Flood Probability Prediction Based on xLSTM

Flood disasters occur frequently and pose significant threats to human life and property, making flood forecasting essential for disaster prevention and mitigation. This paper proposes a probabilistic flood occurrence prediction model based on the extended Long Short-Term Memory (xLSTM) network. Leveraging multi-source time-series data such as historical rainfall, flow rate, and water level, the model captures temporal features associated with flood events and predicts future flood risks in probabilistic terms. Empirical analysis on multiple benchmark flood datasets demonstrates that the xLSTM model outperforms traditional models such as LSTM and GRU in terms of prediction accuracy and generalization capability. The proposed approach shows strong application potential and practical value, providing timely and accurate flood forecasts to support effective flood risk management.