ObjectiveThe underforest terrain is a necessary condition for extracting forest parameters such as individual tree height and crown width. However, due to the large terrain fluctuation of the forest area, it is difficult to obtain a large-scale and high-precision digital terrain model (DTM) of forest area by traditional measurement method. Oblique photogrammetry overcomes the shortcomings of traditional measurement technology and becomes a new method to obtain three-dimensional geographic information. In this paper, UAV oblique photogrammetry technology was used to extract the topography of larch forest, and its accuracy and applicability were evaluated. It provides a reference for subsequent research on extracting individual tree parameters based on oblique photogrammetry technology.

MethodThe typical young, middle-aged and mature Larix forests in the mountainous area of Wangyedian Forest Farm in Inner Mongolia, northern China were selected for UAV flight in the deciduous season. The oblique images of the deciduous season were reconstructed by Context Capture software to generate point clouds in the forest area. Ground points were extracted from point clouds by cloth simulation filtering (CSF), weighted linear least squares (WLS), progressive irregular triangular network filtering (PTIN) and progressive morphological filtering (PMF), and three interpolation methods were used to interpolate ground points to generate complete topography in the survey area. DTM generated from LiDAR data was used to evaluate the extraction accuracy with validated data.

ResultThe results showed that the accuracy of terrain extraction by different algorithms was related to canopy density. In young forest and middle-aged forest area, cloth simulation filter (CSF) had the highest accuracy in extracting ground points from photogrammetric point clouds, with the determination coefficient (R²) reaching 0.999 and the root mean square error (RMSE) reaching 1.61 m and 0.47 m, respectively. In mature forest area, progressive triangulated irregular network (PTIN) had the best effect, with R² 0.999 and RMSE 0.39 m. After selecting the optimal filtering algorithm for different canopy density forest stands, the DTM accuracy of different interpolation methods was compared. The results showed that in young and middle-aged forests, the DTMs both generated by the point clouds of cloth simulation filtering (CSF) and triangulated irregular network (TIN) interpolation had the highest precision, RMSE was 1.58 m and 0.44 m, respectively. In mature forest, the DTM generated by the point clouds of progressive triangulated irregular network flitering (PTIN) and kriging (KRG) interpolation had the highest precision, RMSE was 0.31 m.

ConclusionResearch has shown that oblique photogrammetry can be used for topographic extraction of larch forests.