Flood forecosting model using federated learning

One of the most common natural catastrophes is flooding, which frequently causes serious harm to people's lives, property, crops, and the economy.Flood prediction has long been a major problem for scientists.A flood forecasting model that makes use of the federated learning technique is proposed in this article.Federated learning is a state-of-the-art machine learning (ML) technique that prevents data transfer over the network for model training, thereby addressing network latency issues during flood prediction while guaranteeing data security, privacy, and availability.Federated learning encourages onsite training of local data models, distributing only the trained local models throughout the network, as opposed to sending enormous datasets to a central server for model aggregation.The suggested model identifies stations at danger of flooding, combines locally trained models from 18 clients, and creates flood alerts for particular clients with a five-day lead time.The local feedforward neural network (FFNN) model is trained at the client station where flooding is expected.Based on a number of regional factors, the FFNN model's flood forecasting module forecasts the anticipated water levels.Five distinct rivers and barrages are included in the training dataset, which was gathered between 2015 and 2021 and contains variables including snowmelt, rainfall-runoff, flow routing, and hydrodynamics.