Unbiased Cherry Picking

Introduction

Some weeks ago I was discussing on Facebook some of my previous USHCN temperature adjustment posts (e.g., here and here) with a friend of a friend. His assertion was that my use of "cherry-picked" sites to demonstrate that the adjustments in the temperature record supported the notion that the past was typically cooled and appeared to exaggerate a warming trend was invalid. While "cherry-picking" is an important aspect of non-scientific analysis (i.e., pseudo-science), it also seems to be the knee-jerk response of the Climate Change faithful to facts or analyses that intrude on their beliefs. In fact, those the sites investigated previously were chosen first for convenience (I picked the closest site to me that bears on the National Forest where I worked) and, secondly, systematically (to see whether sites close to one another geographically had temperature records that were correlated with one another).

To be sure, I realize that a small selection of sites may not adequately represent the whole universe of USHCN sites, and I never asserted that the ones I selected did; I merely reported that the adjustments tended to be such that they supported a warming hypothesis and that the "need" for those adjustments was unclear. But I decided to take a look at a more randomized selection to see what that would show and whether it also tended to support the warming narrative.

Methods

USHCN Sites

The first thing I did was devise a means of more-or-less randomly selecting USHCN sites in Idaho. I approached that as follows:

1. I listed all USHCN site identification codes in an Excel column alphabetically as shown on the USHCN web site.
2. There are 29 sites, which I numbered 1 to 29.
3. In another column, I generated a random number between 1 and 29 using Excel's "Rand()" function and formatted each to 0 decimal places.
4. I then selected the sites corresponding to the five lowest numbers (there were 2 numbered "4"). These sites were selected*:

The selected sites seem pretty well distributed geographically and and with respect to elevation. It would have been nice to have a more northerly site given the topography and environmental gradients in Idaho, but this is an unbiased selection.

Although the SurfaceStations project seems moribund, it is referenced in the USHCN documentation pages, particularly with respect to site quality ratings. In the USHCN documentation, only the Salmon site is mentioned, which, according to SurfaceStations has a rating of "4" (poor). The poor rating appears to be a function of proximity to buildings and shading. The other sites are apparently not rated.

These are predominantly rural sites, located in the sorts of areas that urbanites would probably consider "the middle of nowhere"; the most urbanized site is probably Salmon, a "city" of about 3,300 residents. Here are Google Earth images showing the locations of the sites based on their documented Latitude/Longitude locations (I have been unable to locate close-up images of the stations themselves).

Ashton

Glenn's Ferry

Lifton

May

Salmon

Analysis

Next I wanted to set up a comprehensive analysis. Not all sites had completely overlapping data records, so I trimmed them to make the site records comparable. To accomplish this I did the following:

1. I computed the adjustment (TAVG-TRAW) for each temperature observation.
2. I decided to use decadal comparisons among sites; that is, average adjustements by decade (e.g., 1910-1919).
3. I decided to start the comparison in 1910; no sites had complete records before that and this method only removed one site (Lifton) from the 1910-1919 comparison.
4. I decided that I wanted to have at least 7 data points for a site in any decade for comparison; this resulted in dropping some sites (most notably Glenn's Ferry) in several decades; all sites were excluded in at least one decade.

From the results (below), I also took looked at the trends in adjusted raw data and computed a few statistics about infilled (i.e., estimated) data.

This is a simple analysis, not a thorough investigation, but should help add clarity to how and when data adjustments dominate the apparent trends in temperature. It is as unbiased as I could think to make it, though, of course, it's restricted to Idaho.

Results

Here are the basic decadal results displayed graphically:

At first glance, there is no obvious tendency to make decreasing adjustments over time; indeed, the largest adjustments were made toward the end of the 20th century. The relatively large adjustment in the 1930s is significant, however, because that was a period generally recognized as hot and dry. Furthermore, the 1990s were also adjusted down despite the strong El Nino warming toward the end of the decade. Of all the sites, only the record at Ashton was mostly adjusted upward, though it was adjusted downward with the others in the decade of the 1980s.

I like numbers and looking at numbers and statistics, but graphics are clearly more easily understood. So here are graphic representations of the data from each site:

Except for Salmon, trends modeled from adjusted data were more positive (i.e., estimated more warming) than trends modeled from raw data (Note: I made no effort to calculate either statistical significance of models or significance of differences between the two models). Interestingly, all sites have infilled data from before the raw data records begin. Although not shown, two sites (Lifton and May) have more than a decade of pre-record infilled data; Lifton actually has more than two decades of such data. Overall, infilled data accounts for 28% of the official adjusted data record, with a whopping 48% of the data for Glenn's Ferry estimated (which, interestingly, was the only site with a negative trend modeled from the raw data).

Conclusions

Picking sites randomly suggests that, even with raw data alone, most Idaho sites probably have modestly increasing temperatures. As I've indicated in previous posts (e.g., here), I would expect that since most people understand that temperatures seem to have been generally increasing since the Little Ice Age (LIA). Generally speaking, however, data adjustments seem to increase warming trends and potentially reverse some that might be cooling. In my opinion, it seems odd that so much data must be estimated, particularly for very old records; I have not been able to determine the need for or accuracy of the either the adjustment or estimation procedures. These are rural sites, none of which seem likely to have large urban heat island (UHI) effects, particularly during the 1930s. Using the historic imagery in Google Earth, it is clear that the areas have changed very little at least over the past two decades (of course, I can't trace station moves that may have occurred). 

___________
* The official names have been shortened for this post. Officially, they are: Ashton 1N, Glenn's Ferry, Lifton Pumping Stn, May 2SSE, Salmon-Ksra.