Calculating the global temperature
Deriving a reliable global temperature from the instrument data is not easy because the instruments are not evenly distributed across the planet, the hardware and observing locations have changed over the years, and there has been extensive land use change (such as urbanization) around some of the sites.
The calculation needs to filter out the changes that have occurred over time that are not climate related (eg urban heat islands), then interpolate across regions where instrument data has historically been sparse (eg in the southern hemisphere and at sea), before an average can be taken.
There are two main global temperature datasets, both developed since the late 1970s: that maintained by the Climatic Research Unit at the University of East Anglia  and that maintained by the Goddard Institute for Space Studies . Both datasets produce very similar results, and are updated every month with additional data.
In the late 1990s, the Goddard team used the same data to produce a global map of temperature anomalies to illustrate the difference between the current temperature and average temperatures prior to 1950 across every part of the globe.
Uncertainties in the temperature record
A number of scientists and scientific organizations have expressed concern about the possible deterioration of the land surface observing network. Climate scientist Roger A. Pielke has stated that he has identified a number of sites where poorly sited stations in sparse regions "will introduce spatially unrepresentative data into the analyses." The metadata needed to quantify the uncertainty from poorly sited stations does not currently exist. Pielke has called for a similar documentation effort for the rest of the world.
The uncertainty in annual measurements of the global average temperature (95% range) is estimated to be ~0.05°C since 1950 and as much as ~0.15°C in the earliest portions of the instrumental record
. The error in recent years is dominated by the incomplete coverage of existing temperature records. Early records also have a substantial uncertainty driven by systematic concerns over the accuracy of sea surface temperature measurements. Station densities are highest in the northern hemisphere, providing more confidence in climate trends in this region. Station densities are far lower in other regions such as the tropics, northern Asia and the former Soviet Union. This results in less confidence in the robustness of climate trends in these areas.
If a region with few stations includes a poor quality station, the impact on global temperature would be greater than in a grid with many weather stations.