Search

• Provides a protocol for transferring hydrological knowledge between basins. • Proposes a TL-Transformer model for flood simulation and forecasting in data-sparse basins. • Explores the impact of data quality on transfer learning effectiveness. • Explains the manifestation of hydrological similarity in transfer learning through topographic index. • Compares the performance of models including physical and neural network models. There exists a substantial disparity in the distribution of streamflow gauge and basin characteristic information, with a majority of flood observations being recorded from a limited number of well-monitored locations. Transferring hydrological knowledge from data-rich to data-sparse basins has been a persistent issue. While artificial intelligence methods, such as Long-Short-Term Memory Neural Networks (LSTMs) have been attempted for simulating across different basins, the problem of gradient vanishing still inevitably lead to difficulties in responding to new basin datasets. A novel solution is presented in the form of the Transfer Learning Framework based on Transformer (TL-Transformer), which can accurately predict flooding in data-sparse basins (targets) by using models from data-rich basins (sources) without requiring extensive basin attributes at the target location. The framework was demonstrated in the middle reaches of the Yellow River, and performance was evaluated using the Nash Sutcliffe efficiency coefficient, root mean square error, and bias. The results show that TL-Transformer outperforms other models, including TOPMODEL, MLP, TL-MLP, LSTM, TL-LSTM and Transformer in all target basin stations. Compared with those models that without transfer learning, TL-Transformer showed significantly improved performance, where NSE increased by 0.2 and reached above 0.75, RMSE and bias were also controlled within 60 and 20, respectively. Furthermore, pre-training on basins with hydrological similarities increases the benefits of Transfer Learning, and we explained this response phenomenon through the topographic index. These results suggest that deep learning can tap into the commonalities in hydrological data across basins to improve the accuracy of flood forecast applications in areas with limited observations. [ABSTRACT FROM AUTHOR]

• Provides a protocol to optimize hyperparameters of LSTM. • Proposes a PSO-LSTM model for flood simulation and forecast. • Study the effect of hyperparameter changes on LSTM results. • Explores the optimal hyperparameter combination of LSTM at different lead times in flood forecast, and achieve the best simulation results by optimizing model hyperparameters. • Compares the performance of the models including physical and neural network models. Flood forecasting is an essential non-engineering measure for flood prevention and disaster reduction. Many models have been developed to study the complex and highly random rainfall-runoff process. In recent years, artificial intelligence methods, such as the artificial neural network (ANN), have attempted to construct rainfall-runoff models. The more advanced deep learning methods of long short-term memory (LSTM) network have been proved to better predict hydrological time series. However, the selection of LSTM hyperparameters in the past mostly relied on the experience of the staff, which often led to failure to achieve the best performance. The aim of this study is to develop a method to improve flood forecast accuracy and lead time. A deep learning neural network model based on LSTM networks and particle swarm optimization (PSO) is proposed in this paper. The PSO algorithm was used to optimize the LSTM hyperparameter to improve the ability to learn data sequence features. The model focuses on the Jingle Watershed in the Fenhe River and the Lushi Watershed in the Luohe River and was used to predict flood processes using rainfall and runoff observation data from stations in the watersheds. We evaluated the performance of the model with the Nash Sutcliffe efficiency coefficient, root mean square error, and bias. The results show that the PSO-LSTM model outperforms the M-EIES, ANN, PSO-ANN, and LSTM at all stations in the watersheds. The PSO-LSTM model improves the flood forecasting accuracy at different lead times, especially for those exceeding 6 h, and has higher prediction accuracy and stability. The PSO-LSTM model could be used to improve accuracy in short-term flood forecast applications. [ABSTRACT FROM AUTHOR]