Radiative impacts of cloud heterogeneity and overlap in an 2 atmospheric General Circulation Model

II The radiative impacts of introducing horizontal het:erc'geneity vertical overlap of condensate and cloud fraction a package operating in the GEOS-5 Atmospheri examined in terms of diagnostic top-of-t1I,e"1~ibaii!'l cloud radiative effect (CRE) calc specifications about ~_.:.;;;;;.;:, ificantly stronger for the operational cloud scheme for erlap from maximum-random to generalized results to of -4 Wm", and zonal changes of up to -10 Wm". This ·s compared to the other scheme of large layer cloud fractions and of m!llti-Iayer situations with large numbers of atmospheric being simultaneously cloudy, 25 conditions that make overlap details more important. The impact on CRE of the details of condensate distribution overlap is much weaker. Ollce generalized overlap is adopted, both cloud schemes are only modestly sensitive to the exact values of the overlap parameters. We also tind that if one of the CRE components is overestimated and the other underestimated, hoth cannot be driven towards observed values by adjustments to cloud condensate heterogeneity and overlap alone.


3
With new computationally efficient approaches to treat cloud-radiation interactions now 4 available, there are fewer reasons to retain the simplistic cloud descriptions that have persisted 5 in General Circulation Models (GCMs) for many years. Clouds do no longer have to be 6 treated by the radiation schemes of these models as homogeneous slabs within large areas 7 0(10 4 lan 2 ), with fractional coverages and optical depths that have been greatly adjusted to 8 compensate for known biases arising from their nonline~lSJiqtjp!ct"\i"~!h radiation. While 9 capturing the radiative effects of full-blown 3D cloud . be elusive, the 10 representation of in-cloud horizontal heterogeneity 11 statistics of vertical correlations of condensate and cloud 12 radiative transfer framework is now feasible. All a matter of fact, 13 study that amply demonstrates the viability of SU('~Gl.iIliil:!!i.. r1:alCJing.

30
In the following we will present the tools, assumptions, and experimental setup that allow us 3! to examine the degree to which cloud complexity changes the cloud radiative impact (sections 2, 3, and 4). The availability of two cloud schemes in the GCM at hand and our analysis 2 approach provides the opportunity to investigate whether the same assumptions about cloud 3 complexity imposed on ditTerent initial cloud fields can yield notably distinct radiative 4 impacts (section 5) and what causes the contrasting behaviour (section 6). 5 6 2 Implementation of RRTMG into GEOS-5 7 The effects of cloud overlap (fraction and condensate) on the radiative fluxes can be captured 1 point k is paired randomly with one of the N subcolumns, m. Note that when using eq. (1) the 2 computational cost of the calculation over all subcolumns is the same as that of a full spectral 3 integration of a single (sub)column. The performance of this approximation in large scale 4 models has been tested extensively (e.g., Barker et al. 2008). The main issue of concern is 5 whether the conditional random noise, decreasing as the inverse square root of the number of  a cloud generator  the above simple overlap assumptions are inconsistent with cloud fields from observations   4 and cloud resolving models, and that the concept of "generalized" cloud fraction overlap 5 represents observed overlap more realistically. In the generalized overlap paradigm, the 6 combined cloud fraction of two cloudy layers at heights z\ and Z2 with separation distance I1z 7 = Z2-Z\ can be approximated as a weighted average of combined cloud fractions from 8 maximum and random overlap, C m ",(I1z) and C'Qn(I1z), respectively according to: The weighting parameter a(l1z) , is a mea . Norris 2011). Because the fit provided by eq. (4) is usually used in conjunction with eq. (2), generalized overlap has also been termed "exponential-random" overlap (Hogan and The manner in which cloud water contents align in the vertical may also important for 2 processes like radiation (or precipitation To create tbe subcolurnns tbat describe tbe cloud fields within tbe GCM gridcolurnns, two 3 additional pieces of information, besides tbe profiles of cloud fraction C and mean condensate 4 (liquid and ice) are needed, namely to specify tbe decorrelation lengtbs L a and L, and tbe 5 magnitude oftbe horizontal variability of the condensate distributions. We defer discussion of 6 decorrelation lengtbs for tbe next section, and describe variability here. : (7) ,e times (8a) (8b) 16 set as follows, loosely based on Oreopoulos 17 of hydrometeor variability in tbe CloudSat (Stephens airborne and satelllite measurements, such as gamma and lognormal would have been an 23 equally acceptable alternate choice. Eqs. (7) and (9) apply to botb liquid and ice condensate, 24 and in layers where the two phases coexist tbeir ratio is assumed to remain constant across all 25 subcolumns. Since no distinction is made between liquid and ice cloud fraction, the I normalized standard deviation a w /w is de facto the same for liquid and ice condensate 2 distributions. The beta· distribution of normalized condensate x is converted to an actual 3 condensate distribution and then to a cloud optical depth distribution using the AGCM-4 provided effective particle size which is different for each phase, but assumed horizontally 5 homogeneous. 6 Since the specification of the amount of condensate variability via a w does not come explicitly 7 from the host AGCM or derived from rigorous physical principles, and variability is used only 8 to gauge diagnostically the sensitivity of the cloud radiative effe~t, we argue that it is not 9 essential to fully justifY its exact specification. Differe ,

27
We performed such a cloud overlap analysis using CloudSat products for two months, January Our objective for AGCM parameterization purposes is to capture the observed decorrelation 2 length zonal structure shown in Fig. 1. For that purpose, we apply a Gaussian fit (black 3 dashed curves) of the form

12
Our suite of experiments is summarized in

25
We first focus on the sensitivity of globally-averaged eRE to different assumptions about 26 how cloud fields can be generated from profiles of cloud fraction and mean condensate. Fig. 2 and Fig. 3   The CRE response to condensate heterogeneity and generalized overlap when imposed on the 6 cloud fields of an alternate cloud scheme can be substantially different than the one discussed 7 above. This is shown in Fig. 3, which is the same as Fig. 2