Abstract:The accurate measurement of diameter at breast height (DBH) is essential to forest operational management, forest inventory, and carbon cycle modeling. Terrestrial laser scanning (TLS) is a measurement technique that allows rapid, automatic, and periodical estimates of DBH information. With the multitude of DBH estimation approaches available, a systematic study is needed to compare different algorithms and evaluate the ideal situations to use a specific algorithm. To contribute to such an approach, this study evaluated three commonly used DBH estimation algorithms: Hough-transform, linear least square circle fitting, and nonlinear least square circle fitting. They were each evaluated on their performance using two forest types of TLS data under numerous preprocessing conditions. The two forest types were natural secondary forest and plantation. The influences of preprocessing conditions on the performance of the algorithms were also investigated. Results showed that among the three algorithms, the linear least square circle fitting algorithm was the most appropriate for the natural secondary forest, and the nonlinear least square circle fitting algorithm was the most appropriate for the plantation. In the natural secondary forest, a moderate gray scale threshold of three and a slightly large height bin of 0.24 m were the optimal parameters for the appropriate algorithm of the multi-scan scanning method, and a moderate gray scale threshold of three and a large height bin of 1.34 m were the optimal parameters for the appropriate algorithm of the single-scan scanning method. A small gray scale threshold of one and a small height bin of 0.1 m were the optimal parameters for the appropriate algorithm of the single-scan scanning method in the plantation.

This study investigated the implications of different assumptions of 3D forest stand reconstructions for the accuracy and efficiency of radiative transfer (RT) modeling based on two highly detailed 3D stand representations: 3D‐explicit and voxel‐based. The discrete anisotropic radiative transfer (DART) model was used for RT simulations. The 3D‐explicit and voxel‐based 3D forest scenes were used as structural inputs for the DART model, respectively. Using the 3D‐explicit RT simulation as the benchmark, the accuracy and efficiency of the voxel‐based RT simulation were evaluated under multiple simulation conditions. The results showed that for voxel‐based RT simulations: with voxel sizes 0.1, 1, and 10 m and in blue, green, red, and near‐infrared wavebands, the normalized deviations of simulated directional reflectance exceeded the 5% tolerance limit in 89% viewing directions; with voxel sizes 0.2, 1, and 10 m, the normalized deviations of simulated spectral albedo exceeded the 5% tolerance limit in 90.5% wavelengths; for simulated spectral albedo in blue, green, red, and near‐infrared wavebands and fraction of absorbed photosynthetically active radiation, the normalized deviations exceeded the 5% tolerance limit in 65.3% voxel sizes and spatial resolutions. The two major causes for differences in the 3D‐explicit versus voxel‐based RT simulations were: (a) the difference between light interaction in spatially explicit objects and in turbid medium, and (b) the structural difference of 3D contours between voxel‐based and 3D‐explicit models. However, voxel‐based RT simulations were substantially more computationally efficient than 3D‐explicit RT simulations in large voxel sizes (≥1 m) and coarse spatial resolutions (≥1 m).