UC San Diego

The fundamental goals of this study are to 1) quantify the link between aerosols, low-level clouds, and meteorology, and 2) evaluate model representation of aerosol-cloud interactions. Recent in-situ and remote sensing studies indicate that meteorological effects which influence cloud liquid water path dominate the aerosol signal in stratocumulus clouds. To address this issue, Chapter II undertakes a synoptic-scale investigation of aerosol-cloud interactions. Using parcel back-trajectories, we develop a method to isolate meteorological from aerosol impacts on clouds, and evaluate results over the Northeast Atlantic stratocumulus regime. Using MODIS observations and ECMWF analyses, we show that controlling for variations in lower tropospheric stability reduces the dependence of cloud fraction on AOD by at least 24%. We conclude that meteorological forcing must be accounted for in assessing aerosol impacts on cloud forcing, and that doing so requires a Lagrangian analysis of parcel histories. Chapter III extends the analysis in Chapter II by performing an in-depth analysis of the meteorological sensitivities of Northeast Atlantic stratocumulus clouds. Additional satellite observations are obtained from CERES, SSM/I and QuikScat. Compositing is used to quantify the sensitivity of cloud fraction to variations in meteorological state along 72 hour Lagrangian back trajectories. Clouds are found to respond to variations in stability, free tropospheric humidity, and sea surface temperature (SST) advection over long time scales while being influenced by changes in surface divergence over much shorter time scales. Cloud sensitivity to both divergence and stability is shown to be robust and not significantly affected through covariance with other forcings. An additional finding is that the sign of the relationship between cloud fraction and several quantities, including divergence, temperature advection, and surface fluxes, is reversed for long time lags. Along with the differences in stability, SST, and boundary layer humidity that are maintained throughout the trajectories, it is suggested that these point to decoupling of the cloud and sub-cloud layers as a possible cause for cloud dissipation. In contrast, the large cloud fraction composite appears to be more shallow and well-mixed at earlier times in the trajectory, thus maintaining a strong coupling with the surface. In Chapter IV the observational sensitivities observed in Chapters II and III are compared with model representations of aerosol-cloud interactions. Since model parameterizations of aerosol-cloud interactions can be switched on or off, climate simulations can be used to separately quantify the impacts of aerosols and meteorology on cloud cover. Both the ECMWF and GFDL models are analyzed using the trajectory method developed in previous chapters. Both are consistent with Chapter \ref}chapter2} in showing that a significant fraction of the correlation between AOD and CF results from meteorological covariations. However, the two differ significantly in the magnitude of the correction and in their representation of low-level clouds. The ECMWF model, which shows a 43% correction to the AOD-CF sensitivity, also shows a weak correlation (R²=14.4%) when MODIS cloud cover is compared with ECMWF cloud cover predictions. Alternatively, if the dynamical sensitivities of MODIS clouds are compared to those of the ECMWF clouds, the two compare reasonably well. In contrast, the GFDL model shows a 100\% correction, indicating that within the confidence limits of our analysis, low-level cloud cover is not affected by changes in aerosol. However, the GFDL model's treatment of clouds does not compare well with the observational cloud sensitivities identified in Chapter III. Since the model results are only relevant to the observational analysis if the simulations accurately represent cloud cover, from these results it is not possible to quantitatively conclude on a corrected sensitivity of cloud cover to changes in aerosol optical depth. However, the results do qualitatively confirm the results of Chapter II. In addition, the Lagrangian technique developed in previous chapters proves useful in providing detailed diagnostic information on model performance, in particular as a means of testing cloud parameterizations