---

Drought is one of the most destructive natural disasters in human societies that cause irreparable impacts on agriculture, environment, society and economics. So, awareness of occurrence of droughts can be effective in reducing losses. In this study, in order to modeling and forecasting drought severity in a 37 year time period (1971-2007) in 21 meteorological stations, located in the cold semi-arid region of north-west Iran, artificial neural networks was used. The input data was annual rainfall data and annual drought precipitation index for all stations that 80% of the data (1971-2000) used for training the network and other 20% (2001-2007) used for testing it and in the next step drought severity predicted for the years 2008 to 2012 by the trained algorithm without using actual and existed data in this period. The appropriate structure for the network, based on Multi Layer Perceptron with three hidden layer, Back Propagation algorithm, Sigmoid transfer function and 10 neurons in middle layer. The results show that the artificial neural networks are well able to predict the non-linear relationship between rainfall and drought as it can simulate drought precipitation index values largely consistent with the real values with more than 97% regression and less than 5% error. So, drought can be predicted by this method in future and also it is useful in water resources management, drought management and climate change. 

---

 
Suspended particles management is one of the important issues in controlling the air pollution of cities. These particles cause and develop heart and respiratory diseases in people. Mashhad is considered as one of the main and populous cities of Iran. Because of its climatic conditions and its tourism, the city is at the highest risk of this type of pollution. We attempted to use the multi-layer perceptron (MLP) artificial neural network and a Markov chain model to predict PM₁₀ concentrations the city. We applied hourly data of CO, SO₂, PM_2.5 and temperature in late March and April 2015. Out of 1488 data series, 1300 data were used for network training and 188 data were used for validation. The results indicated the optimal performance of these methods in predicting of the amount of pollutants and also the probability of occurrence of hours with different quality of contamination. The best MLP artificial neural network model predicted the amount of pollutant particles with a coefficient of determination (R²) 0.88, index of agreement of  0.91 and a mean square error of  2.26. Also, the Markov model with average absolute error predicted about 0.1 percent of the probability of transferring the condition and the continuation of different states of air pollution caused by suspended particles.