Minutes of the Polar Climate Working Group
22-23 January 2003
NCAR, Boulder, CO
Wednesday, 22 January 2003
1. The meeting was convened by co-chairs Dick Moritz and Elizabeth Hunke. Twenty-four people attended all or part of the meeting.
2. Dick Moritz presented a brief list of topics and tasks that the PCWG had agreed to address after the 2002 CCSM workshop, including:
(i) Experiments to be performed with the model in various configurations;
(ii) Development of a test suite for the sea ice model;
(iii) Release of additional versions of CCSM with improved coupler and other modifications;
(iv) Evaluation of the effects of modifications to the ice ridging formulation on CSIM simulations.
3. Cecilia Bitz presented highlights of SSC discussions that are relevant to PCWG. The SSC proposed that CCSM2 be modified to address high priority problems including the "double ITCZ" and the negative temperature bias in the tropical tropopause. The other model working groups are developing the following potential changes at this time:
ATM - new cloud parameterization (Phil Rasch); new PBL formulation (Chris Bretherton); and further evaluation of the finite volume (FV) dynamical core. Cecilia noted that the bias in atmospheric circulation over the arctic remains a problem important to the PCWG, but it is not currently a high priority for the SSC.
OCEAN - The KPP horizontal viscosity scheme is being adjusted; the parameters determining absorption of shortwave radiation in the upper ocean will be extended to vary with time, latitude, and longitude. There is currently a warming trend in the CCSM2 ocean model, but the cause is unknown.
LAND - The current scheme for assigning grid cell ground albedo depends on the snow-covered fraction, and this may be unrealistic, contributing some 3 -5 degrees C to the warm bias in polar regions.
A new version of CAM (CCSM Atmosphere Model) has a slab ocean.
The CCSM Plan has been outlined and assignments made for writing the sections. The outline will be discussed with the CCSM Advisory Board next week. CCSM is also developing a business plan, which deals with the NCAR portion of CCSM.
The SSC had extensive discussions concerning the Earth Simulator (ES), a vector supercomputer system being developed in Japan. At NCAR, Maurice Blackmon and Frank Bryan are involved in a project to use POP on this vector machine to perform high-resolution eddy resolving ocean simulations. The project also proposes IPCC simulations to be performed by CRIEPI using a vectorized version of CCSM. The discussion within the SSC was spirited, with Maurice advocating collaboration with CRIEPI and with other SSC members asking how this will benefit CCSM. The main concern appears to be that CCSM resources (people's time, expertise) are needed to code and test the vectorized CCSM, but the broad CCSM community would not have vector computing resources available. Also, diversion of resources would slow the pressing tasks associated with getting CCSM ready for the IPCC runs to be performed on NCAR computing machines, while facilitating IPCC runs to be performed by CRIEPI. Of particular interest are efforts by Fujitsu America and LANL to vectorize the LANL sea ice model (CICE). There is general interest in PCWG to consider whether and how to create a single CSIM code that runs efficiently on both vector and parallel machines. This is discussed further in later sections.
The SSC is planning for the 2003 CCSM Workshop, including some changes to the format. This year the PCWG meeting will be run simultaneously with the Paleo WG, so that participants can attend both the Polar and the Climate Change WG meetings.
Phil Merilees added that the SSC and NSF-ATM (Jay Fein) have available $25K to support outreach, especially visits by active CCSM scientists to universities. Such visits would be designed to inform the broader community of CCSM models, research and opportunities, and to drum up interest in the CCSM.
The SSC discussed the Climate Change plan of the current federal administration. It was noted that the CCSM plan would now include a section on "Applied Climate Modeling" that deals with IPCC and similar experiments.
The SSC is promoting collaboration between CCSM and GFDL, which is already active. Mike Winton is trying to identify and hire an appropriate postdoc to implement the CSIM in the new GFDL climate modeling framework. The (old Manabe/Bryan) GFDL model has a large global delta-T under greenhouse forcing, but rather small polar amplification (even though the magnitude of delta-T in the polar regions is quite large). Alex Hall suggests this grassroots effort be expanded to include others, e.g., Tom Delworth.
4. Dick Moritz presented results from simulations of the annual evolution of ocean, atmosphere, and ice variables at the SHEBA experiment site, as simulated by a single column model version of CCSM. Variations of most simulated variables on scales of a few days to one year are encouragingly similar, in a qualitative sense, to observations. The most obvious exception to this is the cloud fraction, which does not capture the observed winter minimum. Quantitatively, there are significant discrepancies in the incoming shortwave and longwave radiation, the onset and duration of the melt season, surface albedo, and the air temperature profile in summer. Sensitivity experiments with a 10% increase in horizontal temperature advection show that perturbations to the surface albedo account for most of the large increase in sea ice ablation.
5. Cecilia Bitz presented results from simulations with the CCM, augmented with a more realistic (but still prescribed) sea ice surface boundary condition, and with horizontal resolutions T42, T85, and T170. The well-known bias in the T42 surface atmospheric circulation over the Arctic Ocean is reduced in the T85 simulation, but remains sufficiently large to adversely affect ice motion and the pattern of mean ice thickness. Additional experiments indicate that the improvement at T85 is due primarily to the effect of more highly resolved orography interacting with the higher resolution atmospheric dynamics.
6. Alex Hall posed the questions "How do we characterize unusual variability, and is the variability of climate in the polar regions unusual?" Criteria include the magnitude of variability, the redness of the spectrum, spectral peaks, and skewed pdf's. He presented results of analysis of GFDL R15 output (15000 years) and CCSM2 output (600 years).
7. Phil Rasch presented an update on his studies of cloud physics in CAM, conducted in collaboration with Byron Boville, and focusing especially on cloud ice processes and effects. Problems that may be affected by the cloud physics include the cold tropopause in the tropics, the warm polar climate, the warm summers in the extratropics, energy conservation problems associated with phase changes that involve the latent heat of fusion in the atmosphere, and temporal discontinuities in upper tropospheric cloud occurrence that depend on static stability which affects the RH threshold for cloud occurrence.
Phil has changed the way ice crystal size depends on T,P using the formulation of Mitchell and Kristjansson (this MK formula came from studies of tropic/mid-latitude cirrus). Phil also implemented sedimentation velocities for ice and liquid particles. The crystal size controls the sedimentation velocities. This differs from the Heymsfield approach in which velocity is a function of the mixing ratio of ice. Phil also implemented a different temperature dependence of ice/water mixing ratio that is consistent with both the cloud microphysics and the radiation schemes. Thus, the only remaining "combined" physics (that does not distinguish processes and physics of ice and liquid condensate) is the condensation term. Phil thinks we can now remove from the model some parameterizations and constraints in the older models that compensated for the absence of feedbacks that will now operate more naturally.
The new model has a single, 90% RH threshold at all x,y,z. Phil notes a paper by Boudal, et al. in the International Journal of Climatology, that reports on the size and shape distribution of ice particles in the polar regions. Byron and Phil plan to submit their revised model to the SSC to be included in the next version of the CAM. They are hoping to write two papers.
Preliminary analysis of simulation results with the new physics indicates it is influential on cloud optical depth and radiation, but perhaps not especially influential on cloud fraction.
8. Keith Hines presented a comparison of clouds and radiation simulated over Antarctica, using a variety of versions of CCM, and also observations. (This was a continuation of work of Briegleb and Bromwich, 1998a,b.) These were AMIP and climatological SST experiments. Model variations included the RRTM radiation code and the old diagnostic and RK prognostic cloud water schemes.
(i) Interior Antarctica (south of 75S) is a region of largest sensitivity to clouds and radiation within NCAR climate models.
(ii) Introduction of prognostic cloud water is found to have a much larger impact than the introduction of RRTM.
(iii) CCM3 and CAM2 have large cold biases during summer in the middle and upper troposphere. The problem is largest near the tropopause.
(iv) The radiative effects of Antarctic clouds appear to be excessive.
(v) Early version of CAM2/CCSM2 does not appear to be improvements over CCM3.
(vi) To improve surface energy balance, the very shallow, stable PBL needs to be better treated in addition to improvements in clouds and radiation.
Jeff Kiehl notes that CAM2 has a problem in that PBL height (which is diagnosed) that can be lower than the lowest model level. When this happens mixing is unrealistically inhibited, and this probably has effects on cloud processes. To remedy this, the AMWG has implemented an interim fix in CAM2.
9. Bruce Briegleb presented a summary of the polar biases seen in the CCSM2 sea ice simulations. He analyzed years 901-1000 of the long control run. Also, he has done a fitting procedure to remove bias from the CCSM2 output fields of radiation and winds, so that he can force the AIO model with CCSM2 output that does not contain these biases. The biases were removed with respect to ECMWF estimates of these quantities. Reductions in these biases separately improved the simulation, but the best simulation resulted from both acting together.
10. Marika Holland presented analyses of the variability of Antarctic sea ice in the CCSM2 control run, years 350-900.
The dominant mode of variability in Antarctic sea ice cover exhibits a dipole pattern with anomalies of one sign in the Pacific and of the opposite sign in the Atlantic. This mode of variability has been documented in observations. Here we examine 600 years of a control simulation of the CCSM2 to determine the simulated sea ice area variability and the mechanisms driving this variability. The dominant modes of simulated variability compare very well to the observations both in their spatial distribution and magnitude. These variations in the ice cover have limited intra-basin eastward propagation that appears to be related to the observed Antarctic circumpolar wave.
The mechanisms driving the simulated sea ice variability are examined. In particular, the interplay of dynamic and thermodynamic processes in forcing the ice variability and how these relate to atmosphere and ocean conditions, including those associated with the southern oscillation and the southern annular mode, are investigated. The relationships found are consistent with the atmosphere and ocean forcing of the sea ice variability, with different processes dominating in the different basins. There is also some indication that positive feedbacks associated with the sea ice conditions influence the atmosphere and ocean temperatures in the regions, acting to prolong the life of the anomalies.
11. Cecilia Bitz presented an analysis of the controls on the position of the sea ice edge. She mentioned CCM simulations by John Chiang (University of Washington) of the last glacial maximum in which sea ice is PRESCRIBED (LGM) and SST is computed using a slab model. The results show that the LGM sea ice is associated with up to 12 degrees C decreased Ts in both polar regions. Results were shown for the energy budget of the ocean and the ocean-ice column at ice edge locations. The variations in ice edge position appear to result mainly from variations in (a) oceanic heat transported into high latitudes and (b) wind patterns that blow the ice poleward, equatorward, or east-west.
12. Elizabeth Hunke presented results from ice-ocean simulations with POP and CICE, motivated by the fact that THC tends to diminish or shut down in POP-CICE simulations unless (a) SSS is restored or (b) the whole thing is coupled to the atmosphere, whereas MICOM (an isopycnal ocean model) tends to oversimulate the THC in such experiments.
Comparison of several runs at different resolutions indicate that mixing and convection at high latitudes are sensitive to model configuration. Two runs at approximately 3 degrees resolution (gx3), one with 180-day surface salinity restoring, the other without restoring, both show anomalous mixing in the Pacific sector of the Southern Ocean that leads to deep water formation. Mixing does not occur in a similar non-restoring run at 0.4 degrees resolution, nor does it occur in the CCSM control run at 1 degree (gx1). While surface temperatures are warmer in the gx3 runs than the higher resolution runs, admitting more evaporation and greater surface salinity, the more critical difference seems to lie below the surface, where the gx3 runs are much fresher. Hence, the gx3 runs are unstable (relatively salty over fresh) where the 0.4 is stable (relatively fresh over salty). These differences could be due to the path of the ACC and/or to differences in ice formation/brine rejection near the coast. Comparison with gx1 indicates the ACC path is less likely the culprit than ice/brine formation.
Within the ice edge, simulations at 3 degree resolution are unable to reproduce polynyas correctly, not producing enough brine in some locales while overturning excessively in others. Surface salinity restoring suppresses mixing in the Southern Hemisphere polynyas by freshening the surface (although it does not suppress mixing in the South Pacific, north of the ice edge).
What is the PCWG plan to address polar biases? Should the focus of the PCWG be on the ice thickness, extent, and volume?
Alex Hall suggests things will go better if we define a focal point for our work, analogous to the focus of the AMWG on the double ITCZ. For example, should we focus on the biases in the wind and radiation fields?
Steve Vavrus suggests that we focus on arctic (or polar) problems that have large, demonstrable effects on the global simulation.
Bill Lipscomb suggests we need to go beyond just ice thickness and extent, and need to include ice motion and fluxes in our focus.
Cecilia Bitz notes that there is a sign error in the mean sensible heat flux over sea ice in CCSM2. Is this a boundary layer problem or a radiation/cloud problem, or some of each?
Marika Holland notes that ocean heat transport into the Atlantic Arctic, especially the Barents, is oversimulated by CCSM2, affecting the ice edge and possibly the troublesome trends in the ocean. Is this a broad scale North Atlantic problem or is it more localized to problems near the ice edge and the arctic marginal seas?
Alex Hall suggests devoting more time/effort to analyze existing simulations and performing experiments with the existing model. He is particularly interested in seeing the CCSM2 control run extended beyond 1000 years.
Cecilia Bitz - There is a problem with the version of CAM that has the slab ocean. The concern is that the Qflux in each grid cell, which varies with position and time (mean annual cycle), is prescribed differently in a given grid cell if the cell is covered by open water or by ice. This introduces a feedback, e.g., when ice covers a formerly open ocean point, the Qflux changes that could act to enhance or damp the ice anomaly.
Thursday, 23 January 2003
1. Julie Schramm - Update on CSIM
(i) Added Lipscomb remapping on 12/02. Other modifications include correction to tilt terms in coupler, which were multiplied by an area correction.
(ii) Added new coupler CPL6 on a branch 01/03
(iii) CSIM "Requirements" document has been drafted and is ready for review. The Developer's Guide is in progress. The scientific document needs to be updated for remapping.
(iv) All components of CCSM must be compliant with ESMF by April, 2004. ESMF is a software infrastructure to enhance ease of use, performance, portability, data communication, and coupling. ESMF is in the design stage, further implementation in CSIM is unclear. Tim Killeen is the PI of ESMF. SSC has not discussed this yet, CC will bring it up with them at the next meeting.
(v) Phil Jones is funded by ESMF to make POP compliant.
(vi) The PCWG computer allocation for January is 2742 GAU's, so far 45% have been used up. We will get a 2.6X increase in GAU in February.
(vii) List of 8 experiments the PCWG agreed to. Phil Merilees - What is the list of scientific questions that goes with these runs? (Julie will provide the info, used in the CSL proposals).
(viii) Need list of proposed changes to CSIM for IPCC deadline of March, 2003:
-- Add constant ice salinity to the sea ice model
-- Make minor changes to albedo formulas (bug)
-- Modify the ridging scheme (how?) to make more realistic thick ice. (But do we know this is a dynamic not thermodynamic problem?).
(ix) CCSM is setting up a Change Review Board (CRB)
CCSM2.1 beta release aiming for spring 2003 with CPL06, CLM2.1, and any physics improvements that are ready. New tests will be applied: error growth tests, bit-for-bit reproducibility on different numbers of processors. The minutes of this group will go on the web.
2. Cliff Chen - Vectorizing CICE
The concepts of vectorization, which includes Amdahl Law, the influence of vector length on performance, and three methods to vectorize IF-branch were introduced. Then we talked about tuning of CICE 3.1. Two of the most time consuming subroutines, stability and stepu, were tuned. The original stability was running in scalar mode and the original stepu has been fully vectorized. We isolated the IF branch first and concated two layers of do-loop into one to increase the vector length. After tuning, we gained more than 50 times performance improvement with stability, and more than 3 times improvement with stepu. The tuned stability still provides bit-for-bit results on SGI IRIX64 system and ran slightly 10% faster. With the experience of these two subroutines, we conclude that vectorization of CICE is feasible.
3. Bill Lipscomb - MPDATA and Incremental Remapping for Ice Transport
Incremental remapping is a transport scheme with several desirable features: it is conservative, monotonicity-preserving, second-order accurate (except where the accuracy is reduced locally to preserve monotonicity), and efficient for solving the large systems of transport equations in multi-category sea ice models. The Los Alamos sea ice model, CICE, has used an incremental remapping scheme for sea ice transport since November, 2001. This scheme was recently added to the CSIM of the CCSM. In a one-year run at NCAR on a 1-degree grid in an active-ice-only configuration, remapping was 55% faster than the current transport scheme, MPDATA, leading to a 21% drop in total time for the ice model. In stand-alone CICE runs at Los Alamos, remapping is about three times faster than MPDATA, and marginal costs for additional categories and ice layers are 4-6 times lower.
Remapping outperforms MPDATA in standard advection tests. Both schemes have some diffusion (much less than the first-order upwind scheme), but MPDATA produces spurious extrema in both transported fields (area, volume, and energy) and passive tracers (thickness and enthalpy), while remapping does not. In active-ice-only runs using CCSM, remapping and MPDATA yield similar results. The largest differences are in the GIN Sea in winter and the Labrador Sea in summer, where MPDATA gives systematically lower ice concentrations. Negative thermodynamic feedbacks appear to compensate for differences in the advection schemes. Differences between the two schemes might be larger in fully coupled runs, which will be performed during the next few weeks.
4. Keith Olsen - The Snow Model in CLM2
Keith was invited to participate in the PCWG by the co-chairs, in order to initiate discussions about upgrading the snow model in CSIM, and how we might use largely existing technology from the CCSM land model.
(i) LSM was developed by Y. Dai, University of Georgia (IAP94)(ii) Liang Yan, University of Arizona (has his own model called VISA)The above are the two main LMWG snow modelers.
There are 5 layers in the snow where we keep track of M(H20), M(ice), h(layer), and T(layer). In the subsurface layers there are melting and freezing and flow. Snow and soil temperatures are solved for simultaneously. The snow skin T is not solved for using the sfc energy budget; instead, the heat capacity of the sfc layer is tuned to achieve a fit to an idealized diurnal cycle function to get skin temperature.
Three processes represent the compaction of snow: crystal metamorphosis; overburden pressure and melt. SCF is the snow cover fraction, and this is a function of snow depth and roughness. It will probably be modified or replaced in the next version.
SCF = hs/(10*zo + hs)
The snow model is verified as a set of 1D models forced by meteorological observations at the local site.
5. Discussion of Additional Problems and Model Physics - Elizabeth Hunke
E. Hunke - There is a problem with the new coupler. CPL5 takes 1178 Mbytes memory and cannot run on the LANL 512 processor machine. EH proposes this as a reason to "subroutinize" the model. This problem comes up at high resolution and on the SGI.
E. Hunke - Marginal ice zone free drift. Right now, when the area of ice, Aice, is less than some threshold, CCSM sets the ice velocity to zero to prevent blowup. The problem is that the model velocity depends on ice area (through the ice mass), but it should not in low concentration areas where floes do not interact. Elizabeth and John Dukowicz are working on a remedy for this problem.
Code vectorization - We need to prioritize the runs that need to be done, and be more specific. Looking ahead, will we have to adopt vectorized CICE or should we mount an effort to vectorize CSIM as it exists now? Should we merge the codes and keep them merged? Shouldn't we separate the vectorization process from the IPCC priorities?
Constant salinity - Cecilia Bitz and Julie Schramm will do this.
Albedo changes - no significant changes to be implemented at this time.
Ridging changes - not for the IPCC.
Table of Modeling Runs for PCWG
#1 - Ice feedback in enhanced CO2 scenario, and a companion run with a fixed ocean. It will take 1 month to get the code in and produce.
#2 - Coupled incremental remapping run.
#3 - Coupled runs with various g(h) categories: 1 cat, 5 cat (control), 10 cat.
#4 - Implement constant ice salinity - CC will code it, Julie will run it. This will be done with AIO and coupled (15 years). It may also require another 1% CO2 run, where the CO2 increase begins after the time at which the shared constants were fixed in the control run. Tony says they are doing a run starting at 950 years. The ocean group will test this out.
AMIP T85 run for the SHEBA year - check with Jim Hack. ARCMIP is interested in this. Analyze the polar atmosphere. No additional resources needed.
Monthly mean values should be processed on the fly: h, Ai, ui, Ts
Bill Large - The coupler group now produces a 2m temperature in the air, but it is not done correctly. It needs to be based on a profile analysis of potential temperature, not temperature. Bill suggests that in doing this correctly, they also get 2 meter values of RH and wind velocity.