A two-stage modelling method for multi-station daily water level prediction

Water level prediction is an essential task in inland water transportation and infrastructure operation. In recent years, the level of uncertainty in the water level variation has increased significantly due to the climate change. Therefore, the need to develop more robust and accurate models for multi-station daily water level prediction along the long and volatile inland rivers has greatly increased. This research proposes a two-stage modelling method to enhance the accuracy and efficiency in simultaneous prediction of daily water levels for multiple stations in inland rivers. Furthermore, taking the Yangtze River trunk line as case study, the daily water data of 19 stations are collected and utilised to verify the performance of the models. First, we divide the 19 stations along the Yangtze River trunk line into 6 clusters by dynamic time warping (DTW) and hierarchical clustering algorithm (HCA). Then, the long short-term memory (LSTM) network and seasonal autoregressive integrated moving average (SARIMA) model are tailored to construct a multi-station daily water level prediction (MSDWLP) model for each cluster. Finally, to validate the proposed method, the daily water level data of 912 consecutive days from the 19 stations are employed. The results demonstrate that the proposed approach can yield more reliable forecasts than traditional deterministic models. Insight from the models can be used to predict daily water levels to better inform decision-making about waterborne transportation, water resources management, and water emergency response.