Assessment of NARX Neural Network in Prediction of Daily Precipitation in Kerman Province

Precipitation is one of important parameters of climatology and atmospheric science that have more importance in human life. recently, extensive flood and drought entered many damage to most parts of the world.
Precipitation forecasting has important role in management and warning of this problem. Due to the interaction of various meteorological parameters in the calculation of rain, leads it to a very irregular and chaotic process.
The purpose of this study, assessment of forecasting precipitation, using data from meteorological stations of the using common statistical period (2012-1989) in Kerman, Baft, Miandeh Jiroft.
In this way, to the training of the artificial neural networks with structure Perceptron, Nonlinear Autoregressive External. Effective Factors in the rain, as input for Artificial Neural Networks and precipitation was considered as the output of the Network. Statistic indicators MSE, R were used for performance evaluation of the models.
The analysis of output results from, Nonlinear Autoregressive External Neural Networks shown that these models have better accuracy and a high ability to forecast precipitation than Perceptron Neural Networks.
The results showed the more exact method concerned to the (NARX) model. The 42 models with all parameters with Levenberg Marquat rule and sigmoid function had the best topology of the model in three stations. Overall, evaluation of NARX results showed that the errors of ANN were negligible. The NARX showed high sensitivity to relative humidity.