Observational and Model Evidence for Positive Low-Level Cloud Feedback

Science, July 24, 2009

Amy C. Clement,^1,* Robert Burgman,¹ Joel R. Norris²

Feedbacks involving low-level clouds remain a primary cause of uncertainty in global climate model projections. This issue was addressed by examining changes in low-level clouds over the Northeast Pacific in observations and climate models. Decadal fluctuations were identified in multiple, independent cloud data sets, and changes in cloud cover appeared to be linked to changes in both local temperature structure and large-scale circulation. This observational analysis further indicated that clouds act as a positive feedback in this region on decadal time scales. The observed relationships between cloud cover and regional meteorological conditions provide a more complete way of testing the realism of the cloud simulation in current-generation climate models. The only model that passed this test simulated a reduction in cloud cover over much of the Pacific when greenhouse gases were increased, providing modeling evidence for a positive low-level cloud feedback.

¹ Rosenstiel School of Marine and Atmospheric Sciences, University of Miami, Division of Meteorology and Physical Oceanography, MSC 362, 4600 Rickenbacker Causeway, Miami, FL 33149, USA.
² Scripps Institution of Oceanography, University of California-San Diego, La Jolla, CA 920930224, USA.

^* To whom correspondence should be addressed. E-mail: aclement@rsmas.miami.edu

Low-level clouds are of great climatic importance because of their net cooling effect on the global climate (1). If the coverage of this type of cloud were to change as the climate warms, it could lead to either an enhancement or a reduction in the warming (i.e., as either a positive or negative feedback, depending on whether cloud cover decreases or increases). At present, the sign of the low-level cloud feedback in climate change is unknown (25).

Previous research on the subtropical stratocumulus decks in the Northeast (NE) Pacific has laid the groundwork for our current understanding of environmental controls on this cloud type (610). These studies have shown that changes in local meteorological conditions can explain much of the variability in low-level cloud cover occurring on daily to interannual time scales. The longer-term variability of these clouds, however, has received much less attention, partly because long-term fluctuations in any particular cloud data set would rightly be regarded with some skepticism. Surface-based cloud observations, for instance, can be problematic due to the subjective nature of the measurement and sparse sampling for large regions of the ocean (11, 12). Satellite-based cloud observations have spurious trends related to instrument drift and calibration (13, 14) and are available for only the past 25 years. Here we examine long-term cloud variations in independent cloud data sets and analyze meteorological data to provide a physical framework for interpreting these variations.

Our principal source of data is monthly mean gridded surface-based observations of total cloud cover from the Comprehensive Ocean Atmosphere Data Set (15) (COADS) during 1952 to 2007. We supplement this with cloud-type information from COADS that has been compiled by Hahn and Warren (16) for the period 1952 to 1997, and in particular, we examine the category of marine stratiform clouds (comprising ordinary stratocumulus, cumulus under stratocumulus, fair-weather stratus, and bad-weather stratus). Additional independent information on total cloud amount, low-level cloud amount, and surface radiative fluxes is provided by the International Satellite Cloud Climatology Project (17, 18) (ISCCP). Before using ISCCP data, we applied some adjustments to remove spurious long-term variability caused by satellite artifacts and to account for erroneous retrievals of low-level cloud-top height (19). Other climate variables used in the analysis are sea surface temperature (SST) (20), sea-level pressure (SLP) from the Hadley center reanalysis (21), and vertical velocity, surface winds, and lower tropospheric static stability (potential temperature at 700 mb minus surface temperature) from the ERA-40 reanalysis (22).

The time series of total and low-level cloud cover averaged over the NE Pacific (115° to 145°W, 15° to 25°N) are displayed in Fig. 1, A and B. Both COADS and adjusted ISCCP data sets show a shift toward more total cloud cover in the late 1990s, and the shift is dominated by low-level cloud cover in the adjusted ISCCP data (bars, in Fig. 1B). The longer COADS total cloud time series indicates that a similar-magnitude shift toward reduced cloud cover occurred in the mid-1970s, and this earlier shift was also dominated by marine stratiform clouds (bars, Fig. 1A). PATMOS-X, a next-generation version (23) of the Advanced Very High Resolution Radiometer Pathfinder Atmosphere (PATMOS) data set with improved algorithms (24), shows similar signals over the 1982 to 2007 period (fig. S1). These cloud changes appear throughout the year, and the shifts are also apparent in SST and SLP time series (Fig. 1, C and D).

The decadal changes in NE Pacific clouds and climate are linked to well-known basin-wide climate shifts (2530). This is illustrated in Fig. 2, A and B, which shows that the regression patterns of SST, SLP, and ERA-40 surface winds on the NE Pacific SST time series resemble the now familiar pattern of Pacific Decadal Variability. The SST signal spans the entire Pacific basin and persists throughout the year, and the SLP pattern comprises a weaker Walker circulation in the equatorial region and a deeper Aleutian low in the North Pacific (Fig. 2B). The extension of the North Pacific SLP low anomaly into the stratocumulus region constitutes a weakening of the climatological high, and trade winds around the high are weakened (hence the anomalous southerly and westerly flow shown in Fig. 2B). The subsidence and lower tropospheric stability (LTS) in the NE Pacific are both weaker when SST is warm there (fig. S2).

The spatial patterns of cloud-cover change (Fig. 3) are physically consistent with the local meteorological changes displayed in Fig. 2 and fig. S2, with reduced cloud cover in the NE Pacific when SST is warm, SLP is low, and subsidence, equatorward advection, and static stability are weak (610). For COADS total cloud, we calculate the regression over the entire time period (Fig. 3A), which includes both the 1976 and late-1990s shifts. The regression for marine stratiform cloud (Fig. 3B), however, includes only the 1976 shift due to lack of the Hahn and Warren data compilation after 1997. A comparison of the patterns in Fig. 3, A and B, indicates that marine stratiform cloudiness dominates the total cloud cover change and that the climate shifts in 1976 and the late 1990s were analogous but of opposite sign (i.e., the earlier shift was a warming and reduction in clouds and the latter a cooling and increase in clouds). The adjusted ISCCP regressions for total and low-level cloud (Fig. 3, C and D) are in agreement with both total and marine stratiform cloudiness from COADS. This concurrence is surprising given the fundamentally different measurement methods (human eye versus satellite retrieval and algorithm). Furthermore, the similarity in pattern and magnitude between adjusted ISCCP low-level cloud cover and the COADS marine stratiform cloud cover is especially impressive considering that they do not occur over the same climate shifts (adjusted ISCCP captures only the late-1990s shift, whereas COADS marine stratiform captures only the 1976 shift). The larger size of both COADS and adjusted ISCCP low-level cloud signals relative to the total cloud signals in the NE Pacific indicates that upper-level clouds increase when low-level clouds decrease. Enhanced upper-level cloud cover is consistent with the weakening of subsidence over the NE Pacific (fig. S2).

We emphasize that the NE Pacific cloud changes described above are tied to cloud changes that span the Pacific basin. Despite much less surface sampling in the Southeast (SE) Pacific, cloud and meteorological changes in that region generally occur in parallel with those in the NE Pacific (Figs. 2 and 3). Also, we find that the leading mode in an empirical orthogonal function analysis (15% of the variance) of global cloud cover (fig. S3) has a spatial pattern similar to that in Fig. 3 and the time series shows the same decadal shifts as in Fig. 1, indicating that the changes in the NE Pacific are part of a dominant mode of global cloud variability.

The regression of adjusted shortwave and longwave cloud radiative effects from the ISCCP Flux Dataset on NE Pacific SST reveals that the change in net cloud radiative effect warms the ocean by about 6 W m^2 K^1 (fig. S4). Despite the weaker winds, latent heat flux anomalies still act to cool the ocean when SST is warmer (31). Model studies have shown a negligible simulated SST response when forced with a wind pattern like that displayed in Fig. 2B (32), suggesting that ocean dynamics play little role in NE Pacific decadal SST variability. Hence, we conclude that a change in solar heating of the ocean due to a change in stratocumulus cloud cover is the principal factor maintaining decadal SST anomalies in the NE Pacific. Previous studies have shown that decreased cloud cover and warm SST additionally promote weaker circulation (3337). This response is caused by a decrease in longwave radiative cooling of the boundary layer by clouds that reduces large-scale horizontal temperature and pressure gradients. The existence of these same relationships among SST, cloud, and circulation on decadal time scales implies that changes in subtropical stratocumulus act as a positive feedback on climate in the region.

Is this feedback present in climate models? To address this question, we analyze the 20th-century climate simulation in 18 coupled ocean-atmosphere general circulation models with comprehensive output available from the World Climate Research Programmes (WCRPs) Coupled Model Intercomparison Project phase 3 (CMIP3) multimodel archive (38, 39). Correlations between cloud cover in the NE Pacific and the local thermal structure (SST and LTS) and circulation (SLP and mid-tropospheric vertical velocity) are computed for each model and compared with observations in Table 1. Models are grouped according to whether they have the wrong sign correlation relative to observations. By eliminating models successively on this basis, we are left with only two that simulate the correct sign correlations for all variables, the INM-CM3.0 and the HadGEM1. Because these two models represent opposite ends of the range of values of equilibrium climate sensitivity (INM-CM3.0 has the lowest value and HadGEM1 has the highest) (5), the cloud-meteorology correlation test alone is not a sufficient metric for global climate sensitivity.

These models are distinct in other ways that are relevant for the simulation of low-level clouds. The INM-CM3.0 adopts a more empirical approach that parameterizes low-level cloud cover as a linear function of relative humidity with coefficients that depend on temperature, altitude, land/ocean, and stratification (40), whereas the HadGEM1 has higher spatial resolution, more explicit cloud microphysics, interactive parameterization of cloudiness as a function of local variability in humidity, and a sophisticated planetary boundary-layer mixing scheme (41, 42). Moreover, the HadGEM1 produces doubled carbon dioxide (2 x CO₂) changes in SST, LTS, and circulation that are consistent with the multimodel mean, but the INM-CM3.0 does not (fig. S5). Unlike the INM-CM3.0, most models simulate a weakening of tropical atmospheric circulation under increased greenhouse gases (43)a phenomenon that appears in 20th-century observations as well (43, 44).

Our observational analysis indicates that increased SST and weaker subtropical highs (Fig. 4A) will act to reduce NE Pacific cloud cover, as indeed occurs in HadGEM1 under increased greenhouse gases (Fig. 4B). Although one might expect an increase in low-level cloud cover from the increase in LTS simulated by all models for 2 x CO₂ (45, 46), the resemblance of the spatial structures of the HadGEM1 2 x CO₂ cloud change and SLP change (Fig. 4) to observed decadal cloud and SLP variability suggests that LTS does not play a dominant role. Although we cannot evaluate the exact causes of these cloud changes without additional experiments, the decreased cloud cover in subtropical stratocumulus regions appears to result from warmer SST and a weakening of the large-scale atmospheric circulation in the Pacific in this model.

The question of whether low-level clouds act as a positive or negative feedback to climate change has been an issue for decades. The analysis presented here provides observational evidence that this feedback is positive in the NE Pacific on decadal time scales. The only model in the CMIP3 archive that properly simulates clouds in the NE Pacific and exhibits 2 x CO₂ circulation changes that are consistent with multimodel mean produces a reduction in cloud throughout much of the Pacific in response to greenhouse gas forcing (i.e., a positive feedback). Evaluating cloud feedback with one model is, however, far from ideal. This presents a clear challenge to develop a larger number of climate models that can pass these and other tests so that we may have greater confidence in the sign of the low-cloud feedback under future changes in greenhouse gas concentrations.

Supporting Online Material

www.sciencemag.org/cgi/content/full/325/5939/460/DC1

SOM Text

Figs. S1 to S5

References

References and Notes

·             1. D. L. Hartmann, M. E. Ockert-Bell, M. L. Michelsen, J. Clim. 5, 1281 (1992). [CrossRef]

·             2. S. Bony, J.-L. Dufresne, Geophys. Res. Lett. 32, L20806 (2005). [CrossRef]

·             3. B. J. Soden, I. M. Held, J. Clim. 19, 3354 (2006). [CrossRef]

·             4. M. J. Webb et al., Clim. Dyn. 27, 17 (2006). [CrossRef] [Web of Science]