A quantitative approach to long‐range prediction

Publisher Website
Google Scholar
Abstract
Considerations are presented which allow for a more quantitative approach to the problem of long range prediction. It is supposed that the interannual variance of time‐averaged meteorological data is composed of approximately separable components. One component called ‘climate noise’ reflects sampling variability of finite time‐averaged weather fluctuations, and it is unpredictable at long lead times. The remaining variance is assumed to be potentially predictable. Climate noise and potential predictability of winter mean temperature are estimated for United States stations. It is further supposed that the potentially predictable part is composed of a component that we cannot now predict and ‘signals’ that we can. A signal is assumed to be in hand through lag correlations associated with the southern oscillation. The impact of the total unpredictable part of the variance, or ‘effective noise,’ on the usefulness of this signal is discussed. The task of improving the value of long‐range forecasts is described as one of transferring more and more of the potentially predictable variations to actually predictable signals.