A Rasterized Lightning Disaster Risk Method for Imbalanced Sets Using Neural Network

Abstract

Over the past 10 years, lightning disaster has caused a large number of casualties and considerable economic loss worldwide. Lightning poses a huge threat to various industries. In an attempt to reduce the risk of lightning-caused disaster, many scholars have carried out in-depth research on lightning. However, these studies focus primarily on the lightning itself and other meteorological elements are ignored. In addition, the methods for assessing the risk of lightning disaster fail to give detailed attention to regional features (lightning disaster risk). This paper proposes a grid-based risk assessment method based on data from multiple sources. First, this paper considers the impact of lightning, the population density, the economy, and geographical environment data on the occurrence of lightning disasters; Second, this paper solves the problem of imbalanced lightning disaster data in geographic grid samples based on the K-means clustering algorithm; Third, the method calculates the feature of lightning disaster in each small field with the help of neural network structure, and the calculation results are then visually reflected in a zoning map by the Jenks natural breaks algorithm. The experimental results show that our method can solve the problem of imbalanced lightning disaster data, and offer 81% accuracy in lightning disaster risk assessment.

Research on Thunderstorm Identification Based on Discrete Wavelet Transform

“…Define R as the label value for each raster, and the value of R for that raster is 1 if the final identified thunderstorm falls within that raster, and 0 if the final identified thunderstorm does not fall within the raster. The accuracy of thunderstorm identification is defined as follows [26].…”