Flood Forecasting Using Artificial Neural Network (ANNs): A Case study of Jamuna River

Jamuna is one of the major rivers in Bangladesh which is highly vulnerable to floods. Bangladesh has faced devastating floods during recent years. As the physical processes responsible for floods are highly complex in nature, it is difficult to simulate the event of a flood and predict the aftermaths. This study summarizes how well feed-forward Artificial Neural Network (ANNs) performs in capturing the interconnection among different variables responsible for flooding at a certain region. This study aims at predicting water level of Jamuna river at Aricha station located downstream using upstream water level at Dalia and Noonkhawa station on Teesta and Jamuna River respectively with ANN as a modeling tool. 3 hourly water level data were collected from 2003 to 2019 and fed into a deep layer of ANNs incorporating several deep learning methods and techniques to predict the flood extents. The deep learning model was validated using observed water level and regulated with statistical indicators like Root Mean Square Error (RMSE). The study concludes that ANN provides a simple, cost efficient, reliable means of predicting flood at Jamuna river with great accuracy and precision (RMSE=0.4, R2>0.95). Further research is required to take into account a lot of unseen and beyond intuition variables responsible for flooding.