Version 5 (modified by maronga, 6 years ago) (diff)

--

The PALM core

The PALM model core is based on the non-hydrostatic, filtered, incompressible Navier-Stokes equations in Boussinesq-approximated form (an anelastic approximation is available as an option for simulating deep convection). By default, PALM has at least six prognostic quantities: the velocity components u, v, w on a Cartesian grid, the potential temperature θ, water vapor mixing ratio qv and possibly a passive scalar s. Furthermore, an additional equation is solved for either the subgrid-scale turbulent kinetic energy (SGS-TKE) e (LES mode, default) or the total turbulent kinetic energy (RANS mode, see PALM-4U components).

In the LES mode, the filtering process yields four subgrid-scale (SGS) covariance terms. These SGS terms are parametrized using a 1.5-order closure after Deardorff (1980). PALM uses the modified version of Moeng and Wyngaard (1988) and Saiki et al. (2000). The closure is based on the assumption that the energy transport by SGS eddies is proportional to the local gradients of the mean quantities.

The following sections outline the most important specifics of PALM. For a more detailed description, see PALM documentation and related publications (Raasch and Schröter, 2001; Maronga et al., 2015).

Discretization in time and space

The model domain in PALM is discretized in space using finite differences and equidistant horizontal grid spacings. The Arakawa staggered C-grid (Harlow and Welch, 1965; Arakawa and Lamb, 1977) is used, where scalar quantities are defined at the center of each grid volume, whereas velocity components are shifted by half a grid width in their respective direction so that they are defined at the edges of the grid volumes. By default, the advection terms in the prognostic equations are discretized using an upwind-biased 5th-order differencing scheme (Wicker and Skamarock, 2002) in combination with a 3rd-order Runge–Kutta time-stepping scheme after Williamson (1980).

Pressure solver

The Boussinesq approximation requires incompressibility of the flow, but the integration of the governing equations does not provide this feature. Divergence of the flow field is thus inherently produced. Hence, a predictor corrector method is used where an equation is solved for the modified perturbation pressure after every time step (e.g., Patrinos and Kistler, 1977). In case of cyclic lateral boundary conditions, the solution of the Poisson equation is achieved by using a direct fast Fourier transform (FFT). PALM provides the inefficient but less restrictive Singleton-FFT (Singleton, 1969) and the well optimized Temperton-FFT (Temperton, 1992). External FFT libraries can be used as well, with the FFTW (Frigo and Johnson, 1998) being the most efficient one. Alternatively, the iterative multigrid scheme can be used (e.g., Hackbusch, 1985). This scheme uses the Gauss–Seidel method for the inner iterations on each grid level.

Boundary conditions

PALM offers a variety of boundary conditions. Dirichlet or Neumann boundary conditions can be chosen for u, v, θ, qv, and p at the bottom and top of the model. For the horizontal velocity components the choice of Neumann (Dirichlet) boundary conditions yields free-slip (no-slip) conditions. Neumann boundary conditions are also used for the SGS-TKE. Kinematic fluxes of heat and moisture can be prescribed at the surface instead (Neumann conditions) of temperature and humidity (Dirichlet conditions). At the top of the model, Dirichlet boundary conditions can be used with given values of the geostrophic wind. Vertical velocity is assumed to be zero at the surface and top boundaries, which implies using Neumann conditions for pressure.

Following Monin-Obukhov similarity theory (MOST) a constant flux layer can be assumed as boundary condition between the surface and the first grid level where scalars and horizontal velocities are defined. In PALM we assume that MOST can be also applied locally and we therefore calculate local fluxes, velocities, and scaling parameters. This scheme involves calculation of the Obukhov length L, which can be either done based on variables of the previous time step ("circular"), via a Newton iteration method, or via a look-up table for the stability parameter z/L.

Parallelization and scaling

The parallelization of the code is achieved by a 2-D domain decomposition method along the x and y direction with equally sized subdomains. Ghost layers are added at the side boundaries of the subdomains in order to account for the local data dependencies, which are caused by the need to compute finite differences at these positions. The number of ghost layers that are used in PALM depend on the order of the advection scheme, with three layers for the 5th-order Wicker-Skamarock scheme. Ghost layer data are exchanged after every time step. Data exchange between proecessor cores is realized using the Message Passing Interface (MPI). Additional loop vectorization via OpenMP is realized which also allows a so-called hybrid parallelization.

PALM shows excellent scaling which was tested for up to 50,000 processor cores (details).

External forcing and nesting

Usually, PALM is used to simulate the flow in the boundary layer which is a certain part of the atmosphere. Processes occurring on larger scales than those in the boundary layer including large scale advection of scalars, large scale pressure gradients or large scale subsidence have also to be considered in the model, especially when focusing on realistic situations observed during measurement campaigns. In limited domain models with non-cyclic boundary conditions the large scale state enters through the boundary conditions at the lateral walls, and is usually taken from larger-scale models. An additional possibility to account for tendencies in the LES model resulting from larger scales than those in the boundary layer is the usage of nudging. Nudging is a (Newtonian) relaxation technique which adjusts the large-eddy simulation to a given, larger scale flow situation (Anthes, 1974; Stauffer and Bao, 1993). Alternatively, a nesting system is available which allows for self-nesting of PALM, i.e. running a parent domain run at coarse grid resolution and large model domain with a nested child model domain run at finer resolution and smaller domain.

Ocean option and coupling to atmosphere

PALM allows for studying the ocean mixed layer (OML) by using an ocean option where the sea surface is defined at the top of the model, so that negative values of z indicate the depth.

A coupled mode for the atmospheric and oceanic versions of PALM has been developed in order to allow for studying the interaction between turbulent processes in the ABL and OML. The coupling is realized by the online exchange of information at the sea surface (boundary conditions) between two PALM runs (one atmosphere and one ocean). The atmospheric model uses a constant flux layer and transfers the kinematic surface fluxes of heat and moisture as well as the momentum fluxes to the oceanic model. Flux conservation between the ocean and the atmosphere requires an adjustment of the fluxes for the density of water.

Embedded models

PALM comes with an increasing number of embedded models:

For more details, see PALM documentation.