ORIGINAL PAPER How robust is the pre-1931 National Climatic Data Center—climate divisional dataset? Examples from Georgia and Louisiana Jason Allard & Clint Thompson & Barry D. Keim Received: 19 November 2013 /Accepted: 1 May 2014 /Published online: 15 May 2014 # Springer-Verlag Wien 2014 Abstract The National Climatic Data Center’ s climate divi- sional dataset (CDD) is commonly used in climate change analyses. This dataset is a spatially continuous dataset for the conterminous USA from 1895 to the present. The CDD since 1931 is computed by averaging all available representative cooperative weather station data into a single monthly value for each of the 344 climate divisions of the conterminous USA, while pre-1931 data for climate divisions are derived from statewide averages using regression equations. This study examines the veracity of these pre-1931 data. All avail- able Cooperative Observer Program (COOP) stations within each climate division in Georgia and Louisiana were averaged into a single monthly value for each month and each climate division from 1897 to 1930 to generate a divisional dataset (COOP DD), using similar methods to those used by the National Climatic Data Center to generate the post-1931 CDD. The reliability of the official CDD—derived from state- wide averages—to produce temperature and precipitation means and trends prior to 1931 are then evaluated by compar- ing that dataset with the COOP DD with difference-of-means tests, correlations, and linear regression techniques. The CDD and the COOP DD are also compared to a divisional dataset derived from the United States Historical Climatology Network (USHCN) data (USHCN DD), with difference of means and correlation techniques, to demonstrate potential impacts of inhomogeneities within the CDD and the COOP DD. The statistical results, taken as a whole, not only indicate broad similarities between the CDD and COOP DD but also show that the CDD does not adequately portray pre-1931 temperature and precipitation in certain climate divisions within Georgia and Louisiana. In comparison with the USHCN DD, both the CDD and the COOP DD appear to be subject to biases that probably result from changing stations within climate divisions. As such, the CDD should be used judiciously for long-term studies of climate change, and past studies using the CDD should be evaluated in the context of these new findings. 1 Introduction The National Oceanic and Atmospheric Administration’ s (NOAA) National Climatic Data Center (NCDC) has subdivided the contiguous USA into 344 climate divisions, which are considered to represent relatively homogeneous climate regions within their boundaries. For each climate division, the NCDC calculated mean monthly temperature and water-equivalent precipitation, as well as ancillary vari- ables such as Palmer drought indices, standardized z scores of temperature and water-equivalent precipitation, degree days, and intermediate Palmer model variables from 1895 to the present (Guttman and Quayle 1996). Given their spatial and temporal completeness and geographic resolution (Guttman and Quayle 1996; McRoberts and Nielsen-Gammon 2011), divisional data are used in climate change studies and applied research (e.g., Muller et al. 1990; Leathers et al. 2000; Svoboda et al. 2002; Wells et al. 2004; McCabe and Clark 2005; Easterling et al. 2007; LaDochy et al. 2007; Seager et al. 2009; Rogers 2013). Although these data have been widely used in many contexts, research has increasingly shown that the NCDC climate divisional dataset, hereafter referred to as the CDD, J. Allard (*) Department of Physics, Astronomy, and Geosciences, Valdosta State University, Valdosta, GA, USA e-mail: jmallard@valdosta.edu C. Thompson Department of Geography, The University of Kansas, Lawrence, KS, USA B. D. Keim Department of Geography and Anthropology, Louisiana State University, Baton Rouge, LA, USA Theor Appl Climatol (2015) 120:323–330 DOI 10.1007/s00704-014-1175-2