Responses of Cloud-Radiative Forcing to Strong El Ni?o Events over the Western Pacific Warm Pool as Simulated by CAMS-CSM

Baichao ZHANG, Zhun GUO, Xiaolong CHEN, Tianjun ZHOU,5*, Xinyao RONG, and Jian LI

1 Nanjing University of Information Science & Technology, Nanjing 210044

2 Climate Change Research Center, Chinese Academy of Sciences, Beijing 100029

3 State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics,Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing 100029

4 CAS Center for Excellence in Tibetan Plateau Earth Sciences, Chinese Academy of Sciences (CAS), Beijing 100101

5 University of Chinese Academy of Sciences, Beijing 100049

6 Chinese Academy of Meteorological Sciences, China Meteorological Administration, Beijing 100081

ABSTRACT Cloud-radiative forcing (CRF) at the top of the atmosphere (TOA) over the western Pacific warm pool (WP) shows unique characteristics in response to El Ni?o events. In this region, the responses of CRF to El Ni?o events have been a useful metric for evaluating climate models. Satellite data are used to analyze the CRF anomalies to El Ni?o events simulated by the new and old versions of the Climate System Model of the Chinese Academy of Meteorological Sciences (CAMS-CSM), which has participated in the Atmospheric Model Intercomparison Project (AMIP). Here, simulations for super El Ni?o years, El Ni?o years, and normal years are compared with observations. The results show that the mean values of both longwave CRF (LWCRF) and shortwave CRF (SWCRF) in CAMS-CSM are weaker than the observations for each category of El Ni?o events. Compared with the old version of CAMS-CSM, the decrease in LWCRF during El Ni?o events is well simulated by the new version of CAMS-CSM. However, both new and old models cannot reproduce the anomalous SWCRF in El Ni?o events. The biases in the CRF response to El Ni?o events are attributed to the biases in the cloud vertical structure because of a weaker crash of the Walker circulation in CAMS-CSM. Due to the modification of the conversion rate from cloud droplets to raindrops in the cumulus convection scheme, the new version of CAMS-CSM has better CRF skills in normal years, but biases in El Ni?o events still exist in the new version. Improving the response of the Walker circulation to El Ni?o events is key to higher skills in simulating the cloud radiative responses.

Key words: cloud-radiative forcing (CRF), El Ni?o, model evaluation, warm pool (WP), western Pacific

1. Introduction

Deep cumulus and cirrus clouds, which are associated with updrafts of the Walker circulation, cover a large area of the western Pacific warm pool (WP). This area attracts worldwide attention because of its unique features in cloud-radiative forcing (CRF) in response to El Ni?o events. During normal years, strong updrafts of the Walker circulation lead to deep convective clouds wrapping over the WP, increasing the shortwave CRF (SWCRF) and longwave CRF (LWCRF) at the top of the atmosphere (TOA). As an offset between negative SWCRF and positive LWCRF, the net CRF (NETCRF) is approximately zero (Kiehl and Ramanathan, 1990; Hartmann et al., 1992; Zhang et al., 1995; Chen et al., 2000;Harrop and Hartmann, 2015). During El Ni?o events,however, the zonal gradient of the sea surface temperature (SST) in the tropical Pacific (TP) weakens, and the Walker circulation could even completely crash in some extreme events (e.g., 1998). The weakened updraft over the WP then leads to a shallower cloud top than that during normal years, producing stronger negative NETCRF(Cess et al., 2001a, b; Burls and Fedorov, 2014; Kiran et al., 2015; Wang and Su, 2015; Wall et al., 2018). Because of the unique cloud vertical structure and the different features of CRF between normal years and El Ni?o events, CRF over the WP has been regarded as a useful metric for evaluating the climate models (Lu et al., 2004;Wu et al., 2010, 2011; Guo and Zhou, 2012).

Climate models are widely used to simulate the current climate and predict future changes. Accurate simulation of the cloud structure is a necessary condition for reliable future projections. CRF, as a useful index for quantifying cloud–climate effects, is poorly simulated by most models because of the complex radiative properties of clouds. Different CMIP5 (Coupled Model Intercomparison Project Phase 5) models have different biases in the simulated CRF in the TP, and the models with a more realistic dynamic framework have better CRF simulation skills (Webb et al., 2001, 2006; Soden and Held, 2006;Williams and Webb, 2009). Thus, improving the simulation of the CRF process is an effective means to improve the performance of the CMIP6 models.

The Climate System Model of the Chinese Academy of Meteorological Sciences (CAMS-CSM) has been officially planned to take part in the CMIP6. The model shows a promising ability to simulate the EASM (East Asian summer monsoon) variability and the ENSO–EASM relationship (Rong et al., 2018). CAMS-CSM can generally reproduce the features of SWCRF and LWCRF, but it underestimates the thermodynamic damping and thermocline feedbacks caused by the biases of the SWCRF feedbacks in ENSO simulations (Chen et al.,2019; Hua and Chen, 2019; Lu and Ren, 2019). A recent study shows that the shortwave radiation flux and the corresponding total cloud cover simulated by CAMSCSM are unrealistic in ENSO events (Hua et al., 2019),but the reason remains to be clarified.

Compared with the CRF in the pre-industrial control simulation (piControl), the Atmospheric Model Intercomparison Project (AMIP) simulation driven by the observed SST exhibits a better performance in simulating the magnitude and spatial distribution for both normal and ENSO years (Chen et al., 2019). A more in-depth analysis of the responses of CRF to El Ni?o events in the AMIP simulation will provide useful references for better understating the bias of the fully coupled model. In the present study, we use the AMIP results from CAMSCSM to assess the simulation skills of WP CRF responses to El Ni?o events. The cause of the CRF simulation biases will also be investigated by comparing the old and new versions of CAMS-CSM, as some improvements related to cloud microphysics and radiation have been implemented in the new version.

The remainder of this paper is organized as follows.Section 2 provides a brief introduction of the data, model,and methods in this study. In Section 3, we assess the simulation skills of CRF in the CAMS-CSM AMIP runs using old and new versions of the model, and determine the cause of the simulated biases between the observations and the CAMS-CSM results. We then discuss how the modification in the new version affects the simulation skills of cloud radiation. Finally, Section 4 summarizes the results of the study.

2. Data, model, and methods

2.1 Data

(1) The observed monthly cloud fraction and gridmean integrated cloud water path (CWP) are derived from the International Satellite Cloud Climatology Project (ISCCP) D2 from July 1983 to June 2007 with a 2.5° × 2.5° latitude–longitude resolution (Rossow and Schiffer, 1991).

(2) Three observed radiation datasets are used. (i)Monthly radiation at the TOA provided by Clouds and the Earth’s Radiant Energy System (CERES) during 2001 to 2005 at a 1° × 1° latitude–longitude resolution(Wielicki et al., 1996); (ii) monthly radiation data derived from ISCCP-FD (Flux Data) on a 2.5° grid from 1984 to 2005 (Zhang et al., 2004); and (iii) monthly radiation data derived from the CERES-TRMM (Tropical Rainfall Measuring Mission) in 1998 at a spatial resolution of 2.5°, which are widely used to analyze the super El Ni?o events of 1998 (Wong et al., 2000).

(3) Wind data are derived from the ERA-Interim atmospheric reanalysis monthly data from 1979 to 2005 on a 1.5° × 1.5° latitude–longitude resolution at standard pressure levels. The spatial resolution of the dataset is approximately 80 km (T255 spectral), with 60 vertical levels from the surface up to 0.1 hPa (Simmons et al.,2006).

All the data are interpolated to 1° × 1° latitude–longitude grids based on bilinear interpolation in this study.Following Lu et al. (2004), we used the mean of the first four months (January, February, March, and April;JFMA for short) in a year to define the strongest period of the El Ni?o events.

2.2 Model and experiments

The CAMS-CSM is a global primitive-equation model based on the ECHAM5 (ECmwf-HAMburg v5.4) atmospheric model developed by the Max Planck Institute for Meteorology and has 31 vertical levels with a top at 10 hPa (Roeckner et al., 2003). The Tiedtke (1989) mass flux scheme, which includes modifications for penetrative convection according to Nordeng (1994), is applied for cumulus convection parameterization. The convective cloud scheme was developed by Tiedtke (1993). The stratiform cloud scheme consists of a cloud microphysical scheme (Lohmann and Roeckner, 1996) and a cloud cover scheme, which diagnostically calculate the cloud fraction as a function of relative humidity (Sundqvist,1978). The subgrid cloud formation is diagnosed as follows: clouds occur where the relative humidity exceeds the prescribed threshold of 85% at the top of the planetary boundary layer (PBL) and increases linearly to the surface layer (99%). When penetrative convection occurs, the threshold decreases further to a minimum value of 0.5 near the tropopause. The radiation scheme adopted in the model was developed by Zhang et al. (2003,2006a, b) and can improve the simulation of SWCRF over East Asia. More details of CAMS-CSM can be found in Rong et al. (2018).

The AMIP run of the two versions of CAMS-CSM is the same as that used for the historical forcing data during 1900–2013, including monthly observations of SST,solar forcing, greenhouse gases (CO2, CH4, N2O, CFC-12, and equivalent CFC-11 that summarizes the effects of all 39 other gases), and anthropogenic aerosols. The differences between the new version and the old version of CAMS-CSM include 1) a modification of the conversion rate from cloud droplets to raindrops in the cumulus convection scheme and 2) addition of an effective solar zenith angle scheme, which accounts for the curvature of the atmosphere and its effect on the length of the optical path of direct solar beam with respect to a plane-parallel atmosphere. In this study, we use the mean of three members of the new version to indicate the new version of CAMS-CSM.

2.3 CRF

SWCRF, LWCRF, and NETCRF at the TOA are defined as follows:

where R denotes all-sky reflected SW at the TOA, Rcrepresents clear-sky reflected SW, F is all-sky reflected LW at the TOA, and Fcdenotes clear-sky reflected LW(Harrison et al., 1990).

We consider the radiation ratio N as follows (Lu et al.,2004):

If SWCRF cooling dominates, both NETCRF < 0 and N > 1 indicate that CRF cools the atmosphere; if LWCRF heating dominates, NETCRF > 0 and N < 1 show that CRF heats the atmosphere; and if NETCRF ≈ 0 and N ≈ 1, the influences of LWCRF and SWCRF are nearly offset.

2.4 Classification of the El Ni?o year

We used the Ni?o3.4 index (5°N–5°S, 120°–170°W)from the NOAA website (https://www.esrl.noaa.gov/psd/gcos_wgsp/Timeseries/Data/nino34.long.anom.data), and the warm and cold episodes in the TP are determined with a threshold of ± 0.5°C of the Ni?o3.4 index. El Ni?o events are defined by a threshold of +1.0°C, while the threshold is +2.0°C for super El Ni?o events. When more than one month has a Ni?o3.4 index higher than +1.0°C from November to April, the year is defined as an El Ni?o year; when the Ni?o3.4 index is higher than+2.0°C, the year is defined as a super El Ni?o year. The El Ni?o events during 1979–2005 are listed in Table 1,including 5 El Ni?o and 2 super El Ni?o events. In this study, the super El Ni?o years are used to indicate El Ni?o because of its strong anomaly. However, there are only two super El Ni?o years; to exclude some specific features, we provide the results of five El Ni?o years to give an auxiliary illustration of the El Ni?o attributes.

2.5 Identification of cloud vertical structure

The relationship between N and NETCRF reflects the connection of the CRF with cloud vertical structure, such as the cloud top and cloud optical thickness (Cess et al.,2001a). Both the cloud fraction and the cloud optical thickness make positive contributions to the CRF. The contribution is linear for the cloud fraction but nonlinearfor the cloud optical thickness. A schematic diagram of the scatter plot of N versus NETCRF is shown in Fig. 1.The blue and red lines and dots are clouds with different cloud top heights. If the cloud optical thickness is small,LWCRF dominates, which is represented by the dots in quadrant 4. When the cloud optical depth increases (arrows in blue and red), both SWCRF and LWCRF simultaneously increase, but SWCRF dominates in the process and causes an NETCRF decrease and the dots move to quadrant 2. The dots in quadrant 4 represent cirrus clouds with a small cloud optical thickness, while the dots in quadrant 2 represent deep convective clouds. The slope of the line reflects the cloud top. The line slope is steeper (red line) for low-top clouds than for high-top clouds (blue line) because N increases more quickly with the cloud optical thickness in low-top clouds due to the small LWCRF.

Table 1. Classification of the 1979–2005 years into super El Ni?o, El Ni?o, and normal years

The above method can capture the characteristics of cloud vertical structure, including the cloud top and the cloud optical thickness (Potter and Cess, 2004). Because the outputs of the model do not include the cloud vertical structure, we used this method to analyze the threedimensional structure of the cloud.

3. Results

3.1 Evaluation of simulated tropical CRF

The CAMS-CSM performance in all-year mean cloud radiative fluxes at the TOA is first evaluated. Figures 2 and 3 show the horizontal distributions of LWCRF and SWCRF, respectively, in two versions of CAMS-CSM and the ISCCP and CERES satellite products. LWCRF in the ISCCP and CERES exhibits maximum centers over the western Pacific WP region (5°S–5°N, 140°–165°E),Maritime Continent, and along the convergence zones;and minimum centers over the subtropical stratocumulus regimes. The SWCRF pattern is nearly a mirror image of the LWCRF pattern (Figs. 3a, d vs. Figs. 2a, d). Therefore, there is an expected cancellation between SWCRF and LWCRF over the western Pacific WP region (Kiehl and Ramanathan, 1990).

However, this balance between SWCRF and LWCRF over the TP is destroyed by the Walker circulation responses during winter and spring of El Ni?o years (Cess et al., 2001a). The cold SST anomalies in the western Pacific WP during El Ni?o years lead to suppressed convection and a lower cloud top and lower cloud amount,reducing both SWCRF and LWCRF (Figs. 3c, f vs. Figs.2c, f). The SWCRF anomalies overwhelm LWCRF in the NETCRF budget over the WP.

Fig. 1. Schematic diagram of the scatter plot for N vs. NETCRF. The numbers 1, 2, 3, and 4 indicate the different quadrants, and the vertical and horizontal lines indicate N = 1 and NETCRF = 0, which means an expected cancellation between SWCRF and LWCRF, respectively. The red(blue) dots indicate the clouds with low (high) cloud tops, and the red (blue) line is fitted from clouds with similar cloud tops using the linear least square method. The arrow on the line indicates the increase in the cloud optical thickness. This schematic diagram was drawn based on Cess et al. (2001a, b).

Fig. 2. Geographical distributions of JFMA (January–April) averaged LWCRF (W m?2) during (a, d, g) normal years, (b, e, h) El Ni?o years,and (c, f, i) super El Ni?o years (details in Section 2.4) from (a–c) ISCCP, (d–f) CERES, (g–i) the old version of CAMS-CSM, (j–l) the new version of CAMS-CSM, and (m–o) the difference between the two versions. The black solid box denotes the warm pool (WP; 5°S–5°N,140°–165°E).

The spatial patterns of the all-year mean LWCRF and SWCRF simulated by CAMS-CSM resemble those of satellite data to a large extent, except for the weaker simulated strength of CRF over the Maritime Continent. For example, the TP (10°S–10°N, 140°E–90°W) mean LWCRF in normal years is 23 W m?2in the old version of CAMS-CSM, in comparison to less than 33 W m?2in ISCCP and 34 W m?2in CERES. LWCRF in the new version is improved to 29 W m?2, which is closer to the observations. A similar bias is seen in SWCRF. The simulated mean SWCRF in the TP is ?34 W m?2in the old version and ?48 W m?2in the new version, in comparison with less than ?55 W m?2in ISCCP and ?54 W m?2in CERES. The biases over the WP are similar to those over the TP (Table 2). Nevertheless, both the old and new versions of CAMS-CSM can generally capture the cancellation between LWCRF and SWCRF, with a regionally averaged NETCRF of approximately 4 W m?2in the old version and ?5 W m?2in the new version over the WP,which are close to the observed ?15 W m?2.

Fig. 3. As in Fig. 2, but for SWCRF.

Table 2. Mean values of LWCRF (W m?2), SWCRF (W m?2), and N with ISCCP, CERES, and CAMS-CSM over the tropical Pacific (TP;10°S–10°N, 140°E–90°W) and warm pool (WP; 5°S–5°N, 140°–165°E) in different years

Both LWCRF and SWCRF decrease in El Ni?o events and especially in super El Ni?o events. The regional mean LWCRF over the WP is reduced from 63 to 29 W m?2in ISCCP and from 64 to 33 W m?2in CERES in response to a super El Ni?o, e.g., during 1998. Similarly, sharp decreases in SWCRF are also seen in both ISCCP (30 W m?2) and CERES (35 W m?2). Such responses in CRF are largely underestimated by CAMSCSM in both the old and new versions, particularly for components of SWCRF. The simulated SWCRF response over the WP is nearly 0 W m?2.

To better demonstrate the response of CRF to El Ni?o events, Fig. 4 compares the spatial pattern of N (N =?SWCRF/LWCRF) among different datasets. The ratio N is approximately 1.6 over the WP but exceeds 3 over the eastern Pacific (10°S–10°N, 135°–90°W; Figs. 4a–e)because of the spatial patterns of SWCRF and LWCRF.Both the old and new versions show that the simulated pattern of N is weaker than that of the observations caused by the weak SWCRF, but the value is greater than 3 over the Maritime Continent (approximately 5°S,140°E), indicating that the model has some biases between the sea and the land.

Fig. 4. As in Fig. 2, but for the CRF ratio N = ?SWCRF/LWCRF.

The interannual variations in CRF and N over the WP from different datasets are compared in Fig. 5. Generally,CRF is weaker in the old CAMS-CSM version than in the satellite datasets, particularly for SWCRF, producing an underestimation in N; and the CRF and N biases are reduced in the new version. Moreover, the simulated interannual variations in CRF values are weaker in both the new and old versions compared to the observations. For example, SWCRF shows a change of up to 50 W m?2between 1998 and 1997 in ISCCP but only a limited variation in the CAMS simulation.

Fig. 5. Four month (JFMA) means of (a) N = ?SWCRF/LWCRF, (b) SWCRF, and (c) LWCRF (W m?2) over the WP (5°S–5°N, 140°–165°E).

Fig. 6. Probability distribution functions (PDFs) of CRF over the WP from observations and CAMS-CSM during (a, b) normal years, (c, d) El Ni?o years, and (e, f) super El Ni?o years. (a, c, e) SWCRF and (b, d, f) LWCRF.

To better understand the model behaviors, Fig. 6 provides the probability distribution functions (PDFs) of CRF over the WP. During normal years, the PDF peaks are populated by SWCRF at approximately ?100 W m?2and by LWCRF at approximately 60 W m?2(Figs. 6a, b).CAMS-CSM fails to capture the observed distributions,with border shapes but lower peaks in both the SWCRF and LWCRF PDFs. During super El Ni?o years, the PDF peaks populated by SWCRF and LWCRF both decrease in the observations; both PDF peaks populated by SWCRF and LWCRF are approximately 40 W m?2. CAMSCSM can provide similar results for LWCRF, but its simulation skills in SWCRF are weak.

In summary, the old version of CAMS-CSM underestimates the strengths of SWCRF and LWCRF over the TP and WP, which are slightly improved in the new version. In addition, both the old and new versions of CAMS-CSM have nearly failed to reproduce the responses of CRF over the WP to the El Ni?o.

3.2 Cloud properties and vertical structure over the WP

Figure 7 shows the relationship between the NETCRF and ratio N (detailed in Section 2.5) over the WP to reveal the vertical structure of clouds. During normal years, ISCCP and CERES data show that N is linear with NETCRF. Moreover, most of the scatters appear in the area where N < 1.3 and NETCRF < ?40 W m?2, indicating that cirrus and deep convective clouds dominate over the WP (Figs. 7a, d). In the two versions of CAMS-CSM,unlike CERES, more scatters appear in the area where NETCRF exceeds ?40 W m?2and N exceeds 2. This indicates that the simulated cloud structure over the WP is more similar to a combination of middle-level and high clouds rather than simply deep cumulus and thin cirrus.Such a bias in cloud vertical structure can partly explain the underestimation of LWCRF during normal years(Fig. 2), as the lower averaged cloud top leads to a higher cloud brightness temperature and stronger outgoing longwave radiation.

For El Ni?o events, satellite data show that scatters are still distributed within the range from the bottom-right(cirrus) to the top-left quadrant (deep convective clouds).Because of the weak updrafts over the WP, however, the amount of the cirrus and deep cumulus clouds and the average cloud height over the WP decrease. The occurrence frequencies of the scatters that have N > 1.3 and a stronger NETCRF slightly increase. The change is more evident during the super El Ni?o years, in which most of the dots exceed N = 1.3, indicating that the proportion of middle-level clouds greatly increases (Figs. 7c, f).CAMS-CSM can capture the above observed El Ni?o-related cloud structure changes over the WP (Figs. 7g–i) as more dots exceed N = 1.3 during El Ni?o events.However, the simulated cloud fractions in CAMS-CSM are higher than those in the observations and the cloud heights are lower than those in the observations, which might be because of the incomplete crash of the Walker circulation during El Ni?o events (see Section 3.3). The cloud height is more realistic in the new version of CAMS-CSM, but the same as the old version, the cloud fractions are still larger than those in observations.

LWCRF and SWCRF are usually proportional to the cloud fractions but are exponential to the cloud optical depth or CWP. To better understand CRF in response to El Ni?o events, we therefore examined the simulation biases of the total cloud fraction, cloud liquid water path(LWP), and ice water path (IWP) in CAMS-CSM, as shown in Fig. 8. The average total cloud fraction is approximately 76.6% in ISCCP, 77.3% in the old version of CAMS-CSM, and 81.8% in the new version of CAMS-CSM (Fig. 8a). During El Ni?o events, ISCCP and CAMS-CSM both agree that the total cloud fraction shows a significant decrease because of the weak Walker circulation and its associated weak updrafts. However,the El Ni?o-related decrease in cloud fraction is weaker in CAMS-CSM than that in ISCCP. The biases in the updrafts are responsible, at least in part, for the underestimation of total cloud fraction changes (see Section 3.3).

Similar to the cloud fraction, the simulated CWP,defined as the sum of the cloud LWP and IWP, also shows significant decreases when El Ni?o events occur,which is probably the result of the weaker response of the Walker circulation to the SST changes in the model.However, the old version generally overestimates the allyear mean CWP by approximately 10 g m?2, but an overestimate of nearly 70 g m?2was found in the new version (Fig. 8d). We further divided the simulated CWP into the cloud LWP and the cloud IWP. It is clear that the interannual variability in the simulated CWP is mainly controlled by the cloud IWP, which decreases during El Ni?o events (Figs. 8b, c). The cloud LWP shows the opposite variability, which partly compensates for the negative contributions from the cloud IWP. Rather than ice clouds, liquid clouds are typically composed of small droplets and are thus efficient in light scattering. In addition, the overestimation of the cloud LWP in the new version caused by the modification of the empirical parameters may have led to the overestimation of CWP.

Fig. 7. Scatter plots of N vs. NETCRF over the WP from (a–c) ISCCP, (d–f) CERES, (g–i) old version of CAMS-CSM, and (j–l) new version of CAMS-CSM for (a, d, g, j) normal years, (b, e, h, k) El Ni?o years, and (c, f, i, l) super El Ni?o years.

Fig. 8. Four month means (JFMA) of (a) total cloud fraction (CF), (b) ice CWP (IWP), (c) liquid CWP (LWP), and (d) CWP over the WP. The horizontal line is the mean value of different datasets.

We therefore suspect that the responses of CRF to El Ni?o events are mainly dominated by the responses of the cloud vertical structure and the cloud fraction.However, the expected overestimation of SWCRF, which should be induced by the overestimation of all-year mean CWP in the old version, is replaced by an underestimation, implying that other factors may compensate for or even overwhelm the effects of cloudiness and CWP in determining SWCRF, such as radiation schemes and cloud microphysics, which need discussion in the future.In the new version, even though the modification of the empirical parameters makes up the underestimation of CRF, this is the result of the overestimation of the LWP.

3.3 Atmospheric circulation interpretation

Previous studies have demonstrated that the variations in CRF and cloud vertical structure over the WP are highly related to the interannual variability in the Walker circulation (Cess et al., 2001a). Since the WP is underneath the updraft branch of the Walker circulation, we compared the JFMA averaged ω over the WP between the simulation and the reanalysis data. In ERA-Interim,the updraft peaks are in the middle-level troposphere(between 500 and 300 hPa) during normal years (Fig.9a). During El Ni?o events, the updraft weakens because of the crash of the Walker circulation (Figs. 9b, c).

The middle-level updrafts in CAMS-CSM are stronger than those in the reanalysis in normal years, which is hardly improved in the new version. The overestimations are even more significant during the super El Ni?o years (Fig. 9c). For example, the simulated ω at 300 hPa is nearly four times stronger in the old version and five times stronger in the new version than that in the reanalysis. In simulations, the induced updrafts during the super El Ni?o years leading to the crash of the Walker circulation are incomplete and finally lead to simulated biases of the cloud vertical structure of the atmosphere.This implies that the simulated Walker circulation remains during the super El Ni?o years, which is inconsistent with the situation in ERA-Interim.

Fig. 9. Four month mean (JFMA) vertical profiles of vertical velocity (ω; hPa day?1) during (a) normal years, (b) El Ni?o years, and (c) super El Ni?o years over the region (5°S–5°N, 140°–165°E) from CAMS-CSM and ERA-Interim.

To better understand the above bias, JFMA-averaged zonal winds along the equator between the simulation and the reanalysis were compared, as shown in Fig. 10.In ERA-Interim, the easterlies (westerlies) prevail in the upper part of the troposphere to the west (east) of 180°.During El Ni?o years, the westerlies over the WP decrease in response to the cold SST anomalies. The replacement of the westerlies by weak easterlies becomes more evident during super El Ni?o events (e.g., 1998).The vertical structure of the zonal winds and the weakening during El Ni?o events are generally reproduced in CAMS-CSM. However, biases remain evident in both the old and new versions, such as the westerlies in the upper part of the troposphere to the west of 120°W (Figs.10f, i) in El Ni?o events.

We further examined the simulation biases of the mean 200-hPa velocity potential and divergent wind(Fig. 11). The simulated large-scale overturning circulations in normal years and El Ni?o events share some common biases. For example, the simulated tropical divergence over the WP shows an identical overestimation during each year (Figs. 11d–i). The overestimation of divergence confirms the stronger simulated updraft over the WP. Moreover, the simulated convergence over the subtropical eastern Pacific is underestimated, and the divergence over the South China Sea is weaker than that in ERA-Interim. The simulation biases can be found in both the old and new versions.

In conclusion, although the changes in empirical parameters in the new version of CAMS-CSM make some improvements to the simulation skills in CRF in normal years, biases in circulations and their induced updrafts can still be found in both the new and old versions,which can further influence the cloud vertical structure,cloud fraction, cloud water content (liquid and ice), and CRF, especially in El Ni?o events when the circulations over the TP are totally changed.

4. Summary

To evaluate the CAMS-CSM simulations of CRF over the WP and its responses to El Ni?o events, we classified years into super El Ni?o years, El Ni?o years, and normal years to compare the simulation results with the observations, and discussed the reasons for biases. The main results of this study are as follows.

(1) The mean values of SWCRF and LWCRF in CAMS-CSM are both weaker than those of the observations over the WP. The weak simulation of LWCRF was caused by the deviation in the cloud vertical structure.For example, the cloud top in CAMS-CSM was lower than that of the observation over the WP.

(2) In response to El Ni?o events, the simulated CRF anomalies are similar to the observations over the TP.However, focusing on the WP, the responses of LWCRF are weaker than those of the observations, and for SWCRF, CAMS-CSM cannot provide similar anomalies compared to the observations. The anomaly biases of SWCRF in CAMS-CSM are mainly caused by the incomplete shutdown of the Walker circulation during El Ni?o events, leading to the existence of updraft over the WP and a weaker decrease in cloud fraction, unlike the observations. For LWCRF, CAMS-CSM can provide a similar response to the observations because the cloud vertical structure response is similar to that of the observations during El Ni?o events. However, because of the influence of the weak response of the cloud fraction, the response of LWCRF is weaker than that of the observations.

Fig. 10. JFMA mean pressure–longitude cross-sections of zonal wind (m s?1) averaged from 5°S to 5°N during (a, d, g) normal years, (b, e, h)El Ni?o years, and (c, f, i) super El Ni?o years from (a–c) ERA-Interim, (d–f) old version of CAMS-CSM, (g–i) new version of CAMS-CSM,and (j–l) the difference between the two versions. The area between the black vertical solid lines is the WP (140°–165°E).

(3) The modification of the empirical parameters between the cloud droplets and the raindrops generates a realistic CRF in the new version of CAMS-CSM. The improvements in CRF are partly caused by the overestimated CWP in the new version. The simulation biases in the circulations and their induced updrafts still exist in the new version, which affect a realistic simulation of CRF responses to El Ni?o events. Hence, a better simulation of CRF over the WP relies heavily on the simulation of the Walker circulation.

Finally, we acknowledge that the reason for the underestimated SWCRF deserves further study. Since the value of the cloud fraction in CAMS-CSM is similar to that of the observations, the weak simulation of SWCRF is hypothesized to be related to the lower cloud molecular number concentration, which will be verified in a future study.

Acknowledgments.We thank the Chinese Academy of Meteorological Sciences for providing the CAMS-CSM model data.

Fig. 11. JFMA mean velocity potential (106 m2 s?1) and divergent wind (m s?1) at 200 hPa over the TP during (a, d, g) normal years, (b, e, h) El Ni?o years, and (c, f, i) super El Ni?o years from (a–c) ERA-Interim, (d–f) old version of CAMS-CSM, (g–i) new version of CAMS-CSM, and(j–l) the difference between the two versions.