Watershed runoff modeling through a multi-time scale approach by multivariate empirical mode decomposition (MEMD).

Area of Science:

• Hydrology and Water Resources
• Environmental Science
• Data Analysis

Background:

• Runoff time series are nonstationary due to climate variability and vegetation dynamics, complicating accurate modeling.
• Understanding temporal features and influencing factors is key to improving runoff prediction accuracy.
• The Yihe watershed in northern China's rocky mountains presents a challenging case for hydrological analysis.

Purpose of the Study:

• To apply Multivariate Empirical Mode Decomposition (MEMD) for analyzing temporal scales of monthly runoff and its drivers.
• To assess the effectiveness of MEMD in enhancing the accuracy of nonstationary runoff modeling.
• To compare modeling performance using decomposed intrinsic mode functions (IMFs) versus original time series data.

Main Methods:

• Multivariate Empirical Mode Decomposition (MEMD) to decompose runoff and influencing factors (precipitation, NDVI, temperature, humidity, evapotranspiration) into intrinsic mode functions (IMFs) and a residue.
• Stepwise Multiple Linear Regression (SMLR) to model the decomposed IMFs and residue.
• Calculation of IMF contribution rates and analysis of predominant trends.

Main Results:

• MEMD decomposed data into six IMFs and a residue, revealing the annual cycle as the primary driver for runoff, precipitation, NDVI, temperature, and potential evapotranspiration.
• Quarterly oscillations significantly influenced relative humidity.
• A decreasing trend was observed in runoff, precipitation, NDVI, temperature, relative humidity, and potential evapotranspiration from 2006 to 2015.
• SMLR models using decomposed data (R²: 0.53–1.0) outperformed models using original data (R²: 0.17–0.6).
• The integrated MEMD-SMLR model improved R² by 24.2% compared to the standard SMLR approach.

Conclusions:

• MEMD is an effective technique for improving the accuracy of nonstationary runoff modeling.
• Decomposition of time series data into IMFs captures essential temporal dynamics for better hydrological predictions.
• The study highlights the importance of considering temporal features and influencing factors for robust water resource management.