humidity vertical structure as compared to ERAIM. Figure 5e shows that the relative humidity rapidly decreases at 850 hPa in ERAIM. However, in CAMS-CSM, the relative humidity decreases at a constant lapse rate from the sea surface to 850 hPa (Fig. 5f). Near the sea surface, the relative humidity in ERAIM is also larger than that in CAMS-CSM. Absence of the large relative humidity vertical gradient in CAMS-CSM reflects the lack of stratocumulus clouds in the model. In addition, CAMS-CSM hardly reproduces the consistent variation in the relative humidity with SWCRE, which is important to the simulation of interannual variations (Figs. 5g, h).

4 Relationships between SWCRE and dynamic–thermodynamic regimes 4.1 Eastern China

Over eastern China, earlier studies have reported that the large-scale environment is important to the generation and maintenance of stratus clouds. Here, the vertical velocity and stability are used as two proxies to construct the dynamic and thermodynamic regimes. To examine the environmental variations associated with the stratus cloud formation, occurrence frequencies of the environmental fields and SWCRE distribution in the environmental regimes are shown. The seasonal and interannual variabilities are represented by the monthly mean and monthly anomalous results during the cold season (November–April; NDJFMA).

The occurrence frequency distributions in ω700 and LTS regimes in CAMS-CSM are similar to that in CERES (Fig. 6). For monthly means, most cases occur at approximately ω700 = −20 hPa day −1 and LTS = 38 K in both CERES and CAMS-CSM. Despite the higher frequency in the center of the regime in CAMS-CSM, the frequency decreases from the center to regime boundary in both the observations and model. The LTS range is narrower, and the ω700 range is wider in CAMS-CSM than that in CERES. The slight range differences reveal errors in the simulated ambient environmental fields.

 Figure 6 Occurrence frequencies of ω700 and LTS over eastern China in (a, c) ERAIM and (b, d) CAMS-CSM. The samples are NDJFMA (a, b) monthly means and (c, d) monthly anomalies. Bin intervals: 2 (1) K for LTS (anomalous LTS) and 20 (15) hPa day−1 for ω700 (anomalous ω700). The percentages denote the total occurrence frequencies in the four quadrants excluding bins with occurrence frequencies lower than 0.1%.

For monthly anomalies, the occurrence frequency ranges in both regimes in CAMS-CSM are close to those of the observations. The frequency is higher when both the anomalous vertical velocity and anomalous LTS approach zero, implying that the vertical velocity and LTS do not deviate greatly from their climatological monthly means in most cases. Additionally, the occurrence frequency in the “positive ω700-negative LTS” and “negative ω700-positive LTS” quadrants accounts for up to 60% in the observation. This result indicates a likely coherent variation, simultaneously becoming favorable or unfavorable in the two anomalies. CAMS-CSM well captures this feature over the interannual time scale.

We further examined the distribution of SWCRE in the ω700–LTS regimes (Fig. 7). In the observations, the low-level lifting encounters a stable stratification, leading to a favorable environmental condition for the stratus cloud formation. In CAMS-CSM, the variation in monthly mean SWCRE values is dominated by ω700, which is similar to that in the observations. However, under the circumstance that both LTS and ascending motion are very strong, SWCRE in CAMS-CSM is not as intense as that in the observations. Considering that the occurrence frequency is quite small under that circumstance, LTS in CAMS-CSM exerts a smaller influence on the variation in SWCRE than ω700.

 Figure 7 SWCRE binned by ω700 and LTS over eastern China in (a, c) CERES/ERAIM and (b, d) CAMS-CSM. The samples are NDJFMA (a, b) monthly means and (c, d) monthly anomalies. Bins with occurrence frequencies lower than 0.1% are removed. Bin intervals: 2 (1) K for LTS (anomalous LTS) and 20 (15) hPa day−1 for ω700 (anomalous ω700). The percentages denote the total occurrence frequencies in the four quadrants excluding bins with occurrence frequencies lower than 0.1%.

For monthly anomalies, higher SWCRE values are found in positive LTS and negative ω700 regimes in both CERES and CAMS-CSM. The SWCRE anomaly evidently becomes stronger as the anomalous ascending motion becomes more intense. The relatively smaller variation in anomalous SWCRE due to anomalous LTS suggests that the high stability is a favorable background for the formation of stratus clouds. The model accurately simulates the generation of stratus clouds and captures the observed interannual variation.

4.2 Marine boundary layer regions

Over five marine boundary layer regions, CAMS-CSM reproduces the occurrence frequency of the ambient environment characterized by the ω500 and EIS regimes (Fig. 8). For monthly means, more cases occur at approximately ω500 = 30 hPa day−1 and EIS = 6 K in both CERES and CAMS-CSM. The variable ω500 is mostly positive in each dataset, implying that the subsiding motion at 500 hPa provides a background for the stratus cloud generation. For monthly anomalies, the frequency pattern in CAMS-CSM agrees well with that in the observations. Similar to eastern China, the vertical velocity and EIS are close to their climatological monthly means in most cases over marine regions. The model reproduces higher occurrence frequency in the quadrants with two same-sign anomalies than that in the quadrants with anti-sign anomalies.

 Figure 8 As in Fig. 6, but for ω500 and EIS over five marine boundary regions. Bin intervals: 1 (0.5) K for EIS (anomalous EIS) and 10 hPa day−1 for ω500 (anomalous ω500).

Figure 9 shows the SWCRE distribution in the ω500–EIS regimes. In the observations, SWCRE evidently strengthens as EIS increases. The simulated SWCRE is generally weaker than that in the observations. Given the weak descending motion at 500 hPa, the underneath high stable stratification contributes to more low-level stratus clouds. The variable ω500 has a much smaller influence on SWCRE than EIS. Overall, CAMS-CSM captures the trend of SWCRE varying with EIS. However, the close relationship between SWCRE and EIS is not well simulated by the model. In addition, when EIS reaches the maximum, the model simulates a greater variation in SWCRE related to ω500. This incorrect relationship results from the spurious strengthening of SWCRE under conditions where the vertical air flow at 500 hPa is nearly static.

 Figure 9 As in Fig. 7, but for ω500 and EIS over five marine boundary regions. Bin intervals: 1 (0.5) K for EIS (anomalous EIS) and 10 hPa day−1 for ω500 (anomalous ω500).

With regard to the monthly anomalies, CAMS-CSM simulates the environmental fields well. The anomalous SWCRE strengthens as the anomalous ω500 becomes more negative in both CAMS-CSM and CERES. This finding implies that the weak descending motion is more favorable for the stratocumulus cloud generation. However, the model can hardly simulate the variation in SWCRE as EIS increases. The spurious strengthening of SWCRE under the low-stability condition reveals that the relation between stability and clouds is unrealistic at the interannual scale.

5 SWCRE simulation sensitivity over the marine boundary region

Previous analyses have demonstrated that CAMS-CSM exhibits a better performance for stratiform clouds over eastern China. For marine boundary layer regions, the model simulates much fewer clouds and less moisture than those in the observations. The weaker SWCRE over these marine regions is largely caused by the cloud deficiency.

For a comprehensive climate model with the relatively coarse resolution, simulations of clouds and their weather/climate impacts (e.g., radiative effects and precipitation) are a consequence of multiple interactive factors (e.g., Zhang et al., 2019). In terms of cloud radiation features, the cloud optical properties are related with both the cloud spatial and temporal coverage, as well as cloud microphysical features (e.g., effective radius and liquid water content). Relations between these cloud physical features and radiation effects are relatively universal. Nevertheless, whether the clouds occur or not largely depends on dynamical and meteorological conditions, which are fundamentally different between eastern China and marine regions. Over eastern China, the accurately simulated SWCRE is largely attributed to the reproduction of large-scale environmental fields. The parameterization of cloud macro/micro physical features acts as a response to the large-scale forcing. For the marine regions, the clouds are largely associated with small-scale turbulent mixing (under favorable large-scale conditions) that can only be approximately parameterized by typical weather and climate models. These complicated underresolved dynamical processes, and their interaction with cloud microphysics, make it hard to adequately parameterize these clouds, which leads us to explore some ad-hoc solutions to this problem.

In this section, we further investigate the model sensitivity by implementing a parameterization scheme for the stratiform cloud cover. The original cloud cover scheme in CAMS-CSM is based on a statistical–dynamical approach (Tompkins, 2002; Roeckner et al., 2003). This scheme assumes that the subgrid scale cloud fraction is caused by the fluctuation in total water content. The probability density function of total water content is obtained by the beta distribution, which is bounded. The distribution of higher-order moments of variance and skewness is directly affected by underresolved processes such as the turbulence and deep convection.

Here, we implement an additional scheme that utilizes an empirical relationship between the marine stratiform cloudiness (Cst) and local LTS (Klein and Hartmann, 1993; Collins et al., 2004):

 ${C_{\rm st}} = {\rm min} \left\{ {1, {\rm max} \left[ {0,\left({{\theta _{700}} - {\theta _0}} \right)\cdot 0.057 - 0.3573} \right]} \right\}.$ (2)

AMIP-type experiments with/without this cloud cover scheme are conducted for a 5-yr period. We compare the two runs in terms of the climatological mean state of SWCRE, vertical structure of cloud cover, and relationships between SWCRE and dynamic–thermodynamic regimes to investigate the SWCRE sensitivity to large-scale environmental fields.

When the stratocumulus cloud scheme is activated, the model shows an improved simulation of the climatological mean state over marine boundary layer regions (Fig. 10). The regionally averaged SWCRE in the revised model becomes stronger over the marine regions during each season, which is closer to that of the observations. The regionally averaged cloud cover increases at low levels in the revised model compared to the control run over the Namibian coast (Fig. 11). This result confirms that CAMS-CSM can well simulate the large-scale thermodynamic environment but fails to produce abundant clouds in a favorable environment. The insufficient stratocumulus clouds in the model lead to a weaker SWCRE. Moreover, via feedback mechanisms, the shortage of clouds may further impact the large-scale structures (e.g., sea surface temperature biases in a coupled mode). Notably, this additional cloud cover scheme does not evidently alter the simulation over eastern China. The small variation in SWCRE over eastern China demonstrates that SWCRE is less sensitive to the thermodynamic factor in that region.

 Figure 10 As in Fig. 2, but for the observations (blue bars), control run (golden bars), and revised model (red bars).
 Figure 11 The zonal cross-sections of the climatological annual mean cloud cover (%) averaged from 10° to 20°S over the Namibian coast in the (a) control run and (b) revised model.

SWCRE in the ω500–EIS regime also becomes more intense in the revised model over the five marine boundary layer regions (Figs. 12a, b). For monthly means, SWCRE in the revised model becomes about 20 W m−2 stronger in each corresponding bin and shows a clear enhancement as EIS increases. The close relationship between SWCRE and thermodynamic environment verifies the accurately simulated large-scale environmental fields over marine regions. The deficiency in the SWCRE simulation over marine regions mainly results from an incomplete representation of the subgrid physics process. The different relationships between SWCRE and dynamic–thermodynamic regimes in eastern China and marine boundary layer regions also imply that the formation mechanisms of stratus clouds are different.

 Figure 12 As in Fig. 9, but for the (a, c) control run and (b, d) revised model. Bins with occurrence frequencies lower than 0.2% are removed. The percentages denote the total occurrence frequencies in the four quadrants excluding bins with occurrence frequencies lower than 0.2%.

For monthly anomalies, the model still hardly reproduces the variation in SWCRE as EIS changes when the cloud scheme is activated (Figs. 12c, d). Although the spurious strengthening of SWCRE under low-stability conditions has been eliminated, little variation in SWCRE is found as the anomalous EIS changes. This insensitivity suggests that the interannual variability in the model may not change by altering only a cloud cover scheme. More efforts are needed to improve the model’s capabilities in this regard.

6 Conclusions

In this paper, the CAMS-CSM performance in simulating SWCRE has been evaluated for global stratus cloud regions. The major points are summarized as follows.

Over eastern China, CAMS-CSM well simulates the climatological mean state and regionally averaged SWCRE for all seasons. The model reproduces the high-value center of relative humidity despite the lower geometric location of the center and insufficient moisture content at 700 hPa. In addition, CAMS-CSM simulates SWCRE in an appropriate manner, where the dynamic regime dominates the change in SWCRE under highly stable conditions.

Over major marine stratus regions, the annual and seasonal means of SWCRE in CAMS-CSM are generally weaker than those in the observations. The absence of a strong vertical gradient and smaller magnitude of the relative humidity reflect that few stratus clouds are generated by the model. The model accurately simulates the occurrence frequency of the ambient field while fails to capture the close relationship between SWCRE and lo-cal stability.

For the marine stratus regions, a stratocumulus scheme that relates the cloud cover with local stability evidently improves the SWCRE simulation. The climatological annual mean and regionally averaged SWCRE in each season become more intense in the revised model. The stronger SWCRE feature under higher stability conditions is captured through the monthly mean results in the revised model.

Acknowledgments. The constructive comments from two anonymous reviewers have greatly helped improve the manuscript.

References