## 1. Introduction

[2] The U.S. Navy has developed a new high-accuracy global atmospheric model, which scales efficiently on current and future state-of-the-art computing platforms: the Naval Research Laboratory (NRL) Spectral Element Atmospheric Model, or NSEAM. Its dynamical core [*Giraldo and Rosmond*, 2004; *Giraldo*, 2005] is based on the “spectral element” method projected to three-dimensional Cartesian coordinates, which effectively eliminates the pole singularity problem of spherical coordinates and combines the local domain decomposition property of finite element methods with the high-order accuracy of spectral transform methods.

[3] An important advantage of spectral element (SE) models is that the solution of the global matrix problem required by the semi-implicit method is straightforward and computationally efficient for either the hydrostatic or nonhydrostatic equations. The semi-implicit method for either an SE or spherical harmonics (SH) model requires the solution of a three-dimensional Helmholtz operator. However, the only reason why SH models are competitive with SE models (or grid point models for that matter) for the hydrostatic equations is that SH models do not require the solution of a matrix problem (the Helmholtz operator is solved exactly for constant coefficients). Still, SE/grid point models are competitive with SH models at high resolution and on distributed memory architectures. In the nonhydrostatic case, the SH models now require the inversion of a matrix which adds additional computational cost to the model. The SE models, on the other hand, already include this overhead and thus transitioning to a nonhydrostatic system incurs only minimal additional computational cost.

[4] NSEAM can adopt any horizontal model grid, fixed or variable, and various time integrators such as Eulerian [*Giraldo and Rosmond*, 2004], semi-implicit [*Giraldo*, 2005], or hybrid Eulerian-Lagrangian semi-implicit [*Giraldo*, 2006] methods. Its spectral element formulation maintains the high-order accuracy of spherical harmonics that the current Navy's global atmospheric model is based on, while it offers flexibility to employ any form of variable grid to enhance horizontal grid resolutions in strategic regions. Its dynamical core scales efficiently, i.e., allows the use of large numbers of processors, and was validated using various barotropic and baroclinic test cases [*Giraldo and Rosmond*, 2004; *Giraldo*, 2005]. The model can be discretized vertically with any grid, but the mass and energy conserving flux form of the finite difference method on the terrain-following (*σ*) coordinate is first selected for comparison with existing models.

[5] For this study, we modify and expand the NSEAM dynamical core to include the physics package utilized in the operational version of the Navy Operational Global Atmospheric Prediction System or NOGAPS [*Hogan and Rosmond*, 1991], but without the land surface parameterization (for recent model physics improvements refer to *Peng et al.* [2004], *Hogan and Pauley* [2007] and *Kim* [2007]). We select the hexahedral grid that consists of six faces of a hexahedron, each of which contains a desired number of quadrilateral elements (see Figure 1) following *Giraldo and Rosmond* [2004].

[6] Among various methods to validate an atmospheric model is to force the model under controlled and simplified sets of boundary and initial conditions so that the results can be interpreted in a relatively straightforward manner and also intercomparable to those of other similar models, although the correct solution is not quite known except by simplified theory and limited indirect observations. A good recent example is the aquaplanet experiments proposed by *Neale and Hoskins* [2001a, 2001b] in which the Earth is covered with water only. These experiments provide a useful and convenient test bed for investigating the interaction between the dynamics and physics in atmospheric models.

[7] In this study, we validate NSEAM by configuring it for aquaplanet experiments mainly following *Neale and Hoskins* [2001a]. We perform various sensitivity experiments in order to understand and improve the aquaplanet simulation. Sensitivity of aquaplanet simulations to horizontal resolution was studied previously. For example, *Lorant and Royer* [2001] found from general circulation model (GCM) experiments that with higher resolution the convective cells are more intensified and concentrated, being accompanied by improved simulation of equatorial waves that modulate near-equatorial convection. Sensitivity of aquaplanet simulations to vertical resolution was also studied earlier. For instance, *Inness et al.* [2001] compared between 19 and 30 (unevenly spaced) layer versions of their GCM and discussed its implications for Madden-Julian Oscillation (MJO) in view of the moisture budget. They reported that the effect of convection is to moisten/dry the lower troposphere in their 19/30 layer simulations, respectively. In the present study, we investigate further sensitivity of the aquaplanet simulations to selected details of the model such as the vertical distribution, as well as the resolution, of the model levels and the way the lifting condensation level (LCL) is calculated or utilized in the context of the frequency and propagation of simulated Kelvin waves and MJO.

[8] In section 2, we outline the coupling of the NSEAM dynamical core with the NOGAPS physics including the new horizontal viscosity that replaces the original horizontal diffusion. In section 3, we describe the setup of the aquaplanet experiments and present the results of the sensitivity experiments in terms of the viscosity, vertical distribution of the model levels and some details of the precipitation physics relevant to LCL. We also discuss our efforts to simulate the Kelvin waves and MJO in this section. We end the discussion in section 4 by giving concluding remarks and a short summary of additional sensitivity experiments performed but not presented in this study. Appendix A includes the derivation of the viscosity operator.