Something curious…

The 20th Century Reanalysis is an effort to use surface pressure observations to create a record of atmospheric conditions through out the 20th Century-in fact, it extends back into the 19th century to some extent. You can in fact download some data from KNMI, but I was messing around with it on the ESRL composites page for it to see if I could investigate what it might hint at for some of those earlier years in my last post (although I was somewhat frustrated by the fact that I can’t change their base period. Anyway, I decided it would be kind of interesting to ask what the near surface layer (1000 mb) temperature trends are over the whole dataset from 1871-2011-or rather, what the difference between the period 2001-2011 and 1871-1881 is in various places. I produced an interesting plot, which I changed the colors on to more clearly show the temperature change’s signs:

20thCenturyReanalysisLongTermChangeInterestingly, it looks like much of the US (and a few other places, including Turkey and much of the region called Levant) show long term *cooling* in this data! Keep in mind, these are *estimates* of the temperature, by a weather model, based on surface pressure measurements and apparently some sea surface temperatures. At any rate, I regarded this as sufficiently surprising as to warrant further investigation: while a reanalysis is likely to contain biases and errors, especially where not tightly constrained by observations or where new data sources are added in or taken out, it nevertheless raises some eyebrows: barometric pressure over the US is presumably very well characterized by an extensive observational network, and presumably has been for a long time. If a realistic simulation of weather processes can then take those measures of surface pressure and accurately estimate the temperature (a vaguely similar idea is not without precedent), then it raises the question whether this could be an indication of a problem with the estimate of the long term temperature trend over the US. At least for the satellite period, I have generally found the US data from NCDC (USHCN) to be pretty good in quality, agreeing well with well supported satellite analyses, but I am open to the possibility that the estimates of long term trends could be off. Now, partly it seemed to me a possibility this was merely because USHCN, going back as it does to 1895, may simply not capture some unusually warm years in the late 19th century in the US (a period of history especially dear to my heart, let me talk about it endlessly to you at some point). Well, NCDC does extend their full temperature dataset for the globe back to 1880, so it is possible to make a comparison of the 20th Century Reanalysis with the NCDC data to find, whence the difference, if any?

First, I focus on the region 24-49N 235-293E, which roughly delineates the US and surrounding coastal areas (and some of Canada and Mexico) and get the NCDC data from KNMI, and the same from KNMI for the 20th Century Reanalysis surface temps-rebaselining both to 1880-2009 to cover all the full years (er, except the last, just realized 20th Century does have all of 2010. Oh well, doesn’t matter much) they share. I then subtracting the NCDC data from the 20th Century Reanalysis data. The result was this residual:

NCDCminus20thCenturyUS1Green is the monthly differences, red is the 12 month running mean, black is the linear OLS trend over the period 1880-2010. Clearly, the 20th Century Reanalysis shows a significantly warmer early US than does NCDC. Specifically, before about 1916, it almost always runs warmer, and it also runs warmer during much of the Great Depression and WWII, but from about the 50’s onwards, it runs consistently cooler, but also pretty flat: in fact, examining the period from 1979-2010, the Reanalsis-NCDC residual warms slightly by ~0.067 K/Century-almost not at all, which is pretty consistent with the conclusion we have had here at Hypothesis Testing: the Data in the US is pretty good over the satellite period. However, because of the differences above, it seems likely to me that the all important question of “warmest year” in the US would probably have been different in this data, at least prior to us having that big temperature spike in 2012-the sort of thing that happens from time to time in most places on Earth, by the way. The question is whether that is an artifact of the reanalysis, or the reanalysis may be capturing a real feature of the climate of the US over the 20th Century. It would be interesting to examine the surface pressure data for evidence of biases and inhomogeneity, and the surface temperature data for possible uncorrected inhomogeneities.

Incidentally, the difference doesn’t seem to be merely a function of the sea surface temperature data or data outside the US. The differences look very similar over the “core” US (31-40 N, 240-280 E):

NCDCminus20thCenturyUS2Anyway, I find these differences quite intriguing. It’s not at all obvious to me why they arise, or which dataset to “believe” better represents the US Temperature record. So anyway. Curious…


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: