Application of Artificial Neural Networks to the Simulation of Climate Elements, Drought Forecast by Two Indicators of SPI and PNPI, and Mapping of Drought Intensity; Case Study of Khorasan Razavi

Abstract

Drought is considered as a destructive disaster that can have irreversible effects on different aspects of life. In this study, artificial neural network was used as a powerful means of modeling nonlinear and indefinite processes in order to simulate drought intensities at 7 synoptic stations of Khorasan Razavi from more than 35 years ago up to the year 2014. Input data were the calculations of the two indicators of PNPI and SPI by DIC software, and the output layer (drought intensity) was taken to the Matlab software and employed as the teaching data (from 25 years), experiment (from 5 years), and validation (from another 5 years). The 3-9-1 structure of the network of layers had the maximum accuracy with the error rate of less than 2 % and high correlation (more than 90 %). After trial and error for each station through sigmoid stimulation function in the Perceptron network, it was observed that the stations of Mashhad and Quchan had the minimum error and the maximum error was related to the station of Neyshabur. The results of comparisons and observations showed that the artificial neural network had high efficiency in simulation of the data. The obtained correlation amount of 0.999 for the base station represented the small error of the model in prediction. Drought forecasting was performed in this study by the trained algorithm in the artificial neural network without using the observation data. The results showed that rainfall, temperature, and speed models had a positive role in forecasting the provinces that would experience drought. Due to its lower amount of error, SPI indicator was selected for mapping, the findings of which showed that the highest drought intensity belonged to the near normal to normal wet lands.