NOAA monthly temperature normals do not match raw GHCN-Daily data - explanation

[wfgarnett3]

Here is NOAA's detailed explanation why the monthly temperature normals do not always match the raw data for any who are interested so you understand why.  Now you know the climate normals charts you see on Wikipedia for cities are based on published normals and do not always match the raw data, and you can see this for yourself at NOWData on weather.gov for the weather stations. 

----------------
The answer to this question is quite straightforward.  Normals are meant to represent the conditions at the time the normals are created. Therefore, they are not constructed directly from the raw data, which often contains issues that make them inhomogenous. In the processing of the 1981-2010 temperature normals, there were three major steps before the monthly normals are processed:
1. Station maximum and minimum temperatures are adjusted for time of observation bias.
2. Station time series are compared in pairwise fashion to surrounding stations to detect inhomogeneities, and are corrected for those that are detected. These inhomogeneities have numerous causes, most commonly shifts in station locations and instrumentation.
3. Station values are then returned to their current times of observation.

After this, both monthly temperature and precipitation time series are "filled" when there are missing values through regression relationships with surrounding stations. Finally, the averages are calculated, and the monthly, seasonal, and annual normals are completed.  Daily normals of temperature and precipitation are then generated through a complex process that constrains the daily normals to the monthly normals, so that they are both smoothed (raw daily averages would be very irregular day-to-day) and add up to monthly precipitation normals or average to monthly temperature averages.

There are other complications with regards to other variables such as degree days, counts above and below thresholds, growing season attributes, etc. All these are described and explained in publications and technical documents on the normals Web site.  The 1991-2020 normals will follow the same approach, and will be available for use by May 1 2021.

The detailed documentation is at: https://www1.ncdc.noaa.gov/pub/data/normals/1981-2010/documentation/
I also attached the main overview publication.

Thank you for your interest and your concerns.

-----------
Original E-mail:

NOAA's own NOWData website shows the published normals do not match the actual data for A LOT of weather stations in the United States - you can see this yourself in one quick minute.

For instance, over here:

https://w2.weather.gov/climate/xmacis.php?wfo=phi

when you choose Philadelphia International as the location, and choose Product "Monthly Summarized Data", with Options 1981-2010, Min Temp, Mean, and click Go you will see the Mean values on your website for each month from January to December exactly match the values from my previous e-mail that was from the raw GHCN data.

However, if you instead choose Product "Daily/monthly normals" on this NOWData website you will see the normal min values are off for all months except October and November just like I said.

When you do "Monthly Summarized Data" for 1981-2010 Max Temp, Mean, the values actually exactly match the published max normals just like I said before.

Thus NOWData itself shows for Philadelphia International the published normals are correct on the Max Temp side for all 12 months, yet are incorrect for 10 months on the Min Temp side.

The situation is reversed for New York's Central Park over here:

https://w2.weather.gov/climate/xmacis.php?wfo=okx

The min temp normals are in line with the mean monthly summarized data but the max temp normals don't match and are clearly incorrect.

You can view many weather stations in the US using NOWData and see varying degrees of inconsistencies between the actual mean monthly summarized data and the published normals both of which reside on the NOWData websites for comparison.

Can someone explain to me why the published normals are not in line with the actual data?