Advanced water level prediction for a large-scale river–lake system using hybrid soft computing approach: a case study in Dongting Lake, China

Water level prediction is vital in developing a sustainable conceptual design of water infrastructures, providing flood and drought control measures, etc. However, due to the complexity and many other inter-related influencing factors within a catchment, water level prediction remains a challenging task. A reliable method that is able to extract the non-linear behaviors of various parameters effectively, and thus enhances the modelling capability in terms of computation time and accuracy is required. Therefore, the Dongting Lake of China, a large-scale river–lake system has been selected for this study. The main aim is to provide a practical method for advanced water level prediction at the downstream outlet of Dongting Lake for flood warning purposes. The novelty of this study is the adoption of a soft computing modelling approach, based on minimum input requirements to reduce its dependency on too many inputs which might limit its functionality in the future. The results obtained show that the model developed can predict the hourly water level in Dongting Lake accurately with an error of 1.2%. It is able to provide an advanced water level prediction of 21 h ahead of the time step, and thus applicable for early flood warning to the surrounding area with densely populated townships.