The LANL AOMIP simulation was performed with an ocean-sea ice coupled model developed at Los Alamos National Laboratory. Detailed documentation for the ocean and ice models can be found in Smith and Gent , Hunke and Lipscomb  and other publications referenced below. The models are configured following the Arctic Ocean Model Intercomparison Project (AOMIP) protocol and integrated using AOMIP-defined forcing fields and parameterizations, except where noted.
 The Parallel Ocean Program (POP [Smith et al., 1992; Dukowicz et al., 1993, 1994]), is a member of the Bryan-Cox class of z-coordinate ocean models that solves the primitive equations for temperature, salinity and the horizontal velocity components. It features an implicit free surface, the UNESCO equation of state formulation of Jackett and McDougall , and two-band shortwave penetration with Jerlov water type IB [Paulson and Simpson, 1977]. The K-profile parameterization (KPP) provides vertical mixing of momentum and tracers, with convection occurring through a high vertical diffusion coefficient (10 m2 s−1), done implicitly in time. Lateral mixing of tracers occurs via the GM mixing parameterization [Gent and McWilliams, 1990] with κ = 2400 m2 s−1. Horizontal friction is biharmonic with a coefficient of −1012 × cos3 ϕ m4 s−1, where ϕ is latitude. A third order upwind scheme is used for horizontal advection of tracers while centered differences are used for momentum; bottom drag is quadratic with a coefficient of 1.225 × 10−3. The salinity-dependent freezing temperature is given by Tf = 0.0575S, where S is the sea surface salinity, but Tf is limited not to exceed the maximum freezing temperature at the ice bottom (−0.184°C). The ocean model provides sea surface temperature, salinity, currents, slope and a freezing or melting potential to the ice model. The freezing potential is converted to “frazil ice” in the ice model; some or all of the melting potential is used to warm and melt the ice.
 The Los Alamos Sea Ice Model (CICE) features the energy conserving thermodynamics model of Bitz and Lipscomb  with four layers of ice and one layer of snow in each of five ice thickness categories [Lipscomb, 2001], the energy-based ridging scheme of Thorndike et al.  and Hibler , an ice strength parameterization given by Rothrock , elastic-viscous-plastic ice dynamics [Hunke and Dukowicz, 1997, 2002] and horizontal advection via an incremental remapping scheme [Lipscomb and Hunke, 2004]. Prognostic variables for each thickness category include ice area fraction, ice volume, ice energy in each vertical layer, snow volume and energy, and surface temperature. The vertical salinity profile is a nonlinear function of depth with a maximum value at the bottom of the ice of 3.2 psu. Temperature dependence of the longwave radiation and sensible and latent heat fluxes is included in the nonlinear flux balance that (iteratively) determines the ice or snow surface temperature. Surface fluxes and temperatures are computed separately for each ice thickness category and merged based on the fractional area covered by that category. All precipitation over sea ice falls as snow in the model, regardless of the air temperature, and the new snow arrives at the same temperature as the existing snowpack. Snow and ice albedos were never agreed upon by the AOMIP participants; the values used here (0.81, 0.77, 0.70, 0.68 for cold snow, melting snow, cold bare ice, and melting bare ice respectively) are used by two other AOMIP modeling groups and are nearly identical to those used by Köberle and Gerdes . CICE has only one parameter for emissivity (0.98), leading to the other exception to the AOMIP parameters (0.97 for ice and 0.99 for snow). The ice model provides fresh water flux, net heat flux and ice-ocean stress to the ocean model.
 When coupled, the ice and ocean models are treated as subroutines, linked through a driver that also reads atmospheric data from files and prepares the data for use by the other components. The driver merges ice and ocean quantities based on the ice area fraction in cells where there is less than 100% ice coverage. The ocean model provides heat for melting the ice based on a turbulent heat flux parameterization whenever the temperature is above the freezing point; new ice formation maintains the ocean temperature at or above the freezing temperature. Virtual salinity fluxes based on the fresh water mass exchange between the ice and ocean contribute to buoyancy forcing at the ocean surface, with a reference ice salinity of 4 psu. Momentum exchange is accomplished through a quadratic ice-ocean drag term computed by the ice model using level 1 (5 m) currents. Turbulent fluxes over ice are computed using bulk formulas following Parkinson and Washington . Over ocean, the formulations of Kauffman and Large  are used.
 The ice and ocean models are discretized for nonuniform, general curvilinear grids in which the north pole has been moved smoothly into a nearby land mass to avoid prohibitively small time steps or filtering associated with converging meridions [Smith et al., 1995]. For the 0.4° experiments described here, we use a 900 × 600 global mesh (resulting in longitudinal spacing), whose north pole is in North America. The horizontal grid is mercator in the southern hemisphere with latitudinal spacing of 0.4°cos ϕ. In the northern hemisphere the grid size also decreases with latitude, resulting in a grid spacing that ranges from 9 km (at high latitudes) to 44 km (at the equator). The vertical grid consists of 40 levels which vary in thickness from 10 m at the surface to 250 m at the bottom. Topography was created by merging the Arctic data from Jakobsson et al. , Smith and Sandwell  data from 72S to 72N, and Southern Ocean data from Lythe and Vaughan , followed by pointwise modifications of important sills and channels that may have been smoothed by interpolation to the grid. Where necessary, atmospheric forcing data sets are merged across 60–68N for the AOMIP run.
 The driver and ice model use a time step of 30 minutes, while the ocean model time step is slightly longer. The ice model exchanges information with the driver once each time step, the ocean model once per day. The 0.4° sensitivity runs described here are all initialized from the end of year 34 of the AOMIP simulation, which began in 1948 following a 5-year spin-up. Additional details regarding the AOMIP protocol can be found at http://efdl.cims.nyu.edu/project_aomip/experiments/coordinated_analysis.
 The 20-year simulations are run on a 320 × 384 (1°) global grid whose north pole is in Greenland [Kiehl and Gent, 2004], using a one-hour time step. The spin-up procedure for the ice initial condition followed the AOMIP protocol, but without surface salinity restoring.