Applications of artificial intelligence and time series models in runoff estimation(Case Study: Part of Halil river basin)

Accurate forecasting of runoff and flooding to avoid human and financial losses is one of the most challenging tasks in hydrological studies of a given locale. Therefore, researchers have paid more attention to the development of accurate flood forecasting models, including the use of artificial intelligence methods.

In this investigation, the efficiency of 3 models, ANN, GMDH and ARIMA, has been investigated in order to simulate the flood of a part of Halil river basin in Kerman province. ANN model is a non-linear modeling method that improves its performance over time. The GMDH composed code is an artificial intelligence model with exploratory self-organizing features, at the conclusion of which a complex system with optimal performance is formed. Composed ARIMA code builds a model to describe the structure of the data and then predict the time series. The input data to the above models included discharge, precipitation, temperature, wind and monthly humidity, and the simulated runoff values ​​were compared with the observed values.

In order to evaluate the accuracy of the models in this research, statistical indices were used and the results showed that the ANN model (RMSE=0.042, MSD=0.001, MAE=0.027) had the possibility to estimate the runoff with higher accuracy compared to the GMDH model (RMSE=0.068, MSD=0.005, MAE=0.056) and the ARIMA time series (RMSE=0.096, MSD=0.009, MAE=0.063) in the studied basin. The mean error in runoff estimation with ANN model has been reduced by 38.23% and 56.25%, respectively, compared to the values estimated with GMDH and ARIMA models. According to the results obtained in this study, the artificial neural network model has been able to show a better performance than the other two models in predicting the outputs due to its suitable structural ability to find the nonlinear relationship between the input and output data.