ESTIMATING FRACTAL DIMENSION OF PROFILES - A COMPARISON OF METHODS

This paper examines the characteristics of four different methods of estimating the fractal dimension of profiles. The semi-variogram, roughness-length, and two spectral methods are compared using synthetic 1024-point profiles generated by three methods, and using two profiles derived from a gridded DEM and two profiles from a laser-scanned soil surface. The analysis concentrates on the Hurst exponent H, which is linearly related to fractal dimension D, and considers both the accuracy and the variability of the estimates of H. The estimation methods are found to be quite consistent for H near 0.5, but the semivariogram method appears to be biased for H approaching 0 and 1, and the roughness-length method for H approaching 0. The roughness-length or the maximum entropy spectral methods are recommended as the most suitable methods for estimating the fractal dimension of topographic profiles. The fractal model fitted the soil surface data at fine scales but not a broad scales, and did not appear to fit the DEM profiles well at any scale.