Assessing the Performance of Machine Learning Algorithms for Water Level Prediction in the Chao Phraya River and its Tributaries: A Focus on Low and High Water Levels

作     者：Priyasiri, Wilmat D.S.M. Rittima, Areeya Kraisangka, Jidapa Sawangphol, Wudhichart Phankamolsil, Yutthana Talaluxmana, Yutthana 

作者机构：Graduate Program in Environmental and Water Resources Engineering Department of Civil and Environmental Engineering Faculty of Engineering Mahidol University Thailand Faculty of Information and Communication Technology Mahidol University Thailand Environmental Engineering and Disaster Management Program Mahidol University Kanchanaburi Campus Thailand Department of Water Resources Engineering Faculty of Engineering Kasetsart University Thailand 

出 版 物：《SSRN》 

年 卷 期：2024年

核心收录：

主　　题：Rain 

摘      要：High precision of hydrological prediction is crucial for real–time operation of flood and drought risk mitigation and strategic planning. This study assessed the predictive performances of three Machine Learning (ML) algorithms;XGBoost, Random Forest (RF), and Multi–Layer Neural Networks (MLNNs) for water level prediction with a focus on low and high water levels. The one–day and one–week water level prediction models for six key gauged stations along the Chao Phraya River and its major tributaries were accordingly developed. Selecting the combination of input features was carried out and structured in the prediction model at all gauged stations based on the physical river–reservoir system using past water level, rainfall, controlled reservoir outflow, and upstream discharges with different travel times. The statistical evaluation indicated that both XGBoost and RF, the tree–based ML models, with rainfall input robustly outperformed than MLNNs, as it strongly achieved higher R2 from 0.957 to 0.999 for model training and from 0.743 to 0.995 for model testing and lower MAE, MSE, and RMSE values for all prediction scenarios. Among these three ML algorithms applied for daily prediction, RF demonstrated the superior performance for low water level prediction exhibiting the smallest percentage error of overestimating lying between +0.0088% and +0.9380%. XGBoost, RF, and MLNNs, demonstrated acceptable performance level for high water level prediction as it shows small percentage error of both overestimating and underestimating lying between –2.2696% and +1.1587%. Additionally, it can capture the entire testing dataset with high precision than weekly model. © 2024, The Authors. All rights reserved.