Automatic Monthly Detection of Maximum and Minimum Temperature Errors and Inhomogeneities in the NOAA/NWS Cooperative Observer Network

A plan has been developed to routinely assess NOAA's observing networks in terms of their climate monitoring capabilities. One element of this plan or "health" assessment is a new approach to automatically detect errors and discontinuities in the maximum and minimum temperatures in the Cooperative Observer Network. The approach is supplemental to existing error detection schemes at the National Climate Data Center and is based on a comparison of a candidate station with neighbors. Time series of daily differences between standarized temperatures from the candidate station and its neighbors are formed at the end of each month. These differences are modeled as approximate white noise when the candidate shows no problems. Cross-correlation and autocorrelation tests are applied to the daily differences such that a large cross-correlation coefficient or lag 1 autocorrelation coefficient indicates the difference series are not white noise and the candidate has an error or discontinuity. When this occurs a flag is raised. Additional steps are taken to limit the number of flags to those occurrences that are clearly not meteorology. We will show examples of applying the correlation tests to monthly time series of daily maximum and minimum temperatures.