Dependence structure analysis of multisite river inflow data using vine copula-CEEMDAN based hybrid model.

Area of Science:

• Hydrology and Water Resources Management
• Time Series Analysis
• Statistical Modeling

Background:

• Existing univariate models fail to capture the dependency structure in multivariate random variables, particularly for multi-site river inflow data.
• Accurate modeling of joint distributions is crucial for managing river basin systems effectively.

Purpose of the Study:

• To propose a novel hybrid approach, Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) Vine copula, for modeling multi-site river inflow.
• To address the limitations of univariate models in capturing the dependency structure of multivariate random variables.

Main Methods:

• A two-stage hybrid model was developed: CEEMDAN for feature extraction and multiple models for component prediction, followed by Canonical Vine for joint uncertainty modeling of residuals.
• Daily river inflow data from the Indus River Basin was used for model application and validation.

Main Results:

• The proposed CEEMDAN-Vine copula model demonstrated superior prediction performance compared to benchmark models (CEEMDAN alone, Vector Autoregressive, and copula-based ARMA).
• The model achieved significantly lower Mean Absolute Relative Error (MARE), Mean Absolute Deviation (MAD), and Mean Square Error (MSE), with higher Nash-Sutcliffe Efficiency (NSE).

Conclusions:

• Modeling the dependence structure of multi-site river inflow data is essential for improving prediction accuracy.
• The CEEMDAN-Vine copula approach offers a robust and effective method for hydrological forecasting, especially in systems with significant joint dependencies.