A new regionalization approach and its application to predict flow duration curve in ungauged basins

Prediction in ungauged basins is an important task for water resources planning and management and remains a fundamental challenge for the hydrological community. Regionalization is typically used to estimate parameter values of hydrological predictive tools for catchments without observed streamflow. This study proposes a new regionalization method, called the index model. The index model establishes a nonparametric relationship between each parameter of predictive tools and a linear combination of predicators. This method is able to describe a wide range of functions, linear or nonlinear, and avoids the potential misspecification which usually occurs as a result of using the ordinary linear regression. We illustrate the method by predicting flow duration curves at 227 unimpaired catchments in southeast Australia. This study also compares results from regional models based on the linear regression, nearest neighbour and hydrological similarity. The results show that the index model produces the most accurate prediction with highest coefficients of efficiency, followed by the linear regression. In particular, the index model improves the model performance substantially at catchments where the linear regression is a fair to poor fit. The index model was also interpretable and showed that potential evapotranspiration and summary statistics of rainfall are predominant in prediction.