Assessment of efficiency of artificial neural network in predicting the trend of desertification processes by using GIS (Case study: Dehloran plain, Ilam)

Desertification is recognized as a main problem in the arid and semi-arid areas. Therefore, identification and prediction of the effective factors in development of desertification are very important for better management of these areas. The main purpose of this study was evaluating the accuracy of an artificial neural network model for predicting the desertification process and selects the most effective criteria on desertification in the Dehloran plain by using the Iranian model for desertification potential assessment (IMDPA). In IMDPA model, water and climatic were selected as effective factors in desertification. In this model, three indicators for climate criteria; annual precipitation, drought index (Standardized precipitation index; SPI and continued drought and for water criteria; ground water table depletion, sodium absorption ratio, Cl, electrical conductivity (EC) and total dissolved solids were evaluated. Each index was rated using of IMDPA model. Then desertification intensity and criteria maps were prepared using a geometric average for predicting period in ArcGIS®9.3. Final data were entered into neural network to predict. The results showed that the neural network model has a high efficiency for predicting the desertification process in the study area. The accuracy of the model was about 80% and mean square error (MSe) was less than one. In addition, the climate factor and the index of EC were found the most effective variables for predicting the desertification process. In 2015-2016 predicted the most important probable criteria affecting the intensity of desertification were climate and water with weighted average 2 (moderate in sub-class1, 2 and 3), 1.84 (moderate in sub-class 1and 2), respectively.