Jordan Journal of Civil Engineering

Paper Detail

Nonlinear Multivariate Rainfall Prediction in Jordan Using NARX-ANN Model with GIS Techniques

Volume 12, No. 3, 2018
Received: 2018/05/14, Accepted:


Omar Arabeyyat; Nawras Shatnawi; Mohammed Matouq;


This paper aims to simulate and predict the amounts of rainfall in semi-arid regions using Artificial Neural Network (ANN) with nonlinear autoregressive exogenous (NARX) input model. The rainfall precipitation readings of 26 stations for the last 30 years were used as an input for the ANN model. A code, developed in MATLAB at different hidden layers and delay times, was used for this purpose to select the best combination that could simulate the case in Jordan. The results revealed that a reduction of 1.4% in the annual average rainfall amounts in the next 10 years might happen. The simulation criteria depended on changing number of delays and number of neurons (hidden layers), which showed that using 2 delay inputs and 8 neurons gives best training for the ANN as per the computed mean squared error (MSE).


Rainfall, NARX, GIS, Time series prediction, Time delay, Neural network.