Abstract:

Obtaining the structural parameters of forests accurately and quickly is of great scientific significance. The technologies of polarimetric and interferometric synthetic-aperture radar, high-resolution optical satellite remote sensing, spaceborne lidar, and multi-angle satellite remote sensing can effectively obtain forest structure regionally. However, it is challenging to obtain structural parameters on a single-tree scale using these technologies. The rapid development of unmanned aerial vehicles (UAVs) recently provides a new means to obtain forest structures. The ground images can be obtained using UAV ordinary cameras, and the ground point cloud can be obtained using structure from motion photogrammetry technology for 3D reconstruction, generating digital surface model, digital elevation model (DEM), and canopy height model (CHM). The local maximum algorithm is used to process CHM data. The key to the local maximum algorithm is to select an appropriate window size and find the largest grid value (seed point) in the window as the highest point of the tree crown to represent the tree. However, it is challenging to calculate the window size. One solution is to use the variable window method. In this study, the height and crown diameter of 49 Populus euphratica trees were measured in the lower reaches of the Tarim River, Xinjiang, China. The Populus euphratica were counted in a 251 hm² grid, and a DJI Phantom 4 Pro was used to acquire the image data. The DJI Phantom 4 Pro is a consumer-grade UAV using oblique and ortho photogrammetry, respectively. An adaptive window maxima algorithm was proposed to analyze and process CHM data, obtaining the structural parameters of the Populus euphratica forest in the study area, such as tree height, crown diameter, and number of trees. The structural parameters extracted using oblique and ortho photogrammetry were compared and analyzed with field-measured data. The results and conclusions are (1) the adaptive window maxima algorithm and fixed window maximum algorithm (1.5-, 2.0-, 2.5-, and 3.0-m fixed windows, respectively) were used to process CHM data obtained using oblique photogrammetry to generate Populus euphratica seed points. The data are compared with the statistical data of 25 sample plots. The R² of fixed window values vary widely from 0.7834 to 0.8261, indicating that we should choose an appropriate window size to achieve higher accuracy according to the actual situation of the region. While the R² of the adaptive window maxima algorithm is the highest (0.8341), it can achieve higher accuracy because the algorithm reduces the window size dependence. (2) The tree height measured using oblique and ortho photogrammetry is overestimated, and the crown diameter is underestimated. The mean, median, and distribution of tree height and crown diameter obtained using ortho photogrammetry are closer to the field-measured values, indicating that oblique photogrammetry could achieve more accurate results than ortho photogrammetry. (3) The oblique photogrammetry and field-measured data were compared to determine the coefficients (R²) of tree height, crown diameter, and tree numbers per hectare (0.9002, 0.8403, and 0.9405, respectively). The root means squared error was 0.4457, 0.6815, and 4.2500, respectively. The results prove that it is effective to use a consumer-grade UAV to obtain forest structure parameters on a single-tree scale using an appropriate measurement method and processing algorithm.

Key words: Populus euphratica, structural parameters, adaptive window maxima algorithm, UAV, oblique photogrammetry