The graphic above, show the “All-Dataset Global Mean Surface Temperature (GMST) Trendline” milestones reaching:

  • 1985 November: Trend line passes 0.50C
  • 2000 April: Trend line passes 0.75C
  • 2012 June: Trend line passes 1.00C
  • 2022 May: Trend line passes 1.25C
  • 2024 September: Trend line reaches 1.32C

Note The “All-Dataset Global Mean Surface Temperature (GMST) Trendline” is calculated by

  • Using all datasets (Noaa, HadCRUT, GISS, Berkeley Earth, Copernicus) – See GMST Data Sets
  • Re-baseline all the datasets to be relative to 1850-1900
  • For each month take the average value of all the datasets
    • Note not all datasets have all the months from 1850
      • E.g. Copernicus starts in 1940 and GISS starts in 1880
  • Use Loess Smoothing, configured to use a 30 year window of data
    • Using python library “loess.loess_1d”
    • Use a factor of 0.1717
      • 30 years is 360 months of data
      • Jan 1850-> Sep 2024 is 2097 months
      • Factor 0.1717 = 360 / 2097

Copernicus Only – Value

Note: The Copernicus Loess-30-year-window value for September 2024 is also listed as 1.35C. This uses the same technique as the “add data” technique, but just uses the Copernicus data, and a factor of 0.354 (360 months / 1017 months).

Trend Line Smoothing options

As per Climate Reporting – Why so many different values – There are multiple global GMST Data Sets sets with slightly different values. The global temperature does jump around year-to-year and therefore it makes sense to apply smoothing techniques to allow the trend to shine through. There are many different “smoothing techniques” available to determine “GMST Trend Temperature”. I am just looking for a value which is calculated deterministically and appears “about right”

The “GMST Trend Temperature” dates I have given above (1985, 2000, 2012, 2022, etc…) are subjective, although after trying various smoothing techniques (using average values over all 5 data sets), the results were always within 1 or 2 years of the values I have put on the graph … for any smoothing techniques that appeared to believably overlap the data points. If you do use a 30-year-running-average and just NOAA / GISSTemp (which run about 0.07C lower than the average since the 1970s), then everything will show up as running around 15 years slower.

Below is the same data, but showing the full 1850-2023 dataset.