Innovative Spiking Neural Network-Based BiLSTM Model for Missouri River Discharge Prediction

Accurate river flow prediction, especially in large basins such as the Missouri River, is of great importance for water resources management, flood control, and agricultural planning. This study evaluated the performance of two advanced neural network models, including bidirectional long short-term memory (BiLSTM) and BiLSTM optimized with spiking neural network (SSN-BILSTM), in predicting daily Missouri River discharge at USGS station ۰۶۶۱۰۰۰۰ Omaha. Six different scenarios, including time lags from one to seven days ago, were used to predict river flow. The results of this study showed that the SSN-BILSTM model performed better than the BILSTM model in all scenarios. Specifically, in the first scenario, the correlation coefficient of the hybrid model was ۰.۹۹۴, which is significantly higher than the BiLSTM model with a value of ۰.۹۸۷. Similar differences were observed in other scenarios, indicating that the SSN-BILSTM model is able to better identify nonlinear relationships in river flow data. This increase in accuracy is due to the advanced structure of the hybrid model, which provides better processing of variable and unstable data. In general, the SSN-BILSTM hybrid model can be used as an efficient tool in water resources management and river flow prediction under variable climatic conditions by providing more accurate predictions and reducing errors.

Keywords:

Flow Prediction , Hybrid Model , Missouri River , Deep Learning