Analysis of Tornado Environments using Convolutional Neural Networks

摘要

Tornadoes are one of the most dangerous and destructive weather-related natural hazards resulting in large property losses in some cases loss of life. Therefore, the prediction of environments that are conducive to tornado genesis is an important problem for meteorologists and climatologists. This paper presents the use of a deep learning model to predict environments that are conducive to tornado genesis. A convolutional neural network was used to predict days with high, medium, and low/no tornado activity. Experiments were conducted to determine meteorological variables and atmospheric pressures that have high prediction power for tornado occurrence. The model produced very accurate results and interesting patterns were found by exploring the resulting predictions.