A Knowledge-Based System for the Diagnosis and Prediction of Short-Term Climatic Changes in the North Atlantic

Abstract

Understanding and predicting climate change is the key problem in climatology. The most well-accepted current approach to this problem involves the development of general circulation models(GCMs).This approach is based on modeling fundamental physical principles in large computer programs. At the same time, however, an increasingly large proportion of the available information regarding the climate system exists in the form of heuristics, or empirical rules of thumb. The objective of the CESNA (Climatic Expert System For the North Atlantic) project is to develop a practical system that can manipulate this qualitative information in such a way as to facilitate insights into observed and anticipated climate changes. The methods used to reach this objective are based on concepts and techniques derived artificial intelligence research on representing and reasoning with uncertain knowledge. A recently completed evaluation of the prototype CESNA measured how well it could predict the sea temperature of the Kola section of the barents sea for the period 1965 to 1991 with a one-year lead time. The system's predictions paralleled the observed temperatures with remarkable accuracy. Similar results were obtained for two other regions, the northwest Atlantic and the southeastern United States. Qualitatively, these experiments show that even though some rules may he poor predictors in a given year, the combined evidence from the remaining rules results in an accurate prediction.