New Diva Features
From GHER
(→4.6.0) |
(→4.6.0) |
||
Line 7: | Line 7: | ||
* Optimisations of file exchanges for use with [http://odv.awi.de/ ODV] | * Optimisations of file exchanges for use with [http://odv.awi.de/ ODV] | ||
* Highly optimised new version of the grid generator (typical reduction of a factor 10 in CPU time) | * Highly optimised new version of the grid generator (typical reduction of a factor 10 in CPU time) | ||
- | * Integration of faster error calculations into the 3D/4D version | + | * Integration of faster error calculations into the 3D/4D version (using new <tt>ispec</tt> values) |
* Clearer log files for debugging | * Clearer log files for debugging | ||
Revision as of 12:14, 26 April 2013
Contents |
4.6.0
To come (expected release date, June 2013; beta testers are welcomed):
- Parallel Solver
- Optimisations for large data sets
- Optimisations of file exchanges for use with ODV
- Highly optimised new version of the grid generator (typical reduction of a factor 10 in CPU time)
- Integration of faster error calculations into the 3D/4D version (using new ispec values)
- Clearer log files for debugging
(note: With the availability of gfortran, we decided to abandon the support of pure f77 compilations and this release needs Fortran 90 or Fortran 95 and -cpp preprocessing tools).
Example of performance gain using divabigtest (test with 1.5 million data points and a very fine mesh):
CPU time in seconds in 4.5.1 4.6.0 for - mesh 27 2 - analysis 318 166
4.5.1
Version 4.5.1 was released in March 2013 includes the following new features, partly based on the feedback we got from the DIVA workshop 2012 in Roumaillac.
- Advection constraint with linear decay rate and local sources (such as found for radioactive tracers and river discharges)
- divadetrend now allows one to change easily the order in which detrending is done (for example first years then months or the inverse)
- two new error calculations are provided:
- one quick version divacpme with better quality than the original quick version of the poor man’s error;
- the other divaexerr is an almost exact error calculation much faster than the exact calculation.
These two options will be implemented into the 4D version for version 4.6.0 so that error fields will be available with more reasonable CPU times for final climatology productions
- Simplification of installation and compilation with additional tests of correct installation
- Housekeeping of the code (simplifications, error messages, cleaning up of code, further optimisations, elimination of depreciated tools)
- New documentation largely augmented with examples and new tool descriptions
- Possibilities to call DIVA from other software via system calls, exemplified by a matlab function divagrid.m
- divadoxml adapted to new specifications from IFREMER
4.4.1
Jan 2012 simple scripts replaced by awk
4.3
- 3D/4D version with
- Adapting and automatisation of data selection for DIVA input (including vertical interpolation from ODV4 spreadsheets)
- Loops on depth, parameters, seasons etc
- Parameter optimisation (robustness and vertical coherency)
- Automated plotting interface
- Preparing NetCDF files with CF compliance
- Faster and more robust mesh generation, optimisation of computational load (for incorporation of stripped version into ODV)
- Adequate contour generation from a topography mask (including tools to retrieve topography fields)
- Automated cross validation for choice of analysis parameters (including fast real cross validation)
- Some additional diagnostic tools (fitting of correlation length)
- Data-quality control (spatial coherence)
- Non-uniform correlation length
- Exact error calculation and poor man's error calculation (for coupling with ODV)
- Advection constraint
- Hydrostatic constraint in 3D mode
- Option to eliminate data that are not on the mesh before estimation of L and S/N
- Option to create a data set from which outliers are eliminated (outliers determined from a preliminary DIVA execution)
- Possibility to create a pseudo-velocity field so as to increase correlations along topography
- Option ot use a relative correlation length spatially changing depending on data coverage
- Option to calculate a lower limit for correlation length (based on average data distance)
- Additional output fields (data coverage)
- Detrending (with help of J. Cartensen) to deal with heterogeneous data distribution
- Error calculation for domain integrals (heat content, total biomass etc)
4.2.1
Release: May 2008
Implementation of new tools:
- divaUVtopo: creates pseudo velocity field based on topography.
- adapation of divafit to take the mean distance between data as an inferior limit.
- divadatacoverage now creates a file containing the mean distance (./output/meandistance.dat)