On January 9, 2007 NOAA provided a press release stating that preliminary temperatures for the United States indicated 2006 was the warmest year on record. Included in the press release was reference to a new method for correcting biases in observations (Version 2) that had a preliminary rank for 2006 as the 2nd warmest year on record. After receipt of additional observations for 2006, temperature statistics were updated on May 1, 2007. The late data changed the rank for 2006 to the 3rd warmest year on record for the old method (Version 1) and the rank remained as 2nd warmest for the new data correction method (Version 2).
Why such changes occur is rooted both in the way the observations are processed for quality and the delay in receipt of data on paper records from many stations. The observations come from the U.S. Historical Climatology Network (USHCN), a network of 1221 climate observing stations in the continental United States. These data are extensively quality controlled for errors and for small biases that may have occurred through time due to artificial changes at each observing station. These artificial changes include station relocations, different instrumentation, and changes in the landscape surrounding the station (e.g. urbanization, removal or planting of vegetation, etc.). Some of these changes may result in "random" changes to the data. For example, even small station relocations can result in temperature readings that are either slightly cooler or slightly warmer than what would have occurred at the former site. Other changes, such changes in urbanization in the vicinity of the station or changes in observing times, can systematically affect temperatures, e.g., add an urban warming bias to the temperature trends. Research has shown that the data from these kinds of changes can be corrected to a large degree based on physical and statistical methods (e.g., see Peterson 2006).
Methods that have been used to correct temperature data are described in more than a dozen peer-reviewed scientific papers by NOAA's National Climatic Data Center (NCDC). A series of data corrections was developed to specifically address potential problems in trend estimation of the rates of warming or cooling in the USHCN. They include:
These data correction schemes have been applied to the USHCN to determine temperature trends across the United States up until the end of 2006. Beginning in 2007 improved correction schemes for items 1 and 3 above have been applied to the USHCN observations (Menne and Williams 2005, Menne and Williams 2007). They have been shown to improve our ability to monitor climate change and variations. Because different algorithms were used in making corrections to the station data in 2007 there are small differences in annual average temperatures between the older corrections (Version 1) and newer Version 2 corrections. These small differences in average temperatures result in minor differences in annual rankings for some years. The new correction scheme has virtually no impact on the long-term temperature trend as annual temperature trends in Version 1 from 1895-2006 were 0.112°F/decade and in Version 2 the trends were 0.110°F/decade.
NOAA continues to work to improve the quality and representativeness of climate data provided to the public and scientific communities. In addition to advanced quality control procedures, these efforts include modernization of the USHCN by installing new, more accurate instrumentation, and ensuring proper station siting in the process. In addition, by the end of next year NOAA should have in place a U.S. Climate Reference Network, a set of 114 very high quality stations optimized for monitoring climate. The operation of the US Climate Reference Network will virtually eliminate the need for the types of corrections that have to be applied to data available today. The modernization of the US Historical Climatology Network will enable trends of regional temperature to be estimated with far fewer data corrections.
http://www.ncdc.noaa.gov/oa/climate/research/ushcn/hcntmptrends.php
