EnKF_analysis

EnKF analysis routines in Fortran 90. Stochastic and SQRT formulations with subspace inversion.

lib: The EnKF analysis routines and a makefile for compiling a library libenkfanalysis.a The analysis library contains the basic update algorithm using a variety of inversion schemes. We currently include stochastic and SQRT formulation with different ways of computing the inverse of (SS'+(N-1)R). Using ensemble subspace inversion with representation of R in terms of measurement perturbations provides for very efficient (linear in measurements) update computations even with correlated measurement errors. An efficient "exact inversion scheme" is also provided for the case with diagonal R. The library also contains support for multiplicative inflation including an adaptive variant where the inflation factor is computed to minimize variance reduction due to spurious correlations.

It is recommended that the analysis routine is called from the enkf routine located in the test directory. This routine includes support for distance based and adaptive localization, it scales matrices for stable computations in the case with variables of varying magnitudes.

support: Some additional routines for defining a model state and reading/writing it to file.

test: A test or example program used to verify the analysis scheme with different options. These routines are used to verify some basic properties of the different analysis schemes as well as studying their numerical behaviour.

Installation:

git clone git@github.com:geirev/EnKF_sampling.git
git clone git@github.com:geirev/EnKF_analysis.git
pushd EnKF_sampling/lib
make BUILD=../../EnKF_analysis/build
popd
pushd EnKF_analysis/lib
make BUILD=../../EnKF_analysis/build
popd

Then compile and run the test program

cd EnKF_analysis/test
make BINDIR=$(HOME)/bin
testEnKF

Note that for the test made here, both lib and test files are compiled in the same build catalog. Thus, as long as lib is compiled first, all the library module files are available for the test program.

In a real application I suggest that you change the build dir in the lib/makefile to correspond to the build dir of your application, and run make to install libanalysis.a and the .mod files in that directory. See, e.g., https://www.linuxquestions.org/questions/programming-9/fortran-90-and-shared-libraries-699152/