The following figure has been commented by Tony Heller and he points out that NASA does not display the cooling with 0.4 deg centigrades from mid 1950 to end 1970 which the NCAR graf displays

failed

Note that it says "Five year average"! This then means that a floating average of five years is used in order to smooth out random variations between years

It further says "Northern Hemisphere only"! This means that the NOAA stations in USA have been used complemented with a few others in Europe and maybe Japan

The yearly mean temperatures over the whole area of USA as provided by NOAA's data base were as follows
The graph displays yearly average temperature for the whole of USA. Data for 1900 - 2018. The rectangle represents 1899 - 2019 and 9 - 15 centigrades The average of these yearly (average) temperatures was 11.75 deg centigrades. The red lines represent 10.75 deg 12.75 deg Data from NOAA "Daily" database (maximal and minimal temperature every day in between 500 and 1000 positions noted and archived)

It can be seen that for 120 years the yearly temperatures have been varying between 10.75 deg and 12.75 deg without any visible tendency to either increase or decrease

In order to detect possible minimal trends one could as in the NASA/NCAR graf use floating averages formed by 5 consequtive years

In that way the following graph is obtained. Now the temperatures are displayed as a curve instead of as dots to be similar to the NASA/NCAR graf

10.5 11.0 11.5 12.0 12.5 13.0 1900.0 1910.0 1920.0 1930.0 1940.0 1950.0 1960.0 1970.0 1980.0 1990.0 2000.0 2010.0 2020.0

It appears to have been a minimal tendency of temperature increase in the period from 1900 to 1934 of roughly 0.5 deg followed by a decrease with the same amount in the period from 1934 to 1970. From 1970 to present time there appears to again have been an increase similar to the one in the period 1900 to 1934

Anyway the NASA/NCAR graf must be "an artist's impression"! No way could such a smooth curves result from forming floating averages over 5 years! Is possibly some kind of "piecewise polynomial approximation" applied?

Computations by Mats Rosengren (Twitter: @stamcose)