From this point on, initialization takes one of two paths. Initial runs are set up in subroutine inital, or continuation runs are initialized in subroutine resume. These two paths are roughly parallel in nature and call many of the same routines. The primary difference is that an initial run reads the initial dataset and spectrally truncates prognostic fields (see inidat, below). A continuation run obtains these fields from the appropriate restart files.
The primary job of subroutine inidat is to read the required fields from the initial dataset and spectrally transform the surface pressure (PS), wind (U,V), temperature (T), and surface geopotential fields (PHIS). Other required fields are the surface type flag (ORO), the sea ice layer temperatures (TS1 through TS4), and the moisture field (Q3). If the slab ocean model is enabled, snow depth (SNOWH) and sea ice thickness (SICTHK) are also read in from the initial dataset.
The amount of local data (stack-based) used in inidat is quite
large. Three-dimensional arrays exist for vorticity, divergence, and the
work array required by the FFT (Fast Fourier Transform) package.
The need for three dimensional arrays is partly due to the fact that the
model can start from an initial dataset randomly ordered in latitude. Since
local memory is allocated on the stack (see "Shared-Memory
Management" ), the large size of the local database does not increase
total memory utilization since the local workspace in inidat will
disappear before the time integration begins.
Subroutine advnce, called from scan1bc each timestep, updates current time information (calendr) and time-variant boundary dataset information (subroutines sstint or somint, and oznint).
Current simulation time information is
computed in calendr. The definition of a "year"
is 365 days. There are no leap years.
Subroutine lsmdrv drives the Land Surface Model calculations (Bonan, 1996). These calculations are not multitasked by latitude band, but rather as one big vector which is divided into several subsections of nearly equal length. Computations associated with each of these subsections are then distributed among the available processors. Refer to the document LSM User's Guide for more details.
Subroutine somoce drives the slab ocean model calculations, if the SOM option is enabled. The cpp token COUP_SOM must be set with a #define in misc.h.
Subroutine scanslt drives the semi-Lagrangian transport (SLT) calculations. This is the only latitude scan in which the full dimensionality of the "extended grid" (plond and platd) is required. In addition to the physical data dimensions (plon and plat), the extended grid includes additional array storage before the start and after the end of the physical data in both the longitudinal and latitudinal dimensions. The extended grid must be initialized at each timestep. This is done in sltini. In scanslt, the wind field at time level "n" is used to predict the evolution of moisture and constituent fields, from time level n-1 to time level n+1 using a semi-Lagrangian transport scheme (Williamson and Rasch, 1989). No I/O is done in scanslt, since the prognostic arrays u3, v3, wfld, and q3 are entirely in-core. The local database of subroutine scanslt contains numerous multi-dimensional arrays. Since many of these arrays contain a constituent index, memory use can increase dramatically as the number of transported constituents is increased.
Routine sltini fills the extensions of certain arrays used by the SLT package in the longitudinal and latitudinal dimensions. Only a small amount of computational work is performed in sltini, but the routine is multitasked over latitude bands since it is called every timestep.
Subroutine scan2 primarily converts spectral space prognostic variables back to gridpoint space and computes global integrals of dry and moist mass. The routine multitasks over latitude pairs rather than individual latitudes. The last step in scan2 toggles the time indices n3 and n3m1 after the conversion back to gridpoint space. No copying of the actual data is necessary. Prognostic data that were referenced by the current time index (n3) will become the previous time index (n3m1) while data just computed in time index n3m1 are referenced by the "next" (n3) time index at the start of the next iteration in stepon.
Gaussian quadrature (subroutine quad),
completion of the semi-implicit timestep (subroutine tstep), and
horizontal diffusion calculations (subroutine
hordif) are all accomplished within multitasked loops in dyndrv.
The quadrature and semi-implicit timestep computations are parallelized
over total wavenumber "n" when the cpp token PVP is defined, and
over Fourier wavenumber "m" otherwise. Horizontal diffusion calculations
are parallelized over the vertical index "k".
Routine aphys applies adjustments directly to the prognostic variables rather than computing tendencies. These adjustments include dry adiabatic adjustment, moist convection, and large-scale condensation.
Routines tphysbc and tphysac (called from linemsbc and linemsac, respectively) call the various physical parameterizations which are applied as tendencies. These include cloud computations, radiation, surface temperature and flux calculations, the planetary boundary layer scheme, vertical diffusion, and gravity wave drag. Cloud calculations are needed internally only by the radiation code. Since radiative computations are normally not performed on every timestep, cloud calculations are performed only on timesteps that the radiative transfer code is exercised.
Routine radctl controls both the longwave and shortwave radiative transfer code. While the rest of the code uses mks units, cgs units are used throughout the radiation computations. Timesteps during which the longwave absorptivity and emissivity calculations are performed is controlled by namelist variable IRADAE. Since absorptivities and emissivities can only be calculated on a longwave radiation timestep, IRADAE must be an even multiple of IRADLW. If IRADAE is greater than IRADLW, absorptivity and emissivity (a/e) values are stored in a buffer and read into memory during non-absorptivity radiation timesteps. It is recommended that the history file write frequency, NHTFRQ(1)be an even multiple of IRADAE to ensure that the absorptivities and emissivities need not be written to the restart dataset.
Routine sltb1 drives the SLT constituent forecast calculations. Each call to this routine fills one latitude line of three-dimensional array qfcst. For more detail on the algorithms of the SLT, refer to (Kiehl et al., 1996) .
Questions on these pages can be sent to... erik@ucar.edu .
Search page