Multifield Analog Prediction of Short-Term Climate Fluctuations Using a Climate State vector

A theoretical framework is developed to consider the abilities of analog techniques for the prediction of short-term climate fluctuations. The basic element of the framework is the definition of a “climate state vector.” This vector points to the position of a “climate particle” whose motion in a multi-dimensional hyper-space represents the time evolution of the climate system. The particle has a number of properties that describe regional covariability of various climatic fields. A series of metrics are assigned to the space in which the climate particle moves. These metrics are used to select past states of the climate system which are analogs to a “current” state. The subsequent prediction is made based on the past evolution of the climate state Vector. Forecasts made with the analog selection techniques are evaluated in terms of the local and global skills that attend them. Thus both the spatial and temporal dependence of the skill score field is examined. Predictions were made for the seasonal average surface air temperature anomaly fields over the North American continent at lead times of one to four seasons in advance. Significant predictive skill was found in the experiments, particularly for the summer season. The result suggests that high predictability is associated with the degree of exactness with which the climate particle retraces its trajectory in hyperspace. This in turn suggests that more accurate predictions can he made with a longer data base than the one used in this study since better analog fits would presumably he found. The results also suggest that both the current state and recent history of the climate system are important in determining the future evolution of climatic anomalies.