Assessment of the performance of eight filtering algorithms by using full-waveform LiDAR data of unmanaged eucalypt forest

Abstract

In this study the strengths and weaknesses of eight filtering algorithms are evaluated by using the mean, standard deviation and RMSE metrics. Seven of these algorithms are implemented in the freeware software ALDPAT (Airborne LiDAR Data Processing and Analysis Tools) and the eighth, known as the Axelsson filter, in the commercial software Terrascan. The referred metrics are calculated by using DTM of topographic surfaces with quite different morphologies and vegetation covers. Forty-three of these surfaces, on circular plots of 400 m2 each, are covered by brushwood and unmanaged eucalypt forest with different stand characteristics. The mean tree density is around 1600 trees per hectare. The reference DTM for assessing the DTM produced by filtering full-waveform LiDAR data using the eight filtering algorithms are created with the help of a total station and geodetic GNSS receivers. The results show that the Axelsson and the so-called Polynomial Two Surface Fitting filters give the best results in terms of RMSE. Nonetheless, the results also show that all the tested filters are suitable for the filtering of full-waveform LiDAR data used in forestry related work, and collected over areas with great amount and high brushwood, chaotic eucalypt tree distribution and high tree density. The results obtained for a forest area with such characteristics – among which it should be mentioned a RMSE of 15 cm - are quite surprising.