Biying XuFaiza Binti Abd Rahman2025-10-152025-10-1520259783031907166978303190717310.1007/978-3-031-90717-3_94https://dspace-cris.utar.edu.my/handle/123456789/11490Urban flood disasters bring huge economic losses to cities every year, and effective runoff prediction is crucial for flood control. However, the complexity and randomness of the runoff process make accurate prediction extremely challenging. Traditional hydrological models require a large amount of meteorological and geographical data, which is particularly difficult in data-scarce areas and results in low prediction accuracy. This paper proposes an intelligent runoff prediction method that combines Long Short-Term Memory (LSTM) neural networks with the CMADS dataset. Using the rainfall data from the CMADS dataset, a rainfall-runoff LSTM model was constructed for the Baihe River Basin in Nanyang, predicting the monthly runoff and daily runoff warning frequency at the Nanyang hydrological station. The results show that the Nash-Sutcliffe efficiency (NSE) coefficient of the LSTM model in monthly runoff prediction can reach 0.8555. During the training period, the accuracy of daily runoff warnings matched the actual warning frequency with 85.5% accuracy, and reached 100% during the testing period. This study provides an efficient tool for urban river runoff prediction, promoting the intelligence and informatization of flood monitoring, and contributes to flood control and water security management in the basin. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.enFlowLSTMrunoffDisastersFlood controlFloodsLossesPrediction modelsRainRiver controlWater managementWeather forecastingEconomic lossFlood predictionMemory modelingMemory networkPrediction methodsRunoff predictionRunoff processShort term memoryUrban flood disastersRunoffbook-chapter