Early Warning System for Flood Disaster Risk Reduction Using Predictive Analytics

Flooding is one of the most common natural disasters today-occurring when a body of water exceeds its capacity due to heavy rainfall over a long time. Many countries have recently experienced severe flooding and other extreme hydrological events. Developing an early warning system to anticipate flood disasters is critical for mitigating the impact of these flood hazards. Several machine-learning (ML) approaches have been proposed for developing flood forecasting models. However, many of these ML models could not provide very accurate predictions because they lack the capacity to analyze large amounts of continuous data. Here, we employed a deep learning technique-Long Short-Term Memory (LSTM) network model, to predict the daily discharge of the Lokoja River in Nigeria. Our results show that LSTM generates a high prediction accuracy of 97%. Also, using the same datasets, the performance of the LSTM model was compared to that of the traditional neural network-Back Propagation Neural Networks (BPNNs) and a hybrid model-that combines the Self Organized Maps (SOM) and BPNNs. The results show that the LSTM model performs better regarding prediction accuracy.