SOME CONSIDERATIONS ON NORMAL MONTHLY TEMPERATURES

Abstract

The concept of a satisfactory normal monthly temperature is considered. January and July mean temperature records at seven stations in the United States are examined. It is found that, except in the western United States in summer, such temperature records may be accepted as constituting a random sample. It is also found that, in general, mean monthly temperatures are not normally distributed. The method of confidence limits is applied to determination of a satisfactory normal temperature. Seasonal and geographical variations in reliability of normal temperatures are observed. Consideration is given to the adequacy with which the normal characterizes the temperature record, and to the influence of trends and cyclic fluctuations on this adequacy. It is suggested that a normal of desired reliability be computed from the most recent portion of the record to be most representative. In general, normals computed for different periods of record will differ; but these differences may have little practical significance when the magnitude of the effect is considered relative to the requirements of the application.