-----
SETUP
-----

This directory is initially created by running preprocess in the
parent directory.

That creates:

raw-isti/
synoptic/
timeseries/
network.switzerland
station.data

Then, to cut the analysis down to 1950 onwards, cd into synoptic/ and
delete or relocate the files 1[78]* 19[01234]*


----------------
KRIGING RAW DATA
----------------

From ., you can then run ../krig.R and ../test.R (by starting R and
executing source("../krig.R"), etc.)

The ../test.R script generates the six-panel plot used in our
presentation.  It operates on the first file in synoptic/ (so 195001).

The ../krig.R script krigs each timestep file in synoptic/, generating
corresponding timestep files in outkrig0/.  Running ../convert.syn.ts
will then transpose all the synoptic files in outkrig0 and create
timeseries files in eta0/.


The data variables in the raw data files (inputs) are tmax, tmin, and
tavg.  The kriging operates on tavg.  The data variables in the kriged
files are zed, etazed, and eta (zed is the covariate effect, eta is
the smoothed climate field, and etazed is their sum).


!!!!!! NB: There's a typo in ../krig.R; the 'zed' field it outputs is
actually a second copy of eta.  This is fixed in krig.est1.R, but will
not be in any of the copies of krig.R floating around, so check them
before using.



-------------------------
KRIGING TREND-FITTED DATA
-------------------------

The output from the temporal model comes as timeseries files in a
directory named estimate1/. 

First, move everything in that directory to estimate1/timeseries.

Then run ./convert.ts.syn and it will transpose everything in
estimate1/timeseries to create files in estimate1/synoptic.  This is
nearly the same as a chunk of the preprocess script, but it also
filters out the stations that have no input data at a given time.

These are used as inputs to ./krig.est1.R, which is just the same as
../krig.R except for its inputs and outputs.  (And a fixed typo bug;
see NB above.)  ./krig.est1.R will krig each of the files in
estimate1/timeseries and write the results to outkrig1/.

Finally, run ./convert.syn.ts to convert the synoptic files in
outkrig1/ to timeseries files in eta1/.  These can then be passed back
to the temporal model.


The data variables in the temporal model output are y, trend, and
seas.  The kriging operates on trend.  The data variables in the
kriged files are again zed/etazed/eta, which are the
covariate/sum/smoothed-climate, this time of the fitted trend.

