An LSTM deep learning framework for history-based tornado prediction using meteorological data and damage assessment using NDVI anomalies

Abstract

Extreme weather patterns can affect ground and satellite sensors before and after their occur. This study focused on tornadoes that occurred on December 10 and 11, 2021 in the state of Kentucky. The main goal of this research was to develop a deep learning algorithm based on history to predict this phenomenon. Four scenarios were created based on temperature, air pressure, wind speed and their combination. The temperature-based scenario shows high accuracy and shows the time series of temperature rise several degrees before the tornado. In the second step, the normalized difference vegetation index (NDVI) anomaly was calculated and classified for Mayfield city. Severe NDVI anomalies showed high consistency with enhanced Fujita scale and ultra-high-resolution satellite imagery, with a correlation greater than 0.9.