Radar remote sensing has been widely used to estimate moisture and surface roughness due to its sensitivity to the physical and geometrical parameters of the soil. There are different models to explain the relationship between radar backscattering, surface and sensor parameters. The most important of these models are the integral equation model (IEM) and the small perturbation model (SPM). Due to the complexity of these models, in order to estimate roughness and moisture a neural network is used for inversion of these models. In this article, the X-band of the SAR image is used to estimate the surface roughness. One of the innovations of this research is the use of fractal SPM model in surface roughness estimation. To evaluate the accuracy of roughness estimation from SAR image, digital terrain model (DTM) that prepared using lidar data is used. For calculation of field roughness, Euclidean geometry and fractal geometry have been used, and they have been compared with roughness estimated from SAR image using two fractal SPM and IEM models. The results of this research have shown that the best accuracy is related to the estimation of the surface roughness with the fractal SPM model, which is compared with the ground roughness measured by the fractal geometry method. The accuracy of this method is 22% better than the similar method with the IEM model. The results of this paper showed that the use of fractal physical model as well as fractal geometry significantly increases the accuracy of roughness estimation from SAR images.