(1) The ATLMIX version of the Miami isopycnic model contains a Kraus-Turner 
mixed layer, accomodates 2 independent thermodynamic variables (T,S), and 
supports variable bathymetry and irregular coastlines. The model is 
documented in a paper entitled "Salinity-driven thermocline transients in 
a wind- and thermohaline-forced isopycnic coordinate model of the North 
Atlantic" by Bleck/Rooth/Hu/Smith. (JPO, 1992, in press, Aug. issue). 

(2) Model versions 1.x are snapshots taken at various stages of model 
development as bugs were discovered and loose ends were tied up. In versions 
1.x, temperature and salinity are advected and diffused independently 
throughout the domain. Model versions 2 treats temperature as a diagnostic 
variable in isopycnic layers (but not in the mixed layer!). This cuts 
advection time in half and eliminates the need for a coordinate maintenance 
("cabbeling") algorithm. Detrainment is patterned after Appendix E in BRHS.
Versions 2.2 and higher include diapycnal mixing of buoyancy and salinity.

(3) Horizontal array dimensions are uniformly set to (IDM,JDM), even though 
most fields do not make use of the full allotted space due to horizontal
staggering considerations (Arakawa C grid). Specifially, there are only 
(IDM-1)x(JDM-1) mass points, (IDM)x(JDM-1) U-velocity points and (IDM-1)x
(JDM) V-velocity points. U and V points addressed as (I,J) are actually 
located at grid locations (I-1/2,J) and (I,J-1/2), respectively, relative to 
mass point (I,J). Vorticity and Coriolis parameter addressed as (I,J) are 
at (I-1/2,J-1/2). The distribution of water and land within the rectangle 
spanned by (IDM-1,JDM-1) is defined by the DEPTHS array. DEPTHS(I,J)=0 means 
land.

(4) The two time slots required for all prognostic and some diagnostic 
variables are incorporated into the third (vertical) array dimension. The 
vertical grid index KM [as in TEMP(I,J,KM)] means model level K, time level 
M (as in "mid" time). Letter N [as in TEMP(I,J,KN)] indicates the "new" and 
"old" time level.

(5) We recommend that the code be compiled with all variables preset to 
negative-indefinite. On Cray machines, this is accomplished by compiling 
with the "-ei" option and loading with the "-f indef" option. Grid values 
whose use in the various finite-difference operators is anticipated (and 
only those) are set to zero during model initialization. This provides some 
safeguard against contamination by "spurious" land data in case of 
unforeseen complexities in the coastline shape.

(6) If compiled with the "-ei" compiler option, model version 1 will crash 
upon encountering single-grid-interval coastal inlets or channels. Version 2 
contains code to detect and eliminate such features from the DEPTHS array. 
(With a smarter initialization procedure we should be able to overcome this 
restriction.) Single-point promontories and ultra-slim islands are okay.

(7) There are no restrictions regarding the number of islands, but the 
parameter MS must be adjusted to the maximum number of "interruptions" of 
grid rows or columns by land. Specifically, MS-1 interruptions are allowed.
Interruptions in diagonal direction are governed by the parameter MD. If MS 
and/or MD are chosen too small, the program will stop in Subroutine BIGRID, 
INDXI, or INDXJ during the initialization phase.

(8) The model does  n o t  use a rigid lid; instead, it advances the baro-
tropic and baroclinic solutions in split-explicit fashion. The barotropic 
streamfunction is therefore computed for display purposes alone. If the 
iteration count in the Poisson solver is set too low for the solution to 
converge, the model run itself will not be affected. No attempt is presently 
made to determine the correct streamfunction value along island boundaries. 
We suggest the following quick fix to prevent the Poisson solver from 
setting the streamfunction to zero along island boundaries:
  (a) Change the name of the GINDEX Common block in Subr. POISND to GINDX1.
  (b) Make a copy of Subr. BIGRID which references GINDX1 instead of GINDEX.
  (c) In a copy of the DEPTHS array, replace zero values indicating the 
      presence of islands by small nonzero numbers.
  (d) Call the modified BIGRID routine with the modified DEPTHS array.
The solution produced by the above procedure (which amounts to covering 
islands with a thin sheet of water) should be sufficiently accurate for most 
display purposes.

(9) Numerous diagnostic messages in the code are presently commented out 
("CDIAG") but can be activated as needed. Many of these messages refer to 
a specific grid point (ITEST,JTEST) which the user must define in a DATA 
statement. Subroutine PRTIJ prints clusters of 5x5 grid variables centered 
on ITEST,JTEST in their proper spatial context.

(10) The model grid is defined on a Mercator projection of the earth's 
surface. Provisions for rotating the poles 90 degrees, i.e., defining a 
true meridian as the "equator" of the projection, are made. Subroutine 
NEWOLD/OLDNEW gives the location of (rotated) model grid points in relation 
to a lat/lon grid and vice versa. However, interpolation routines for 
translating fields of variables back and forth are the responsibility of the 
user.

(11) Versions 1.3 and 2.x read 2-degree North Atlantic basin depths from a 
file named DEPTH.51x57 and 7 sets of monthly forcing fields from files 
named FORCING.TAU_X, FORCING.TAU_Y, FORCING.WINDSPD, FORCING.RADFLX, 
FORCING.T_HAT, FORCING.BOWEN, FORCING.PRECIP. These form a complete set of 
boundary conditions that can be used to test-run the model. (Initial con-
ditions -- based on zonally averaged Levitus climatology -- are generated 
internally.) A batch submit file for the NCAR Cray Y-MP is given in 
DEMORUN_2.0.

(12) To eliminate the line numbers appended to the beginning of each line 
in the source code, enter the UNIX editor vi and issue the command 
:1,$s/.....//  (the line numbers are useful for communication between users 
and model administrator).

(13) The model runs on 32-bit workstations, but roundoff errors will be 
noticeable in the "cumulative mass gain" diagnostics. Also, strange salinity 
values may be created on the fringe of massless regions by the transport 
scheme. If these are considered objectionable, the variable CUSHN in 
Subroutine ADVEM should be inflated somewhat (see Appendix D in BRHS paper). 
The cubic polynomial solver, which in version 2.x is part of the detrainment 
algorithm, also may have occasional difficulties on 32-bit machines (which 
have no lasting effect, however).

                                                            Rainer Bleck