A new short-wave randomized parameterization for the scalar optical properties of cirrus and its impact on an operational GCM

The most recent 2007 report of the Intergovernmental Panel on Climate Change (IPCC) concluded that one of the most significant remaining uncertainties in climate models is cloud feedback. One cloud that has the potential to vary significantly in terms of the magnitude of its feedback is ice crystal cloud or cirrus. The physical reasons for this are that cirrus can permanently cover 30% of the Earth's surface and in the tropics this coverage can reach up to 70%. Cirrus is composed of non-spherical ice crystals which can vary from simple pristine hexagonal ice columns, plates, bullet-rosettes composed of n-branches, to roughened forms of these particles, which can also combine to form complex ice aggregates. The sizes of these ice crystals can also vary over orders of magnitude as well as their total ice mass or ice water content and in-cloud temperature. With such an array of possible ice crystal size and shape the total or scalar light scattering properties of cirrus such as the mass extinction coefficient, single-scattering albedo and asymmetry parameter may also significantly vary and it is this variability in their light scattering properties that adds to the uncertainty of the cloud feedback recently highlighted in the IPCC report. The question is how to best represent this variability in the scalar optical properties of cirrus in operational general circulations models (GCMs)?

In this paper an ensemble model of cirrus ice crystals is used to predict the randomized scalar light scattering properties of cirrus in the short-wave and these scalar optical properties are parameterized as a function of ice mixing ratio and in-cloud temperature. The ensemble consists of six ice crystal geometrical models which vary as a function of ice crystal maximum dimension, the first of which is the hexagonal ice column of aspect ratio unity representing the smallest ice crystals in the particle size distribution function (PSD). The second element is the six-branched bullet rosette, the third is a three-branched hexagonal ice aggregate and thereafter hexagonal elements are arbitrarily attached until a ten-branched hexagonal ice aggregate is constructed, representing the largest ice crystals in the PSD. In the short-wave the technique of Monte Carlo ray tracing is applied to each member of the ensemble to predict the scalar optical properties, to mimic the impact of surface roughness and/or ice crystal distortions the internal ray path directions are randomized as well as including the largest ice crystals with randomly distributed spherical air inclusions. The randomizations applied vary between 0 (i.e., the scalar optical properties remain pristine meaning that no internal ray path randomizations have been applied) to fully randomized (i.e., the ray paths have been fully randomized including spherical air inclusions). The resulting pristine and randomized scalar optical properties are integrated over 2740 tropical and midlatitude PSDs to obtain the randomized short-wave bulk scalar optical properties of cirrus.

The 2740 midlatitude and tropical PSDs are generated from a parameterization of the PSD that relates the shape of the PSD to the ice water content (IWC) and in-cloud temperature (Tc). The in situ estimates of IWC and measurements of Tc are obtained from a number of field campaigns that cover the tropics and midlatitudes. The bulk scalar optical properties are then related to the IWC and Tc, the IWC being an important prognostic variable of GCMs. The same parameterization of the cirrus PSD used for the optical properties is also used in the cloud physics scheme of the GCM, therefore in this paper we will demonstrate that physical consistency between the cloud physics and radiation schemes has been achieved for the first time in an operational configuration of a GCM. It is common practise to relate the bulk optical properties in a GCM to the ice crystal effective dimension (De- based on PSDs not related to the cloud physics scheme of the GCM) which is diagnosed through some assumed deterministic polynomial fit to Tc and/or IWC. We demonstrate that such dependence on De is not necessary in GCMs.

It is shown that for simplicity and computational speed the randomized bulk scalar optical properties can be parameterized in terms of the prognostic variable ice mixing ratio (I) and Tc as a linear combination with I and the mass extinction coefficient transformed to log10 space. The accuracy of the parameterization is shown to fit the mass extinction data too generally within ±10%, the single-scattering albedo and asymmetry parameter are fitted to well within ±1%, respectively. The parameterization of the mass extinction coefficient is tested against midlatitude and tropical in situ estimates obtained from a number of field campaigns and it is shown that the parameterization is generally well within current experimental uncertainties (±50%).

The new short-wave randomized parameterization has been implemented into an operational configuration of the Met office HadGEM3 global model. We present results from an atmosphere only 10 year climate run of HadGEM3, and we contrast results between the old parameterization (based on I and De) with the new parameterization, as well as the impact on the GCM of the various optical randomizations. We attempt to validate the top-of-atmosphere cloud solar radiative forcing against ERBE, ISCCP and CERES observations. The general conclusion is that the new parameterization tends to darken the clouds relative to the former parameterization, and this is especially apparent in the tropical warm pool region where maximum differences occur. The physical reasons for these differences are explored. The usefulness of directly linking the prognostic variable I and Tc with the cirrus bulk scalar optical properties that is consistent with the cloud physics scheme in the GCM is manifestly demonstrated through helping to diagnose potential GCM error. Such a diagnosis will ultimately help in reducing GCM uncertainty in predicting the cloud feedback.