Performance of Alternative “Normals” for Tracking Climate Changes, Using Homogenized and Nonhomogenized Seasonal U.S. Surface Temperatures

Abstract

leven alternatives to the annually updated 30-yr average for specifying climate ?normals? are considered for the purpose of projecting nonstationarity in the mean U.S. temperature climate during 2006?12. Comparisons are made for homogenized U.S. Historical Climatology Network station data, corresponding nonhomogenized station data, and spatially aggregated (?megadivision?) data. The use of homogenized station data shows clear improvement over nonhomogenized station data and spatially aggregated data in terms of mean-squared specification errors on independent data. The best single method overall was the most recent 15-yr average as implemented by the Climate Prediction Center (CPC15), consistent with previous work using nonhomogenized and spatially aggregated data, although ?hinge? functions with the change point fixed at 1975 performed well for the spring and summer seasons. A hybrid normals-specification method, using one of these piecewise continuous functions when the regressions are sufficiently strong and the CPC15 otherwise, exhibits a favorable trade-off between squared error and bias that may make it an optimal choice for some users.