Attending: Ray Najjar (minutes taker), Jim Orr, Mick Follows, Ken Caldeira, Jorge Sarmiento, Scott Doney, Matt Hecht, Chris Sabine, Bob Key and John Bullister.
Jim gave us an update on the development of graphical and analysis software packages being developed by him and Patrick Brockmann at LSCE. Data storage will likely be in the form of NetCDF. Graphical packages could vary. Currently LSCE uses TecPlot, which has the advantage of handling a variety of grids. One must purchase this, however. Ferret, on the other hand, is free, but cannot handle any grid.
There was some discussion as to what aspects of the model simulations should be saved. We reached no firm conclusion on this. A proposal will be made by Jim and Ray and sent to OCMIP participants. For the steady-state runs, the choice of what to save is not critical because the model can be run out a little longer and additional information saved.
An issue that came up repeatedly was defining how long runs should be in order to be considered at a steady state. Jorge showed an interesting figure of a tracer study in which an ocean GCM was initialized at zero and restored towards unity at the surface with a 30-day time scale. Even after 6000 years of integration, there was a 6% north-south difference in the deep Pacific. The only part of the model that was above 0.99 was the thermocline. These results suggest that it may take longer than previously thought to reach equilibrium. It was decided that criteria need to be developed to define "equilibrium," even if some models, because of computational cost, can not be run out to equilibrium. It is therefore critical that some information about temporal trends in the models be recorded so that these could be taken into account when the models are analyzed and compared.
It was suggested that some effort be put into looking at the equilibration issue. Jorge's group will continue the simulation described above to see how long convergence takes. Ken and Scott agreed to look at simple models to try to understand the problem. (Ken and Rick Murnane have already started an electronic dialogue about this on the ocmip-all mailing list.) Ray suggested that one group do very long runs of the equilibrium runs in OCMIP: the solu- bility pump, natural C-14 and the biological model.
Ways to speed up models are to use different timesteps for momentum and tracers, as well as increasing the timestep with depth for some tracers. Matt has also experimented with using larger timesteps for passive vs active tracers. In any event, it is advisable to run the model towards the end of the run with the same timestep for all prognostics for a fixed amount of time. Scott's experience with the NCAR ocean model is that only 20-30 years are needed for running synchronously.
Jim also mentioned a scheme that he and others (Aumont et al., 1998, Climate Dynamics) developed to accelerate offline model runs. The scheme is called DEGINT. This can amount to large savings in computer time. An issue here is whether offline models can give the same results as online models. Jorge's group has essentially abandoned the offline model because of this.
We agreed that model SSTs were the appropriate choice for computing gas solubilities and Schmidt numbers for gas exchange. It was also agreed to take into account climatological sea level pressure and surface humidity variations in relating the dry air mixing ratio to the saturation concentration.
One interesting point brought up by Jim is that a CFL-like criterion can be violated in the computation of air-sea gas fluxes, particularly for non-CO2 gases, which have relatively short equilibration times. He presented a algorithm for adding in a correction term that relieves this instability.
We all agreed on adopting the recommendation of Chris and Ray that the Dickson and Millero refit of the Mehrbach constants K1 and K2 are the best choice for OCMIP. Many studies are converging on the opinion that these constants are the most accurate for computing pCO2 from DIC and Alk, which all OCMIP models need to do for solubility pump, biological pump and CO2 uptake simulations. All other aspects of the equilibrium constants protocol were also agreed upon. Chris and Bob agreed to make a subroutine based on this protocol that computes the CO2 aqueous concentration given, as inputs, the DIC, Alk, phosphate, silicate, temperature and salinity. It was also noted that very useful information about the carbonate system is available from the "DOE Handbook" edited by Dickson and Goyet. The web sites for this document is www-mpl.ucsd.edu/people/adickson/CO2_QC
We agreed to adopt the CFC protocols designed by Ray and John, with one exception: an alternative formulation describing the meridional variation of CFCs should be considered. In the protocol, a straight line fit between two points was proposed; a piecewise linear fit, or something else, may be preferable. John and Scott agreed to look at observations and determine what would be the best shape to pass through the two "observations."
John also gave an update on the WOCE CFC data. The global survey is now almost complete and the spatial resolution is remarkable, with zonal and meridional transects approximately every 20-30 degrees. It is clear that this is going to be an extremely valuable data set for OCMIP. John and others (Rana Fine, Bill Smethie, Mark Warner and Ray Weiss) have a proposal submitted to work on the synthesis of the data. A number of issues need to be addressed in the synthesis, such as quality control and the non-synoptic nature of the data--it was collected over a 12-year period. By the end of this year, John felt that OCMIP should be in a position to make some initial evaluations of the models with WOCE CFC data.
Jim suggested, in addition to the protocols that Ray has been sending out, that we describe in more detail the step-by-step process of incorporating the protocols into the models. We agreed that this was a good idea, and Jim passed out an example for the CFC simulations.
Most aspects of the nutrient-restoring model proposed by Ray were accepted. Some exceptions are:
Jorge volunteered to have his group do a "trial" run of the nutrient-restoring model to check to see if the model is properly formulated and gives reasonable results. It was decided that, so long as this could be done quickly, that the other groups would hold off on doing the nutrient-restoring run until we are satisfied with the Princeton results. We did not decide what was meant by quickly. Ray will clarify this.
Ken showed that there are significant differences in D14C in simulations using the method of Toggweiler, which carries the 14C/12C ratio as the only variable, in comparison to a "full-blown" method which carries, 12C, 13C and 14C separately. Pre-bomb distributions show a consistent offset of 15 permil and post-bomb distributions show a 6 permil offset. Ken also showed some box- model studies that qualitatively supported these results. These results suggest that OCMIP should not simulate D14C by carrying the 14C/12C ratio as a tracer.
Jim then showed some 3D model studies by Bacastow and Maier-Reimer (Climate Dynamics, 1990) in which it was clear that biological processes have a negligible impact on the the distribution of D14C. In their model, a solubility pump model was run for both 12C and 14C and the ratio was computed.
The above suggests the following protocol for simulating C-14:
We decided to adopt this protocol for simulations of radiocarbon. Jim agreed to write up this protocol and distribute it to ocmip-all for final approval.
Bob Key gave an overview of the WOCE 14C data, showing the variety of ways the data could be used to evaluate the models. For example, there are useful comparisons of WOCE data with earlier (e.g., GEOSECS) data, showing the evolution of the bomb signal. There is also a way to extract the pre-bomb signal with greater accuracy than previous efforts. Bob noted that Broecker's method of using the silicate distribution to isolate pre-industrial 14C has serious problems. A better solution appears to be the use of "potential alkalinity" (salinity- and nitrate-corrected alkalinity).
Chris gave us an update of the CO2 Survey data. The Indian Ocean QC and preliminary analysis is essentially done: there are two north-south sections and three east-west sections. Chris has made estimates for the spatial distibution of natural and anthropogenic CO2 in the Indian Ocean. He expects that the Pacific should be done by the end of this year.
Chris and Bob will not only provide analyses to the modelers, but will also contribute to OCMIP model evaluation, as they have been doing for the Princeton/GFDL model. As was argued in our proposal, we will produce much better science if we have close collaboration between modelers and measurers.