Application of Artificial Neural Network and Fuzzy regression in remotely sensed monitoring of drought

Drought monitoring is a major issue for agricultural water management and environmental protection .In this study, artificial neural network and fuzzy regression models have been used to evaluate the performance of several remotely sensed indices retrieved from MODIS images including NDVI, VTCI, VHI, NVSWI, TCI and TVX for monitoring drought in 7 meteorological station across Iran namely Kermanshah, Tabriz, Kerman,Mashhad, Urumia,Yazd and Zanjan. The VHI, NVSWI, TCI and TVX had the highest number of significant coefficient of correlation with amount of rainfall in study stations. Based on error measures, the Fuzzy regression approach had the least error in modeling correlation of VHI, TCI and NVSWI with rainfall amount. Using the ANN model, the TVX found to be the best index in monitoring drought with highest accuracy. The results revealed that in the case of symmetric membership functions, changing the value of confidence level parameter would affect the value of fuzzy spread coefficient. For example increasing the confidence level parameter in case of VHI from 0.7 to 0.8 led to 50% increase of spread. In case of non-symmetric fuzzy coefficient, the peak point is sensitive to skewness factors; such that its value was increased for 22.2% moving from minimum to maximum skewness factor in case of TVX index. The decrease in confidence level parameter of TVX, which represents the degree of fuzziness, confirmed the better performance of artificial neural network in correlating TVX index and rainfall.