A sequential iterative dual-filter for Lidar terrain modeling optimized for complex forested environments

This paper introduces a sequential iterative dual-filter method for filtering Lidar point clouds acquired over rough and forested terrain and computing a digital terrain model (DTM). The method belongs to the family of virtual deforestation algorithms that iteratively detect and filter objects above-the ground surface. The method uses both points and raster models to do so. The algorithm performance was first tested over a complex badlands environment and compared to a reference model obtained using a traditional TIN-Iterative approach. It was further tested on a benchmark site of the ISPRS (site 5) representing mainly forests and slopes. Over badlands, the resulting DTM elevation RMSE was 0.14 m over flat areas, and increased to 0.28 m under forested and rough terrain. The later value was 12.5% lower than the one obtained with a TIN-Iterative approach. Over the ISPRS site, the TIN-Iterative model provided better results for 3 out of the 4 sample sites. But the proposed algorithm, still worked fairly well provided a total classification error of 5.52%, and is well ranked compared with other algorithms. While the TIN-iterative approach might work better with low density, the proposed one is a good alternative to process high density point cloud and compute DTMs suitable for modeling either hydrodynamic or morphological processes under forest cover at a local scale.

► We developed a Lidar filtering method to compute DTM under complex and forested areas. ► The method belong to the family of virtual deforestation algorithms. ► Algorithm performance was tested against other methods for a badlands and a ISPRS reference site. ► The proposed method performed better than the reference algorithm using high density Lidar data. ► under low point densities TIN-iterative method perform better.