3.2. The Climate Effects of Anthropogenic Aerosols
Figure 2 shows anthropogenic aerosol effective radiative forcing (ERFall = F
ALL−CTL) and its two main components, i.e., RF from direct radiative forcing (RFari = Faerosol
ALL−CTL) and effects on cloud radiative forcing (Fcloud
*ALL−CTL). The global mean ERFalls from the GAMIL, ECHAM, and CAM models were −0.27, −0.63, and −0.54 W∙m
−2, respectively. The ERFall from the GAMIL model was clearly weaker (less negative) than that from the ECHAM and CAM models. The global mean RFari from GAMIL was −0.21 W∙m
−2, which was also weaker than that from the ECHAM (−0.33 W∙m
−2) and CAM (−0.29 W∙m
−2) models. However, the regional distribution pattern of RFari was generally similar for all models. As expected from the regional maximum in AODa (
Figure 1), east and south Asia were the largest contributors to globally averaged RFari. The global means of Fcloud
*ALL−CTL from the GAMIL, ECHAM, and CAM models were −0.05, −0.28, and −0.24 W∙m
−2, respectively. As compared to RFari, the difference in Fcloud
*ALL−CTL among these models contributed most to the model diversity in ERFall. For all models, there were few regions of Fcloud
*ALL−CTL that could pass the 10% significance level test. In addition to the instantaneous Twomey effect (i.e., RFaci), Fcloud
*ALL−CTL also includes the rapid adjustments from “ari” and the Twomey effect. The complex rapid adjustments might be the main source of uncertainty. In contrast, the diagnosed Faerosol can be well constrained because it is not closely related to rapid adjustments [
25,
33]. The RFari was statistically significant over high anthropogenic aerosol burden areas (AODa > 0.1). Because of the uncertainty in Fcloud
*ALL−CTL, ERFall was generally not statistically significant, expect for in some regions of east and south Asia.
Figure 3 shows the clear-sky effective radiative forcing (ERF
call = F
cALL−CTL) and clear-sky instantaneous radiative forcing from “ari” (RF
cari = Faerosol
cALL−CTL). For all models, the ERF
call was almost the same as the RF
cari. The global annual mean difference between ERF
call and RF
cari (i.e., F
c*ALL−CTL) was small and negligible (≤0.02 W∙m
−2,
Table 3). In other words, the RF
cari clearly dominated the ERF
call magnitude. The spatial patterns of negative RF
cari were all roughly consistent with AODa (
Figure 1), as RF
cari depends primarily on “ari”. The global means of RF
cari from the GAMIL, ECHAM, and CAM models were estimated at −0.45, −0.73, and −0.74 W∙m
−2, respectively. The RF
cari from GAMIL was substantially weaker than that from the ECHAM and CAM models. This result is consistent with RFari (
Figure 2). Note that the clear-sky RF
cari from CAM was a little stronger than that of ECHAM, whereas the all-sky RFari from CAM was a little weaker than that of ECHAM. This suggests that the cloud masking effect (RFari − RF
cari) in the CAM model was stronger due to its higher cloud optical property (COD,
Table 3). The statistically significant regions of negative RF
cari were substantially larger than seen for RFari (
Figure 2). After excluding the cloud masking effect, the RF
cari was considerably less variable than the RFari.
Here, the anthropogenic aerosol climate effects (combined “ari” and “aci”) based on the CMIP6 protocol are compared with those based on the models’ own default treatments.
Table 3 lists the differences between the CTL and ALL experiments and the differences between the CTL and OLD experiments. Firstly, we analyzed the GAMIL model experiments. The AOD
ALL−CTL and AOD
OLD−CTL were 0.032 and 0, respectively. The default GAMIL model does not consider anthropogenic aerosol direct radiative forcing. Thus, the Faerosol
ALL−CTL was −0.21 W∙m
−2, whereas the Faerosol
OLD−CTL was 0.06 W∙m
−2. Note that the positive value of Faerosol
OLD−CTL was caused by the impact of anthropogenic aerosol “aci” on diagnosing Faerosol. This is well discussed in
Section 3.4. Because only the Twomey effect is considered in the MACv2-SP, the relative changes in the column-integrated grid-mean cloud droplet number concentration (CDNC
ALL−CTL) and the liquid water path (LWP
ALL−CTL) were very small. However, CDNC
OLD−CTL and LWP
OLD−CTL were obvious because a stronger Twomey effect and subsequent lifetime effect are considered in the default GAMIL model. This is one reason for the fact that the COD
ALL−CTL (0.051) was clearly less than the COD
OLD−CTL (1.985). As expected, the Fcloud
*ALL−CTL (−0.05 W∙m
−2) was clearly weaker than the Fcloud
*OLD−CTL (−1.95 W∙m
−2). Hence, the ERFall estimated based on the CMIP6 protocol (F
ALL−CTL, −0.27 W∙m
−2) was clearly weaker than that based on the model’s own default treatment (F
OLD−CTL, −1.98 W∙m
−2). Secondly, we analyzed the ECHAM model experiments. The AOD
ALL−CTL (0.025) was a little larger than the AOD
OLD−CTL (0.021). This explains why the Faerosol
ALL−CTL (−0.33 W∙m
−2) and Faerosol
cALL−CTL (−0.73 W∙m
−2) were stronger than the Faerosol
OLD−CTL (−0.26 W∙m
−2) and Faerosol
cOLD−CTL (−0.65 W∙m
−2). Because the default ECHAM model does not consider the aerosol indirect effect, the Fcloud
*OLD−CTL was almost zero (0.02 W∙m
−2). However, the Fcloud
*ALL−CTL was −0.28 W∙m
−2. Thus, the ERFall estimated based on the CMIP6 protocol (F
ALL−CTL, −0.63 W∙m
−2) was clearly stronger than that based on the model’s own default approach (F
OLD−CTL, −0.21 W∙m
−2). Finally, we analyzed the CAM model experiments. The AOD
ALL−CTL (0.027) was larger than the AOD
OLD−CTL (0.018). This is the primary reason for the fact that the Faerosol
ALL−CTL (−0.29 W∙m
−2) was stronger than the Faerosol
OLD−CTL (−0.07 W∙m
−2). However, the Fcloud
*ALL−CTL (−0.24 W∙m
−2) was clearly weaker than the Fcloud
*OLD−CTL (−2.12 W∙m
−2). As a result, the ERFall estimated from the MACv2-SP (F
ALL−CTL, −0.54 W∙m
−2) was clearly weaker than that from the model’s own default approach (F
OLD−CTL, −2.22 W∙m
−2). In short, the difference in ERFall based on the models’ default approaches among these three models (−1.98 W∙m
−2 GAMIL, −0.21 W∙m
−2 ECHAM, −2.22 W∙m
−2 CAM) was clearly greater than that based on the CMIP6 protocol. The model diversity in ERFall was dramatically reduced after using the same anthropogenic aerosol forcing. The ERFall based on the CMIP6 protocol from the ECHAM (−0.63 W∙m
−2) and CAM (−0.54 W∙m
−2) models fell within the range of ERFall with the five climate models (from −0.9 to −0.4 W∙m
−2) shown in Fiedler et al. [
29]. However, the GAMIL model produced a very weak ERFall (−0.27 W∙m
−2). Excluding this outlier, the difference in ERFall between the ECHAM and CAM models was 0.10 W∙m
−2, which was much less than that based on the models’ own default approaches (2.01 W∙m
−2).
3.3. Contributions from “Ari”
As discussed in
Section 2.4, there are two ways to estimate the contributions from “ari”, i.e., the difference between the ARI and CTL experiments and the difference between the ALL and ACI experiments.
Table 4 lists these two kinds of changes. For all models, under clear-sky conditions, the global annual mean F
cARI−CTL was almost identical to the F
cALL−ACI. In other words, the values for ERF
cari calculated by the above two methods were almost identical. Furthermore, the standard deviations of ERF
cari (i.e., F
cARI−CTL and F
cALL−ACI) from all model experiments were very small. After considering cloudy skies, the standard deviations of ERFari (i.e., F
ARI−CTL and F
ALL−ACI) were clearly increased. Therefore, the F
ARI−CTL might be quite different from F
ALL−ACI, e.g., the F
ARI−CTL (−0.21 W∙m
−2) from the GAMIL model was 50% stronger than the F
ALL−ACI (−0.14 W∙m
−2). This indicates that the radiative effect of cloud-relevant rapid adjustment is the main contributor to the perturbation of modeled ERFari. The averaged ERFari results (0.5F
ARI−CTL + 0.5F
ALL−ACI) from the GAMIL, ECHAM, and CAM models were −0.18, −0.28, and −0.23 W∙m
−2, respectively. The maximum difference among these three models was 0.10 W∙m
−2.
The Faerosol
ALL−ACI and Faerosol
cALL−ACI were almost identical to the Faerosol
ARI−CTL and Faerosol
cARI−CTL (not shown). This again indicates that the diagnosed RFari and RF
cari can be well constrained. Here, only the RFari (Faerosol
ARI−CTL) and RF
cari (Faerosol
cARI−CTL) from the ARI and CTL experiments are shown in
Figure 4. It is obvious that both RFari and RF
cari from the GAMIL model (−0.21 and −0.45 W∙m
−2) were substantially weaker than those from the ECHAM (−0.35 and −0.73 W∙m
−2) and CAM (−0.31 and −0.74 W∙m
−2) models. The main reason for this is that the all-sky and clear-sky natural aerosol radiative forcings (Faerosol and Faerosol
c) from the GAMIL model (−5.74 and −8.58 W∙m
−2) were stronger (more negative) than those from the ECHAM (−2.47 and −4.44 W∙m
−2) and CAM (−1.49 and −2.70 W∙m
−2) models (
Table 3). A stronger Faerosol (Faerosol
c) can result in a weaker modeled RFari (RF
cari) [
33]. A sensitivity test (not introduced in this study) showed that the RFari from the GAMIL model can be enhanced to −0.30 W∙m
−2 by reducing the background natural aerosol radiative effect.
The semi-direct effects from all models are shown in
Figure 5. Both Fcloud
*ARI−CTL and Fcloud
*ALL−ACI can represent the semi-direct effect. The global mean semi-direct effect ranged from 0 to 0.08 W∙m
−2. The positive sign is consistent with the fact that the rapid adjustments reduce the ERF of black carbon [
54]. As compared with RFari, the spatial distribution of the semi-direct effect was very disorderly. The regional distribution of Fcloud
*ARI−CTL was not close to Fcloud
*ALL−ACI. There were few regions that could pass the 10% significance level test. In terms of global mean values, the magnitudes of Fcloud
*ARI−CTL and Fcloud
*ALL−ACI were statistically non-significant as compared with their standard deviations (
Table 4). Taking the ECHAM model experiments, for example, the global means for Fcloud
*ARI−CTL and Fcloud
*ALL−ACI were 0.08 and 0.02 W∙m
−2, respectively. These values were not clearly larger than the corresponding standard deviations (0.07 and 0.10 W∙m
−2). The differences between Fcloud
*ARI−CTL and Fcloud
*ALL−ACI from the GAMIL (0.07 W∙m
−2) and ECHAM (0.06 W∙m
−2) models were very noticeable. This suggests that 10 member ensembles with the 10-year averages are not enough to get a stable estimate of semi-direct effect for these two models.
3.4. Contributions from “Aci”
Like “ari”, there are also two ways to estimate the contributions from “aci” (
Table 5). For all models, the changes in Fcloud were very close to the changes in Fcloud
*, because of the fact that “ari” was excluded. Note that the Faerosol might be non-zero, e.g., the Faerosol
ALL−ARI from the ECHAM model was 0.02 W∙m
−2 and larger than its standard deviation (0.01 W∙m
−2). The reason for this is that “aci” could also impact the diagnosis of Faerosol. Compared with the impact of “ari” on diagnosing Fcloud, this impact was very small and negligible. It is obvious that ERFaci mainly depended on the changes in Fcloud or Fcloud
*. The standard deviations of the changes in Fcloud, Fcloud
*, and F were noticeable. As a result, the F
ACI−CTL may be clearly different from F
ALL−ARI. The averaged ERFaci (0.5F
ACI−CTL + 0.5F
ALL−ARI) from the GAMIL, ECHAM, and CAM models were −0.09, −0.35, and −0.32 W∙m
−2, respectively. The maximum difference in the averaged ERFaci among the three models was 0.26 W∙m
−2. This was the dominant source of the model diversity in ERFall as compared with ERFari.
The spatial patterns of ERFaci were almost identical to changes in Fcloud
* (not shown) because the ERFaci mainly depends on the Twomey effect. The anthropogenic aerosol effects on cloud forcing from the Twomey effect can be estimated as Fcloud
*ACI−CTL or Fcloud
*ALL−ARI (
Figure 6). For all models, there were few regions that could pass the 10% significance level test. The main reason for these uncertainties is discussed in the next paragraph. For the ECHAM and CAM models, the spatial patterns of Fcloud
*ACI−CTL were generally similar to Fcloud
*ALL−ARI, but one should keep in mind that the local Fcloud
*ACI−CTL and Fcloud
*ALL−ARI were almost not statistically significant. The common negative regions were located over high anthropogenic aerosol burden areas (dN > 1.25,
Figure 1). As compared with the ECHAM and CAM models, the spatial patterns of Fcloud
*ACI−CTL and Fcloud
*ALL−ARI from the GAMIL model seem disordered. This phenomenon might be explained by the weak Twomey effect. In terms of global mean values, the Fcloud
*ACI−CTL and Fcloud
*ALL−ARI from the GAMIL model (−0.12 and −0.06 W∙m
−2) were clearly weaker than that from the ECHAM (−0.30 and −0.36 W∙m
−2) and CAM (−0.31 and −0.30 W∙m
−2) models. The Twomey effect works by changing cloud optical depth (COD). The COD
ACI−CTL and COD
ALL−ARI from the ECHAM model (0.175 and 0.198) were clearly larger than that from the GAMIL (0.096 and 0.089) and CAM (0.120 and 0.110 W∙m
−2) models. This is not consistent with the difference in the Twomey effect among these three models. Furthermore, both the GAMIL and CAM models use a two-moment stratiform cloud microphysics scheme, whereas a one-moment cloud microphysics scheme is used in the ECHAM model. This is also not consistent with the difference in the Twomey effect among these three models. It seems that the reason for the model diversity in the Twomey effect is very complex and difficult to figure out.
The standard deviations of Fcloud
*ACI−CTL and Fcloud
*ALL−ARI from the GAMIL model were 0.08 W∙m
−2 and 0.15 W∙m
−2, respectively. These standard deviations were close to their corresponding ensemble averages (−0.12 and −0.06 W∙m
−2). The Fcloud
*ACI−CTL and Fcloud
*ALL−ARI not only include the instantaneous Twomey effect (i.e., RFaci), but also subsequent changes in cloud optical properties from rapid adjustments. It is necessary to figure out the relative contribution of RFaci and rapid adjustments.
Figure 7 shows RFari from the ACI and ALL experiments. Note that both RFaci
CTL and RFaci
ARI were zero. The RFaci
ACI and RFaci
ALL showed identical magnitudes and regional distributions. The global magnitude of RFaci was −0.10 W∙m
−2. There were only negative regions of RFaci, and the maximum negative region was located in east and south Asia and adjacent oceans. All regions of RFaci were statistically significant. In short, the modeled RFaci was very stable. This indicates that the rapid adjustments induced by the Twomey effect contributed to the standard deviations of ERFaci, Fcloud
*ACI−CTL, and Fcloud
*ALL−ARI. In terms of global mean, the sign of this rapid adjustment was not clear (Fcloud
*ACI−CTL − RFaci
ACI = −0.02 W∙m
−2 or Fcloud
*ALL−ARI − RFaci
ALL = 0.04 W∙m
−2). It is interesting to point out that the impact of model internal variability on RFari was obvious (
Figure 2 and
Figure 4) as compared with RFaci. The reason is that RFari is derived from the difference between two experiments (e.g., Faerosol
ARI−CTL), whereas RFaci is derived from one experiment (e.g., RFaci
ACI). As compared with RFaci, RFari includes the impact of the differences in background atmospheric state between two experiments caused by model internal variability.