Hybrid Hydrological Data-Driven Approach for Daily Streamflow Forecasting

Abstract

Hydrological forecasting is key for water resources allocation and flood risk management. Although a number of advanced hydrological forecasting methods have been developed in the past, daily (or subdaily) forecasting remains a major challenge in engineering hydrology. The uncertainties associated with input data, model parameters, and model structure necessitate developing more robust modeling techniques. In this study, a hybrid machine-learning approach based on hydrological and data-driven modeling is developed for daily streamflow forecasting. The proposed hybrid hydrological data-driven model (HHDD) approach succeeds in improving daily prediction compared to that predicted by the standard conceptual hydrological model (HYMOD). In addition, the developed HHDD model is more robust in terms of providing direct uncertainty analysis results. The results indicate that a better resemblance of streamflow pattern is achieved by integrating physically based and data-driven approaches into the developed HHDD model.