Assaying of ability of statistical models for remote sensed vegetation cover maping of desert areas

Measurement and monitoring of vegetation cover fraction is very important in many environmental studies. Although remote sensing techniques have shown successful to estimate the parameter, but still remote sensed measurements of it encounter to special problems in desert areas. In this research has tried, the relationship between reflections in the different bands of LISSIII and TM sensors and percentage of vegetation cover be investigated using field measurements of vegetation cover fraction in a desert area. Accordance with the findings of this research, if we consider the relationship linear, only correlation between the percentage of vegetation and the near-infrared band has shown significant at the 0.05 level (2-tailed). While if we consider it nonlinear, correlation between reflections of all bands and the percentage of vegetation cover have shown significant at the 0.01 level (2-tailed). Also according to the results was determined using a linear multivariate regression equation and using all the bands, it is possible to make the relationship significant at all levels and to improve accuracy of estimations of the model. Even was observed with the change in the type of used regression model and covert it to a non-linear multivariate regression model, it is possible to improve the correlation coefficient and the coefficient of determination between the observed and estimated vegetation cover percentage significantly, In a way that correlation coefficient and the coefficient of determination between the observed and estimated vegetation cover percentage at best condition is calculated 0.917 and 0.841 respectively. Based on the results obtained in this study to increase the accuracy of modeling the nonlinear relationship between percent vegetation cover and the band satellite reflections using artificial neural network is the best way. In this research using artificial neural networks increase the accuracy of modeling significantly, so that the the coefficient of determination was passed 0.9 at best. Also based on neural network modeling hundred repetitions of this research can be refer to a neural network with radial function has better accuracy of modeling than multi-layered neural network to determine the percentage of vegetation using satellite data in drylands.