In this section, we describe how the time integration
is started from data consistent
with the spectral truncation.
The land surface model requires its own initial data, as described by
Bonan [22].
The basic initial data for the model
consist of values of
and
on the Gaussian
grid at time
From these,
, and
are
computed on the grid using (3.11), and (3.49). The Fourier
coefficients of these variables
and
are determined via an FFT subroutine (3.149), and the
spherical harmonic coefficients
, and
are determined by Gaussian quadrature
(3.150). The relative vorticity
and divergence
spherical harmonic coefficients are determined directly from the
Fourier coefficients
and
using the relations,
Once the spectral coefficients of the prognostic variables are
available, the grid-point values of
and
may be calculated from (3.180), the gradient
from (3.183) and (3.184), and
and
from (3.189) and
(3.190). The absolute vorticity
is determined from the
relative vorticity
by adding the appropriate associated
Legendre function for
(3.117). This process gives grid-point
fields for all variables, including the surface geopotential, that are
consistent with the spectral truncation even if the original
grid-point data were not. These grid-point values are then
convectively adjusted (including the mass and negative moisture corrections).
The first time step of the model is forward semi-implicit rather than
centered semi-implicit, so only variables at are needed. The
model performs this forward step by setting the variables at time
equal to those at
and by temporarily dividing
by 2 for this time step only. This is done so that formally
the code and the centered prognostic equations of chapter 3
also describe this first forward step and no additional code is needed
for this special step. The model loops through as indicated sequentially in
chapter 3. The time step
is set to its original value
before beginning the second time step.