Application of Machine Learning Approaches in Rainfall-Runoff Modeling (Case Study: Zayandeh_Rood Basin in Iran)

Run off resulted from rainfall is the main way of receiving water in most parts of the World. Therefore, prediction of runoff volume resulted from rainfall is getting more and more important in control, harvesting and management of surface water. In this research a number of machine learning and data mining methods including support vector machines, regression trees (CART algorithm), model trees (M5 algorithm) and artificial neural networks have been used to simulate rainfall- runoff process in Zayandeh_rood dam basin in Iran. Data used in this research included 9 years of daily precipitation, minimum temperature, maximum temperature, mean temperature, mean relative humidity of daily times 6:30, 12:30 and 18:30 and run off. A number of 3294 lines of data were totally used, and simulations were carried out in two different conditions: without previous run off data as input vectors (M1 condition), and with previous runoff data as input vectors of the models (M2 condition). Results show that machine learning techniques used in this research are not able to present acceptable predictions of runoff in M1 condition (without previous runoff data). However, predictions are considerably improved when previous runoff data are used as input beside other inputs (M2 condition). Between the models used in this research support vector machines (SVM) presented the most accurate results, as the values of RMSE for results presented by SVM, regression tree, model tree and artificial neural network are 2.4, 6.71, 3.2 and 3.04, respectively.