ChangeLog

2012/03/04 Reto Stockli (MeteoSwiss)

This file documents changes since the Pheno Analysis Version 1.0

TODO:
- test: weight observations by global pft occurence???
- implement monthly aggregated output
- F90-based pft map and topo map generation
- implement MLEF
- R-based diagnostics instead of IDL (get rid of IDL alltogether)
- generate local-scale PFT adjustment: PFT in state vector!
- add support for GEO and proj.4 projections

2.0.0
-----
- forward model is now in single precision, EnKF runs in double precision
- implemented reprojection module with support for specific (MODIS, rotated pole) projections
- rewritten MODIS module
- included new COSMO driver data
- revised model geographic information to not be based on lon/lat only. Right now
  the model grid operates in the native grid of the specified projection in the site file
- packed meteorological driver data into derived type
- reorganized output to correctly include topography, topography distribution and pft distribution
- rewritten topo, pft, ecmwf, alma and COSMO reading routines with new projection routine
- removed reading of NARR and GEOS data
- tested single precision EnKF. Didn't work. So switched back to double precision
- evergreen species received yearly averaged meteorological stats, so they do
  not obey any seasonality but rather the mean leaf state responds to the mean
  climatological state by e.g. lat and lon

1.84
----
- adjusting parameter sets for better global prediction skill

1.83
----
- revised IDL-based PFT-map generation scripts for Collection 5 MODIS LC/VCF data
- pft map now uses ECMWF-based climatology of Temperature and Precipitation
- revised IDL-based Topography-map generation scripts
- ecmwf path (era interim or era 40 or operational) has to be given in namelist
- FPAR/LAI assimilation data is more thoroughly screened for water pixels
- revised parameter min/max separation: minimum separation of ensemble mean depends on initial variance
- generated script to geographically subset ECMWF netcdf files
- pft and topography data is now part of output files
- switched from unlimited to fixed time dimension (due to bug in Parallel NetCDF 4)
- namelist variable stateobs guarantees that model states are saved with the actual observation pft and hgt coverage
  this allows a fair comparison of obs and mod for analyzed fields.
- merged meteorology daily and multi-day average code into module (avoid temp arrays with intel compiler)
- merged phenology prediction code into module (avoid temp arrays with intel compiler)
- removed tave column in force, observation, state and parameter namelists
- added a per-parameter min/max in init_module.F90

1.81
----
- implemented chilling model for dbf pft's
- adjusted parameters
- removed pft-specific hard coded parameters (only gave trouble)
- rewritten indexing of parameters/states/forcings
- tested with example1 and 2
- added yearly, monthly and daily resubmission to shell scripting
- changed date_start and date_end from integer to character
- bug corrected restart file is always written, except when restartdate 
  is set to 1, then it is only written after the analysis step

1.80
----
- removed src tree
- rewritten observation gridding
- modified subgrid-scale dis-/aggregation method
- improved memory performance
- redone submission scripts
- redone restart procedure
- redone ini/site namelist file structure
- fixed bug in NetCDF output
- many more things

1.72
----
- corrected bug in netcdf_module.F90 : unlimited time dimension works now in Parallel NetCDF4 I/O
- adjusted all ini files for new configuration on CSCS julier system (/store ...)
- added new configuration script: config_julier.sh

1.71
----
- added IDL and R analysis scripts for Swiss and Global climate change study (in progress)
- added option in init_driver.F90 to select a subset of PFT's (e.g. only tall tree's)
- added support for SLURM job submission system
- improved performance of reading MODIS L3 tile data (modis_module.F90)

1.70
----
- implemented fully new Parallel I/O for NetCDF output. In distributed mode, a parallel NetCDF-4 is written.
- removed bug that prevented calculation of hgtlevels in no-prediction processes
- removed bug that prevented to have the model's version number in processes > 0
- parameters are not written in output file when distributed mode is used (nxprocs>1 or nyprocs>1)

1.69
----
- adapted diagnostics to climate experiment

1.68
----
- adapted Lapack detection on the IBM xlf&xlc compiler (bluefire)
- adjusted new photoperiod calculation to restrict daylength to times where sun is at least 5deg over the horizon
- re-tested examples 1 and 2
- created and validated new global parameter set

1.67
----
- completely rewritten phenology prediction and meteorology averaging codes to be self-consistent
  and independent from other modules so that they can be implemented in other models

1.66
----
- moved parameter min/max separation to the assimilation_module.F90 so that the
  separation is also done on processes that do not perform prediction.

1.65
----
- my second son Erich was born. We're very happy!
- adjusted all prior parameter sets to converge to "realistic" solutions with the new LAIMIN/LAIMAX framework
- adapted the Examples to perform a yearly analysis, collecting all observations from a full year, to yield a parameter
  set that reflects constant parameters with seasonal variability of states
- tested examples 1 and 2

1.64
----
- changed from prediction of FPAR to prediction of LAI. This implies that now instead of FPARMIN/FPARMAX
  the two parameters LAIMIN and LAIMAX are needed. FPAR is diagnosed from LAI by use of FPARSAT and LAISAT.
- all parameter files have been adapted in data/ini and data/parameters. GROWTHRATE and DECAYRATE need to be
  scaled since they now describe the dynamics of LAI instead of FPAR
- a minor bug in enkf_module.F90 was corrected that resulted in a crash with nrobs=1 and use_sqrt=.true.
- all examples were rerun and tested. It is recommended for any phenological data assimilation to use
  a yearly analysis (and not at every time step) in order to be consistent with a seasonal cycle. This
  however implies a longer assimilation period and a slower convergence.
- adapted to up-to-date F90 coding styles: character*4 -> character(len=4) for use with gcc 4.5.x and higher

1.63
----
- adapted configure script for better handling of NetCDF, HDF and Math libraries on different platforms
- automatic inclusion of current code version number in files
- option in state, parameter, forcing and observation output for NetCDF add_offset and scale_factor
  and output variable type (byte, short compression etc.!)
- added option to read scale_factor and add_offset in diagnostic scripts

1.62
----
- modified init_driver.F90 to allow global prediction of fractional land area (for the assimilation only
  land points with at least 50% land are chosen).
- updated documentation and two examples

1.60
----
- completely changed file structure, run scripts, configuration data
- implemented the GNU autoconf / automake configuration
- re-tested on Linux, OSX, AIX and other systems including several MPI distributions

1.50
----
- improved idl analysis scripts for the global phenology paper that is currently in preparation.
  this includes pft-specific global maps and the evaluation of the global parameter uncertainty and prediction error
- successfully tested global data assimilation and prediction by use of photoperiod instead of global radiation
- tested all example scripts (4 local and 1 regional data assimilation).
- corrected pft short and long names in utils/pft/makepftclim.pro (the last 4-5 pft's (crops) were misnamed)
- cleaned up directories for upload to sourceforge
- updated documentation

1.40
----
Too many changes have happened without intermediate versions during the period 2009/2010. They can be summarized:
- 35 plant functional types included in regional experiments
- code adapted for better scalability from Laptop to supercomputer scale
- global parameter estimation rewritten for single solution and redistribution to regional domains
- weighted and gridded observations for global / regional parameter estimation
- full 10 year MODIS data assimilation possible (2000-2009)
- publication about global phenological reanalysis with this code is in preparation

1.10
----
- re-tested the 4 local and 1 regional experiments (regional experiment is shifted eastwards by 0.25 degrees)
- added better VPD downscaling (first downscale temperature, then calculate downscaled VPD with large scale mixing ratio)
- corrected bug in ERA-40 reading (ERA Interim was fine)

1.09
----
- added support for Intel Compiler and Intel MPI
- added GEOS4 reanalysis support
- time averaging during assimilation for all parameters possible via parameter file (run/data/parameters.dat)
  higher averaging numbers enable to slow down convergence for parameters that converge too quickly.
- reprocessed example datasets with new IDL codes in utils/ (pft's, topography, ECMWF Interim) 

1.08
----
- removed bug in MODIS observation error calculation
- removed bad NetCDF output name (double definition of RAINMIN corrected)
- new met driver module: cosmo_module.F90 reads COSMO NWP NetCDF surface weather forcing
- added driver_module variables: pollon & pollat for treatment of rotated pole grids
- pft, topo boundary conditions and met driver grids can be on rotated pole grids
  --> If both, pft and topo BC are rotated, then the model and the assimilation 
      also runs on the rotated pole grid, and the output is on that grid as well

1.07
----
- Non-F90 standard system() command removed from all modules. Fully replaced by the glob.c routine 
- Makefile PGI and XLF define statements removed (obsolete, since code is fully portable now)
- fixed bug in pheno_analysis.pl script: did not correctly create runtime directory if it already was present from previous (aborted) run
- all photoperiod parameters are now in units hours [h] instead of seconds [s]
- optional flag simplifying the use of rainfall instead of vpd; or photoperiod instead of rg as predictors in phenology_module.F90
  --> when those are selected then the PHOTOMIN/PHOTOMAX, or RAINMIN,RAINMAX parameters need to be selected for perturbation as well

1.06
----
- length of sitename and gsitename(:) increased from 50 to 100 characters in driver_module.F90 and init_driver.F90

1.05
----
- Automatic yearly job resubmission on NCAR bluefire (AIX, PBS) simplified
- It is possible to run multiple instances of Pheno Analysis on a single host as long as they have a different experiment name. Before, Pheno Analysis had a single fixed temporary I/O directory, while now for each experiment a temporary output directory is created and driver/boundary/output data linked to this directory. Upon model termination, the temporary output directory is removed again
- a new sites file: sites/sites_global.dat with 64 global sites, 4 sites per pft (except water) distributed over all continents that each contain an area with a pft majority for the respective pft
- state assimilation disabled temporarily for both local and regional mode. We now only have parameters in the ensemble matrix since the main aim of this framework is to constrain parameters
- several modifications in the analysis routines phenology_regional.pro and phenology_local.pro were performed

1.01
----
- posteriori state and parameter inflation decreased from 1e-3 to 1e-6 times the initial state/parameter variance in assimilation_module.F90
- total time diagnostics overflow for pgf90 fixed (imcomplete yet) in driver.F90