Comparison between global surface temperature anomaly datasets – Take 2

Guest essay by David Dohbro

Recently WUWT published my comparison of several different land-based and satellite-based global surface temperature anomaly (GSTA) datasets (1, and data sets’ descriptions and references therein). However, my analysis was critiqued for not using the same baseline period for each data set. Namely, HadCRUT4 refers to the official WMO period 1961-1990, while NCDC and GISS instead use the periods 1901-2000 and 1951-1980, respectively, which results in higher positive temperature anomalies for these two estimates. In addition, UAH references to 1981-2010 while RSS’ output is calculated relative to a 1979–1998 average.

Paul Homewood assisted me in re-analyzing the data by showing how to adjust a GSTA data set to the same baseline/reference period, for which I am very thankful. In addition, it needs to be restated here that this is not an exercise to determine which dataset is better than the other, it is merely assessing how (well) each compare and if there are temporal trends in the difference between each data set.

Here UAH’s data set is used as a reference. Any other data set could have been used as a reference, but since the satellite-based datasets start in the year 1979, whereas the land-based datasets in some cases go back to the year 1850, comparison of the different types of data sets prior 1979 is impossible. As mentioned, UAH references their GSTAs to the 1981-2010 period. All other 4 datasets’ GSTA were first adjusted to that same period for each monthly GSTA. Then the UAH monthly GSTAs were subtracted from the corresponding monthly GSTAs of each other dataset. From that, a 12-month running average (12-mra) was calculated (Figure 1). A value of 0 means that the UAH data and the other dataset are similar, a value >0 means the other dataset report a higher monthly GSTA compared to UAH and vice versa. The 12-mra data was analyzed for normality (Kolmogorov-Smirnov test, Anderson-Darling test, and Lilliefors-van Soest test), and regression analyses (none-parametric: Spearman Rank, parametric: linear-regression to determine slope) were performed on the data for trend analyzes (table 1). Figure 1 shows the linear-regression lines’ slope and R-square value only.

Figure 1.

None of the data was normally distributed (Table 1), hence none-parametric regression analyses was conducted to assesses how well the relationship between difference and time can be described using a monotonic function. It follows from figure 1, that the land-based GSTA datasets compared here follow the same general trend over time; increasing positive divergence with UAH, whereas the other satellite-based GSTA dataset (RSS) shows an opposite and even stronger trend.

Table 1: Statistics of the 12-mra from the difference between several GSTA data sets and the UAH’s GSTA data set.

However, the 12-mra of the difference between UAH and NCDC, when normalized to 1981-2010, is not increasing statistically significant indicating that the difference between these two datasets to date has not increased over time (linear regression slope is also close to 0). The 12-mra of the difference between GISS, HadCRUT4 and UAH, respectively, is, however, increasing with almost identical slopes (figure 1, table 1).

This means that if this trend continues, these two datasets will in the future report monthly GSTA that are increasingly higher than UAH’s. RSS on the other hand will in the future report monthly GSTA that are increasingly lower than UAH’s and that of the three land-based data sets. Albeit the fact that RSS and UAH use data retrieved from the same systems; the National Oceanic and Atmospheric Administration satellites microwave (and Advanced) sounding unit (MSU, and AMSU), the trend observed here suggests that computational and/or other data-treatment differences may play a role. In addition, this disagreement is likely a problem related to under-constraining such that one or both of the measurements is (are) biased relative to the true state of the measurand (2).

In conclusion, all five GSTA datasets analyzed here show an average GSTA over the past 35 years of between 0.01°C to 0.42°C above their respective baseline period temperature, depending data set (1). The land-based data sets report in all most all cases monthly GSTA that are higher than the satellite based GSTAs (1). In addition, the land-based monthly GSTAs analyzed here (GISS, NCDC, HadCRUT4) are increasingly positively diverging from the two satellite-based GSTAs used (RSS, UAH) over time, whereas the two-satellite-based data sets are increasingly negatively diverging from each other. These trends, and especially the opposing trend with RSS, warrant more in-depth analyses, continued scrutiny and attention.

References

(1) http://wattsupwiththat.com/2014/05/19/a-comparison-between-global-surface-temperature-and-satellite-anomaly-datasets/#more-109546

(2) Mears, C. A., F. J. Wentz and P. W. Thorne, (2012) Assessing the Value of Microwave Sounding Unit-Radiosonde Comparisons in Ascertaining Errors in Climate Data Records of Tropospheric Temperatures, J. Geophys. Res., 117(D19), D19103, doi:10.1029/2012JD017710.