Data-Based Models for Hurricane Evolution Prediction: A Deep Learning Approach

Abstract

We show that the use of 6 − hr displacement probabilities as input features to Recurrent Neural Networks (RNNs) increases accuracy of the storm trajectory forecasting• We demonstrate compounded error accumulation in forecasts for the Many-To-One type RNN prediction models applied to storm trajectory forecasting. As a remedy, we highlight the use of the Many-To-Many type RNN prediction models.