CCSM AMWG Intercomparison: Notes on the Simulated Arctic Climate

11 December, 2000

• 1. Introduction
• 2. The Key Arctic Quantities
  • 2.A. Climatological Seasonal Mean Sea-Level Pressure
    • 2.A.i. Winter
    • 2.A.ii. Summer
  • 2.B. Climatological Monthly Mean Cloudiness
  • 2.C. Climatological Monthly Mean Radiation Budgets at TOA and SFC
• 3. Summary

The CCSM Atmosphere Model Working Group (AMWG) is conducting an intercomparison study of AMIP-style simulations performed with the CCM version 3.10, using several convection schemes, several dynamical cores, and two formulations for the emission spectrum of water vapor. The CCSM Polar Climate Working Group is very interested in the performance of the CCSM atmosphere model, because it has a large impact on polar feedbacks and on the coupling among the ocean, sea ice and atmosphere. With this in mind, we inspected plots of climatological seasonal and monthly mean quantities simulated by the suite of models under consideration. The simulated quantities were compared with observationally-based analyses listed in Table 1.

TABLE 1: List of Observational Analyses in the Comparison

Variable	Source
Sea Level Pressure	NCEP Reanalysis Data, 1979-1996
TOA Radiation	ERBE data, 1985-1989 (Barkstrom and Smith, 1986)
Cloudiness	Climatology of Surface Obs, Hahn, et al., 1987
Surface Radiation	Synthesis of published data, Curry and Ebert, 1992

The suite of models is listed in Table 2:

TABLE 2: List of Models in the Comparison

Identifier	Model
ccm3527	CCM version 3.6, which was used in CSM1
ccm3911	CCM version 3.10, PCW, 30 levels, New overlap
lin-rood/e0220	CCM 3.10 with the Lin-Rood dynamical core
H20abs01	CCM 3.10 with modified H20 absorption band for LW radiation
sld031	CCM 3.10 with semi-Lagrangian dynamical core
e0224	CCM 3.10 with the Lin-Rood dynamical core and increased horizontal diffusion & longer Rayleigh Friction timescale
r1up	CCM 3.10 standard model using a 1-digit reduced grid
ras03	CCM 3.10 physics with relaxed Arakawa-Schubert convection
trig6	CCM 3.10 using Zhang-McFarlane convection with physical triggers
vdt1	CCM 3.10 physics with vertical diffusion of dry static energy
zhang	CCM 3.10 using Zhang-McFarlane convection with modified closure

To prioritize effort given limited time, we have focused much of our analysis on the ccm3911 version, which represents the "standard" model, and the lin-rood version, because it shows the greatest improvement in the simulation of arctic SLP. Comparisons between models and the observationally-based analyses provide a basis for judging model performance. Comparisons between ccm3527 and any other model provide a basis for judging progress since CSM1 was released. Comparisons between ccm3911 and any of the models with modified dynamical core, convection scheme and water absorption band provide a basis for distinguishing the effects of each such modification on the basic (CCM 3.10) model.

Based on the analysis of CCM3.6 (Briegleb and Bromwich, 1998a) and our own analyses of CCM output fields, we have selected three aspects of the Arctic climate that merit attention: Surface geostrophic winds (which derive from the sea-level pressure (SLP) field), because of their impact on the motion and mass balance of sea ice; Surface and TOA radiation, because of their impact on atmospheric temperature, ice mass balance, and climate feedbacks; and cloud cover, because of its impact on the radiation. The differences between the simulated and analyzed features stand out even when the quantities are averaged over a month or a season, and over a multi-year climatology (i.e. over the years of the AMIP integrations). In the case of cloudiness and radiation, the differences are pronounced even after averaging around latitude circles. In the case of SLP, it is important to resolve spatial variations in the climatological average field.

It is now well known that the fields of arctic sea-level pressure simulated by most GCM's have the following deficiencies.

• (i) In winter the ridge of high pressure extending across the Arctic Ocean from North America to Siberia is displaced poleward from its observed position, resulting in stronger than observed onshore winds in the East Siberian Sea, and absence of onshore winds at the Canadian Archipelago coast. A consequence of this pattern is that sea ice models driven by such GCM winds produce thickest ice in the East Siberian Sea, and insufficiently thick ice on the North American side. This is opposite to the observed spatial pattern of ice thickness (Bitz, et al., submitted). These authors show that the ice thickness pattern depends critically on getting the climatological mean annual cycle of SLP right.
• (ii) In summer, the observations show a climatological low pressure center on the north pole, flanked by relatively small-scale high pressure centers along the 45 E and 135 W meridians, respectively. By contrast, most of the GCM's simulate an extensive high pressure cell centered on the North Pole.
• (iii) GCM-simulated surface geostrophic winds tend to have too large an onshore component along the East coast of Greenland, where the sea ice exits the arctic into the North Atlantic.

The CCM3.10 SLP and winds are typical of many other GCM's with regard to the features of arctic climate mentioned above. In this respect CCM3.10 is not much different than CCM3.6. These features are illustrated by Figures 1 and 2.

Figure 1 NCEP (a) Winter and (b) Summer SLP: Colored lines on winter map indicate axes of ridge (dashed lines) and trof (solid lines) from NCEP, CCM3.10, Lin-Rood and SLD. Contour interval is 2mb for winter and 1mb for summer. (Click on thumbnail to expand.)

Figure 2 CCM3.10 fields of SLP: (a) Winter and (b) Summer. Colored arrows on winter map indicate the NCEP (red) and CCM (green) surface geostrophic wind vectors at four key points.

The simulations of arctic SLP by CCM3.6 (not shown) and the standard model CCM3.10 are generally similar, with perhaps a slight degradation for CCM3.10. Inspection of the variants of CCM3.10 indicates that the changes in convection schemes and dynamical cores are all influential (subject to verifying statistical significance of the differences relative to base CCM3.10), and the arctic surface circulation is improved in all cases. In one case, the Lin-Rood dynamical core, the improvement is substantial, and produces much better agreement with the NCEP winter SLP field in the central Arctic (Figure 1 and Figure 3). Based on a rough estimate for the statistical significance of the distribution of difference between two means, we find the winds at the four points in Figures 3a differ at the 95% confidence level.

Figure 3 Lin-Rood fields of SLP: (a) Winter and (b) Summer. Colored arrows on winter map indicate the NCEP (red) and Lin-Rood (blue) surface geostrophic wind vectors at four key points.

Following the methods of Bitz, et al. (submitted), we plan to use the mean annual cycle of SLP simulated by the Lin-Rood version to drive a sea ice model. The results will indicate whether and by how much this new SLP field improves the simulated field of sea ice thickness in the Arctic. The Lin-Rood version appears to improve slightly the direction of the simulated geostrophic wind along East Greenland, though the magnitude of the simulated wind is a bit larger than observed here.

2.A.ii. Summer

None of the GCM's simulate the summertime low pressure cell centered on the North Pole in the NCEP analysis. Instead, the models exhibit a dominant high pressure cell over the Arctic Ocean. The one exception to this latter statement is the Lin-Rood version (Figure 3). Here the high pressure is reduced in spatial extent and shifted off the pole, towards Fram Strait. A small low pressure cell appears over western Alaska, but the Lin-Rood summer simulation of SLP still does not resemble the NCEP climatology. All of the GCM's, including Lin-Rood simulate a stronger than observed northward extension of the Pacific high pressure, a spurious low pressure cell over Scandinavia, and stronger than observed zonal geostrophic flow south and east of Iceland into Europe. On the good side, all models exhibit an Icelandic low and an East Asian low, more or less as seen in the NCEP analysis.

The features of the mean seasonal SLP suggest that perhaps synoptic low pressure systems are better able to penetrate the Arctic in the Lin-Rood version of the model. It would be very interesting to see statistics of the synoptic variability of SLP simulated by this version.

Most GCM's tend to simulate too much cloud cover over the Arctic Ocean during winter. Observations show about 50-60% climatological mean total cloudiness and 10-30% low cloudiness in winter, compared to 80-90% and 70-90%, simulated e.g. by CCM3.6 at the respective levels (Briegleb and Bromwich, 1998a). In summer, there is more variation among the cloud climatologies simulated by different GCM's. Observed summer total cloudiness is in the 70-80% range. A priori we don't expect much improvement in the simulation of winter arctic cloud, because the primary change from CCM3.6 to CCM3.10 is the prognostic cloud water (PCW) scheme introduced by Rasch and Kristjanssen (1998). These authors found little difference in the climatological arctic cloudiness simulated with and without PCW.

Figure 4 compares observed climatological January mean low (top panel) and total (bottom panel) cloudiness, averaged around latitude circles, with values simulated by the models. In the central Arctic, all the models perform similarly, greatly oversimulating both total and low cloudiness. The Lin-Rood simulation stands out because total cloudiness is oversimulated by an additional 5% above the other models, between 65 and 85 north latitude. Apparently the different dynamical cores and convection schemes do nothing to remedy the oversimulation of winter cloud in the arctic. Note that the additional cloudiness simulated by the Lin-Rood version is consistent with the conjecture that cyclonic storms are more influential in the Arctic under this dynamical core.

Figure 5 shows the corresponding statistics of cloudiness in July. The simulated total cloudiness in all the models decreases about 10% from winter to summer, and the observed cloudiness increases greatly, bringing the two into reasonable agreement. The clouds simulated by the Lin-Rood scheme do not stand out from those simulated by the others models at this season.

Previous studies (e.g. Briegleb and Bromwich, 1998a) have shown that CCM3.6 oversimulates the downward longwave radiation at the surface of the Arctic during winter, and undersimulates the downward shortwave radiation at the surface during summer. At TOA, CCM3.6 undersimulates the net shortwave radiation as well as the outgoing longwave radiation throughout the year. In clear sky situations, CCM3.6 simulates surface incoming shortwave irradiance well, but undersimulates the downward surface longwave radiation. Briegleb and Bromwich ascribe the oversimulation of downward LW during winter to the oversimulation of winter cloud cover. The summer undersimulation of downward SW is attributed to an oversimulation of liquid water content in the clouds.

Figure 6 shows the simulated and observed longwave (LW) downward surface irradiances at 80 degrees north latitude. The surface incident LW, simulated by all of the models tends to be larger than observed, in spring, autumn and winter. The different physics embodied in the various model versions are influential on this flux, with model-to-model differences ranging up to 20 W/m2. Model minus observed differences of 10-20 W/m2 are typical. Interestingly the Lin-Rood version simulates the lowest incident longwave during winter, and in this season it nearly matches the observations. This is surprising, because the Lin-Rood version simulates more total cloud and more low cloud than all the other models, which would tend to increase surface LW if temperature and humidity profiles remained constant. Indeed Briegleb and Bromwich (1998a) attribute the CCM3.6 oversimulation of surface incident longwave in winter to the oversimulation of cloudiness.

To investigate this further, we compare in Figure 7 the vertical profiles of temperature (not yet available) and specific humidity simulated by ccm3911, e0220 (aka Lin-Rood) and the observed profiles at two arctic RAWINSONDE stations. All data are climatological average profiles for January. We see that the Lin-Rood profile is significantly colder and drier than both the observations and ccm3911. To quantify the implied difference in surface flux, we will perform simulations with the CRM using these T,q profiles, but these simulations may not be complete in time for the AMWG meeting.

Briegleb and Bromwich also note that under clear skies, the CCM3.6 undersimulates incident surface LW. Moritz and Rivers (in prep.) have performed instantaneous radiative flux (IRF) experiments using a column radiation model (CRM) version of CCM3.6, with high quality input and verification data from the SHEBA experiment. For clear skies the CRM undersimulates incident LW radiation when it is given accurate profiles of humidity, temperature and cloud. NCAR investigators are aware of this bias, and think the culprit may be the representation of the water vapor band emissivity in the longwave radiative transfer model. They have modified this emissivity for version H20abs1. As expected, this model has the largest downward LW throughout the year (Figure 6). Of course correcting this problem with the water vapor absorption band adds to the already positive bias in the total downward LW. Further analysis of this bias requires investigation of the simulated humidity, temperature and cloud properties.

The downward shortwave at the surface (Figure 8) is undersimulated throughout the sunlit season by all models. None of the model variations makes a significant dent in this bias. To get further insight into possible causes, in Figure 9 we show the TOA radiation budgets. The lower panel shows net SW at TOA, and indicates that during summer, all models are biased low by some 10-15 W/m2 over the Arctic Ocean. From this and the surface results, it follows that the simulated atmospheres are reflecting a bit too much SW radiation back to space, and they are absorbing far too much SW radiation. CRM simulations performed by Moritz and Rivers (in prep.) indicate that with accurate profiles of humidity and temperature, the CCM3.6 radiative transfer code does a good job in clear sky cases. So it would seem that the main problem is too much absorption of shortwave radiation in the cloudy arctic atmosphere. This is interesting in the case of Lin-Rood, which exhibits a fairly dry atmosphere in the Arctic. What is absorbing all this shortwave radiation, and how is it related to the simulated clouds?

The outgoing longwave radiation at TOA (the OLR) is biased low in all models, by just a few W/m2 in summer. In winter some of the models simulate too much OLR and some too little. Notably, the Lin-Rood simulates too little, which indicates colder cloud tops emitting the radiation to space, consistent with the lower simulated tropospheric temperatures shown on Figure 7. It is also noted that the Lin-Rood and e0224 versions simulate more high cloud amount than any of the other models at 80 north, but only by a few percent more than ccm3911. Examination of profiles at Resolute and Thule indicates that in January, e0224 simulates a troposphere about 2 degrees colder than the observations, and than the base CCM3.10. Thus radiation emanating from clouds at a given level will be reduced by about 8 W/m2 in this model. This by itself would presumably go a long way to explaining why surface and TOA LW fluxes are smaller in Lin-Rood than in the other models.

The Lin-Rood version of CCM3.10 provides room for optimism that the arctic SLP can be simulated well enough to drive the CCSM sea ice model to a fairly realistic solution. But none of the models makes much progress on the outstanding problem of simulating winter cloudiness in the arctic. The Lin Rood atmosphere does improve the wintertime longwave radiation at the surface, and this is consistent with its colder and drier troposphere. The problem here is that it is also colder and drier than the observed troposphere. Thus if wintertime cloud cover simulations are improved in Lin-Rood, the surface downward LW would presumably be lower than observed, instead of too high as it is in CCM3.10. It will probably take additional ice-phase microphysics, linked to precipitation processes, to address the problem of winter cloudiness, and this may be the key to getting the LW radiation right. Likewise, none of the models takes much of a bite out of the large undersimulation of incident SW radiation at the surface. Simple budget analysis indicates it may be fruitful to look into how absorption is happening under cloudy skies.

Further analysis is also needed to determine how the surface circulation is affected by Lin-Rood: Is this a horizontal resolution effect? Does the change in surface circulation really represent the effect of more cyclonic storm presence in the central Arctic? And if so, wouldn't that imply a greater poleward heat transport across, e.g. 70 degrees north, and how does that square with the colder, drier arctic troposphere simulated by Lin-Rood?

Bruce Briegleb provided data sets for the observational analyses of cloudiness, surface radiation, and TOA radiation. This material is based upon work supported by the National Science Foundation under Grant No. 0004261. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.

Barkstrom, B.R. and G. L. Smith, The earth radiation budget experiment: Science and implementation. Rev. Geophys., 24, 379-390, 1986.

Briegleb, B. and D. Bromwich, Polar Radiation Budgets of the NCAR CCM3. J. Climate, 11 (6), 1246-1269, 1998.

Curry, J.A. and E. E. Ebert, Annual cycle of radiation fluxes over the Arctic Ocean: Sensitivity to cloud optical properties. J. Climate, 5, 1267-1280, 1992.

Hahn, C.J., S. G. Warren, J. London, R.L. Jenne and R. M. Chervin, Climatological data for clouds over the globe from surface observations. Rep. NDP-026, 57 pp. , 1987. [Available from Carbon Dioxide Information Center, Oak Ridge, TN, 37831-6050].

Rasch, P. and Kristjanssen, 1998: A comparison of the CCM3 model climate using diagnosed and predicted condensate parameterizations, J. Climate, 11, 1587-1614.