Widespread thermometer measurements of sufficient accuracy to reliably estimate large-scale changes in near-surface air temperature over land areas did not become available until the mid-19th century, and routine measurements of ocean temperatures did not become available until the late 19th century. In addition to missing data and individual measurement errors, there are a variety of artificial biases present in long-term temperature records that must be removed to yield records of sufficient accuracy to evaluate climate trends. Equipment and measurement procedures have changed over time—for example, ocean temperatures have been measured by satellites, buoys, and ships, and the ship-based measurements have included readings taken by hull sensors, in water drawn in to cool the engines, and in buckets pulled up by hand from the water surface. Temperature measurements are also not evenly distributed in space or time; observing stations are common in densely populated land areas, while the southern oceans were only sparsely observed before satellite measurements became available in the late 1970s. Finally, temperature measurements can be affected by a number of local factors, such as the "urban heat island" effect (see
Chapter 12) and other changes in land use; although these changes represent real changes in local climate, they need to be quantified and corrected when evaluating large-scale changes in climate.
Several research groups around the world, including NASA's Goddard Institute for Space Studies (GISS), NOAA's National Climatic Data Center, and the Climate Research Unit at the University of East Anglia in the United Kingdom, collect and maintain databases of both historical and present-day meteorological data and use them to produce estimates of regional and global climate change. Producing these estimates requires each individual record to be quality-controlled and corrected to remove the artificial biases described above, and then additional steps are needed to convert the assemblage of individual records into representative large-scale averages. Each group uses somewhat different data sources and analysis procedures (see, e.g., Hansen et al., 1999, 2001; Karl and Williams, 1987; Menne and Williams, 2009; Menne et al., 2009). Most of these data and methods are publicly available.
Each of the research teams that produce large-scale temperature estimates has developed methods for dealing with the potential biases and sources of error such as those described in the preceding paragraph. For example, NASA GISS uses a linear interpolation procedure to "fill in" missing data and temperatures in areas between observing stations, and data from urban stations (which are identified based on either population density data or "nightlight" levels observed by satellite) are adjusted so their long-term trends match those of neighboring rural stations. The University of East Anglia instead corrects the station-level data first and then uses a simple averaging procedure to combine the data. These procedures have been developed over several decades (e.g., Hansen and Lebedev, 1987) and are constantly reevaluated to identify and correct for additional sources of error. It was recently determined, for example, that a change in the way that certain ship-based temperatures were treated introduced a spurious signature into the mid-20th-century temperature record including an abrupt drop of ~0.5°F (0.3°C) in 1945 (Thompson et al., 2008).
For the GISS data, the uncertainties associated with corrections to the raw data and with the underlying raw data themselves are estimated to yield a total uncertainty in global-average surface temperature estimates of about 0.09°F (0.05°C) during the past several decades. During the first few decades of the record, the estimated uncertainty is twice as large (0.18°F or 0.10°C), as might be expected due to the smaller number of measurements and their lower precision relative to modern instruments (Hansen et al., 2006; see also Thompson et al., 2009). Global temperature estimates produced by other research teams yield results that agree within these estimated uncertainties. Changes in temperature, or other climate variables, are typically reported as anomalies
(differences) relative to a specified time period because this minimizes errors associated with calibration to absolute temperature.
Was this article helpful?