In your Kalispell plot, do both datasets you used to compare begin in 1896 or 1899? I ask because the monthly data I find begins in 1896, but the daily data found in http://www1.ncdc.noaa.gov/pub/data/ghcn/daily/42500244558.dly begins in 1899. Right now I am assuming that you are ignoring the years 1896 through 1898, but it is hard to tell from the plot.

In your Kalispell plot, do both datasets you used to compare begin in 1896 or 1899? I ask because the monthly data I find begins in 1896, but the daily data found in http://www1.ncdc.noaa.gov/pub/data/ghcn/daily/42500244558.dly begins in 1899. Right now I am assuming that you are ignoring the years 1896 through 1898, but it is hard to tell from the plot.

John,

If you are looking in hcn_doe_mean_data.Z , the fourth line
labeled 3C is a “confidence factor”, and is not related to
the urban adjustment. The USHCN urban adjusted mean temps
are in a different file: urban_mean_fahr.Z .

Jerry

The 128°F “outlier” from McConnellsville Lock OH back on Jan 2, 1900 would never have been a problem for contemporary weather plotters and analysts back then. They would have simply noted that the three nearest first order stations…Columbus OH, Parkersburg WV and Cincinnati OH all reported a max temperatures of 28°F for that date and you can guess what they plotted for McConnellsville.

This sort of illustrates the value of what was lost when computers programmed without AI oversight replaced thinking humans. Of course computers can do a lot more and much more quickly than people, but the point is that before the computer age, with much less to do and more time to do it, weather observers on the whole were more precise and fussier about their observations then than they are today when the dependence is on machines and electronics. The McConnellsville record was clearly a typo rather than a faulty observation. Where and how it entered into the NCDC data base is a matter of speculation, but you can bet that no observer ever entered that number onto the original record.

When I look at the records for Kalispell, the “4. Urban: these data have all adjustments including the urban heat (UHI) adjustments.” records all have a small magnitude. Does this represent the amount of adjustment? It clearly does not represent the adjusted temperature.

Thank you for pointing me to exactly what I was looking for.

Moving from 138 stations to 1062 by “changing a few” quality rules seems to me to be
a step change of sorts.?

One criteria that interested me was attempt to select stations that had minimal TOBS
adjustments ( not in magnatude but in number) I think it would be interesting from a purely
theoretcial standpoint to look at what trends and errors we get in the US record if we select
Homogeneous stations at the start, rather than trying to homogenize bad stations to
good ones and pretending that we gain “area coverage” thereby. ( I think willis and I have beat this dead horse and are ready to mumify it)

I have been aware of the 138 station collection (NPD042), and
have saved a copy of it, but have not done anything else with it.
I have tinkered a bit with the 1062 station collection (NDP070).

It was in NDP070 that I first noticed my favorite “outlier”:
128 F TMAX in McConnellsville Lock, Ohio, on January 2, 1900.

ftp://ftp.ncdc.noaa.gov/pub/data/ushcn/daily/README

It would appear that in 1992 (H92) they selected the best 138 stations, and then later expanded that to 1062, changing the quality guidelines to gain spatial coverage.

The selection of stations for inclusion
in H92 was performed with the following data quality issues in mind.

1. The degree to which each station maintained a constant
observation time for maximum and minimum temperatures,
excursions from a station’s predominant observing time of no
more than four years being desired.

2. At least 95% of a station’s pre-1951 data should be contained
in NCDC digital daily archives.

3. A station’s potential for heat island bias over time should be
low.

4. Quality assessments based upon the decile ranking assigned by
Karl et al. (1990) to the stations’ monthly maximum/minimum
temperature data for certain quality characteristics.

Since the release of H92, much more work has been conducted at NCDC
involving compilation and digitizing of daily data. However, to enable
the compilation of a database providing better spatial coverage of the
contiguous United States, the four station selection criteria listed
above were not strictly adhered to in the current version of the HCN/D
presented here.”

A revision, and a bit more:

“…, and could not be the source for stations for which TMAX/TMIN data
are not reported.”

should have been:

“…, and could not be the source for stations for which the “official”
TMEAN is based on something other than TMAX and TMIN.”

Regarding USHCN data getting to GHCN, see the update to
http://www.climateaudit.org/?p=2711

Various parts of the US NCDC route data to various collections such as
GHCN and USHCN. The USHCN data go through USHCN processing before they
become “official” and go to GHCN, or get published in the USHCN
directory, on the NCDC FTP server. Then GISS, or whoever wishes to do
so, e.g. CDIAC, can pick them up.