I was just messing about with some model-observation comparisons and just thought I would post some of the results. I don’t claim that I’ve done these properly, so use with care. I will, however, explain what I did, so that it should be clear (and so that people can highlight any errors in what I’ve done). I went to KNMI Climate Explorer and downloaded the monthly tas data (near surface air temperature) for the CMIP5 RCP4.5 runs (selecting one member per model which produces 42 model outputs). Once I selected this, I produced two different outputs; one baselined to 1951-1980 (to compare with GISTemp) and one baselined to 1961-1990 and masked to be -70S to 80N (to compare with HadCRUT4).
I then went to the Met Office and downloaded the monthly HadCRUT4 data. I then went to GISTemp and downloaded the monthly mean global surface temperatures (which requires selecting this at the bottom of the page and then saving the ouput as plain text). I then simply plotted the surface temperature data over the model ouput and also plotted a multi-model mean. The resulting figures are below.
A few additional comments. I don’t know if I’ve done this correctly (these kind of comparisons are invariably a bit more complicated than it may at first seem) but I have tried to compare like with like. Although I have tried to take coverage bias into account when comparing the models with HadCRUT4, I haven’t used blended model output – I’m only using the near surface temperature from the models, while the temperature datasets are a combination of near surface temperatures and sea surface temperatures. I also haven’t tried to produce any kind of uncertainty interval for the models; I’ve simply plotted the monthly model outputs for all 42 models. Therefore, as I said above, if you do use these, use them with care.