[Dart-dev] [4039] DART/trunk/DART_LAB: Additional corrections and documentation for DART_LAB
nancy at ucar.edu
nancy at ucar.edu
Fri Sep 4 15:22:34 MDT 2009
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/dart-dev/attachments/20090904/e5c69e1b/attachment.html
-------------- next part --------------
Modified: DART/trunk/DART_LAB/matlab/gaussian_product.m
===================================================================
--- DART/trunk/DART_LAB/matlab/gaussian_product.m 2009-09-04 20:42:25 UTC (rev 4038)
+++ DART/trunk/DART_LAB/matlab/gaussian_product.m 2009-09-04 21:22:34 UTC (rev 4039)
@@ -1,7 +1,8 @@
function varargout = gaussian_product(varargin)
% GAUSSIAN_PRODUCT demonstrates the product of two gaussian distributions.
%
-% This is fundamental to ensemble data assimilation. Change the
+% This is fundamental to Kalman filters and to
+% ensemble data assimilation. Change the
% parameters of the gaussian for the Prior (green) and the Observation
% (red) and click on 'Plot Posterior'.
%
Modified: DART/trunk/DART_LAB/matlab/oned_ensemble.m
===================================================================
--- DART/trunk/DART_LAB/matlab/oned_ensemble.m 2009-09-04 20:42:25 UTC (rev 4038)
+++ DART/trunk/DART_LAB/matlab/oned_ensemble.m 2009-09-04 21:22:34 UTC (rev 4039)
@@ -4,14 +4,28 @@
% Click on the 'Create New Ensemble' button to activate the interactive
% observation generation mechanism and lay down a set of 'observations'
% representative of your ensemble. (Think: Some H() operator has
-% converted the model state to an expected observation.)
+% converted the model state to an expected observation.) This is done by
+% placing the cursor near the axis in the plot and clicking. When you
+% have all the ensemble members you want, click in the grey area of
+% the window outside of the white axis plot.
%
-% After you have an ensemble and an observation, choose an assimilation
-% algorithm and click 'Update Ensemble'. The algorithm is applied and the
-% Posterior (blue) is plotted below the Prior (green). Choose 'EAKF'
-% and click 'Update' ... multiple times. Do the same for 'EnKF' and
-% 'RHF'. Hmnnnn ....
+% After you have an ensemble and an observation, click 'Update Ensemble'.
+% The algorithm is applied and the Posterior (blue) is plotted below the
+% Prior (green). The mean and standard deviation of the posterior are
+% also printed on the plot.
+%
+% The type of ensemble Kalman filter update can be chosen using the
+% pulldown menu at the bottom.
%
+% Checking the 'Show Inflation' box will also apply inflation to the
+% prior before doing the update and will print the mean and standard
+% deviation of the inflated prior and the resulting posterior. The
+% inflated prior and posterior are plotted on an axis below the
+% axis for the uninflated ensemble.
+%
+% The 'EAKF' is a stochastic algorithm so repeated updates can be done
+% for the same prior and observation.
+%
% change the Observation Error SD, lay down an ensemble pretty far away
% from the observation - have fun with it.
%
Modified: DART/trunk/DART_LAB/matlab/oned_model.m
===================================================================
--- DART/trunk/DART_LAB/matlab/oned_model.m 2009-09-04 20:42:25 UTC (rev 4038)
+++ DART/trunk/DART_LAB/matlab/oned_model.m 2009-09-04 21:22:34 UTC (rev 4039)
@@ -7,17 +7,22 @@
% assimilation. It is possible to explore assimilation algorithms,
% ensemble sizes, model biases, etc. on-the-fly. The posterior
% of the state is indicated by blue asterisks, the states evolve along
-% a tajectory indicated by the green lines to wind up at a prior state
+% a trajectory indicated by the green lines to wind up at a prior state
% for the assimilation - indicated by the green asterisks. After the
% assimilation, the (posterior) state is indicated in blue and the
% process is ready to repeat.
+%
+% ONED_MODEL opens two windows. A gui control window that also plots
+% the most recent prior, posterior, and observation, and a figure
+% window that plots time sequences of the assimilation, the RMS error,
+% spread and kurtosis, and prior and posterior rank histograms.
%
% The top button alternates between "Advance Model" and "Assimilate" to
% single-step the model. The "Start Free Run" button is useful to watch
% the system evolve and generate estimates from many assimilation cycles.
%
% Since this is a 'perfect model' experiment, we know the true state,
-% the amount of noise added to the observations, etc.; so it is possible
+% the amount of noise added to the observations, etc.; so it is possible to
% calculate the error of the ensemble in addition to the spread. The
% Truth is not (in general) the same as the observation!
%
Modified: DART/trunk/DART_LAB/matlab/run_lorenz_63.m
===================================================================
--- DART/trunk/DART_LAB/matlab/run_lorenz_63.m 2009-09-04 20:42:25 UTC (rev 4038)
+++ DART/trunk/DART_LAB/matlab/run_lorenz_63.m 2009-09-04 21:22:34 UTC (rev 4039)
@@ -16,7 +16,8 @@
% the model space in the immediate vicinity of the True State. The
% smaller view provides the context of the entire model space.
% Both views are fundamentally views 'from above' ... looking down
-% on the Z-axis.
+% on the Z-axis although they can be rotated when the assimilation is
+% stopped.
%
% After you get the feel for a few single steps through the process
% (by repeatedly pressing the 'Advance/Assimilate' button), select
Modified: DART/trunk/DART_LAB/matlab/run_lorenz_96.m
===================================================================
--- DART/trunk/DART_LAB/matlab/run_lorenz_96.m 2009-09-04 20:42:25 UTC (rev 4038)
+++ DART/trunk/DART_LAB/matlab/run_lorenz_96.m 2009-09-04 21:22:34 UTC (rev 4039)
@@ -4,17 +4,22 @@
%
% To demonstrate the analogue to the atmosphere, the model is a cyclic
% 1D domain with equally-spaced nodal points. There are 20 ensemble
-% members in this example, each with 0.1% noise from a random normal
-% distribution.
+% members initially in this example.
%
-% There are several experiments to perform - including generating the
-% true state with one forcing and assimilaing with a model that has
-% the wrong forcing. If there's such a thing as a perfect model
-% experiment with an imperfect model, this is IT!
-%
-% This utility also explores the effect of Localization - the ability
-% to restrict the impact of an observation to a subset of the state
-% vector.
+% The model can be single-stepped through model advance and assimilation
+% steps using the top pushbutton, or allowed to run free using the
+% 'Start Free Run' button. A variety of assimilation algorithms can
+% be selected from the first pulldown. Model error in the assimilating
+% model (an imperfect model assimilation experiment) can be selected
+% with the second pulldown. The localization, inflation and ensemble
+% size can be changed with the three dialogue boxes. Changing the
+% ensemble size resets the diagnostic displays. The figure window
+% displays time sequences of the prior and posterior error and prior
+% and posterior (if assimilation is on) rank histograms.
+%
+% It takes about twenty timesteps for the intially small ensemble
+% perturbations to grow large enough to be seen using the default
+% settings.
%
% See also: gaussian_product, oned_model, oned_ensemble, twod_ensemble,
% run_lorenz_63
Modified: DART/trunk/DART_LAB/matlab/twod_ensemble.m
===================================================================
--- DART/trunk/DART_LAB/matlab/twod_ensemble.m 2009-09-04 20:42:25 UTC (rev 4038)
+++ DART/trunk/DART_LAB/matlab/twod_ensemble.m 2009-09-04 21:22:34 UTC (rev 4039)
@@ -4,11 +4,14 @@
% on unobserved state variables.
%
% Click on the 'Create New Ensemble' button to activate the interactive
-% observation generation mechanism and lay down a set of 'observations'
-% representative of your ensemble. Start out small, say 6 or so.
+% observation generation mechanism and lay down a set of ensemble
+% samples of an unobserved variable (vertical axis) and an observed
+% variable (horizontal axis). The ensemble members are created by
+% left clicking in the central portion of the figure window.
+% Start out small, say 6 or so.
% In this case, some H() operator would generate the Observed Quantity.
% The Unobserved State Variable could simply be some portion of the
-% model state that is not needed by the H() operator, for example.
+% model state.
%
% After creating the ensemble, the correlation between the Observed
% Quantity and the Unobserved State Variable is calculated.
Modified: DART/trunk/DART_LAB/presentation/DART_section1.pdf
===================================================================
(Binary files differ)
Modified: DART/trunk/DART_LAB/presentation/DART_section1.ppt
===================================================================
(Binary files differ)
Modified: DART/trunk/DART_LAB/presentation/DART_section2.pdf
===================================================================
(Binary files differ)
Modified: DART/trunk/DART_LAB/presentation/DART_section2.ppt
===================================================================
(Binary files differ)
Modified: DART/trunk/DART_LAB/presentation/DART_section3.pdf
===================================================================
(Binary files differ)
Modified: DART/trunk/DART_LAB/presentation/DART_section3.ppt
===================================================================
(Binary files differ)
Modified: DART/trunk/DART_LAB/presentation/DART_section4.pdf
===================================================================
(Binary files differ)
Modified: DART/trunk/DART_LAB/presentation/DART_section4.ppt
===================================================================
(Binary files differ)
More information about the Dart-dev
mailing list