Surface weather station monthly mean temperatures estimated to differ
from 1961-1990 station monthly mean by at least 10 C
 
This file (and others) may be updated from time to time.
Last updated September 3, 2005
 
Using simplifying assumptions to deal with the gaps, and conflicts, in
GHCN "raw" mean monthly temperature data from January 1800 until mid
2005, the program bd.for estimated 1,809 monthly anomalies of 10 C (18 F)
or more.
 
1451 ranged from -10 C through -21.147 C, and 358 ranged from 10 C
through 19.47 C, in data from 568 stations in 34 "countries" (GHCN lists,
for example, Antarctica as a country), none of the stations being within
28 degrees latitude of the equator.
 
Over half, 938 instances, occurred among 358 USA stations including 106
at 16 stations in Alaska.  About half of the remainder, 436 instances,
occurred in 50 stations in what GHCN calls the Russian Federation Asian
sector.  Only six Southern Hemisphere stations made the cut with a total
of 12 instances.  (Of the 7280 GHCN stations, 1079 are in the Southern
Hemisphere.)
 
The monthly distribution of the 1,809 instances is:  Jan 617 Feb 607
Mar 118 Apr 20 May 3 Jun 6 Jul 8 Aug 2 Sep 2 Oct 2 Nov 117 Dec 307.
 
Various distributions may be skewed by the relatively large number of
weather stations in the "lower 48" states of the USA that are included in
GHCN.
 
As an indication of conflicting information in historical temperature
data, a tabulation of 2,005 conflicting monthly means which differ by 4 C
or more is in the file multidat.ge4.
 
Files:
 
readme.txt    this file
bigdif.out    List of large monthly anomalies (see below for format)
stations.bd   GHCN station data for stations in bigdif.out
country.bd    GHCN country codes for stations in bigdif.out (except 701)
bigdif.loc    combination of bigdif.out, stations.bd, and country.bd
stations.ll   Like stations.bd but preceded with counts per station, and
              sorted loosely by latitude and longitude
bd.for        fortran program that did the calcs (see caveats below) (updated 8/17/2005)
**** bd.for was missing the last year of each station data. corrected 8/17
rawmean.zip   PKZIP compressed copy of input file: GHCN "raw" monthly
              mean temps sorted by station and date
multidat.ge4  List of conflicting monthly means in GHCN "raw" means file
md.for        fortran program that produced multidat.ge4 (updated 9/3/2005)
**** md.for was missing numbers such as 3.9999. corrected 8/18
**** added counts of means, and counts of different means, for each month/"station" 9/3
 
Format of bigdif.out:
 
Station     Year Mo Num> 19611990  ThisObs   Difference
                    Obs  Mon Mean
 
Station    GHCN station number
 
Year       Year
 
Mo         Month
 
Num>       Number of months of data for calculating 1961-1990 mean
Obs
 
19611990   Calculated 1961-1990 mean for that calendar month
MonMean    at that station
 
ThisObs    Mean temperature calculated to differ from 1961-1990 monthly
           mean by at least 10 C
 
Difference Calculated difference from 1961-1990 monthly mean
 
Caveats:
 
Over a period of years, a "station" may have been in different places,
in other words, may have been several stations.
 
GHCN temperature files have many gaps among the data, and also often
have multiple "observations" for the same month/station.
 
For comparisons of monthly observations with a 30 year monthly mean, the
most critical gaps are those in the years used to calculate that 30 year
mean, 1961-1990 in this case.  Of the 1,809 instances reported, 341 are
based on "30 year" monthly means for which there are fewer than 30 years
of data for that calendar month during the years 1961-1990.  Instead of
skipping such instances, the program bd.for reports the number of months
on which the "30 year" monthly means was based.  On the other hand, if
the data for a station begin after 1961, or end before 1990, the program
skips that station.
 
Another problem is how to deal with conflicting data, i.e. multiple
"observations" for the same station/month.  The program bd.for simply
averages them.
 
The input file is sorted slightly differently than the original GHCN
"raw" monthly mean file.  The original file is sorted so that
multiple observations for each station are treated as successive series.
In order to facilitate averaging of multiple observations, the input
to the program is sorted in strictly station, date sequence.
 
The input file is a sorted copy of v2.mean.zip at:
ftp://ftp.ncdc.noaa.gov/pub/data/ghcn/v2/zipd/
or v2.mean.Z at:
ftp://ftp.ncdc.noaa.gov/pub/data/ghcn/v2/
 
Considering the problems of the input data, and the simpifying
assumptions of the program bd.for, the output should be regarded as
rather loose estimates.
 
Jerry Brennan
August 16, 2005