Below is the table that seemed to disappear from my previous post.

My point, David, is that the global and NH trends for the difference between GISS and UAH/RSS temperature anomalies are statistically different than zero for the period 1998-2008 (through September) and that with the agreement of UAH and RSS over that time period would make, in my view, further analysis of these differences critical for climate science. The trend differences are not small.

It also appears that the trend differences (GISS – UAH/RSS) are relegated to the NH. Note also that the correlation for GISS to UAH using monthly data for the SH is relatively much less than for global and NH comparisons – even though the difference trend line for the SH is flat. I rechecked my calculations for the SH and found that the GISS versus UAH correlation (0.55) was correct for that region and the GISS versus RSS correlation was 0.47.

My curiosity centers on 2008, which is on the far right side inside the green oval (the January, 2008 value is +0.31, for orientation). Did something happen around January, 2008 to cause UAH and GISS to diverge? Too early for conjecture but it does deserve some tracking over the coming months.

Re: Kenneth Fritsch (#135),

In the linked post above, David Smith looked at monthly differences in anomaly temperature trends between satellite and surface measurements for the period 2000-2008 (current).

In my previous analysis linked above, I used annual data for a comparison of satellite to surface measurements presumptively to avoid the larger autocorrelations associated with the use of monthly data in regressions. Based on David’s analysis, I calculated much the same statistics as I did in the one linked above. The analysis covers the period 1998-2008 (Sept) and uses monthly data and regresses the data set differences over the 129 months.

Reading the table from left to right I have the temperature anomaly measurement difference used for the regression, the region measured, the adjusted R^2 from the regression, the trend of the difference in degrees C per decade, the unadjusted standard deviation of the trend, the lag 1 auto correlation of the regression residuals and finally the adjusted trend standard deviation based the Santer et al. (2008) correction.

Before discussing the results of the regressions, it is important to point to the much closer match of the monthly data between the satellite (UAH and RSS) than the satellite measurements to the GISS surface measurements. I calculated the monthly correlations for UAH versus RSS and UAH versus GISS and RSS versus GISS for the global, NH and SH regions. The results reported below show evidence for the matches and mismatches noted above.

Correlations:

UAH vs RSS (global) = 0.95; UAH vs GISS (global) = 0.76; UAH vs RSS (NH) = 0.92;
UAH vs GISS (NH) = 0.72; UAH vs RSS (SH) = 0.94; UAH vs GISS (SH) = 0.55

Since the RSS measurements correlate well with UAH, I compared UAH versus GISS and assumed that RSS would compare very much the same. Using the adjusted trend standard deviation one can see that the trends for all the differences are statistically significant except that for the GISS-UAH difference for the SH where the difference shows no trend (p = 0.15).

Steve McI inline comments:

Notwithstanding this insult, on Aug 11, I published a balanced assessment at Climate Audit and concurrently at Watts Up, providing a balanced assessment, leading as follows:
There’s been quite a bit of publicity about Hansen’s Y2K error and the change in the U.S. leaderboard (by which 1934 is the new warmest U.S. year) in the right-wing blogosphere. In contrast, realclimate has dismissed it a triviality and the climate blogosphere is doing its best to ignore the matter entirely. My own view has been that matter is certainly not the triviality that Gavin Schmidt would have you believe, but neither is it any magic bullet. I think that the point is significant for reasons that have mostly eluded commentators on both sides.

And yet Hansen was on record as saying the following:

“The U.S. annual (January-December) mean temperature is slightly warmer in 1934 than in 1998 in the GISS analysis (Plate 6). This contrasts with the USHCN data, which has 1998 as the warmest year in the century. In both cases the difference between 1934 and 1998 mean temperatures is a few hundredths of a degree. The main reason that 1998 is relatively cooler in the GISS analysis is its larger adjustment for urban warming. In comparing temperatures of years separated by 60 or 70 years the uncertainties in various adjustments (urban warming, station history adjustments, etc.) lead to an uncertainty of at least 0.1°C. Thus it is not possible to declare a record U.S. temperature with confidence until a result is obtained that exceeds the temperature of 1934 by more than 0.1°C.”

I have seen statements from John Christy that the UAH data set is independent of the land and ocean surface measurement, like GISS and HadCRUT. I have seen comments from posters here at CA that claim there is a connection of UAH to the surface measurements no matter how vaguely defined, but I have never seen evidence for a connection unequivocally presented. Obviously UAH and RSS use the same MSU measurements and make different adjustments.

The trends are differences are the largest between UAH and GISS with the RSS trends lying somewhere in between. In order to do a proper and simple minded statistical analysis one would be required to determine the more valid data set between the dependent sets UAH and RSS. I used the surface record of GISS because it is the set most analyzed at CA and data from it were more readily available to me on my computer.

I am not prepared to distinguish between the UAH and RSS data sets here, so instead I use a comparison of UAH with GISS as the two data sets giving the largest differences in trends. It should be noted that from a quick perusal of the first listed graph below that subtle differences between the UAH and RSS data sets leads to a substantial difference in the trend, i.e. 0 142 and 0.183 degrees C per decade, respectively. Notice also that the GISS trend differs from both the RSS and UAH trends most visually in the period from 1998-2007. The second listed graph shows that the UAH and RSS trend measurements are in excellent agreement from 1998-2007 while that between UAH/RSS and GISS is considerably less. As a matter of fact, the agreement between GISS and HadCRUT for that period is not that good considering that these measurements are not independent.

The table compares the data source or set, the region and the season for the statistics, listed in order left to right, of the adjusted R squared for the regression of the annual temperature anomalies versus the years from 1979-2007, the trend in degrees C per decade, the standard deviation of the trend, the lag 1 auto correlation of the residuals from the regressions and the standard deviation of the regression residuals (or the de-trended anomalies).

There are the obvious differences noted between the data sets for the trends and R^2 that in turn vary with region of the globe and the seasons. The NH shows statistically significant (using rejection at p equal or less than 0.05 that I use throughout my comments here) larger trend than the SH for all the data sets. A breakdown of the global trends by seasons shows none of these seasonal differences are statistically significant. The difference between GISS and UAH for the NH trend is statistically significant (and after adjusting the standard deviation of the trend for the auto correlation using the factor at the bottom of the table after the treatment used by Santer et al. (2008)).
The largest differences in seasonal trends for the NH occur between winter (DJF with the greater trend) and summer (JJA with the lesser trend). This seasonal difference for the UAH data set becomes very close to being statistically significant after adjusting for auto correlation.

I looked at the standard deviation of the regression residuals with an eye to explaining, perhaps, differences in measuring techniques. Interesting was the small differences between the UAH and RSS residual deviations for all regions and seasons as compared to that with GISS and in the single case of HadCRUT. An F test showed that the differences were statistically significant for only UAH or RSS versus HadCRUT for the global annual trends. The UAH/RSS to GISS differences in some of the comparisons did approach statistical significance.

I conclude this post by suggesting that climate scientists or the organizations in which they are represented if truly interested in getting the temperature trends correct would take a long and hard look at these difference in temperature data sets. Otherwise I see it as “its good enough government work” – and that may well be the case.

For you or anyone looking for the actual location of the Verhojansk station, I finally spotted the buildings up north of town, nowhere near the locations previously given. Enter 67.565295,133.41239 into Google maps. The main building and the instrument field just to the south should be visible at maximum zoom. See the discussion over on WUWT for further on the site.

If three or four changes come out of this incident, what would you have them be? Precisely what changes would we all like to see, and how many changes would it be?

1) Transparency in all things
2) Verification on temperature data from outside source required before data is released
3) Love of science more than love of money, politics, fame, and environmentalism

p.s. I know these are not going to happen. Don’t worry, I’m not holding my breath.

I know I would like to see the mercator projection maps done away with.