Hybrid modeling in meteorological applications Part I: Concepts and approaches |
| |
| Authors: | Prof Dr E R Reiter |
| |
| Institution: | (1) Present address: WELS Research Corporation, 3300 28th Street, Suite 200, 80301 Boulder, CO, USA |
| |
| Abstract: | Summary Hybrid modeling entails the combination of a numerical weather prediction model and a  symbolic model. The latter uses symbolic objects, their characterizing attributes and sets of behavioral constrains which prescribe changes in the states of these objects as functions of time, space, and other imposed quantitative or heuristic conditions. Integration of these two modeling components for an on-line, real-time, operational system is feasible only if both the numerical and the symbolic model can be executed in a distributed mode, i.e. at a user's location rather than in a central weather service office. This condition entails the design of a numerical model that can run on relatively inexpensive desktop workstations or high-end PCs. Given such a capability, the output from the numerical model can be used to satisfy a number of behavioral constraints of objects (such as thunderstorm, blizzard, etc.), defined in the symbolic model. These constraints can be embedded invisibly as functions of time and pixel location on the computer screen, to be called upon as soon as the respective object is activated, e.g. by placing an icon on the screen.To make such a hybrid weather prediction model responsive to details in topography, it will have to be able to interface with a geographic information system (GIS) database. Since such databases can be very voluminous, management procedures for indexing and rapid information retrieval have to be instituted. The approach discussed here involves restructuring of given GIS data into B+ trees.The hybrid prediction model whose design is described in this paper, executes very quickly on a PC (e.g. a 33 megahertz Intel 80486 chip based machine). It allows assimilation of locally generated observational data to improve forecast quality, and can respond to queries of a highly specialized nature in support of tactical decisions within the time frame between nowcasting (3 hours) and 24 hours.With 6 Figures |
| |
| Keywords: | |
| 本文献已被 SpringerLink 等数据库收录！ |
|
