Efficient Prediction of Discharge and Water Levels Using Ensemble Learning and Singular-Spectrum Analysis-Based Denoising

This work addresses forecasting two essential factors in river hydrodynamics, which are discharge (Q) and water (H) levels. The accurate forecast of the two has long been a challenge in hydrological researches and flood prediction. While the traditional statistical models fail to capture the peak discharge during flooding seasons (i.e., due to the excessive level values), the simulation's numerical models face the difficulty of precise input parameters (e.g., measured values of surface zones, root zones, etc.). The emerging deep learning shows a lot of potential in solving the challenges of Q and H prediction. However, applying deep learning in such a context is not straightforward due to the following critical issues. First, the amount of training data is insufficient due to the data collection is non-trivial. Second, although lacking, the collected data incurs noises (e.g., measurement errors). We aim to overcome those shortcomings in a newly proposed deep learning model that accurately predicts Q and H. The model is a new ensemble of the one-dimensional convolutional neural network (1D-CNN), long short term memory (LSTM) models, to handle the insufficient data issue. Moreover, we adopt the Singular-Spectrum Analysis technique to eliminate noise from the collected data. The experimental results show that our proposed approach outperforms existing methods.