For the digital terrain modelling from airborne laser scanning data the identification of points that are reflections from bare earth is the main issue of process of elaboration ALS data. In this work the hierarchic method of identification of points belonging to terrain surface was presented. This method is based upon the approximation of data using moving polynomial surface. Moving polynomial parameters are estimated locally based upon M-estimators of robust estimation method. In the estimation process the depended on the distance weighting function and asymmetrical damping function were used. A lot of numeric tests on the real airborne laser scanning data were executed. This data had a form of 15 testing samples contained referenced data as correctly classified terrain and objects’ points. The method was verified based upon the comparison of data after automatic filtration with referenced sets of points. In the result of comparison the percentage errors of automatic filtering were determined. Total percentage filtering error was evaluated on the level from about 1% to about 12%. These values depended mainly on the terrain form and terrain coverage. It has been noticed that in the critical places e. g. along the dykes the including to the algorithm additional information a-priori as correct terrain points make the automatic filtering more accurate.

15-27