Technical Comments

Comment on “Observational and Model Evidence for Positive Low-Level Cloud Feedback”

See allHide authors and affiliations

Science  16 Jul 2010:
Vol. 329, Issue 5989, pp. 277
DOI: 10.1126/science.1186796


Clement et al. (Reports, 24 July 2009, p. 460) provided observational evidence for systematic relationships between variations in marine low cloudiness and other climatic variables and found that most current-generation climate models were deficient in reproducing such relationships. Our analysis of one of these models (GFDL CM2.1), using more complete model output, indicates better agreement with observations, suggesting that more detailed analysis of climate model simulations is necessary.

Clement et al. (1) found that decadal variations in low cloudiness in the subtropical Northeast (NE) Pacific were strongly correlated with several climatic variables. They noted that the negative correlation between low cloud amount and sea surface temperature (SST) in this region was suggestive of a positive low-cloud feedback on decadal time scales and proposed that the ability to reproduce the observed relationship would constitute a useful test of low cloud-climate relationships in climate models. Using output archived by the Coupled Model Intercomparison Project phase 3 (CMIP3), they examined a set of current-generation climate models and found that most models did not reproduce the observed relationships.

We suggest that the assessment of Clement et al. regarding one such model, GFDL CM2.1 (2), is too pessimistic because they did not directly examine the relationship between low cloudiness and regional climate variables. Because low cloud amount was not archived in the CMIP3 database, total cloud amount was used instead, a choice that was presumably based on the similarity of the total cloud–climate and low cloud–climate relationships in the observed data. (Three-dimensional monthly mean cloud amount distributions were archived by CMIP3, but approximating low cloud amount from them would have been problematic.)

Using output from a GFDL CM2.1 20th-century climate simulation (GFDL CM2.1 20C3M, run 1), we first computed the correlations between total cloud amount and regional climate variables and compared them with the results of Clement et al. (1) as a quality check. Both sets of correlations agree to within 0.03 (Table 1), which is an acceptable deviation given that different software and analysis periods were likely used.

Table 1

Correlation between cloud quantities and other climatic variables in the NE Pacific for observations and the GFDL CM2.1 climate model. Observed correlations are taken from Clement et al. (1). Cloud variables are correlated with SST, LTS, SLP, and mid-tropospheric pressure vertical velocity (ω500). ISCCP, International Satellite Cloud Climatology Project; COADS, Comprehensive Ocean Atmosphere Data Set; MSC, marine stratiform cloud cover.

View this table:

Using low cloud amount (defined as the cloud fraction in the layer below the 680-hPa pressure level) from GFDL CM2.1, correlations with all of the regional climatic variables become much closer to the observed values. The model simulates correlations with the correct sign for all four variables, suggesting that the feedback process identified by Clement et al. (1) also operates in this model.

Teleconnections between SST in the NE Pacific and other climate variables provide further evidence of the similarity of the simulated and observed feedback processes. When the NE Pacific is warm, low cloud anomalies are negative in situ and west of the South American coast and positive in a band extending east-southeastward from the west central Pacific (Fig. 1A). These features are broadly similar to those diagnosed by Clement et al. (1), although a region of simulated positive cloud anomalies just west of Baja, California, is not consistent with observations. The simulated total cloud anomaly pattern (fig. S1) is also in broad agreement with the observed pattern, although there are regional discrepancies in the eastern portion of the basin that may be associated with the tendency for convection to be displaced to the west in GFDL CM2.1 as a result of a cold bias in equatorial SST (3).

Fig. 1

Regression of climate variables on the time series of SST averaged over the NE Pacific (115° to 145°W, 15° to 25°N). All variables are annual means with a 1-2-1 smoothing applied. Values are shown per degree change in NE Pacific SST. (A) Low cloud amount (%). (B) SLP (colors, in hPa) and surface winds (vectors; standard vector length is 1.5 m/s).

Similarities are also evident between the simulated and observed atmospheric circulation patterns associated with variations in NE Pacific SST (Fig. 1B). A negative sea-level pressure (SLP) anomaly and cyclonic circulation in the central North Pacific occurs when the NE Pacific is warm, with a weakening of the Walker circulation and westerly wind anomalies in the western equatorial Pacific. Southerly wind anomalies in the NE Pacific stratocumulus region represent a weakening of the climatological trade winds, as in observations [figure 2B in (1)]. The teleconnections of NE Pacific SST with local SST, mid-tropospheric vertical motion, and lower tropospheric stability (figs. S2 to S4) also resemble the corresponding observed patterns.

The most noteworthy discrepancy between the observed feedback processes and the GFDL CM2.1 simulation involves the sensitivity of the in situ correlations to the metric of cloud amount that is used. In the model, the correlations with vertical velocity and lower tropospheric stability change sign when total clouds are used instead of low clouds, and the correlation with SLP essentially vanishes. This does not happen in the real climate system, although the correlations with total clouds are weaker, especially for lower tropospheric stability (LTS) and vertical velocity (Table 1). Using total cloud amount as a surrogate for low cloud amount apparently makes a larger difference in the model because the simulated variability of high clouds over the NE Pacific is larger than in observations.

Models are imperfect representations of the climate system, and simulating the relationships between clouds and climate is particularly challenging. We readily acknowledge that much work remains if models are to represent such interactions with greater fidelity. For example, the relationship between NE Pacific SST and the in situ net cloud radiative effect is only about half as large in the model as in observations. This discrepancy is mainly in the longwave component of the cloud radiative effect and is likely due to a difference in high-cloud variability. In addition, the spatial pattern of SST variability in the model is not identical to the observed pattern, which is evident in the local differences in the regressions of SST, clouds, and LTS just west of Baja, California. Despite these differences, we find that the representation by GFDL CM2.1 of decadal variations in NE Pacific low-level clouds, when analyzed with the method of Clement et al. (1), is considerably more realistic than their analysis would suggest. We also note that GFDL CM2.1 and HadGEM1 (the only model found by Clement et al. to successfully represent these relationships) share a common formulation of boundary layer turbulent mixing (4), suggesting that this parameterization may contain the necessary ingredients to successfully simulate these variations.

Although the ability to simulate interdecadal variations in cloudiness constitutes a useful test of climate models, we note that this may not be a sufficient condition for accurately predicting the low-cloud feedback in response to global warming. Despite having similar cloud-climate correlations in the context of NE Pacific variability, the estimated doubled-CO2 climate sensitivity of 4.4 K for HadGEM1 (5) is considerably larger than the 3.4 K sensitivity of GFDL CM2.1 (6). Although the results of Clement et al. are suggestive of a positive low-cloud feedback, it may not be possible to robustly establish the sign of this feedback based on the evidence currently available.

The analysis we have performed on GFDL CM2.1 may be more broadly relevant to other CMIP3 models, although it remains to be determined whether other models would have appeared more realistic had low cloud amount been available. With the more complete cloud diagnostics that will soon be available through the next CMIP intercomparison, it will be possible to better analyze decadal cloud variability in model-simulated marine low cloudiness.

Supporting Online Material

Figs. S1 to S4

References and Notes

  1. The authors acknowledge the contributions of the anonymous reviewers, which were helpful in improving this manuscript. The contribution of A.J.B. to this work was supported by the New Jersey Agricultural Experiment Station. The contribution of S.A.K. to this work was funded through the Atmospheric System Research and Regional and Global Climate Modeling Programs of the Office of Science in the U.S. Department of Energy and was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under contract DE-AC52-07NA27344.
View Abstract

Navigate This Article