Model and Generation of Weather Forecast Analytic Data

To solve the problem of "information exploration" in operational weather forecast, building a data warehouse to help forecaster's analysis is necessary. The key and most valuable idea is to change raw data to analytic data, include extracting useful data, making data clean, and aggregating data to rough granularity data. Usually the meteorological data got in operational weather forecast is processed, clean and canonical. So the main process is "aggregation" to concentrate the weather information to fewer data which have clear physical meaning.A conceptual model of weather analytic data is suggested with a pentagon tuple considering the spatial, transitional, physical and multi-scale natures of meteorological data. The pentagon tuple refers to ID (identification), SA (spatial attributes), EA (entity attributes), TA (time attributes) and PA (physical attributes), including several detailed attributes set each. Although meteorological data is field data, forecasters usually use spatial object data to analyze the weather systems. So the main work of changing raw data to analytic data is identifying spatial objects from field data.Four aggregations arithmetics to change raw data to analytic data are suggested: Statistics for fixed region, statistics for given spatial or temporal partitions, identification of basic weather systems and identification of weather conceptual models. The former two are relatively simple statistics, while the latter two are complex for mutative spatial object and they are discussed in detail.Basic weather systems include region of high/low, center of high/low and trough/ridge in a data field. A filtering-dividing-measuring arithmetic is suggested. Filtered with a Mexican-hat function, the trough/ridge become high/low region and easier to identify, and then the high/low region are divided from the filtered field, with some arithmetics adopted to tread with multi-scale problems of meteorological field. At last the divided regions are measured to get area, extreme value, length, width, aspect ratio (width/length), geometry center, extreme data location, points of central line, including all attributes of SA, EA, TA and PA. If the aspect ratio is smaller than a threshold, the region will be identified as a trough or ridge, and the central line is the trough or ridge line.A knowledge base system with spatial fuzzy production rule is suggested for identifying weather conceptual models (e.g., cold front), and the rational process of this rule is described. 4 topological relations, several order relations, measure relations and their subjection functions are suggested. The conclusion of the rules is expanded to spatial objects with a result-spatial-object.