This directory contains results of an atmospheric classification project
supported by the Department of Energy Atmospheric System Research (ASR)
Program.

The project involved subdividing reanalysis output (centered on ASR 
facilities in the U.S. Southern Great Plains (SGP) and Darwin, 
Austrialia) into a finite set of weather-types or, as I call them,
atmospheric states.   The identification of the states is based on
a neural-network clustering algrorithm and statistical tests which
ensure that each state has a distinct profile of cloud occurrence
and a temporally stable profile over time.  In effect, the states 
represent a mapping between large-scale dynamical conditions and 
clouds-observed at the ASR sites on much smaller spatial scales.

Details on the algorithm have been published by Marchand et al. 2009 
and Evans et al. 2012. These papers and updated descriptions of the 
atmospheric states are included.   
 
The two principal outputs of the classification are (1) a time series
of states and (2) a set of state definitions.

(1) Time Series

The time series simply indicates which state the atmosphere was 
found to be "in" at a specific time.  The construction of the
states and the time series provided here are both based on 
ECMWF interm reanalysis.  A comparison of results using
states based on NCEP reanalysis showed little difference in
either the identified states or the resulting time series.  These
data are provided as a simple text file and as a Matlab file.

(2) State Definitions

Each state represents a specific weather pattern, and we say the
atmosphere is "in" state #X when then weather pattern at a specific 
moment more closely match that of state #X than any other state.
Also included in the SGP and Darwin directories are figures and
data files (with NetCDF and Matlab versions) showing the composite 
reanalysis data associated with each state on a set of fixed-pressure 
surfaces.

NOTE:  While in principal one could use these "fixed pressure"
definitions to identify states in climate or other large-scale 
model output, these figures represent a mean picture rather than
the true state definiitions which involve a normalization (or weighting
factor).  Those users interested in apply these state definitions
to specific model output (e.g. a GCM) should contact
Dr. Roger Marchand (rojmarch@u.washington.edu) for additional
data and information on how to do so correctly.