The problem of auto‐correlation of meteorological time series, illustrated by examples from some South African stations

Abstract

AbstractThe main object of this paper is to present a number of characteristic values of the auto‐correlation coefficient as actually computed from observed series of temperature and pressure. For this purpose the temperature and pressure records of some South African stations were selected.In the case of temperatures, it was found that, provided any daily or seasonal fluctuations are first eliminated and the standard deviation is constant, the coefficient of auto‐correlation R' has the analytic form R' = e‐ky, in which k is a constant and y is the time‐lag. If, however, there is a marked variation in the standard deviation, such as is nearly always present in hourly values of temperature, R' is a function not only of the time‐lag, but also of the actual times marking the beginning and end of the time‐lag under consideration.In the case of a series of pressure readings R' is not a simple exponential function, but may contain one or more damped harmonic terms.