Global warming involves warming of the whole globe (the clue is in the name), but it does not necessarily affect every part of the globe at the same rate.
Different parts of the globe can experience very different changes under greenhouse warming. As a result, if you want to measure global warming, you have to measure the whole globe, not just a part of it. But of the three main in situ temperature records (GISTEMP, NCDC and HadCRUT), only GISTEMP is near-global in coverage over recent decades, and only by means of allowing each weather station to cover a larger region of the map. The incomplete coverage of the Hadley and NCDC datasets may be seen in Figure 1, along with three global temperature reconstructions.
Figure 1: Coverage maps for various temperature series. Colors represent mean change in temperature between the periods 1996-2000 and 2006-2010, from +2C (dark red) to -2C (dark blue). Note that the cylindrical projection exaggerates the polar regions of the map.
In this article I will attempt to redress the balance by presenting 16 new versions of the temperature record (10 of which I consider to be realistic), each of which has global or near-global coverage. Thirteen of these are created by creating a composite of a non-global series (HadCRUT3/4, NCDC or BEST) with a global series (GISTEMP, UAH or the NCEP/NCAR reanalysis). The remaining three series are derived by extrapolating the HadCRUT3/4 and NCDC data.
As with my previous article, these results have not been subject to peer review and should be treated as tentative. However the diversity of methods and consistency of the results are strongly suggestive.
Summary of results
I will start by summarizing the results, and then describe the methods and data in detail.
Since the main impact of incomplete coverage is seen in recent temperature trends, we will start by looking at these. The trend over the 15 year period 1996-2010 is shown in Figure 2 for each of the 16 new temperature series (colored bars), compared to the official versions (white bars with coverage percentages superposed). Some of the temperature series have known problems, however two clusters of likely trends have been identified; I have named these the GISS cluster (due to similarity with GISTEMP) and the Had4 cluster (for series based on HadCRUT4).
Figure 2: Global temperature trends 1996-2010
The global temperature records from these two clusters are shown in Figure 3, along with GISTEMP, NCDC and HadCRUT3 using a 60 month moving average. Use the buttons below the graph to select between 30- and 60- year views. The records are also available as a .CSV file which may be read into any spreadsheet program.
Figure 3: Comparison of temperature series (X) 1950-2010 ( ) 1980-2010
The 15 year trend (1996-2010) for the GISS cluster of 6 temperature series is 0.147°C/decade (this is the value for NCDCext, chosen because it is unaffected by the HadSST2 1998 discontinuity). However there has been a tendency towards more frequent La Nina conditions over the period, which has impacted this trend. Foster and Rahmstorf (2011) estimate this influence at -0.026°C/decade, so the underlying warming trend is 0.173°C/decade. This is comparable to the 0.17°C/decade long term trend found by Foster and Rahmstorf. (They also estimate a net cooling effect due to the solar and volcanic influences, however these terms are less well determined and so I have omitted them.)
The 15 year trend (1996-2010) for the Had4 cluster of 3 temperature series, each of which includes the new HadSST3 bias corrections, is 0.179°C/decade (for HAD4ext). Including the El Nino term, the underlying warming trend is 0.205°C/decade.
The two representative trends quoted above are conservative estimates and include a known cool bias due to smoothing – the other trends variously include known cool and warm biases. There is additional evidence for each of these biases, but they have not yet been quantified.