Abstract ATM-9530518 Martin, James H. Rodionov, Sergei N. University of Colorado Title: A Knowledge-Based Approach to Climate Analysis and Forecasting Information regarding the climate system is often imprecise, incomplete, vague, or uncertain. A significant part of this information exists as empirical facts and rules, expressed in qualitative form. The objective of this study is to develop a system that can manipulate qualitative information is such a way as to provide understanding about observed climate changes. The appropriate method to reach this objective is based on concepts and techniques of expert systems and artificial intelligence. The information about parameters of the climate system will be presented both in quantitative and qualitative forms. Relationships between climatic variables will be expressed as production rules written in IF-THEN format. Each rule will contain a certainty factor that represents a numerical measure of the confidence one has in the validity of a rule. Two methods will be used to combine certainty factors. The first method is based on the Dempster-Sharper theory, and the second one on the Bayesian scheme. The system will be able to evaluate the future state of various climatic variables (such as temperature, precipitation, atmospheric pressure, etc.) and of macroclimatic features (such as centers of action, jet streams, upper atmospheric troughs and ridges, etc.). It will also explain why the conclusion was reached and how certain it is. This system may also be considered as a decision support system valuable for recording ideas and retrieving necessary information that will facilitate discussions about observed and anticipated climate changes. The grant will support the prototype development which will focus on the Northern Atlantic. Then the system will be expanded to the entire Northern Hemisphere. The system will be useful both for climatolog ists and decisionmakers. It will find different applications in fisheries, agriculture, navigation, recreation and other climate-related industries.
