A Deep Learning Model for Effective Cyclone Intensity Estimation

Abstract

Tropical cyclones are extremely dangerous weather phenomena that cause significant damage to human life, property, economy, agriculture, and development. Currently, various methods are being used to estimate cyclone intensity. One such method is the objective deviation angle variance technique, which estimates the intensity of tropical cyclones from satellite infrared imagery by performing statistical analysis of the brightness of those images. The limitation of this method is that it requires images with properly marked cyclone centers. Another method to estimate cyclone intensity involves feature engineering and machine learning, however, which is a manual process. To address the limitations associated with the above conventional methods, a new approach is being proposed that uses a deep learning mechanism to design a CYCLONE NETWORK (CY-Net) model. This model will estimate the cyclone intensity by using INSAT-3D infrared (IR) images. The CY-Net model is developed based on structural, intensification, and landfall features along with biasing parameters such as wind speed, sea level pressure, and sea surface temperature. The INSAT 3D data is given as input to the model for training, testing, and validation. It undergoes convolution along with the Re-lu activation function to generate feature maps, max pooling, sub-convolution, and fully connected layers. The stochastic gradient descent factor is measured to implement the backpropagation. The Stochastic gradient and backpropagation network are implemented to obtain the best filter coefficients. The trained, tested, and validated model is deployed in Python-flask for web application and then hosted using the web servers for web application. This approach uses advanced machine learning techniques to estimate cyclone intensity and has the potential to improve accuracy and reduce manual effort.