The two versions of the BCC-CSM for CMIP5 and CMIP6 are used to compare the ECS, TCR, radiative forcing and feedback processes. The analysis period includes all the integration periods (150 yr) of the two idealized experiments from which the results of the piControl run are subtracted in the corresponding time periods to reduce the effects of climate drift.
The piControl simulation, as the baseline of abrupt4×CO2 and 1pctCO2, is first analyzed. As shown in Fig. 1, the global mean net TOA radiation and GMST in BCC-CSM2-MR are –0.51 W m–2 and 14.4°C, respectively, with corresponding trends of 0.06 W m–2 per century and 0.02°C per century. The values in BCC-CSM1.1m are –0.76 W m–2 and 14°C, with trends of 0.11 W m–2 per century and –0.17°C per century, respectively. BCC-CSM2-MR shows improvement in simulating the equilibrium states in piControl with smaller climate drifts than those in BCC-CSM1.1m.
Figure 1. Time evolution of BCC-CSM model simulated (a) net radiation at the TOA (W m−2) and (b) surface air temperature (°C) in piControl (bold lines), 1pctCO2 (medium bold lines), and abrupt4×CO2 (thin lines) experiments. The red lines denote BCC-CSM2-MR simulations for CMIP6 while the blue lines denote BCC-CSM1.1m simulations for CMIP5.
The responses in abrupt4×CO2 and 1pctCO2 have different features. The net radiation at the TOA shows an approximately linear increase in 1pctCO2, whereas in the abrupt4×CO2 experiment, it quickly decreases in the first 30 years, and the rate of decline becomes very slow (Fig. 1a). Consistent with the responses of the net TOA radiation, the GMST linearly increases in 1pctCO2, while in the abrupt4×CO2 experiment, it increases quickly during the first 10–20 years and then slowly afterwards (Fig. 1b).
Using the Gregory-style regression based on Eq. (1), the equilibrium GMST under abrupt quadrupling of CO2 concentration forcing is estimated in the abrupt4×CO2 experiments, as well as the corresponding ERF and net climate feedback (λ).
As shown in Fig. 2, the quadrupling forcing of BCC-CSM2-MR (F4×) is 5.57 ± 0.24 W m−2 (± 1σ), the equilibrium temperature change (ΔTeqm) is 6.08 ± 0.21 K, and the slope of the line, i.e., the feedback parameter, λ is −0.92 ± 0.06 W m−2 K−1 (Fig. 2a). The corresponding values for BCC-CSM1.1m are 6.90 ± 0.24 W m−2, 5.83 ± 0.21 K, and −1.18 ± 0.04 W m−2 K−1 (Fig. 2b). The ECS and ERF can be estimated as half of ΔTeqm and F4×, respectively, based on the logarithmic relation between CO2 concentration and forcing (Myhre et al. 1998). The ECS derived from the quadrupled CO2 scenario has been found to be overestimated compared to that from the doubled CO2 scenario but may still underestimate the real ECS due to the strong positive feedback under a warming climate (Meraner et al., 2013; Rugenstein et al., 2020).
Figure 2. Linear fitting estimations of 4 × CO2 ERF (F4×, y-axis intercept; W m−2), ΔTeqm (i.e., 2 × ECS, x-axis intercept; K), and λ (slope; W m−2 K−1) under constant forcing of quadrupled CO2 (abrupt4×CO2) based on Eq. (1). (a) BCC-CSM2-MR in CMIP6 and (b) BCC-CSM1.1m in CMIP5.
In addition to the results of BCC-CSM2-MR and BCC-CSM1.1m, 23 other CMIP5 (Chen et al., 2019) and 9 CMIP6 models are also used for references and comparisons. The climate sensitivity and feedback parameters of the CMIP6 models are listed in Table 1.
Model ERF (F2×) ECS λ TCR κ λLWCS λSWCS λLWCL λSWCL λCL BCC-CSM1.1m (CMIP5) 3.45 2.91 −1.18 2.19 0.25 −1.97 0.76 0.16 −0.13 0.03 BCC-CSM2-MR 2.78 3.04 −0.92 1.40 0.40 −1.91 0.72 0.13 0.15 0.28 CESM2 3.22 5.19 −0.62 2.04 0.49 −1.80 0.53 −0.13 0.78 0.65 CESM2-WACCM 3.27 4.72 −0.69 1.97 0.55 −1.86 0.31 −0.17 1.02 0.85 CNRM-CM6-1 3.66 4.88 −0.75 2.26 0.48 −1.76 0.82 0.31 −0.11 0.20 CNRM-ESM2-1 2.99 4.76 −0.63 1.86 0.42 −1.60 0.78 0.25 −0.05 0.20 GISS-E2-1-G 4.03 2.69 −1.50 1.73 0.21 −1.59 0.58 0.25 −0.73 −0.49 GISS-E2-1-H 3.51 3.12 −1.13 1.90 0.40 −1.53 0.82 0.21 −0.62 −0.41 IPSL-CM6A-LR 3.36 4.60 −0.73 2.45 0.33 −1.54 0.79 −0.04 0.06 0.03 MIROC6 3.72 2.58 −1.44 1.59 0.53 −1.94 0.78 −0.05 −0.24 −0.29 MRI-ESM2-0 3.44 3.13 −1.10 1.70 0.60 −1.94 0.83 0.02 −0.01 0.01 MME (CMIP6) 3.40 3.87 −0.95 1.89 0.44 −1.75 0.69 0.08 0.02 0.10 StdDev (CMIP6) 0.36 1.04 0.33 0.31 0.12 0.17 0.17 0.17 0.55 0.43 Note: BCC-CSM1.1m: Beijing Climate Center Climate System Model version 1.1 with medium resolution;BCC-CSM2-MR: Beijing Climate Center Climate System Model version 2 with medium resolution;CESM2: Community Earth System Model version 2;CESM2-WACCM: CESM2 interactive with the Whole Atmosphere Chemistry Community Climate Model;CNRM-CM6-1: Centre National de Recherches Météorologiques Climate Model version 6.1;CNRM-ESM2-1: Centre National de Recherches Météorologiques Earth System Model version 2.1;GISS-E2-1-G: NASA Goddard Institute for Space Studies model version E2.1 with the GISS ocean model;GISS-E2-1-H: NASA GISS model version E2.1 with the Hybrid Coordinate Ocean Model;IPSL-CM6A-LR: Version 6 of the Institut Pierre-Simon Laplace (IPSL) climate model with low resolution;MIROC6: The sixth version of the Model for Interdisciplinary Research on Climate;MRI-ESM2-0: Meteorological Research Institute Earth System Model version 2.0.
Table 1. Climate sensitivity and feedback indices of BCC-CSM1.1m (CMIP5) and 10 CMIP6 (including BCC-CSM2-MR) models, including ERF of doubled CO2 concentration (F2×; W m–2), ECS (K), net climate feedback (λ; W m–2 K–1), TCR (K), OHU efficiency (κ; W m–2 K–1), clear-sky/cloud and longwave/shortwave components of λ (λLWCS, λSWCS, λLWCL, and λSWCL; W m–2 K–1), and cloud feedback (λCL = λLWCL + λSWCL; W m–2 K–1)
The ECS, TCR, ERF, and feedbacks are compared between BCC-CSM2-MR and BCC-CSM1.1m, as well as with the multimodel ensembles in CMIP6 and CMIP5. The ECS differs little in the two versions of the BCC model, that is, the values differ by 3.04 and 2.91 K in the BCC-CSM2-MR and BCC-CSM1.1m, respectively. The ECS values of the BCC models are slightly lower than the multimodel mean results, which are mostly evident for BCC-CSM2-MR due to an overall shift in the CMIP6 models to higher ECS values than those in the CMIP5 models (Fig. 3a). The small changes in the ECS from BCC-CSM1.1m to BCC-CSM2-MR result from compensation between the decrease in ERF (Fig. 3c) and positively increased net feedback (Fig. 3d). In contrast, the TCR is significantly different between the two versions. This value is 1.40 K in BCC-CSM2-MR, whereas it is 2.19 K in BCC-CSM1.1m (Fig. 3b). The distribution of TCR seems to follow that of ERF (Figs. 3b, c). The extremely small ERF in BCC-CSM2-MR (Fig. 3c) can explain the very low TCR, which is located at the lower bound of the CMIP6 model spread (Fig. 3b). Although the ERF of BCC-CSM1.1m is close to the CMIP5 multimodel mean (Fig. 3c), the TCR is slightly larger than the mean (Fig. 3b) due to an extremely low OHU efficiency (κ) with reference to the CMIP5 model spread (Fig. 3e), which is consistent with the evidently small ocean heat content that increases with global warming compared to that in BCC-CSM2-MR (Fig. 4). The OHU efficiency of BCC-CSM2-MR is close to the CMIP6 multimodel mean. The estimated κ based on observational data indicates a value of at least 0.5 W m−2 K−1 (Watanabe et al., 2013). Hence, κ in BCC-CSM2-MR shows an evident improvement from BCC-CSM1.1m.
Figure 3. Probability density distributions of (a) ECS (K), (b) TCR (K), (c) ERF (W m−2), (d) λ (W m−2 K−1), and (e) OHU efficiency (κ; W m−2 K−1) derived from 24 CMIP5 models and 10 CMIP6 models. Ranges between 5% and 95% (shading) for each parameter are shown. Corresponding values of BCC-CSM2-MR and BCC-CSM1.1m for corresponding parameters are marked by dashed lines in red and blue colors, respectively.
Figure 4. Responses of ocean heat content to global mean warming in the 1pctCO2 experiment of the two versions of BCC-CSM in CMIP6 (red) and CMIP5 (blue).
In summary, the two versions of the BCC model simulate different forcings and responses. The ECS in BCC-CSM2-MR is close to that in the previous version (Fig. 3a), whereas the TCR is evidently smaller than that in the previous version (Fig. 3b) due to a larger κ (Fig. 3e) and weaker ERF (Fig. 3c). The nearly unchanged ECS in BCC-CSM2-MR is the compensated result from more positive feedback (Fig. 3d) and evidently weakened effective forcing (Fig. 3c).
The four feedback components based on Eq. (3) are listed in Fig. 5 for the two BCC model versions, along with the multimodel results for CMIP5 and CMIP6. Consistent with Fig. 3d, the net feedback λNet exhibits a large spread across models, in which the two BCC-CSM versions are not located far from the multimodel mean. Evidently, the more positive net feedback in BCC-CSM2-MR (−0.92 W m−2 K−1) than that in BCC-CSM1.1m (−1.18 W m−2 K−1) mainly comes from λSWCL (0.15 vs −0.12 W m−2 K−1), which also has the largest uncertainty across models. Hence, understanding the changes in λSWCL is the key to explaining the changes in the net feedback of the BCC model system from CMIP5 to CMIP6.
Figure 5. Net climate feedback λ and its decompositions (λLWCS, λSWCS, λLWCL, and λLWCS; W m−2 K−1) of CMIP5 (light blue cross) and CMIP6 (light red cross) models, including BCC-CSM2-MR (red diamond), BCC-CSM1.1m (blue diamond), CMIP6 multi-model mean (dark red cross), and CMIP5 multi-model mean (dark blue cross).
To clearly show the contributions of regional processes to global mean feedbacks, the spatial distributions of longwave and shortwave components under clear-sky and cloud conditions are analyzed. Although the global means of λLWCS, λSWCS, and λLWCL change very slightly from BCC-CSM1.1m to BCC-CSM2-MR (Fig. 5), the regional variations may not be negligible. For example, as shown in Fig. 5d, the difference in λSWCS between BCC-CSM2-MR and BCC-CSM1.1m is only −0.04 W m−2 K−1 for the global mean, which actually results from compensation for prominent increases in high latitudes in the Northern Hemisphere and decreases over the Southern Ocean around Antarctica.
The differences in λLWCL and λSWCL between the two BCC-CSM versions also have evident regional distribution features. Compared with BCC-CSM1.1m, λLWCL in BCC-CSM2-MR increases in the western Indo-Pacific and decreases in the central Pacific, which leads to a global mean difference of nearly zero (−0.03 W m−2 K−1; Fig. 6f). The pattern of the λSWCL changes between BCC-CSM2-MR and BCC-CSM1.1m is almost opposite to the global changes in λLWCL, and the increasing λSWCL is stronger than the decreasing λLWCL in most regions (Figs. 6f, h). The changes in λLWCL and λSWCL reflect different cloud responses in the two BCC-CSM versions.
Figure 6. Geographical distributions of feedback components (W m−2 K−1) in BCC-CSM2-MR (left) and differences between BCC-CSM2-MR and BCC-CSM1.1m (right). (a, b) Clear-sky longwave feedback (λLWCS), (c, d) clear-sky shortwave feedback (λSWCS), (e, f) cloud longwave feedback (λLWCL), and (g, h) cloud shortwave feedback (λSWCL). The values on the top-right corner are global mean. Dotted shadings exceed the 5% significance level.
The variables λLWCL and λSWCL are mostly contributed by the responses of high-level and low-level clouds, respectively, because high-level clouds are effective in absorbing longwave radiation from the surface, while low-level clouds are good at reflecting incident solar radiation. For simplicity, we use the cloud ice path (CIP) and cloud water path (CWP) to represent high-level (cold and frozen) and low-level (warm and liquid) clouds, respectively. More high-level clouds under warming can lead to a positive λLWCL, whereas more low-level clouds cause a negative λSWCL. The patterns of the changes in CIP and CWP between the two BCC-CSM versions are similar (Figs. 7d, f), indicating that the responses of local high-level and low-level clouds are enhanced or weakened jointly in BCC-CSM2-MR, which is consistent with the opposite patterns of the λLWCL and λSWCL changes (Figs. 6f, h). As expected, a larger decrease in low-level clouds than in high-level clouds under warming (Figs. 7d, f) leads to a strengthened enhancement of the positive λSWCL rather than the negative λLWCL (Figs. 6f, h) in the BCC-CSM2-MR.
Figure 7. Geographical distributions of responses of cloud properties in BCC-CSM2-MR (left) and differences between BCC-CSM2-MR and BCC-CSM1.1m (right). (a, b) Cloud fraction (CF; % K−1), (c, d) cloud water path (CWP; g m−2 K−1), and (e, f) cloud ice path (CIP; g m−2 K−1). Dotted shadings exceed the 5% significance level.
The compensation between the reduced ERF and enhanced λ results in nearly no change in ECS from BCC-CSM1.1m to BCC-CSM2-MR. We mainly focus on the λ differences in this study because the forcing components derived from the Gregory method show patterns similar to those of λ but with the opposite sign (Fig. 8), with pattern correlation coefficients of approximately −0.85 globally (left columns in Figs. 6, 8). In fact, the ERF and λ derived from the Gregory method are not independent since the intermodel correlation coefficient between the two quantities listed in Table 1 is −0.69, exceeding the 5% significance level. FSWCL also mainly contributes to the reduced ERF (Fig. 8h), suggesting that the changes in ERF and λ may share similar mechanisms.
Figure 8. Geographical distributions of ERF components (W m−2) in BCC-CSM2-MR (left) and differences between BCC-CSM2-MR and BCC-CSM1.1m (right). (a, b) Clear-sky longwave forcing (FLWCS), (c, d) clear-sky shortwave forcing (FSWCS), (e, f) cloud longwave forcing (FLWCL), and (g, h) cloud shortwave forcing (FSWCL). Dotted shadings exceed the 5% significance level.
Approximately 60% of the increases in λSWCL (0.15 W m−2 K−1) in BCC-CSM2-MR are contributed by those in the Southern Ocean (Fig. 6h), which is caused by the decreasing CF and CWP with warming (Figs. 7b, d). It is interesting that the pattern of changes in λSWCL around Antarctica is nearly identical to that of λSWCS (Fig. 6d), suggesting a close relation between the changes in responses of sea ice and low clouds from BCC-CSM1.1m to BCC-CSM2-MR.
A previous study indicated that models with greater (lesser) ice coverage in piControl generally possess a colder (warmer) and drier (moister) climate, exhibit stronger (weaker) ice-albedo feedback and experience greater (weaker) warming under greenhouse gas forcing (Hu et al., 2017). We find that the control-run sea ice concentration (SIC) in the Southern Ocean in BCC-CSM2-MR is evidently less than that in BCC-CSM1.1m throughout an annual cycle, especially in the cold season (June–November), the period with maximum sea ice coverage (solid lines in Fig. 9). As suggested by the above mechanism, the GMST in BCC-CSM2-MR is 0.4°C higher than that in BCC-CSM1.1m in piControl (Fig. 1b), and the positive ice-albedo feedback (λSWCS) is weakest in the Southern Ocean (Fig. 6d), corresponding to slow sea ice melting with warming (dashed lines in Fig. 9). However, in contrast to the weak warming expected from the weak ice-albedo feedback, the warming in BCC-CSM2-MR is increased (Fig. 1b) under the quadrupled CO2 forcing because the weak ice-albedo feedback (Fig. 6d) is largely offset by the strong λSWCL in the Southern Ocean (Fig. 6h).
Figure 9. Annual cycle of Antarctica sea ice concentration (SIC; %; left y-axis) in piControl (solid lines) and its response to per 1-K global warming (% K−1; right y-axis) in abrupt4×CO2 experiment (dashed lines) of BCC-CSM2-MR (red) and BCC-CSM1.1m (blue).
To clarify the mechanism by which weakened λSWCS could lead to positively strengthened λSWCL, the response differences between BCC-CSM2-MR and BCC-CSM1.1m are analyzed in the cold season (Fig. 10) when the sea ice responses are the most distinct (Fig. 9). The results are masked south of 66.5°S (Fig. 10) because the shortwave processes can be neglected due to the limited solar radiation in the cold season. The response of the Antarctic SIC to per 1-K global warming in BCC-CSM2-MR shows slower melting than that in BCC-CSM1.1m (Fig. 10b). The pattern is nearly identical to that of λSWCL (Fig. 10a), indicating a strong linkage between sea ice response and cloud shortwave processes. When the sea ice melts slowly in BCC-CSM2-MR, more sensible and latent heat fluxes from the ocean to the atmosphere are inhibited (Fig. 10d). Meanwhile, the static stability in the boundary layer, represented by the air temperature difference between 925 and 850 hPa, is enhanced due to the cold surface (Fig. 10c). As a result, the CWP, representing low clouds, decreases around Antarctica (Fig. 10c) due to weakened vertical upward motion (Wall et al., 2017). The decreasing response of low clouds in BCC-CSM2-MR leads to increased shortwave radiation into the air-earth system to warm the climate, largely offsetting the relatively weak ice-albedo feedback in the Southern Ocean. Therefore, compared with BCC-CSM1.1m, the strong positive λSWCL around Antarctica in BCC-CSM2-MR and its close relation with weak ice-albedo feedback is well clarified, which can be attributed to the reduced sea ice coverage in the piControl simulation.
Figure 10. Differences of (a) λSWCL (W m−2 K−1), (b) response of SIC (% K−1), (c) responses of CWP (g m−2 K−1; shadings) and static stability in the boundary layer (temperature difference between 925 and 850 hPa; K K−1; contours drawn for ±0.2, ±0.6, and ±1.0), and (d) surface sensible (SH; W m−2 K−1; contours drawn for ±2, ±6, and ±10) and latent heat (LH) responses (shadings) in cold seasons (June–November) between BCC-CSM2-MR and BCC-CSM1.1m. The solid and dashed contours denote the positive and negative values. Dotted shadings exceed the 5% significance level.
|Note: BCC-CSM1.1m: Beijing Climate Center Climate System Model version 1.1 with medium resolution;BCC-CSM2-MR: Beijing Climate Center Climate System Model version 2 with medium resolution;CESM2: Community Earth System Model version 2;CESM2-WACCM: CESM2 interactive with the Whole Atmosphere Chemistry Community Climate Model;CNRM-CM6-1: Centre National de Recherches Météorologiques Climate Model version 6.1;CNRM-ESM2-1: Centre National de Recherches Météorologiques Earth System Model version 2.1;GISS-E2-1-G: NASA Goddard Institute for Space Studies model version E2.1 with the GISS ocean model;GISS-E2-1-H: NASA GISS model version E2.1 with the Hybrid Coordinate Ocean Model;IPSL-CM6A-LR: Version 6 of the Institut Pierre-Simon Laplace (IPSL) climate model with low resolution;MIROC6: The sixth version of the Model for Interdisciplinary Research on Climate;MRI-ESM2-0: Meteorological Research Institute Earth System Model version 2.0.|