A Case Retriever for Historical Meteorological Data

Large archives of historical meteorological data offer novel opportunities for weather forecasting and meteorological research. However, effective exploitation of such an archive is impeded by the logistics of managing large volumes of data and the difficulty of identifying relevant data. MetVUW Workbench is a case retriever that allows meteorologists to identify relevant past weather situations (cases) given a description of the important meteorological phenomena. This paper presents an overview of the architecture and key components of the system. The MetVUW research initiative (METeorology at Victoria University of Wellington) aims to develop tools to assist weather forecasters and meteorological researchers. MetVUW Workbench is a tool for intelligent retrieval of meteorological data constructed as part of the MetVUW research initiative. The system is designed to help meteorologists rapidly locate relevant meteorological data in large archives. Past situations are retrieved by specifying high-level descriptions of their content, using a graphical query language. This paper presents an overview of MetVUW Workbench. Section 1 describes the architecture of MetVUW Workbench. Section 2 describes the user interface for constructing queries. Section 3 sketches MetVUW Workbench’s representations for cases. Section 4 describes how cases that are good matches to the query are efficiently selected from the case base. Section 5 describes how the selected cases are evaluated for their similarity to the query. Section 6 briefly mentions the facilities for displaying the retrieved cases. R ecent advances in remote sensing pro vide meteorologists with large quanti ties of useful data, including satellite imagery and relatively high resolution, three-dimensional numerical data (Papathomas, Schiavone, & Julesz, 1988). Furthermore, a considerable historical record of this data is accumulating. Archives of meteorological information consequently offer novel opportunities for weather forecasting and meteorological research. However, effective exploitation of such archives is impeded by the logistics of managing large volumes of data, and the difficulty of identifying relevant data. Although the processing power of computers continues to increase rapidly, increases in on-line storage capacity and access speed have been relatively slow. Larger data sets must frequently be relegated to off-line or slow secondary storage, such as magnetic tape or CD-ROM. To quickly locate and retrieve relevant data from slow storage, indexing strategies are required that can determine which data are relevant without accessing the full data sets.