Evaluating Different Machine Learning Models for Runoff Modelling

Moscow State University of Civil Engineering, 26, Yaroslavskoye shosse, Moscow, 129337, Russia

¹ Corresponding author: alaa-slieman@hotmail.com

Abstract

Estimation and forecasting of hydrological factors are of particular importance in hydrological modelling, and surface runoff is one of the most important of these factors. Machine learning (ML) models have attracted the attention of researchers in this field. So, this article aims to evaluate several types of ML models such as autoregressive integrated moving average (ARIMA), feed forward back propagation artificial neural network (FFBP-ANN), and adaptive neuro-fuzzy inference system (ANFIS) models in order to estimate runoff values at Al-Jawadiya meteostation in the Orontes River basin in Syria. A large number of ARIMA models were built and the seasonal effect on the models also verified. After that, FFBP-ANN models were used with the change in the number of inputs, the number of hidden layers, and the number of neurons in the hidden layer. Also, a large number of FIS models have been built and artificial neural algorithms have been used in the process of model parameters optimization. The results showed a preference for artificial intelligence models in general over ARIMA models, as well as a slight preference for FFBP-ANN models over ANFIS models. This study recommends expanding the use of ML models to reach the best models for forecasting hydrological factors.