Report of the CCSM Polar Climate Working Group Meeting
24 June 2003 Breckenridge, Colorado
Joint Session with the Climate Variability Working Group
Richard Cullather presented "An Appraisal of Storm Track Depictions in Coupled and Uncoupled Simulations of the CCSM Version 2." A feature-tracking algorithm of intermediate complexity is applied to the once-daily output sea level pressure fields of the CCSM2 millennial integration for the years 561-580, and to an ensemble member of the CAM2.0.1 20th century observed SST and sea ice-forced simulation for the years 1980-1999. These calculations are compared to a similar analysis using daily NCEP/NCAR reanalyses. A primary motivation of this study is to assess the relation between deficiencies in model storm tracks and biases in the monthly mean fields, which were briefly reviewed. In general, the synoptic variability was found to be larger in both model simulations than in the observations. Frequency statistics for cyclonic and anticyclonic systems were presented separately. In the Northern Hemisphere, a wintertime bias pattern in both models known as the "pole problem" is characterized by the poleward displacement of the Siberian High. Moreover, the annual maximum sea level pressure tends to occur in the wrong month in the models (May/June rather than March). Diagnosed wintertime model cyclone tracks indicate subtle differences with the reanalyses. In particular, the North Atlantic track in coupled and uncoupled simulations has a more zonal orientation with an abrupt northward turn near the Norwegian coastline. This general character in the models of fewer cyclone tracks crossing through the Denmark Strait is maintained even when the sign of the North Atlantic Oscillation is considered. Additionally, wintertime anticyclones in coupled and uncoupled simulations over the Canada Basin generally have a shorter life span and have lower maximum pressure values than in the reanalyses. For the Arctic in summer, the reanalyses depict cyclones that form within the Siberian Arctic Front baroclinic zone and propagate into the central Arctic. In contrast, coastal Siberian cyclone tracks in the coupled simulation are more zonal. For the Southern Hemisphere, discrepancies between the models and reanalyses are characterized by the absence of the previously documented cyclone track bifurcation in the New Zealand sector, and fewer instances of cyclogenesis near the Antarctic Peninsula.
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 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 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. (This talk was a reprise of Marika's presentation to the PCWG in January 2003, for the benefit of the Climate Variability Working Group.)
Alex Hall, in "Assessing Climate Feedbacks Using Models and Observations," proposed that climate models could be used to identify which feedbacks might be measurable in observations. The presence of internal variability and the fact that the observational record is not in equilibrium with the concurrently applied forcing present difficulties in extracting information about the strength of the real climate's feedbacks from the observed record. However, an analysis of the GFDL climate model demonstrates that these difficulties are surmountable for continental snow albedo feedback, offering the possibility of constraining the snow albedo feedback by examining the satellite record.
Following these three presentations, the PCWG took a short break and reassembled separately from the Climate Variability Working Group.
Session on Model Physics and Experiments (D. Moritz, chair)
Cecilia Bitz presented "Effects of Fixed and Interactive Sea Ice in Enhanced CO2 Simulations with CCSM and CAM-SOM," examining the modeled climate sensitivity in four configurations: (a) full atmosphere with a slab ocean, (b) full atmosphere and ocean, and both (a) and (b) with and without an interactive sea ice model. In (a), the active sea ice model accounts for 40% of the global sensitivity, compared with the run without interacting sea ice; feedbacks in the Arctic double the climate sensitivity in that area, and even the tropics see changes due to sea ice feedbacks. In (b), only 18% of the global sensitivity is associated with the interactive sea ice, and there were no changes observed in the tropics. The difference in the tropics can be attributed to differences in the atmospheric heat transport as well as in the ocean, where warming in the Southern Hemisphere occurs unless sea ice is fixed (non-interactive). The thermohaline circulation decreases in the North Atlantic in all configurations, most notably when sea ice is fixed. Questions remain as to whether this study should be continued, since fixing sea ice in the non-interactive runs violates conservation of water and energy.
Marika Holland provided an update on simulations intended to test the model sensitivity to the resolution of the ice thickness distribution, comparing runs with 1 and 10 categories to the standard 5 category case. The 1-category case was run in both the "control" and 1% increasing CO2 scenarios. In the Northern Hemisphere, in the 1-category case, the ice is 25 to 50 cm thinner than in the 5-category case, the seasonal amplitude is reduced, and the ice coverage is slightly larger. All of these changes are consistent with changes in ice growth/melt rates due to reduced thin ice. In particular, the air temperature is colder (associated with increased ice cover), even though the mean ice thickness is smaller. Differences in the Southern Hemisphere are along the same lines but even larger because the mean ice thickness is too large. Air temperatures are as much as 6 degrees colder around Antarctica. Because the mean ice thickness in the Arctic is too thin, missing thin ice in the 1-category case has less influence on the simulation there. The 10 category run, which has not finished, is showing less of an effect than between the 1-category and 5-category cases, as expected. However, thickness changes are as large as 50 cm in the Arctic. Effects on transients have not been examined yet.
Four posters were advertised by Elizabeth Hunke, Richard Grotjahn, Jiping Liu, and Bill Lipscomb.
Dick Moritz led a discussion concerning biases in the CCSM control climate simulations pertaining to the polar regions, asking the question whether any of these biases could be attributed to deficiencies in the sea ice model. Biases include:
1. Spatial distribution of ice (thickest ice is on Siberian side, which is opposite to the location of thickest ice in observations; simulated ice in the Weddell Sea is much thicker than observed).
2. Hemispheric mean sea ice thickness (Arctic smaller than observed, Antarctic larger than observed).
3. Ice extent in Antarctic is significantly larger than observed.
4. Surface air temperature higher than observed over land and sea ice
5. Open water fraction larger than observed in the Arctic Ocean.
6. Arctic tropopause temperature is lower than observed.
- The spatial distribution of mean ice thickness in the Arctic Ocean has long been attributed to errors in the atmospheric surface winds and sea level pressure field.
- Several possible reasons for the large ice thickness in the Antarctic were offered, including a potential bug in the sea ice ridging mechanism, weak wind forcing from the Antarctic continent, weak cyclogenesis in the Antarctic peninsula region, and resolution of the Antarctic coastal current. All but the first are attributable to other model components.
- Large ice thickness in the Weddell Sea is generally blamed for the large ice extent in the South Atlantic.
- It was noted that the new version of CAM mediates some of these problems, such as by producing lower air temperatures that lead to thicker ice in the Arctic (2, 4, 6). We still hold out hope that CAM will adopt the FV dynamical core, which has been shown to improve the surface pressure simulation in the Arctic.
Related to the discussion of biases in the control simulation, Cecilia Bitz compared the CSIM albedo parameterization with unpublished albedo data (Warren and Brandt). The CSIM albedo parameterization had modest biases of either sign depending on the surface conditions. Cecilia recommends that the data be used in conjunction with the SHEBA data to construct a parameterization that is good for both hemispheres.
Session on Model Code (E. Hunke, chair)
Elizabeth Hunke outlined the PCWG procedures for testing and adopting code changes in CSIM, which are designed to ensure that the new code meets CCSM standards, improves the climate simulation or computational performance, and can be trusted. All substantial code changes must go through the PCWG process. "Substantial" means all physics changes and any software engineering that results in major changes to code organization and/or efficiency. Cosmetic changes generally are unimportant. The procedure may involve some iteration but typically follows the following steps:
1. Modeler develops improved code, tweaking a few lines, replacing a parameterization or providing an alternative physics module.
2. Modeler tests changes within CCSM framework.
3. Outcome of tests is posted on PCWG web site and modeler provides explanation to PCWG members.
4. PCWG members peruse results (and code if desired).
5. PCWG members decide whether to recommend adoption of change to CCSM Code Review Board.
Defining requirements and tests that are suitable for all possible code changes is impossible; however, we have developed drafts for these lists that are being followed loosely. DRAFT CSIM Requirements List (or "Targets" or "Wish") list - standard requirements that CSIM must fulfill:
1. any applicable CCSM requirements
2. compiles and runs with -r8 (double precision)
3. restarts exactly on specified machine
4. any change to computational efficiency due to code changes is documented
5. runs on the standard grids released with CCSM
6. runs on standard computer platforms
7. runs in both fully coupled and AIO mode
8. certain internal configurations work
9. produces history files containing specific fields
10. runs with any number of categories specifically tested with 1, 5, 10
11. meets requirements of a Test Suite (Examples---under development):
- In 1-year tests:
(1) transport (advection, ridging, ITD changes) conserves global sums of volume, internal energy (and area where necessary) to 10 digits
(2) thermodynamic changes to energy, volume is compatible with fluxes
(3) satisfies certain nonglobal conservation constraints (for example, the ridging parameterization conserves volume locally)
- In 1-timestep tests:
(4) dynamics produces zero motion given zero momentum forcing
(5) runs with no ice present
Issues include code familiarity (i.e., trust, which the PCWG procedure is intended to address), acknowledgment of authors (which we do), and ownership/control of the code. Regarding the latter, neither LANL nor NCAR enjoys any special privileges with regard to code adoption into CSIM. The PCWG retains the final decision on what recommendations will be made to the Code Review Board, and PCWG members work under the assumption that the Code Review Board will not adopt code into CSIM that has not been recommended by the PCWG. (Members of the Code Review Board, Marika Holland and Julie Schramm, assured us that this would not occur.) In practice, the requirements list not complete or formalized, CSIM4 does not satisfy all requirements even now, the test suite is still under development, particular tests depend on character of the code change, and there will always be bugs to fix. In conclusion, we forge ahead anyway by gaining consensus on what's "reasonable." The PCWG has been wildly successful so far, largely because of wide community participation in the code development process.
Julie Schramm provided a quick update on changes to the code since the previous PCWG meeting. These include:
1. incremental remapping (tested in 15 year M run)
2. incremental with open water advection (15 year M and B runs)
3. nonzero sea ice salinity (with respect to ice-ocean exchanges) (15 year B run)
4. alteration of wind and ice-ocean stress terms for the free drift regime (15 year M run)
5. calculation of Tref and Qref (short restart tests; these are merely diagnostic)
6. minor bug fixes
a. do not allow albedo less than zero over thin, bare, melting iceb. use Lvap instead of Lsub for calculating ice-ocean fluxesc. include penetrating shortwave in ocean mixed layer model heat balance.
Cecilia Bitz reiterated that the PCWG wishes to maintain the history of the CSIM code through documentation and acknowledgment of authors. Changing the thermo code does not mean that the documentation needs to be changed, since the underlying equations and theory will be the same. On a related topic, she recommends we maintain usability of the code by keeping the structure and names of things as similar as possible.
Bill Lipscomb outlined the path to CSIM vectorization currently envisioned. CICE vectorization is nearly complete, as presented in Bill's poster. The parts of CICE and CSIM that are almost identical (EVP dynamics, incremental remapping advection, and ITD) will be easily transferred from CICE to CSIM. The ridging (ice_mechred) and utility (history, diagnostics, etc) modules will be vectorized independently in the two codes, with CSIM following the CICE example. The tricky part is the thermodynamics. Bill and Cecilia have agreed to make modifications to the vectorized thermo module in CICE to bring it more in line with CSIM with regard to physics, numerical algorithms where necessary, and general "look and feel." This module will be presented to the PCWG for adoption following testing. A recommendation was made that the Software Engineering Working Group provide a list of standard code conventions for vectorizing CCSM component modules, recognizing that the different components may need to employ different approaches to vectorization.
Elizabeth Hunke asked the group for updates on certain CCSM problems that have been noted in previous meetings.No one has looked into the sign error in the mean sensible heat flux, which could be originating with CSIM or CAM. The Atlantic-to-Arctic heat transport in the ocean is excessive; ocean modelers attribute this to low resolution (the North Atlantic current does not turn left into the Labrador Sea, instead heading straight across into the GIN and Barents Seas, leading to excessive ice in the Labrador Sea). Cecilia Bitz has fixed the erroneous Qflux parameterization in the slab ocean model used in CAM, but the change has not made it into CAM. Jim Hack needs to be contacted about this.
Additional CCSM/CSIM configurations were discussed. Julie Schramm has begun work on developing a stand-alone version of CSIM. Reaction by the PCWG was generally positive, since this will entrain more CSIM users, but there are outstanding questions: How much extra support (i.e., Julie's time) will this entail? Should forcing data sets be provided for each model configuration/grid? A thermo-only ice model has been requested for use in CAM. They are not satisfied with the prescribed ice option in CSIM, which does not allow ice thickness changes (duly noted). The active atmosphere and ice model configuration using the CSIM's ocean mixed layer model does not work now; the PCWG recommends that it be fixed. Finally, Rick Smith asked whether the PCWG is interested in using a tripole grid that has improved properties in the Arctic. The response was overwhelmingly positive. LANL will incorporate this change into CICE, and then transfer the changes to CSIM. Most of the changes will be to the boundary update routines, which are common to both models. This change will go through the usual, full testing process.
Before adjourning, Bill Lipscomb gave a quick outline of the changes to the thermodynamics that he and Cecilia envision.
PCWG "Action Items" (grouped by type, not in any particular order):
1. Communicate PCWG needs to SSC/WG co-chairs, especially with respect to CCSM biases that affect the sea ice/high latitude simulation. (Dick, Elizabeth, CC)
2. Track down and fix strength/ridging problem causing CFL violations. (Bill, Elizabeth, others)
3. Determine why the SH ice is so thick (CSIM problem or other component?)
4. Revisit albedo with new observations. (CC, Bruce)
5. Get CAM to adopt CC's fixed Qflux for slab ocean. (CC, Dick)
6. Fix thermo-only CSIM for CAM? ("duly noted")
7. Get atm+ice+MLocn working. (CC)
8. Vectorize CSIM. (Bill, Cliff, Julie, others, including CC for thermo)
9. Develop stand-alone CSIM capability. (Julie)
10. Add tripole grid capability. (LANL)
11. Further define and refine list of core requirements (Elizabeth, Dick, Marika)
12. Identify/fix CSIM non-compliance. (Julie, others)
13. Automate a basic test suite (parameterization independent). (Julie)