FIGURE 1: UAH & BOM TEMPERATURE ANOMALIES COMPARED

FIGURE 2: UAH & BOM DECADAL WARMING RATES COMPARED

FIGURE 3: THE CORRELATION TEST

FIGURE 4: THE VARIANCE TEST FOR TEMPERATURE

FIGURE 5: THE VARIANCE TEST FOR DECADAL WARMING

FIGURE 6: A TEST FOR BIAS: TEMPERATURE 

FIGURE 7: A TEST FOR BIAS: DECADAL WARMING

[LIST OF POSTS ON THIS SITE]

1. The University of Alabama Huntsville (UAH) assessments of lower troposphere temperatures from satellite microwave sounding data since December 1978 are widely accepted as the most reliable estimates of regional and global temperatures in the study of global warming and climate change. In addition to global and zonal mean temperature estimates, the UAH also publishes country-specific land temperature means for the USA lower 48 states and for Australia. The utility of these country specific mean temperature estimates is that they may be used to test temperature reconstructions for the same region from the instrumental record particularly so in the context of widespread allegations by climate change skeptics of data tampering by the relevant meteorological organizations to show hotter conditions and/or higher rates of warming as a way of climate change activism. It is universally accepted that the UAH data are direct measurements and not reconstructions and that as such they contain no such reconstruction bias. It is therefore a useful exercise to compare the UAH data with regional reconstructions for evidence of data tampering in the reconstructions. This post is a comparison of the UAH and reconstruction temperatures for Australia in this context.
2. The Australian temperature reconstruction by the Bureau of Meteorology (BOM) is “calculated from a homogeneous temperature dataset (known as the Australian Climate Observations Reference Network-Surface Air Temperature, or ACORN-SAT, dataset) developed for monitoring climate variability and change in Australia.” In a recent work it was found that there is a high correlation of ρ=0.7 between  the two temperature anomaly series that rises to ρ=0.8 when precipitation corrections are made. The distance between the two temperature data sets is estimated with a standard deviation of their differences of σ=0.6. This work is available online [LINK] . The conclusion paragraph of the work says that “The UAH tropospheric temperatures and BOM surface temperatures in Australia are correlated, with similar variability (0.70 correlation). Accounting for anomalous rainfall conditions increases the correlation to 0.80. The Tsfc trends have a slightly greater warming trend than the tropospheric temperatures, but the reasons for this are unclear. Users of the UAH data should expect monthly differences between the UAH and BOM data of 0.6 deg. C or so on a rather routine basis (after correcting for their different 30-year baselines used for anomalies: BOM uses 1961-1990 and UAH uses 1981-2010).” 
3. In this post we extend this analysis several ways. (1) The analysis is carried out for each of the twelve calendar months separately. It is shown in a separate post that the the behavior of temperatures in the calendar months contain significant differences [LINK] (2) A bias test is included because the standard deviation of σ=0.6 found in the reference study does not contain information about whether these differences are symmetrical or whether there is a bias such that one source tends to be higher than the other. (3) The standard deviation test is extended to compare the standard deviation of both sources taken together with the average standard deviation of the two sources taken one at a time. The test determines whether the two sources are very different (4) The study is carried out for both temperature and for decadal trends (the OLS trend in a moving ten-year window that moves one year at a time. (5) Because the study is carried out separately for the calendar months such that all twelve calendar months may be compared, the study is limited to whole years of data with a revised sample period of 1/1/1978 to 31/12/2018.
4. The data to be analyzed are presented one calendar month at a time in Figure 1 and Figure 2. These Figures are GIF images that cycle through the twelve calendar months. Figure 1 presents temperatures and Figure 2 displays their decadal moving trends. The blue lines in these charts are BOM data and the red lines are the UAH data. Some differences between the two series can be visualized in these charts along with significant differences in these results among the calendar months – underscoring the utility of studying temperature phenomena one calendar month at a time.
5. Figure 3 presents the results of the correlation test. The left frame of Figure 3 displays correlations between BOM and UAH temperatures and the right frame displays the correlations between their decadal trends for each of the twelve calendar months. As reported in the reference study (paragraph#2), very high correlations are seen. Also seen in these charts are large differences among calendar months and significant differences between temperature correlations and decadal trend correlations. Generally temperature correlations tend to be higher than decadal trend correlations and the austral spring months of July to November show high correlations in both temperature and decadal trends while February, June, and December tend to show lower correlations. Of particular note is the oddity of a complete collapse in December trend correlations. The relevant source data for December (low correlation) and September (high correlation) are displayed below.
   

   

    

6. Figure 4 and Figure 5 present a standard deviation test similar to that in the reference studies but with several important differences. The left frame of these charts shows the standard deviation for the two series taken together in red and the average of the standard deviations taken separately in blue. The comparison shows whether the temperatures in the two series are very different (stdev together >> stdev separately) or whether they tend to have similar values (stdev together ≈ stdev separately). In the case of temperatures (Figure 4) we find that the values tend to be different but the same test for decadal trends shows no real difference in the magnitudes of decadal trends.
7. The variance tests in Figure 4 and Figure 5 tell us whether the values are different but provide no information on whether the differences are in random directions or whether there is a bias such that the data from one source tends to be higher or tends to be lower than the data from the other source. This information is extracted in the bias tests in Figure 6 and Figure 7. The left frame of these charts shows plots the data for each calendar month on the dame chart across the sample period 1979-2018. The right panel plots the sum of the difference (DIFF=BOM-UAH) for each calendar month across all 40 years in the sample period. If the differences BOM-UAH are randomly distributed around zero their  sum will be close to zero but if there is a bias BOM>UAH the sum will be positive and for the bias BOM<UAH the sum will be negative. The bias test for temperature shows clear evidence of a positive bias such that BOM temperatures tend to be higher than UAH temperatures particularly in April and May and July through September where very high bias values of 17 to 27 are seen. However, the bias test for decadal trends shows no bias with very small bias values < 1 both positive and negative. This result indicates that there may be a systematic methodological difference in how temperature anomalies are computed such that the the bias in temperature anomalies tend to be higher for BOM in a way that does not affect decadal trends. A possible source for this bias is a difference in the reference period used computing temperature anomalies. This finding is supported by the conclusions in the reference study described in paragraph#2.
8. CONCLUSION: Mean temperatures for Australia computed by the BOM in a reconstruction from the instrumental record for the period 1979-2018 are compared with UAH satellite temperatures, a known reliable and unbiased source. The comparison does not show systematic differences that one would expect to find if the BOM reconstruction had been arbitrarily altered perhaps to show higher rates of warming. The bias in the values of the temperature anomalies can be explained in terms of a difference in the reference period used for computing temperature anomalies. It is possible that allegations of temperature tampering by the BOM relate to the data earlier than 1979.