科研之友
微信
新浪微博
Facebook
分享链接
摘要
Sophisticated numerical computation is widely used in weather forecast. Various computation models exist, and each of them is designed for different meteorological elements, areas, or seasons. Therefore, multimodel ensemble forecast is aimed at integrating various numerical model forecasts to obtain an accurate forecast result. The existence of deviations compels forecasters to calibrate ensemble forecast result. Traditional calibration methods are strongly dependent on the experience of the forecaster and are usually performed inefficiently with grid-by-grid or station-by-station mode. Such calibration methods experience several drawbacks, including the high complexity and the low comprehensiveness of searching for similar historical forecast results in a limited dataset and the lack of real-time interactive analysis. To solve these problems and improve the efficiency and the accuracy of calibration, we propose a visualization system for calibrating multimodel ensemble forecast result. This system can assist forecasters to quickly locate potential calibrating stations by visualizing the deviation of historical forecast and observation result. Then, associated regions are calculated automatically under similarity and connectivity constraints, and corresponding coefficients between associated stations are generated based on parameter learning of the factor graph. These associated regions and coefficients are used as a guide for forecasters in setting unified deviation and weight for each numerical model. Finally, automatic calibration is conducted on the rest stations in the associated area. The performance of the visualization system is evaluated with real meteorological data.
- 出版日期2016-11
- 单位清华大学
