Drifts 12: analysis of barometers based on KNMI series as reference

[DavorJ]

This analysis is similar to #48, but here we use the #56 KNMI series as reference.

Here is a sample plot:

The top left is the comparison between the KNMI series (red) and some barometer (black). Each series is centered around the median for a better comparison. The black is plotted over the red, so that if there is no visible shift, the red should hardly be visible.

The bottom left curve is the difference between the two. This is a good visualization for the drift. For visual convenience I have added a black horizontal line which is the median of the first 25 % of the differences: assuming the first 25 % is correct, the rest should not deviate much from it.

The bottom right histogram is a summary of the bottom left differences. And the top right is a summary of the histogram.

This is how a good barometer looks like:

I have compiled all the graphs in a single pdf, going from worst case to best case (based on whisker range of the differences).

ToDo: compare this analysis with #51 and #55.

[fredericpiesschaert]

I like these graphs, they seem to give a good indication when things start to go wrong. Perhaps outliers should be excluded because they obscure possible drift signs, e.g. I see something happening at the end of the series but the outliers at the start spoil the fun

[fredericpiesschaert]

I see at least four different things: 1) all well, 'perfect' series, difference and variance are stable and acceptable 2) steady difference but increasing variance this seems to happen quite a lot and often very suddenly: 3) seasonal component 4) what I would call true drift:

it would be nice to understand the physical meaning of these cases, especially case 2 and 3 (on the assumption that it is not some kind of a data-artefact that we are overlooking)