Generative Adversarial Fusion for Supercell Thunderstorm Forecasting

This paper proposes a novel machine learning model that attempts to address this problem by taking advantage of two different types of crucial data.

Supercell thunderstorms have long evaded forecasters, leading to inadequate warning and preventable deaths. Current prediction methods use direct weather observation and algorithmic computer models, while they are not very computationally intensive, they lack in their ability to maintain accurate predictions. Machine learning has been showing some promising results with generative networks proving to show the most promise. This paper proposes a novel machine learning model that attempts to address this problem by taking advantage of two different types of crucial data. My proposed architecture combines three distinct Machine Learning models: a Recurrent Deep Neural Network, a Wasserstein Generative Adversarial Network, and a Transformer Generative Adversarial Network to leverage numerical and visual data. While my research is purely theoretical at this stage, I believe that my proposed machine learning architecture has the potential to pave the way for future studies and practical implementations in the field of supercell thunderstorm prediction, providing significant improvements in forecast accuracy and lead time.

Explore More!

Research Paper