# The WSF: The Advanced Research WRF (ARW) solver

The

Weather Research and Forecasting (WRF) model is a numerical weather prediction

(NWP) model designed to carry out research operations on the atmospheric

system. The spectrum of options available in it for performing operations

pertaining to physics and dynamics is very broad. It is suitable for a broad

span of applications across scales ranging from large-eddy to global

simulations. Such applications include real-time NWP, data assimilation

development and studies, parameterized-physics research, regional climate simulations,

air quality modeling, atmosphere-ocean coupling, and idealized simulations.

The

principal components of the WRF system are depicted in Figure 1. The WRF

Software Framework (WSF) provides the infrastructure that accommodates the

dynamics solvers, physics packages that interface with the solvers, programs

for initialization, WRF-Var, and WRF-Chem. There are two dynamics solvers in

the WSF: The Advanced Research WRF (ARW) solver (originally referred to as the

Eulerian mass or “em” solver) developed primarily at NCAR, and the NMM

(Nonhydrostatic Mesoscale Model) solver developed at NCEP. Community support

for the former is provided by the MMM Division of NCAR and that for the latter

is provided by the Developmental Testbed Center (DTC).

The

ARW is the ARW dynamics solver together with other components of the WRF system

compatible with that solver and used in producing a simulation. Thus, it is a

subset of the WRF modeling system that, in addition to the ARW solver,

encompasses physics schemes, numerics/dynamics options, initialization

routines, and a data assimilation package (WRF-Var). The ARW solver shares the

WSF with the NMM solver and all other WRF components within the framework.

Physics packages are largely shared by both the ARW and NMM solvers.

The

ARW solver is comprised of the following options –

•

Equations: Fully compressible, Euler nonhydrostatic with a run-time hydrostatic

option available. Conservative for scalar variables.

•

Prognostic Variables: Velocity components u and v in Cartesian coordinate,

vertical velocity w, perturbation potential temperature, perturbation

geopotential, and perturbation surface pressure of dry air. Optionally,

turbulent kinetic energy and any number of scalars such as water vapor mixing

ratio, rain/snow mixing ratio, cloud water/ice mixing ratio, and chemical

species and tracers.

•

Vertical Coordinate: Terrain-following, dry hydrostatic-pressure, with vertical

grid stretching permitted. Top of the model is a constant pressure surface.

•

Horizontal Grid: Arakawa C-grid staggering.

•

Time Integration: Time-split integration using a 2nd- or 3rd-order Runge-Kutta

scheme with smaller time step for acoustic and gravity-wave modes. Variable

time step capability.

•

Spatial Discretization: 2nd- to 6th-order advection options in horizontal and

vertical.

•

Turbulent Mixing and Model Filters: Sub-grid scale turbulence formulation in

both coordinate and physical space. Divergence damping, external-mode

filtering, vertically implicit acoustic step off-centering. Explicit filter

option.

•

Initial Conditions: Three dimensional for real-data, and one-, two- and

three-dimensional for idealized data. Digital filtering initialization (DFI)

capability available (real-data cases).

•

Lateral Boundary Conditions: Periodic, open, symmetric, and specified options

available.

•

Top Boundary Conditions: Gravity wave absorbing (diffusion, Rayleigh damping,

or implicit Rayleigh damping for vertical velocity). Constant pressure level at

top boundary along a material surface. Rigid lid option.

•

Bottom Boundary Conditions: Physical or free-slip.

•

Earth’s Rotation: Full Coriolis terms included.

•

Mapping to Sphere: Four map projections are supported for real-data simulation:

polar stereographic, Lambert conformal, Mercator, and latitude-longitude

(allowing rotated pole). Curvature terms included.

•

Nesting: One-way interactive, two-way interactive, and moving nests. Multiple

levels and integer ratios.

•

Nudging: Grid (analysis) and observation nudging capabilities available.

•

Global Grid: Global simulation capability using polar Fourier filter and

periodic east-west conditions.

Model

Physics has the following options –

•

Microphysics: Schemes ranging from simplified physics suitable for idealized

studies to sophisticated mixed-phase physics suitable for process studies and

NWP.

•

Cumulus parameterizations: Adjustment and mass-flux schemes for mesoscale

modeling.

•

Surface physics: Multi-layer land surface models ranging from a simple thermal

model to full vegetation and soil moisture models, including snow cover and sea

ice.

•

Planetary boundary layer physics: Turbulent kinetic energy prediction or

non-local K schemes.

•

Atmospheric radiation physics: Longwave and shortwave schemes with multiple

spectral bands and a simple shortwave scheme suitable for climate and weather

applications. Cloud effects and surface fluxes are included.

Model

initialization –

The

ARW may be run with user-defined initial conditions for idealized simulations,

or it may be run using interpolated data from either an external analysis or

forecast for real-data cases. Both 2D and 3D tests cases for idealized

simulations are provided. Several sample cases for real-data simulations are

provided, which rely on pre-processing from an external package (usually the WRF

Preprocessor System, referred to as WPS) that converts the large-scale GriB

data into a format suitable for ingest by the ARW’s real-data processor.

The

programs that generate the specific initial conditions for the selected

idealized or real data case function similarly. They provide the ARW with:

•

input data that is on the correct horizontal and vertical staggering;

•

hydrostatically balanced reference state and perturbation fields; and

•

metadata specifying such information as the date, grid physical

characteristics, and projection details.

Initialization

using the real data –

The

initial conditions for the real-data cases are pre-processed through a separate

package called the WRF Preprocessing System (WPS, see Fig. 2). The output from

WPS is passed to the real-data pre-processor in the ARW— program real— which

generates initial and lateral boundary conditions. This section is primarily

about the steps taken to build the initial and the lateral boundary conditions

for a real-data case. Even though the WPS is outside of the ARW system, a brief

description is appropriate to see how the raw meteorological and static terrestrial

data are brought into the model for real-data cases.

The

WPS is a set of programs that takes terrestrial and meteorological data

(typically in GriB format) and transforms them for input to the ARW

pre-processor program for real-data cases (real). Figure 2 shows the flow of

data into and out of the WPS system. The first step for the WPS is to define a

physical grid (including the projection type, location on the globe, number of

grid points, nest locations, and grid distances) and to interpolate static

fields to the prescribed domain. Independent of the domain configuration, an

external analysis or forecast is processed by the WPS GriB decoder, which

diagnoses required fields and reformats the GriB data into an internal binary

format. With a specified domain, WPS horizontally interpolates the

meteorological data onto the projected domain(s). The output data from WPS

supplies a complete 3-dimensional snapshot of the atmosphere on the selected

model grid’s horizontal staggering at the selected time slices, which is sent

to the ARW pre-processor program for real-data cases.

The

input to the ARW real-data processor from WPS contains 3-dimensional fields

(including the surface) of temperature (K), relative humidity (and the

horizontal components of momentum (m/s, already rotated to the model

projection). The 2-dimensional static terrestrial fields include: albedo,

Coriolis parameters, terrain elevation, vegetation/land-use type, land/water mask,

map scale factors, map rotation angle, soil texture category, vegetation

greenness fraction, annual mean temperature, and latitude/longitude. The

2-dimensional time-dependent fields from the external model, after processing

by WPS, include: surface pressure and sea-level pressure (Pa), layers of soil

temperature (K) and soil moisture (kg/kg, either total moisture, or binned into

total and liquid content), snow depth (m), skin temperature (K), sea surface temperature

(K), and a sea ice flag.

Nesting

–

The

ARW supports horizontal nesting that allows resolution to be focused over a

region of interest by introducing an additional grid (or grids) into the

simulation. In the current implementation, only horizontal refinement is

available: there is no vertical nesting option. The nested grids are

rectangular and are aligned with the parent (coarser) grid within which they

are nested. Additionally, the nested grids allow any integer spatial (?xcoarse/?xfine)

and temporal refinements of the parent grid (the spatial and temporal

refinements are usually, but not necessarily the same). This nesting

implementation is in many ways similar to the implementations in other

mesoscale and cloud scale models (e.g. MM5, ARPS, COAMPS). The major

improvement in the ARW’s nesting infrastructure compared with techniques used

in other models is the ability to compute nested simulations efficiently on

parallel distributed-memory computer systems, which includes support for moving

nested grids. The WRF Software Framework, described in Michalakes et al.

(2004), makes these advances possible. In this chapter, we describe the various

nesting options available in the ARW and the numerical coupling between the

grids.

1-way

and 2-way grid nesting –

Nested

grid simulations can be produced using either 1-way nesting or 2-way nesting as

outlined in Fig. 3. The 1-way and 2-way nesting options refer to how a coarse

grid and the fine grid interact. In both the 1-way and 2-way simulation modes,

the fine grid boundary conditions (i.e., the lateral boundaries) are

interpolated from the coarse grid forecast. In a 1-way nest, this is the only

information exchange between the grids (from the coarse grid to the fine grid).

Hence, the name 1-way nesting. In the 2-way nest integration, the fine grid

solution replaces the coarse grid solution for coarse grid points that lie

inside the fine grid. This information exchange between the grids is now in

both directions (coarse-to-fine for the fine-grid lateral boundary computation and

fine-to-coarse during the feedback at each coarse-grid time step). Hence, the

name 2-way nesting.

The

1-way nest set-up may be run in one of two different methods. One option is to

produce the nested simulation as two separate ARW simulations as described in

the leftmost box in Fig. 3. In this mode, the coarse grid is integrated first

and the coarse grid forecast is completed. Output from the coarse grid

integration is then processed to provide boundary conditions for the nested run

(usually at a much lower temporal frequency than the coarse grid time step),

and this

is followed by the complete time integration of fine (nested) grid. Hence, this

1-way option is equivalent to running two separate simulations with a

processing step in between.

The

second 1-way option (lockstep with no feedback), depicted in the middle box in

Fig. 3, is run as a traditional simulation with two (or more) grids integrating

concurrently, except with the feedback runtime option shut off. This option

provides lateral boundary conditions to the fine grid at each coarse grid time

step, which is an advantage of the concurrent 1-way method (no feedback).

Possible

grid configurations –

A

simulation involves one outer grid and may contain multiple inner nested grids.

In the ARW, each nested region is entirely contained within a single coarser

grid, referred to as the parent grid. The finer, nested grids are referred to

as child grids. Using this terminology, children are also parents when multiple

levels of nesting are used. The fine grids may be telescoped to any depth

(i.e., a parent grid may contain one or more child grids, each of which in turn

may successively contain one or more child grids; Fig. 4a), and several fine

grids may share the same parent at the same level of nesting (Fig. 4b). Any

valid fine grid may either be a static domain or it may be a moving nest (with

either prescribed incremental shifts or with automatic moves via a vortex following

algorithm, such as tracking the minimum of the 500 mb height). The ARW does not

permit overlapping grids, where a coarse grid point is contained within more than

a single child grid (i.e., both of which are at the same nest level with

respect to the parent; Fig.

4c). In addition, no grid can have more than a single parent (Fig. 4d). For

global domains, a fine grid domain cannot cross the periodic lateral boundary

of the parent domain.

For

both 1-way and 2-way nested grid simulations, the ratio of the parent

horizontal grid distance to the child horizontal grid distance (the spatial

refinement ratio) must be an integer. For 2-way and concurrent 1-way nesting,

this is also true for the time steps (the temporal refinement ratio). The model

does allow the time step refinement ratio to differ from the spatial refinement

ratio. Also, nested grids on the same level (i.e., children who have the same

parent) may have different spatial and temporal refinement ratios. For example,

in Fig. 4b, the horizontal grid resolution for domain 1 could be 90 km, for

domain 2 could be 45 km, and for domain 3 could be 30 km.