Regime-Wise Genetic Programming Model for Improved Streamflow Forecasting

Abstract

Forecasting of stream flow plays a vital role in flood forecasting studies, design, and operation of reservoirs. Several approaches such as physical models, conceptual models and statistical/black-box models are used to model complex uncertain peak flows in rivers. In the past, Genetic Programming (GP) have been a widely used for different hydrological applications. In this study we propose a regime-wise genetic programming model for efficient forecasting of streamflow during peak flows. In this approach, we first classify the flows into three regimes such as low, med and high based on their flow magnitude and develop separate GP models. The proposed approach was applied to a case study from Godavari River Basin, India. The results obtained show that the proposed approach of separate models for high flows performs better than the single model for all regimes.