readme

集合卡尔曼滤波(EnKF) 数据同化方法可以避免了EKF 中协方差演变方程预报过程中出现的计算不准确和关于协方差矩阵的大量数据的存储问题,最主要的是可以有效的控制估计误差方差的增长,改善预报的效果。

%% Copyright (C) 2008 Pavel Sakov
%% 
%% This file is part of EnKF-Matlab. EnKF-Matlab is a free software. See 
%% LICENSE for details.

This directory contains Fortran90 code for the QG model for building

(1) a standalone binary;
(2) Matlab binary for conducting a single step of the QG model in when used
    for data assimilation (DA) experiments with the package kalman-matlab.

To build a standalone binary one needs a F90 compiler (tested with g95) and 
netcdf libraries. Edit the Makefile and run:
	make

This should compile the binary "qg" which then can be used for making QG model 
runs. This is a necessary prerequisitive for conducting DA experiments because 
it the initial ensemble is composed of dumps from a long run.

To bild a Matlab binary, in Matlab run the command from mexcmd.m.

To create a collection of samples necessary for DA experiments (by memory):

1. Build "qg" binary
2. Run:
	./qg prm.txt
	./qg prm-ens.txt
3. In Matlab run:
	psi = qgread('qg_samples-r=2e-5,dt=1.0,RKH2=1e-12.nc');
	[n_sample, ni, nj] = size(psi);
	n = ni * nj;
	S = reshape(psi, n_sample, n)';
	save 'QG_samples-true.mat' S n_sample n;
	psi = qgread('qg_samples-r=2e-5,dt=1.0,RKH2=1e-11.nc');
	[n_sample, ni, nj] = size(psi);
	n = ni * nj;
	S = reshape(psi, n_sample, n)';
	save 'QG_samples-ens.mat' S n_sample n;
4. ln -s QG_samples*.mat ../../../samples

The output of the model can be visualised in Matlab e.g.:
	fname = 'qg_samples-r=2e-5,dt=1.0,RKH2=1e-12.nc';
	[psi, q, psimean, qmean, t] = qgread(fname);
To see dump #10:
	nr = 10;
	qgplot(psi(nr, :, :), q(nr, :, :));
To see the mean fields:
	qgplot(psimean, qmean);
To see the movie:
	qgplay(psi, q, t, 0.2);