A robust and flexible algorithm to study spatial series like soil roughness profile has been introduced. It avoids using classic spectral analysis, considers the profile first and foremost as non-stationary and makes it possible to identify the separate domains inside the profile where chosen statistical parameters and roughness indexes have their own value. The method analyses a roughness profile considering it as an assemblage of several entities that may differ in terms of statistical properties and length, without establishing constraints as to number and extension. The method derives the variability of statistical and roughness properties along the profile and extracts the possible components - random and oriented - detectable inside the sample. Some examples of application illustrate the possibility offered by the method to study real roughness profiles recorded in the field by a portable laser microprofilometer. The procedure proposed allows the investigation of local roughness properties with varying degrees of accuracy and should be useful to monitor the differential evolution of roughness on patterned soil surface, increasing the overall information content. A general definition of 'ordered roughness' is introduced. The definition proposed seems more suited to current techniques for the numerical treatment of digital profiles and for the existing physical relationships between the scale of observation of roughness and the scale of the process investigated (hydraulic resistance, water storage in depressions).
