基于决策树的雷暴天气短临预报

Thunderstorm Short-Term Forecast Using Decision Tree Model

雷暴天气是指伴有雷电、冰雹、大风和强降水的局地强对流性天气,它对航空运输、电力设施、通讯设备与建筑物等均可造成不同程度的破坏,严重时甚至造成人员伤亡。开展雷暴天气过程的短临预报具有重要的科学意义与实用价值。本文对我国2010~2015年雷暴天气事件的时空分布特征进行了统计分析,结果表明:雷暴天气事件集中分布于国内少数区域夏季7~8月份的14:00~18:00点。在此基础上,基于2010~2015年的地面气象观测资料建立决策树模型,预测未来3~4小时雷暴天气的发生概率。模型对雷暴事件的漏报率和误报率均低于10%,这一结果表明:本文所建立的模型能够较为准确地进行雷暴天气短临预报,能够为保障飞行安全提供较为可靠的决策支持。而对特征的相对重要性排序结果表明:测站的地理环境特征和气象条件对雷暴天气过程的发生具有显著的影响。

The thunderstorm is local strong convection weather with light, hail, strong wind and heavy precipitation.It can lead to different degree of damages to air transportation, electric power facilities, communication equipment and buildings, even serious casualties. It is of important scientific significance and practical values to the thunderstorm forecast. This study analyzed the temporal-spatial distribution of the civil thunderstorm event in 2010~2015, indicating that the thunderstorm event mainly located in limited regions at 14：00~18：00 in July to August of the summer. The decision tree model was built based on the civil surface meteorological observations in 2010~2015, to predict the probability of the thunderstorm event in 3~4 hours. The model predict the positive and the negative samples with the error lower than 10% respectively, showing that the model was capable of the correct thunderstorm short-term forecast, and can provide reliable decision support for flight safety. Moreover, the feature importance shows that the local geographical characteristics and meteorological conditions have significant influence on the thunderstorm event.