Scholarlink

<p>Conventional data-driven models require a substantial volume of meteorological and hydrological variables as input for the accurate prediction of river water level, which may not be feasible for many river basins having data scarcity. More particularly, the accurate prediction of river water level using a univariate approach often becomes difficult employing the conventional models. Placing this challenge as the key motivator, this study aims to propose a novel data-driven approach by combining five sophisticated preprocessing techniques, EMD, EEMD, CEEMDAN, VMD, and SVMD with a recently proposed model, namely cross-variable linear integrated enhanced Transformer (CLIENT), for improving the univariate predictions of river water level. The Rupsa-Pasur River, located at the southwestern coastal region of Bangladesh, was chosen as the study region due to its highly stochastic hydrological dynamics. Four statistical metrics, i.e., NRMSE, MAE, KGE, and NSE, were computed for all the developed models at both the training and testing phases following a uniform train-test ratio of 8:2. The Borda count strategy was implemented with a view to identifying the optimal data-driven model. Compared to the traditional lag features-based input combinations, the proposed framework achieved superior performance at both stations. Results exhibited that the SVMD-enhanced CLIENT model (referred to as the C6 model) outperformed all other models for both stations. The C6 model demonstrated better capability of recognizing the major portion of the hidden and advantageous signals within the time series. This led to in an NSE value of 0.999 during both the training and testing phases, indicating a notable enhancement compared to existing literature, while also achieving considerably lower computational costs for both stations. The framework’s capability to provide precise predictions based solely on historical water level data underscores its applicability in regions with limited data availability. Thereafter, an interactive graphical user interface (GUI) was developed with the C6 model following a five-fold cross validation and external validation.</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

A novel integration of cross variable transformer and signal decomposition for real-time prediction of river water level: an implication for sustainable water resources management

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月16日

查看全文
  • Md. Jobayer Parvez Ratul,
  • Usmi Akter,
  • Tajrian Mollick,
  • Md. Mahmudul Hasan,
  • Md. Tasim Ferdous,
  • N M Refat Nasher,
  • Md. Jahir Uddin

摘要

Conventional data-driven models require a substantial volume of meteorological and hydrological variables as input for the accurate prediction of river water level, which may not be feasible for many river basins having data scarcity. More particularly, the accurate prediction of river water level using a univariate approach often becomes difficult employing the conventional models. Placing this challenge as the key motivator, this study aims to propose a novel data-driven approach by combining five sophisticated preprocessing techniques, EMD, EEMD, CEEMDAN, VMD, and SVMD with a recently proposed model, namely cross-variable linear integrated enhanced Transformer (CLIENT), for improving the univariate predictions of river water level. The Rupsa-Pasur River, located at the southwestern coastal region of Bangladesh, was chosen as the study region due to its highly stochastic hydrological dynamics. Four statistical metrics, i.e., NRMSE, MAE, KGE, and NSE, were computed for all the developed models at both the training and testing phases following a uniform train-test ratio of 8:2. The Borda count strategy was implemented with a view to identifying the optimal data-driven model. Compared to the traditional lag features-based input combinations, the proposed framework achieved superior performance at both stations. Results exhibited that the SVMD-enhanced CLIENT model (referred to as the C6 model) outperformed all other models for both stations. The C6 model demonstrated better capability of recognizing the major portion of the hidden and advantageous signals within the time series. This led to in an NSE value of 0.999 during both the training and testing phases, indicating a notable enhancement compared to existing literature, while also achieving considerably lower computational costs for both stations. The framework’s capability to provide precise predictions based solely on historical water level data underscores its applicability in regions with limited data availability. Thereafter, an interactive graphical user interface (GUI) was developed with the C6 model following a five-fold cross validation and external validation.