SST data sets form a major part of global surface temperature analyses considered in this assessment report. To use an SST data set as a boundary condition for atmospheric reanalyses products (Box 2.3) or in atmosphere-only climate simulations (considered in Chapter 9 onwards), gridded data sets with complete coverage over the global ocean are typically needed. These are usually based on a special form of kriging (optimal interpolation) procedure that retains large-scale correlation structures and can accommodate very sparse data coverage. For the pre-satellite era (generally, before October 1981) only in situ data are used; for the latter period some products also use AVHRR data. Figure 2.18 compares interpolated SST data sets that extend back to the 19th century with the uninterpolated HadSST3 and Had- NMAT2 products. Linear trend estimates for global mean SSTs from those products updated through 2012 are presented in Table 2.6. Differences between the trends from different data sets are larger when the calculation period is shorter (1979–2012) or has lower quality data (1901–1950); these are due mainly to different data coverage of underlying observational data sets and bias correction methods used in these products.
In summary, it is certain that global average sea surface temperatures (SSTs) have increased since the beginning of the 20th century. Since AR4, major improvements in availability of metadata and data completeness have been made, and a number of new global SST records have been produced. Intercomparisons of new SST data records obtained by different measurement methods, including satellite data, have resulted in better understanding of uncertainties and biases in the records. Although these innovations have helped highlight and quantify uncertainties and affect our understanding of the character of changes since the mid-20th century, they do not alter the conclusion that global SSTs have increased both since the 1950s and since the late 19th century.