Forecasting Lake Water Levels Under Global Warming Using BiGRU with Quantile Regression

Abstract

Climate change and global warming exacerbate fluctuations in lake water levels, complicating disaster prevention and water management. Developing high-quality forecasting models crucially enhances decision-making and preventive actions. However, current studies focus on point estimation forecasts, neglecting crucial prediction intervals that can provide insightful knowledge for decision-making in real-life applications. This study introduces a method integrating a Bidirectional Gated Recurrent Unit with quantile regression to predict water levels and quantify uncertainty with a 90 % confidence interval defined by the 5th and 95th percentiles. Bayesian optimization was used to optimize hyperparameters of the deep learning model and each optimization examined 2000 hyperparameter combinations. Forecasting horizons are set to 1 day, 3 days, and 7 days to support both immediate actions and longer-term strategic management. The models demonstrate robust performance, with R2 of 0.98, 0.96, and 0.89, and RMSE of 0.02, 0.03, and 0.05 for the respective horizons using testing data.