Runoff prediction using black and gray box models

In the past decade, machine learning for empirical rainfall–runoff modeling is considered to be a promising approach as a useful complement to hydrologic models, particularly in basins where data to support process-based models are limited. In this paper, we used black-box models (i.e. neuro-fuzzy and support vector machine) and gray-box models (i.e. TOPMODEL and HBV) for simulating the transformation of daily rainfall-runoff process in the Nodeh khormaloo watershed located in Gorganrood River Basin and compare their performance in terms of predictive accuracy. For the black-box models, the three input vectors including discharge, temperature and rainfall are selected in nine different scenarios based on the sequential time series data. Our result show that the neuro-fuzzy model which consists of three antecedent values of flow and one antecedent values of temperature outperforms other models when the root mean square error and coefficient of determination are used as quality indicators. In general, the black- box models outperformed the HBV and TOPMODEL simulations for the calibration and validation data sets. A detailed comparison of the overall performance indicated that the neuro-fuzzy and SVM models predicted runoff in warm months were consistently lower than that in the cold months.