CODE AND DATA TO REPRODUCE BUOY-ONLY COMPARISON TO ERSST V3/4

The code assumes a unix-like (linux or mac) system, with c-shell
(csh/tcsh), python 2.6 or 2.7 and the numpy python library.
wget is required if you want to work from the raw data.

For windows users there are various Unix toolkits which may work,
otherwise install Linux in a virtual machine such as VirtualBox
or WMware.

The intermediate gridded data files for ERSSTv3, ERSSTv4 and the
buoy-only reconstruction are provided so that you can perform a
quick check with no additional downloads, however if you want to
reproduce the calculations from the raw data then you will need
to run the get scripts. The icoads data in particular are very
large (44GB).

The scripts are as follows:

compare.csh
  - produce the buoy-ERSST comparisons using the pre-prepared grids
get_ersst3.csh
  - download and prepare the ERSSTv3 gridded data
get_ersst4.csh
  - download and prepare the ERSSTv4 gridded data
get_icoads.csh
  - download and prepare the buoy-only gridded data (SLOW)

Run with:
 csh compare.csh >& compare.log &
or similar.


FUTURE DEVELOPMENT

The current buoy-only calculation is probably the simplest credible
calculation, with the aim of enabling others to develop it further, as
well as making the analysis of possible causes of bias as simple as
possible.

The main sources of bias are probably:
 - the diurnal cycle (small for most SSTs, large cycle records omitted)
 - the annual cycle (currently handled by post-gridding anomaly calc)
 - gridding bias (due to large grid cells, not handled but should be random)
 - inter-buoy bias (initial survey suggest most but not all buoys agree well)
 - coverage bias (avoided in this study by masking ERSST and matching baseline)

The annual cycle still plays a role for buoys with incomplete record (rare)
or which change cells during a month (common, but in the simplest case the
bias in the two cells will cancel, so noise like).

Annual cell bias and gridding bias could be better handled by converting
to anomalies *before* averaging. This requires the construction of an SST
climatology. My current plan is to do this using spherical harmonics at
month centres, and interpolating between months. This could also
potentially be used to investigate inter-buoy bias, however the
divergent regional behaviour of ENSO etc may prevent this.
