Scholarlink

<p>Reliable and continuous streamflow prediction is a critical goal in hydrology, particularly for water resources assessment and flood management. Here, we evaluate the effectiveness of five regression-based methods, including least median squares regression (LMSR), support vector regression (SVR), pace regression (PR), linear regression (LR), and isotonic regression (IR), in predicting monthly streamflow with limited hydrological data. We utilized historical streamflow and meteorological data from an ungauged watershed in Algeria to train and validate the models. Six physical factors affecting the streamflow (e.g., flow rate, flow velocity, depth, width, mean surface velocity, and hydraulic radius) were considered in the modeling process. Model performance was assessed using metrics such as Nash-Sutcliffe efficiency (NSE), root mean square error (RMSE), and coefficient of determination (R<sup>2</sup>). The IR model demonstrated superior performance across all metrics, achieving the lowest RMSE of 0.814 and 2.559, and the highest R<sup>2</sup> values of 0.999 and 0.993 during the training and testing phases, respectively. The SVR, PR, and LR models exhibited comparable performance, with RMSE around 2.8–3.3 and 2.7–3.0, and R<sup>2</sup> around 0.996 and 0.987–0.990 during the training and testing phases, respectively. In contrast, the LMSR model performed worse than the other models, with the highest RMSE of 33.596 and 19.276, and the lowest R<sup>2</sup> of 0.992 and 0.981 during the training and testing phases, respectively. The robustness and generalizability of the IR model were evident as it consistently outperformed the other models in both the training and testing phases. The SVR, PR, and LR models, while closely ranked due to similar performance across both phases, were significantly outperformed by the IR model. The LMSR model lagged, emphasizing the importance of refining or considering alternative models for better performance. Our study contributes to the field by providing a detailed comparison of different regression-based machine learning models for streamflow prediction. The findings underscore the effectiveness of the IR model and highlight the need for further exploration of alternative models to improve performance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Investigation into the Prediction Success of Different Regression-Based Models for Monthly Streamflow Prediction in an Ungauged Watershed

  • 期刊论文

  • 发表日期：

    2025年8月26日

查看全文
  • Yahi Takai Eddine,
  • Zeghmar Amer,
  • Marouf Nadir,
  • Abolfazl Jaafari

摘要

Reliable and continuous streamflow prediction is a critical goal in hydrology, particularly for water resources assessment and flood management. Here, we evaluate the effectiveness of five regression-based methods, including least median squares regression (LMSR), support vector regression (SVR), pace regression (PR), linear regression (LR), and isotonic regression (IR), in predicting monthly streamflow with limited hydrological data. We utilized historical streamflow and meteorological data from an ungauged watershed in Algeria to train and validate the models. Six physical factors affecting the streamflow (e.g., flow rate, flow velocity, depth, width, mean surface velocity, and hydraulic radius) were considered in the modeling process. Model performance was assessed using metrics such as Nash-Sutcliffe efficiency (NSE), root mean square error (RMSE), and coefficient of determination (R2). The IR model demonstrated superior performance across all metrics, achieving the lowest RMSE of 0.814 and 2.559, and the highest R2 values of 0.999 and 0.993 during the training and testing phases, respectively. The SVR, PR, and LR models exhibited comparable performance, with RMSE around 2.8–3.3 and 2.7–3.0, and R2 around 0.996 and 0.987–0.990 during the training and testing phases, respectively. In contrast, the LMSR model performed worse than the other models, with the highest RMSE of 33.596 and 19.276, and the lowest R2 of 0.992 and 0.981 during the training and testing phases, respectively. The robustness and generalizability of the IR model were evident as it consistently outperformed the other models in both the training and testing phases. The SVR, PR, and LR models, while closely ranked due to similar performance across both phases, were significantly outperformed by the IR model. The LMSR model lagged, emphasizing the importance of refining or considering alternative models for better performance. Our study contributes to the field by providing a detailed comparison of different regression-based machine learning models for streamflow prediction. The findings underscore the effectiveness of the IR model and highlight the need for further exploration of alternative models to improve performance.