EEMD- and VMD-based hybrid GPR models for river streamflow point and interval predictions

Abstract

Reliable river streamflow (RSF) forecasting is an important issue due to its impact on planning and operation of the water resources system. In this study, based on Lower Upper Bound Estimation (LUBE), hybrid artificial intelligence methods were used for point and interval prediction of monthly RSF. Two states based on stations' own data and upstream stations' data were considered for RSF modeling of the Housatonic River during the period of 1941–2018. Ensemble Empirical Mode Decomposition (EEMD) and Variational Mode Decomposition (VMD) methods were used for enhancing the streamflow point forecasting accuracy. Interval Prediction (PI) was applied for tolerating increased uncertainty. Results showed that in state 1, the error criterion value for the superior model decreased from 0.155 to 0.082 and 0.09 for the EEMD- and VMD-based models, respectively. Generally, hybrid models increased the modeling accuracy between 20% and 40%. Via the integrated approaches, the upstream stations' data was successfully used for streamflow prediction of stations without data. In this state, the PI coverage probability values for the VMD-based model were approximately 12% higher than the single model. Generally, the VMD-based model led to more desirable results due to having higher PI coverage probability and lower mean PI width values. HIGHLIGHTS AI methods were applied to model the RSF in successive hydrometric stations.; To obtain a model with higher efficiency, EEMD and VMD techniques were applied.; Interval prediction was applied for providing more details for practical operation decisions.;

Keywords

• interval prediction
• lower upper bound estimation
• streamflow
• successive points
• variational mode decomposition