Tests results of several experiments on temperature network Tm2

Several experiments to test the influence of different factors have been carried out on the temperature network Tm2, which has an intermediate level of cross-correlations:


Different shapes of the seasonality

Anomalous biases in the series can be constant or show a seasonality (have different values along the year). Four different scenarios have been tested to check the performance of the homogenization packages under different conditions:

  1. Biases without seasonality. Example case: A thermometer replacement with a different calibration. (Already shown in the "first five experiments", these results are displayed here for comparison with the other cases.)
  2. Biases with a sinusoidal seasonality. Example case: The surroundings of the station were irrigated in the past, but no any more, due to a change in land use. (Already shown in the "first five experiments", these results are displayed here for comparison with the other cases.)
  3. Biases change along the year with the form of a squared wave. Example case: A change of albedo with a different effect in the dry and rainy season in a savanna climate.
  4. Biases with a sinusoidal seasonality. Example case: A change of albedo in a climate with two distinct maxima in the monthly precipitation.
None Sinusoidal Squared Double sinus.
Root mean square errors
Trend errors

Influence of sample size

The influence of sample size is tested with samples of 10, 20, 40 and 80 series of the Tm2 master network, with random inhomogeneities in all series with sinusoidal seasonality of random amplitude. (USHCN could not be tested for 80 series because the compilation was done for a maximum of 40 series, and new compilation attemps were unsucessful.)

10 series 20 series 40 series 80 series
Root mean square errors
Trend errors

Concentrated shifts of 2°C

Changes in observational methods or instruments can take place during a short period of time and affect a high proportion if not all of the stations in a network. Examples can be the transitions from an old fashioned thermometer screen to the Stevenson type or, more recently, from manual to automated stations. When these changes are applied simultaneously to all the network, no relative homogenization procedure can detect the biases, since they would appear as part of the climate variability. Here 2°C shifts have been added, uniformely distributed over a decade, to 40, 70 and 100% of the Tm2 series with random inhomogeneities and random amplitude sinusoidal seasonalities. (Results without these added shifts are shown in the left column for comparison.)

0 % 40 % 70 % 100 %
Root mean square errors
Trend errors