Selection of a Proper Model to Predict Monthly Drought in Semnan Using Weather Data and Linear and Nonlinear Models

Abstract

Nowadays, greater recognition of drought and introducing its monitoring systems, particularly for the short-term periods, and adding predictability to these systems, could lead to presentation of more effective strategies for the management of water resources allocation. In this research, it is tried to present appropriate models to predict drought in city of Semnan, Iran, using time series, adaptive neuro-fuzzy inference system (ANFIS) and artificial neural networks (MLP and RBF). For these modeling processes, average monthly meteorological parameters of rainfall, temperature, minimum temperature, maximum temperature, relative humidity, minimum relative humidity, maximum relative humidity and SPI drought index were used during the period 1966 to 2013. The results showed that among the many developed models, the ANFIS model, with input data of average rainfall, maximum temperature, SPI and its last-month value, 10 rules and Gaussian membership function, showed appropriate performance at each stage of training and testing. The values of RMSE, MAE and R at training stage were 0.777, 0.593 and 0.4, respectively, and at testing stage were 0.837, 0.644 and 0.362, respectively. Then, the input parameters of this model were predicted for the next 12 months using ARIMA model, and SPI values were predicted for the next 12 months. The ANN and time series methods with low difference in error values were ranked next, respectively. The input parameters SPI and temperature had better performance and rainfall parameter had weaker performance.

Keywords

• adaptive neuro-fuzzy inference system
• artificial neural network
• prediction of drought
• time-series.