Seaice-only estimation in the Labrador Sea
------------------------------------------

Notes:

- The offline code is enabled during compilation with:
  ALLOW_BULK_OFFLINE in EXF_OPTIONS.h

- At runtime turn off time-stepping in data PARM01 using:
  saltStepping = .FALSE.,
  tempStepping = .FALSE.,
  momStepping  = .FALSE.,

- specify SST in data.exf using:
  climsstfile
  climsststartdate1
  climsststartdate2
  climsstperiod

- and turn off surface relaxation to avoid biasing output diagnostics:
  climsstTauRelax = 0.0, 

==============

09-Sep-2005
heimbach@mit.edu, ifenty@mit.edu, menemenlis@jpl.nasa.gov

This is a 2x2 degree configuration of a seaice-only model
in the Labrador Sea (Nx*Ny*Nz = 20*16*23).
The seaice model is forced by monthly-mean atmospheric state.
The lower boundary condition is SST, also provided as monthly-mean
fields (rather than being computed by an ocean component).
In its current form the seaice model is thermodynamic-only, i.e.
#undef SEAICE_ALLOW_DYNAMICS
The estimation problem consists of a least-squares fit of
daily-mean modeled vs. observed seaice fractional area.
Observations are from satellite-borne sounding microwave radiometers.

The setup consists of 2 parts:
1. the forward/adjoint model component which computes the gradient of
   the model vs. data mistif costfunction w.r.t. to the time-varying
   atmospheric state.
2. the optimization/linesearch routine that uses the gradient information
   to provide an improved guess of the control vector.

Instructions to configure:

0. Preparation
--------------
# from model base dir., e.g. MITgcm/
set basedir = `pwd`
cd ${basedir}/verification/lab_sea/
mkdir exe
mkdir exe/optim

1. forward/adjoint model component
----------------------------------
cd ${basedir}/verification/lab_sea/build/
../../../tools/genmake2 -mods ../code_ad_seaice_only
make depend
make adall

2. optimization/linesearch routines
-----------------------------------
# This part needs more attention:
# Compiler options in the Makefile are hard-coded,
# make sure you choose the right ones for your platform
#
# Assume you are in model base dir., e.g. MITgcm/
#
# A: lsopt
# --------
cd ${basedir}/lsopt/
# edit Makefile to choose right compiler
make
#
# B: optim
# --------
cd ${basedir}/optim/
# Makefile in ${basedir}/optim/ needs 4 adjustments:
# 1: -DMAX_INDEPEND=1000000
#    make sure this number is >= number of control elements
# 2: Make sure you've chosen the right compiler for your platform
# 3: Make sure you've chosen the right INCLUDEDIRS for the header files,
#    in this case it is 
#    -I../verification/lab_sea/build/
# 4: Make sure you've got the right set of BLAS libraries
#    some are provided in lsopt/
#    See the README file in lsopt/ on how to get platform-optimized
#    BLAS libraries, and how to best install them.
# If you're done with all these, just type
make
#
# C: move executable and parameter files to executable directory
# -------
cp optim.x ../verification/lab_sea/exe/optim/.
cp ../lsopt/data.ecco ../lsopt/data.ctrl ../verification/lab_sea/exe/optim/.

3. Run state estimation
-----------------------
cd ${basedir}/verification/lab_sea/exe/optim
ln -s ../data.optim .
cd ${basedir}/verification/lab_sea/exe
ln -s ../build/mitgcmuv_ad .
ln -s ../input_ad_seaice_only/* .
ln -s ../input/*.labsea1979 .
#
./run_estim.csh &