Mar 02 2008
Correcting for Microsite Errors Using Regional Averages: A Case Study
One of the issues that comes up frequently at Watts Up With That is that siting of the surface stations introduces a bias into the temperature record. All things being equal, any changes in location should have an equal probability of being warmer or cooler than the previous location. The contention is that all things are not equal; there are several reasons why it is not reasonable to assume that siting biases average out. The merits of such arguments will not be discussed here. Instead, I will assume that there are microsite biases in the surface temperature record, and provide a procedure to calibrate individual stations with nearby stations.
[Added: Steve McIntyre has posted on Climate Audit about this post. There have been several threads at CA that have addressed this topic as well.]
[Added 03/Mar/2008: See also Two Wrongs Don’t Make a Right… Unfortunately.]
First, let’s look at the raw temperature record of a random station in the USHCN. When I say random, I mean I had a computer generate a random number, X, from 1 to 1221 (the number of stations in the USHCN), and then chose the station that was in the Xth position in the list of stations. The station chosen was Saguache, Colorado. Unfortunately, this site has not yet been surveyed by surfacestations.org. I then obtained the station history from NCDC.
I used the same methodology as described in a previous post. Basically, I look at the monthly temperature anomaly at station X and compare it to the anomalies at surrounding stations. I chose the closest 10 stations to station X to represent the regional average, as long as they were within 250 kilometers. By subtracting the regional average from the station X data, we’re left with a spatial anomaly due to local effects.
|
This plot show the local effects at Saguache in the raw temperature data. The red lines indicate dates when the station moved. Qualitatively, there was a change in temperature each time the station moved. Some moves had a greater affect on the temperature than others. The station move in the late 1940s had a much smaller change in temperature than move in the early 1980s. This is not unexpected, and NOAA provides a fix for this effect, called SHAP. The SHAP data is no longer available on the NCDC site, so I need to look at the Filnet data (basically SHAP data and filling the bad values).
|
This is the same plot at above, except using the Filnet data instead of the raw data. Some of the station moves have disappeared - like they were supposed to do. However, most of the recent temperature steps still appear in the data. For instance, there were two station moves in 1995 and 1996 that introduced a large positive bias into the record. Then there were three moves in 2002-2004 that cooled the recorded temperature at the station. It’s clear that the SHAP correction did not properly adjust for at least 2 of the station moves since 1995.
Methodology
In order to quantify the correction needed due to local effects, I applied a Gaussian filter with a half-width of 12.5 months. This will remove any month-to-month variations (noise), but keep the longer-term trend (signal). A plot showing the effects of the smoothing can be seen below.
|
The red line is my correction factor. What this plot says is that the some local effect in 1996-96, probably related to the station moves, caused the temperature anomaly at this station to be higher than at the surrounding stations. The correction factor will tend to decrease the Filnet reported temperatures during the period from 1995-2004. The results of the correction can be seen in the plot below. I plotted yearly temperatures here to help remove some of the noise.
|
The correction has increased the temperatures slightly from about 1960 to 1995, and decreased the temperatures from 1995-2004. This will dramatically reduce the temperature trend seen from 1980 to present. The effect this will have on the longer term temperature trend is unknown, and was beyond the scope of this work.
Conclusions
In this post, I examined one surface station record to determine the effects of microsite bias. In doing so, I found that the SHAP adjustment as applied by NOAA does not account for all the station moves in the station history. A simple and tractable correction method is outlined in this case study which uses regional anomalies to correct stations for local effects.
This is the third occasion [Lampasas, TX; Miami, AZ] where SHAP corrections have been documented to not fully account for station moves. Furthermore, this analysis was done on a random station in the USHCN; it was not cherry-picked to prove a point. This suggests the SHAP algorithm does not correct for all microsite issues related to station moves. People using the SHAP-corrected data should be aware that not all microsite biases have been removed, and they should attempt to account for these issues themselves.
(No Ratings Yet)
Loading ...Related Posts:
9 Responses to “Correcting for Microsite Errors Using Regional Averages: A Case Study”
To reduce spam, comments are automatically closed 30 days after the last comment. If you would like to comment on any closed thread, please use the contact form at the top of this page.
Thank you for this investigation, I will endeavor to get this station surveyed as soon as is practical.
There is another station in Colorado of interest that has been surveyed, Dillion. I’ve never written it up, though both the original surveyor Bob Thompson and myself both visited it unaware of each other on the very same day, just a couple of hours apart.
His survey has been posted, mine has not.
Without any comment on my part, or to look at the site survey, I invite you to apply the same methodology there on the data.
Later we can compare notes.
Thank you for your consideration.
Saguache, Colorado. Been there many times on motorrad tours/rides. One time I was trying to make my eighth or ninth pass of the day by heading over North Pass out of Saguache toward Gunnisun. I could see the very dark clouds up ahead along the intended route of travel. Decided to make a run for it anyway as a personal high for passes in a single day was within reach. Got a little ways down the road and the electrical storm started. The once-dark cloud were now deep black and the lightening was truly frightening. Tried to blast thru but then the hail and rain started. The road was hard to see but the ice was rolling all around and making small accumulations at some spots on the road. Finally decided to give up as I was getting wetter and the road slicker and the sky darker ( between the lightening). Dialed back on the throttle and went back to Saguache to try to find a motel room. Was very wet when I arrived. But I was alive.
An extremely tiny place and one for which I suspect there has been reverse urbanization. The station moves are very likely due to station-watchers moving on to greener pastures, as they say. Saguache has nothing to offer relative to long-term personal success and happiness.
BTW, what are the effects of the changes in elevation among the stations that were selected to make the ‘regional average’?
Thanks
Anthony: Absolutely. I’ll post about Dillion sometime this week.
Dan: Part of the process that makes this “simple and tractable” is that I assume that any changes in temperature anomalies (e.g. elevation, latitude, other microsite) will average to zero for the regional average. Thus, I didn’t even look at the temperature time series of the other stations. My reasoning is that since I’m averaging 10 station anomalies, any signal due to a local effect at a station in the regional average is going to be 1/10 the signal of the station of interest. Whether this is a good assumption is open to debate, and should definitely be quantified before being seriously used. Your question is not an unimportant question, it’s just not what I was trying to do here. Perhaps I’ll discuss it in another post sometime soon.
Now this is real science.
The red line is my correction factor. What this plot says is that the some local effect in 1996-96, probably related to the station moves….
Should that be 1995-96?
Atmoz, check out Menne’s paper on change point analysis.
This is supposed to be used sometime soon in ushcn data.
( maybe it already is, it’s not exactly clear)
[...] This bug should not invalidate previous work that used the FILNET data, since that should have all the bad values filled in already. But I’ll need to check that too I guess. [...]
Atmoz, are you sure Mountainair, NM is within 250 km? I make it to be about 250 miles. Haven’t checked any others.
[Reply: No, I'll re-check the distance calculating thingy. Although it would be odd, but not unimaginably so, that I'd select miles over kilometers.]
[...] station that is classified as “good” is Saguache, Colorado. I have previously shown that this station experienced a series of station moves in the 1990s and early 2000s that have not [...]