Drought prediction using artificial intelligence models (case study of Fars province)
Subject Areas : BIABANZODAEImohammad ebrahim afifi 1 , vahid sohrabi 2
1 - Member of the Faculty of Geography Department of Islamic Azad University, Larestan Branc
2 - Geography and Urban Planning, Faculty of Humanities, Larestan Branch Azad University, Iran
Keywords: Drought, artificial intelligence, Fars province,
Abstract :
Drought is one of the most costly natural disasters in Iran. The necessary step to deal with it and adjust its consequences is to know and understand the vulnerabilitydimensions of each region.The suitable method for predicting drought for the next month,four artificial intelligence methods including deep learning (Alexnet network is used, which is one of the convolution networks).K-nearest neighboralgorithm, multi-layer support vector machine and decision tree areconsidered.became.The monthly rainfall data of 11 synoptic stations of Fars province during a 29-year statistical period (1988 to 2017) were used as experimentaldata. The standardized precipitation index (SPI) was calculated to show the drought situation in terms of intensity and duration in time scales of 1, 3, 6, 9, 12 and 24 months. At first, the precipitation data was placed as the input of the neural networks and the classification of the standardized precipitation index was placed as the output of the networks. 80% of the data was used for training and 20% of the data was used for testing the networks. The results showed that all the networks had the ability to predict drought, based on the evaluation criteria of the macro-f1 deeplearningnetwork in the time scale of 1 month with 22.71%, the most inefficient method and the decision tree with 64.65%, the most efficient method, with increasing scale Over time, the deep learning network improved its performance, in the 24-month time scale, with 35.65%, the best performance related to the deep learning network, followed by the support vector machine network with 57.40%
