4.2. Dynamics

The CICE cicecore/ directory consists of the non icepack source code. Within that directory there are the following subdirectories

cicecore/cicedyn/analysis contains higher level history and diagnostic routines.

cicecore/cicedyn/dynamics contains all the dynamical evp, eap, and transport routines.

cicecore/cicedyn/general contains routines associated with forcing, flux calculation, initialization, and model timestepping.

cicecore/cicedyn/infrastructure contains most of the low-level infrastructure associated with communication (halo updates, gather, scatter, global sums, etc) and I/O reading and writing binary and netcdf files.

cicecore/drivers/ contains subdirectories that support stand-alone drivers and other high level coupling layers.

cicecore/shared/ contains some basic methods related to grid decomposition, time managers, constants, kinds, and restart capabilities.

4.2.1. Dynamical Solvers

The dynamics solvers are found in cicecore/cicedyn/dynamics/. A couple of different solvers are available including EVP, EAP and VP. The dynamics solver is specified in namelist with the kdyn variable. kdyn=1 is evp, kdyn=2 is eap, kdyn=3 is VP.

Two alternative implementations of EVP are included. The first alternative is the Revised EVP, triggered when the revised_evp is set to true. The second alternative is the 1d EVP solver triggered when the evp_algorithm is set to shared_mem_1d as oppose to the default setting of evp_standard_2d. The solutions with evp_algorithm set to standard_2d or shared_mem_1d will not be bit-for-bit identical when compared to each other. The reason for this is floating point round off errors that occur unless strict compiler flags are used. evp_algorithm=shared_mem_1d is primarily built for OpenMP. If MPI domain splitting is used then the solver will only run on the master processor. evp_algorithm=shared_mem_1d is not supported with the tripole grid.

4.2.2. Transport

The transport (advection) methods are found in cicecore/cicedyn/dynamics/. Two methods are supported, upwind and remap. These are set in namelist via the advection variable. Transport can be disabled with the ktransport namelist variable.

4.3. Infrastructure

4.3.1. Kinds

cicecore/shared/ice_kinds_mod.F90 defines the kinds datatypes used in CICE. These kinds are used throughout CICE code to define variable types. The CICE kinds are adopted from the kinds defined in Icepack for consistency in interfaces.

4.3.2. Constants

cicecore/shared/ice_constants.F90 defines several model constants. Some are hardwired parameters while others have internal defaults and can be set thru namelist.

4.3.3. Dynamic Array Allocation

CICE v5 and earlier was implemented using mainly static arrays and required several CPPs to be set to define grid size, blocks sizes, tracer numbers, and so forth. With CICE v6 and later, arrays are dynamically allocated and those parameters are namelist settings. The following CPPs are no longer used in CICE v6 and later versions,

-DNXGLOB=100 -DNYGLOB=116 -DBLCKX=25 -DBLCKY=29 -DMXBLCKS=4 -DNICELYR=7 -DNSNWLYR=1 -DNICECAT=5 -DTRAGE=1 -DTRFY=1 -DTRLVL=1 -DTRPND=1 -DTRBRI=0 -DNTRAERO=1 -DTRZS=0 -DNBGCLYR=7 -DTRALG=0 -DTRBGCZ=0 -DTRDOC=0 -DTRDOC=0 -DTRDIC=0 -DTRDON=0 -DTRFED=0 -DTRFEP=0 -DTRZAERO=0 -DTRBGCS=0 -DNUMIN=11 -DNUMAX=99

as they have been migrated to Table of namelist options

nx_global, ny_global, block_size_x, block_size_y, max_blocks, nilyr, nslyr, ncat, nblyr, n_aero, n_zaero, n_algae, n_doc, n_dic, n_don, n_fed, n_fep, numin, numax

4.3.4. Time Manager

Time manager data is module data in cicecore/shared/ice_calendar.F90. Much of the time manager data is public and operated on during the model timestepping. The model timestepping actually takes place in the CICE_RunMod.F90 file which is part of the driver code.

The time manager was updated in early 2021. Additional information about the time manager can be found here, Time Manager and Initialization

4.3.5. Communication

Two low-level communications packages, mpi and serial, are provided as part of CICE. This software provides a middle layer between the model and the underlying libraries. Only the CICE mpi or serial directories are compiled with CICE, not both.

cicedyn/infrastructure/comm/mpi/ is based on MPI and provides various methods to do halo updates, global sums, gather/scatter, broadcasts and similar using some fairly generic interfaces to isolate the MPI calls in the code.

cicedyn/infrastructure/comm/serial/ support the same interfaces, but operates in shared memory mode with no MPI. The serial library will be used, by default in the CICE scripts, if the number of MPI tasks is set to 1. The serial library allows the model to be run on a single core or with OpenMP parallelism only without requiring an MPI library.

4.3.6. I/O

There are three low-level IO packages in CICE, io_netcdf, io_binary, and io_pio. This software provides a middle layer between the model and the underlying IO writing. Only one of the three IO directories can be built with CICE. The CICE scripts will build with the io_netcdf by default, but other options can be selecting by setting ICE_IOTYPE in cice.settings in the case. This has to be set before CICE is built.

cicedyn/infrastructure/io/io_netcdf/ is the default for the standalone CICE model, and it supports writing history and restart files in netcdf format using standard netcdf calls. It does this by writing from and reading to the root task and gathering and scattering fields from the root task to support model parallelism.

cicedyn/infrastructure/io/io_binary/ supports files in binary format using a gather/scatter approach and reading to and writing from the root task.

cicedyn/infrastructure/io/io_pio/ support reading and writing through the pio interface. pio is a parallel io library (https://github.com/NCAR/ParallelIO) that supports reading and writing of binary and netcdf file through various interfaces including netcdf and pnetcdf. pio is generally more parallel in memory even when using serial netcdf than the standard gather/scatter methods, and it provides parallel read/write capabilities by optionally linking and using pnetcdf.