Application of geostatistical approach in generating DEM for relief studies using UAV in forest areas

Abstract

In relief studies, and especially when performing terrain analysis, a key product is the Digital Elevation Model (DEM), which is often generated from data collected using an Unoccupied Aerial Vehicle (UAV). However, areas covered by high vegetation, mainly forests, are a significant problem when acquiring aerial photography data, since UAV Structure-from-Motion (UAV-SfM) photogrammetry does not provide information from sites located under the tree canopy. This article aims to demonstrate how to improve a DEM when generating it in areas covered with high vegetation based on an aerial photography dataset acquired by UAV missions, using geostatistical methods. The study was conducted in central Poland, where a parabolic dune covered by dense forest has formed, along with a blowout occurring in its vicinity, which is periodically filled with water. The study showed that the use of Ordinary Kriging (OK) in the reconstruction of relief covered with high vegetation provided better results compared to interpolation methods implemented in the photogrammetry software used. This is confirmed, by the computed low error values of Mean Error (ME): −0.0003 m, Root Mean Square Error (RMSE): 0.0777 m from Cross-Validation (CV) and low values of 0.002–0.44 m standard deviation of kriging (SD_OK). The resulting DEM shows high similarity of 69.17 % (coefficient of determination) and a strong correlation (0.8316) with the reference model used. The applied geostatistical approach to generating the DEM for relief study demonstrates significant potential for application in other regions with similar conditions, offering an effective alternative for areas with dense vegetation where traditional photogrammetric methods encounter limitations.