Observed and modeled relationships among surface temperature, cloud properties, and longwave radiation over the Arctic Ocean

Arctic surface temperature is evaluated using satellite retrievals from the TIROS Operational Vertical Sounder (TOVS) and surface-based observational products. In general, winter retrievals are consistent with surface-based measurements. In summer, there is a cold bias of approximately 2K in the TOVS data over sea ice owing to uncertainties in detecting stratus clouds. TOVS satellite retrievals and surface observations from the Surface Heat Budget of the Arctic Ocean (SHEBA) field campaign are then used to evaluate the performance of a global climate model (GCM). In addition to the traditional approach of validating individual model variables with observed fields, the model's relationships and sensitivities among interrelated fields are examined by a linear regression method. This provides additional information on how well the model represents feedbacks. Three climate variables are considered: surface temperature (Ts), total cloud cover (CLD), and downward longwave flux (DLF). The GCM provides a reasonable representation of both the annual cycles and the variability of these climate variables. There is also good agreement between the modeled and observed relationships between pairs of climate variables. A neural network (NN) approach is applied to investigate and quantify non-linear relationships between longwave cloud forcing (CFL) and cloud properties. A distinct bimodal distribution of sensitivities characterizes the relationships between pairs of variables be tween CFL and other cloud properties. Although the mean states of the relationships agree well with a previous study, the mean states often do not exist. For example, the mean sensitivity of CFL to cloud cover is 0.68Wm -2%-1, but in reality it is dominated by a low sensitivity of 0.15Wm-2% -1 for clear-sky conditions, and a high one of 0.85W m-2%-1 for cloudy conditions. The sensitivity of CFL to liquid water path and to cloud-base height decreases as these two variables increase. The sensitivity to cloud fraction increases as cloud cover increases. The sensitivity to cloud-base temperature is low for very cold or very warm clouds. The neural network approach captures the sensitivities of longwave cloud forcing to cloud properties, and provides a new way to evaluate quantitatively the relationships in climate feedbacks and to identify different modes of variability.