date: Wed, 28 Oct 2009 06:03:40 +1100
from: "Don McNeil" <dmcneilatXYZxyz.mq.edu.au>
subject: additional outliers
Following your suggestion, we've found about 20 more unusually high SST values between 1909 and 1923 in our region 78 (between latitudes 45 and 65 south and longitudes 20 degrees west and 20 degrees east) and when these are omitted the graph is much clearer. We're now finding that the global temperature increase from 1909 to 1944 almost exactly matches the increase from 1977 to 2008 in terms of its overall gradient (0.167 degrees per decade) but the individual regions have different patterns in the two periods, and the correlation between them is not high. We're still working on this and will send you a copy of the statistical analysis in due course.
>>> Phil Jones <p.jonesatXYZxyz.ac.uk> 10/27/09 7:35 PM >>>
There are good physical arguments for omitting SST values near to
or taken under sea ice. Determining SST normals (for 1961-90)
anywhere near sea ice is a problem, as there are few measurements and
they are more likely to be from warmer years/seasons.
You could flag outliers in terms of standard deviation units (all
based on the common period of 1961-90) and remove them in this way.
Over land we've been through many of those that exceed absolute
values of 4 sigma. Many are correct as they are confirmed by
neighbours. Many occur near sea-ice boundaries. Stations on the north
coast of Iceland have some very large sigma departures when very
occasionally sea ice reached the north coast in the past.
Also the more continental the land temperatures the more likely
it is in higher latitudes for the distribution to be slightly
negatively skewed especially in winter months.