The Use of Time Series and Artificial Neural Networks in Drought Simulation (Case Study: Bojnourd City)

Drought is a climatic phenomenon and is actually considered a part of the climate of a region. Drought has a hidden nature and the duration of its occurrence is long, and its effects appear in a non-structural way and as a result, the damages caused by it in various sectors such as agriculture, social, economic, environmental, etc. gradually appear. In this research, artificial neural network was used as a powerful tool in simulating the drought of Bojnourd city. For this purpose, the statistical data of precipitation, relative humidity, and temperature from ۱۹۹۷ to ۲۰۱۴ are the basis of the upcoming research. SPI drought index was used as sample output. ۷۰% of the data were considered as training data and ۳۰% as testing data. The networks used are of backpropagation type and radial basis function with error backpropagation algorithm and Lunberg-Marquardt learning method. Box Jenkins method from MINITAB software and BPI to BP۲۴ models from MATLAB software were used in the branch for drought simulation. The value of the correlation coefficient for the training phase (R) was ۰.۹۵ and for the test phase it was ۰.۸۱, which has the lowest error in the test phase.the results of the training phase were close to each other in most of the models. Among the selected models of the post-release network, the BP۱۹ model was selected as the selected example. RMSE determination coefficient was estimated with a value of ۰.۱۶ for the training stage and MAE (mean absolute error) was estimated as ۰.۰۰۷۱.

کلیدواژه ها:

Minitab ، Box Jenkins ، post release networks ، SPI ، Bojnourd