Predictive Modeling of Hourly Water-Level Fluctuations Based on the DCT Least-Squares Extended Model

Water-level is one of the critical parameters for a river. It has a close relation to human living & production and socio-economic sustainability development. WLF (water-level fluctuation) evaluation and forecasting for a river is then becoming increasingly important. For water resource planning and management, traditionally, mathematical models are separately developed and designed for sectorial applications. As predictions utilizing more different forecast variables require additional efforts and costs to acquire and predict the variables, advantages of time-series-based or data-driven modeling lie on its conciseness and good performance even higher accuracy. The Fourier-based analysis technology is a classical tool widely used for time-series analysis. However, the Fourier-related approach in its conventional form is not directly applicable to prediction. Addressing hourly WLF prediction from the viewpoint of time-series analysis, a called DCT-LS-extended (“discrete cosine transform (DCT)-based least-squares-extended”) forecast model is presented in this study. The DCT coefficients for the proposed DCT-based forecast modeling are determined in the least-squares sense on the basis of previous hourly WLF observations. Experiments at hydrological monitoring stations in the XiangJiang River of China yield stultifying results. Potentiality of the proposed method is demonstrated by further analysis. The proposed DCT-LS-extended model forecasts hourly WLFs best fitting with less than 12-term DCT coefficients. The proposed method may benefit other applications.