Optimum length of record for climatological estimates of temperature

Abstract

The n-year average of a meteorological element is used to predict itself one or more years ahead, n is varied, and the value giving the best predictions is defined as the optimum length of record. Monthly temperatures and daily maximum temperatures at stations in the United States are analyzed in this manner. It is concluded that the optimum is about twenty years. This conclusion is supported by some experiments with random numbers and by agreement with other investigations.