In yesterdays posting, we saw how an excursion of 80% of record ranges would provide more than enough tracking error to cause biased analysis. (Tracking error of the ‘excursion’ from a ‘full average’ for temperatures in a city such that the analysis done by folks using The Reference Station Method or comparing a “Grid Box” in one time to a box made from different thermometers in another time could easily have large errors).
But what if the excursion is smaller? I mean, really, how likely is it that you could ever get to 80% of a record?
OK, here are the graphs for a 20% excursion regime. Just a little bit away from the overall average. I’m going to start with the graphs that show the ‘tracking error’ as yellow and magenta with the overall average as green and the 20% of record excursion averages just each side of it. To understand what these graphs are showing, you will need to read the prior posting. Here I’m just going to assume you’ve done that.
San Francisco vs Sacramento
We will start with the comparison that had the weakest power in the prior 80% charts. What happens between these two nearly sea level cities just 1/10 th the maximum distance used for data creation with The Reference Station Method?
Even with a measly 20% excursion, and event that ought to happen with some fair frequency, we have an effect. That “cold phase PDO” when we have lots of volatile stations vs warm phase when at the peak we move to low volatility stations has ‘something to grab on to’ even from these two cities. The yellow and magenta range out to about 10 F each side of 0 at peak, with an average of about +/- 5 F over the whole year. More than enough to create a 1/2 C ‘error’ in the analysis as done by other folks. If they do not allow for the volatility effect, they have no idea what they are measuring with their ‘in-fill’ and “homogenized” fabricated data products.
Sacramento vs Reno
How about comparing that inland near sea level with just a couple of hours drive “up slope” in the mountains? I, like millions of others, have driven from Sacramento to Reno for an evening and driven home the same day (or night, or weekend… depending on how fast you gamble your money away 8-)
Our annual profile swaps so that the winters now have the ‘nearly 10 F’ tracking error while the summers are much closer (they both get darned hot in the summer sun and heat tends to self limit about 105-110 F while in winter Sacramento can sit under “tule fog” for weeks while Reno can be clear skies in a high desert radiating away heat down to very very cold). But we still have a tracking error that runs to both sides of zero during cold and hot phases.
San Francisco vs Reno
And what will happen when GHCN drops out ‘nearby’ cities like Sacramento so codes like GIStemp need to “reach” further, out to as much as 1000 km, to create missing data via The Reference Station Method? Here we look at San Francisco (where the thermometer survives) vs Reno (well inside the GIStemp data fabrication range).
A fairly consistent 10 F of tracking error between either cold phases or warm phases and the overall average. So we have a load of high volatility stations in the GHCN during a cold phase of the PDO, then leaving as we move toward the top of a hot phase. And this is substantially ignored. It would be nearly trivial to get 1 F of “warming” out of this tracking error. Simply mitigate only 90% of it. Even a slight failure to be perfect is sufficient to “create Global Warming” out of nothing but station volatility changes and natural hot / cold cycles like the PDO, AMO, AO, etc.
An interesting “dig here” would be to see if the European station arrival / departure dates match the AMO and if other countries match their local oscillations. That kind of ‘perfect accident’ would be a very interesting bit of evidence. An innocent process ought to have calendar dates of thermometer inclusion / drop that are not so ‘tuned’.
But in any case, the use of a simple average for fabricating missing data is clearly subject to significant tracking errors if it does NOT take into account the variations in correlation between the two locations as major meteorological regimes shift. Regime change matters.