Predictability of seasonal climate variations: a pedagogical review

Abstract

Introduction It is well known that the day-to-day changes in the large-scale atmospheric circulation are not predictable beyond two weeks. The small-scale rainfall patterns associated with the large-scale circulation patterns may not be predictable beyond even a few days. However, the space–time averages of certain atmospheric and oceanic variables are predictable for months to seasons. This chapter gives a pedagogical review of the ideas and the results that have led to our current understanding and the status of the predictability of seasonal climate variations. We first review the current status of the understanding of the limits of the predictability of weather. We adopt Lorenz' classical definition of the predictability of weather as the range at which the difference between forecasts from two nearly identical initial conditions is as large in a statistical sense as the difference between two randomly chosen atmospheric states. With this definition of predictability, it is implied that the upper limit of predictability depends on the saturation value of the maximum possible error, which, in turn, is determined by the climatological variance. Lorenz provided a simple conceptual model in which the upper limit of weather prediction skill is described by three fundamental quantities: the size of the initial error, the growth rate of the error and the saturation value of the error. This simple model is able to explain the current status of the seasonal, regional and hemispheric variations of numerical weather prediction (NWP) skill.