Scholarlink

<p>This study evaluates the performance of hybrid models combining vine copula with wavelet decomposition and complete ensemble empirical mode decomposition (CEEMD) for daily river discharge simulation at two hydrological stations (Abajalo and Tapik in Iran). The proposed approaches—C-vine-wavelet (decomposition levels 2, 3, and 4) and C-vine-CEEMD—were assessed using multiple statistical metrics, including the root mean square error (RMSE), the Bayesian information criterion (BIC), the Akaike information criterion (AIC), and the Nash–Sutcliffe statistic (NSE), Willmott’s Index, and Explained Variance Score, along with Taylor diagram analysis for comprehensive model comparison. Results indicate that the C-vine-wavelet (level 3) model consistently outperformed other methods, achieving the lowest RMSE (2.99&#xa0;m<sup>3</sup>/s at Abajalo and 4.81&#xa0;m<sup>3</sup>/s at Tapik) while maintaining high accuracy in trend detection (NSE &gt; 0.6, Willmott’s Index &gt; 0.9). The C-vine-CEEMD&#xa0;model demonstrated superior capability in capturing temporal variability (highest NSE = 0.72 at Abajalo) but exhibited higher errors in peak flow estimation. Taylor diagram analysis confirmed that both top-performing models maintained a strong correlation (<i>R</i> ≈ 0.8–0.9) with the observed data, although with a slight overestimation in extreme events. A key finding was the station-dependent performance of the models, with CEEMD showing better results at Abajalo than Tapik, suggesting that hydrological characteristics influence model effectiveness. The wavelet-based approach at level 3 decomposition provided an optimal balance between noise reduction and signal preservation, making it the most reliable choice for discharge simulation. The findings support the use of C-vine-wavelet (level 3) as a robust method for daily discharge prediction, while emphasizing the need for site-specific model calibration in practical applications.</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 hybrid framework: vine copulas synergized with multiscale decomposition for enhanced daily river discharge simulation

  • 期刊论文

  • 发表日期：

    2026年5月21日

查看全文
  • Majid Rastgari,
  • Samad Emamgholizadeh,
  • Emad Mahjoobi,
  • Mohammad Nazeri Tahroudi

摘要

This study evaluates the performance of hybrid models combining vine copula with wavelet decomposition and complete ensemble empirical mode decomposition (CEEMD) for daily river discharge simulation at two hydrological stations (Abajalo and Tapik in Iran). The proposed approaches—C-vine-wavelet (decomposition levels 2, 3, and 4) and C-vine-CEEMD—were assessed using multiple statistical metrics, including the root mean square error (RMSE), the Bayesian information criterion (BIC), the Akaike information criterion (AIC), and the Nash–Sutcliffe statistic (NSE), Willmott’s Index, and Explained Variance Score, along with Taylor diagram analysis for comprehensive model comparison. Results indicate that the C-vine-wavelet (level 3) model consistently outperformed other methods, achieving the lowest RMSE (2.99 m3/s at Abajalo and 4.81 m3/s at Tapik) while maintaining high accuracy in trend detection (NSE > 0.6, Willmott’s Index > 0.9). The C-vine-CEEMD model demonstrated superior capability in capturing temporal variability (highest NSE = 0.72 at Abajalo) but exhibited higher errors in peak flow estimation. Taylor diagram analysis confirmed that both top-performing models maintained a strong correlation (R ≈ 0.8–0.9) with the observed data, although with a slight overestimation in extreme events. A key finding was the station-dependent performance of the models, with CEEMD showing better results at Abajalo than Tapik, suggesting that hydrological characteristics influence model effectiveness. The wavelet-based approach at level 3 decomposition provided an optimal balance between noise reduction and signal preservation, making it the most reliable choice for discharge simulation. The findings support the use of C-vine-wavelet (level 3) as a robust method for daily discharge prediction, while emphasizing the need for site-specific model calibration in practical applications.