天气预报聚类分析的可视化研究

Visualization in weather forecast based on custer analysis

聚类分析经常用于选定区域中代表性天气趋势的确定,然而所识别的聚类可能对所选区域的微小变化非常敏感,导致聚类结果的可信度和鲁棒性不佳。为此文中提出一种针对天气预报聚类结果的交互式可视化界面,以分析聚类结果对所选区域变化的敏感性。所提出的交互式可视化界面可以显示已识别聚类的主成分变化、个体聚类成员的可变性以及已识别聚类的空间位置的不确定性趋势。试验评估表明,将使用所述的方法应用于对热带气旋预测中不同区域的聚类可视化分析,能够直观展示聚类过程中的可信度和鲁棒性信息。

Cluster analysis is often used to determine the representative weather trends in selected regions.However,the identified clusters may be very sensitive to small changes in selected regions,causing poor reliability and robustness of clustering results.To this end,an interactive visual interface for weather forecast clustering results is proposed to analyze the sensitivity of clustering results to selected region changes.The proposed interactive visualization interface can display the principal component changes of the identified clusters,the variability of the individual cluster members,and the uncertainty trend of the spatial locations of the identified clusters.The experimental evaluation shows that the method will be applied to cluster analysis of different regions in tropical cyclone prediction,which can visually show the credibility and robustness information in the clustering process.