云环境下基于朴素贝叶斯的雷暴预报初探

Thunderstorm Prediction Based on Bayesian Classification in Cloud Computing Environment

雷暴是一种灾害性天气系统,对建筑和人类造成极大的伤害。因此,对雷暴的预报具有重要的意义。将传统的朴素贝叶斯分类方法进行并行化,在Hadoop云计算平台下利用NCEP 1.0×1.0历史再分析资料和江苏省闪电定位资料进行雷暴的预报,并与逐步回归分析法和神经网络方法进行比较。实验结果表明,云环境下的朴素贝叶斯分类方法在预报准确率和空报率等方面均优于逐步回归分析法,但其空报率略高于神经网络方法。总体而言,文中的算法对雷暴的预报有较好的效果,这为雷暴预报提供了一个新的思路。

Thunderstorm is a disaster weather system. It has important significance for thunderstorm prediction. In this paper, we used NCEP FNL Operational Global Analysis data which are on 1.0 ＊ 1.0 degree grid to prepare operationally every six hours and the lightning strike positioning data in Jiangsu province. The traditional Bayesian classification methods were parallelized to predict the thunderstorm in Hadoop cloud computing platform. Comparing with the tradi tional stepwise regression analysis and the neural network algorithm, the experimental results show that the Naive Bayesian classification method in the cloud environment is better than the stepwise regression analysis in forecast accuracy, false alarm rate and CSI, but the false alarm rate is a little bit higher than the neural network algorithm. In summary, the new algorithm has a good performance on the thunderstorm prediction. It provides a new method for forecasting thunderstorm.