Global surface temperature data sets are an essential resource for monitoring and understanding climate variability and long-term trends. The most commonly used data sets combine historical observations of near surface air temperatures at land stations with global data sets of sea surface temperatures (SST) obtained from a changing mix of ship-based and buoy measurements. While the concept of these data sets is fairly simple, their construction is challenging due to difficulties in obtaining data, documenting and accounting for changes in instrumentation and observing practices, addressing changes in station location and local land use, understanding random measurement errors, and deciding where and how to infill missing data in space and time.

The four most highly cited data sets are NOAA's GlobalTemp (Formerly MLOST), NASA's GISTEMP, Berkeley Earth, and the UK's HadCRUT. HadCRUT also has a land-only version, CRUTEM. In addition, a state-of-the art dataset, the The Dynamically Consistent ENsemble of Temperature (DCENT) was released in 2024. Each group has approached the above challenges somewhat differently. The final data sets differ in their spatial coverage, spatial resolution, starting year, and degree of interpolation (DCENT is the only major uninterpolated product). Most of these data sets are presented as anomalies (departures from baseline, long-term mean temperatures); only the Berkeley Earth data provide absolute temperatures for each timestep, while the other projects provide a baseline climatology to which the anomalies may be compared. Numerous comparisons of global and hemispheric mean temperature anomaly timeseries calculated from these data sets have been made, showing highly consistent variations and trends. Nonetheless, users doing more analysis than the global mean temperature will find important distinctions among the data sets.

Besides the in-situ based data sets summarized here, other estimates of global temperatures (since the 1970s) are based on satellite measurements such as from the MSU and AIRS instruments, or on atmospheric reanalyses.

It is best practice to use multiple products as well as incorporate their uncertainty estimates in any analyses involving surface temperature. In addition, recent science suggests that there are biases in the sea surface temperature record in the first half of the 20th century that are not fully accounted for in existing products (see Sippel et al. 2024 or the more accessible news piece for more information).