Global Analysis - February 2004

An omission in processing a correction algorithm led to some small errors on the Global Historical Climatology Network-Monthly dataset (GHCN-M v3.2.0). This led to small errors in the reported land surface temperatures in the October, November, December and Annual U.S. and global climate reports. On February 14, 2013, NCDC fixed this error in its software, included an additional improvement (described below), and implemented both changes as GHCN-M version 3.2.1. With this update to GHCN-M, the Merged Land and Ocean Surface Temperature dataset also is subsequently revised as MLOST version 3.5.3.

The net result of this new version of GHCN-M reveals very small changes in temperature and ranks. The 2012 U.S. temperature is 0.01°F higher than reported in early January, but still remains approximately 1.0°F warmer than the next warmest year, and approximately 3.25°F warmer than the 20th century average. The U.S. annual time series from version 3.2.1 is almost identical to the series from version 3.2.0 and that the 1895-2012 annual temperature trend remains 0.13°F/decade. The trend for certain calendar months changed more than others (discussed below). For the globe, ranks of individual years changed in some instances by a few positions, but global land temperature trends changed no more than 0.01°C/century for any month since 1880.

NCDC uses two correction processes to remove inhomogeneities associated with factors unrelated to climate such as changes in observer practices, instrumentation, and changes in station location and environment that have occurred through time. The first correction for time of observation changes in the United States was inadvertently disabled during late 2012. That algorithm provides for a physically based correction for observing time changes based on station history information. NCDC also routinely runs a .pairwise correction. algorithm that addresses such issues, but in an indirect manner. It successfully corrected for many of the time of observation issues, which minimized the effect of this processing omission.

The version 3.2.1 release also includes the use of updated data to improve quality control and correction processes of other U.S. stations and neighboring stations in Canada and Mexico.

Compared to analyses released in January 2013, the trend for certain calendar months has changed more than others. This effect is related to the seasonal nature of the reintroduced time-of-observation correction. Trends in U.S. winter temperature are higher while trends in summer temperatures are lower. For the globe, ranks of individual years changed in some instances by a few positions, but global temperature trends changed no more than 0.01°C/century for any month since 1880.

More complete information about this issue is available at this supplemental page.

NCDC will not update the static reports from October through December 2012 and the 2012 U.S and Global annual reports, but will use the current dataset (GHCN-M v. 3.2.1 and MLOST v. 3.5.3) for the January 2013 report and other comparisons to previous months and years.