Extreme Weather Event (Cyclone) Detection in India Using Advanced Deep Learning Techniques
Atharv Dabhade, Samshitha Roy, Marwa S. Moustafa, Sayed A. Mohamed, Reda El Gendy, Shovan Barma
Abstract
Detection and Prediction of extreme weather events have always been important and challenging task, which has become very significant while the whole world is experiencing adverse effect of climate change. The cyclone, one of the extreme weather events has become very often and intense in Indian sub-continent areas in last few decades and its accurate detection and prediction has become very relevant. In this direction, machine learning (ML) based approached has become very popular than other conventional methods. However, the performance of the traditional ML based approaches is highly dependent on hand crafted feature engineering. In contrast, deep learning, an advance ML eliminates such barrier and selects the feature set automatically in hierarchical way, which has already displayed significant performance in various domain of research including weather forecasting. In this purpose, convolutional long-short term memory (ConvLSTM) can efficiently extract spatial and temporal information, which are very important for accurate detection of extreme weather events. Therefore, a model with ConvLSTM has been proposed in this work to detect and predict the cyclone in India. For validation, Community Atmospheric Model (CAM5.1) image data considering seven variables have been generated and employed. Two matrices, peak signal to noise ratio (PSNR) and structural index similarity (SSIM) have been used to validate the model performance. The results and analysis show that the proposed idea can effectively improve the cyclone detection and predication.