Optimizing the Long Short-Term Memory (LSTM) model by Bayesian method for salinity intrusion forecasting: a study at Dai Ngai station, Soc Trang province, Vietnam

Abstract

Salinity intrusion forecasting is essential and challenging for hydrometeorology, especially in climate change. Employing machine learning (ML) algorithms and conventional forecasting techniques are gaining popularity and providing high performance. This study presents a method to optimize a machine learning model based on the Long Short-Term Memory (LSTM) algorithm for multistep-ahead salinity forecasting (up to 7 days) at Dai Ngai station, Soc Trang province. The optimization method based on the Bayesian algorithm for hyperparameters optimization and input predictors optimization has been highly effective for predicting salinity with a lead time of 1 day to 7 days. Specifically, the forecast results evaluated by the R2 and RMSE indexes both give satisfactory results on the test data set (with lead time from 1 day to 7 days, R2 ranges from 0.9 to 0.54, and RMSE ranges from 0.27 to 0.53). This study is a premise for improving machine learning models for short-term and long-term salinity intrusion prediction in the Mekong delta and Vietnam.