【Objective】 The accurate estimation of forest aboveground biomass is important to determine changes in global carbon reserves and the corresponding climate change in real time. Combining a variety of remote sensing data, feature optimization, and classification modeling is an effective means to improve the accuracy of estimating forest aboveground biomass.【Method】 In this study, the research object was defined as the temperate natural forest at the Daxinganling ecological observation station in Genhe City, China. Additionally, fifty-five ground survey data containing airborne light detection and ranging (LiDAR) and Landsat8 operational land imager (OLI) remote sensing imagery were utilized. The partial least squares algorithm was used to optimize the selected variables, and the model was constructed by linear multiple stepwise regression and constructed by the k-nearest neighbor algorithm (KNN-FIFS) for fast iterative feature selection; the forest aboveground biomass was retrieved under different combinations of the two data sources.【Result】 The inversion accuracy of the single LiDAR data based on linear multiple stepwise regression model had a R2 of 0.76, and a root mean squared error (RMSE) of 21.78 t/hm2. The inversion accuracy of the single Landsat8 OLI data had a R2 of 0.24, and a RMSE of 39.27 t/hm2. The accuracy of LiDAR and Landsat8 OLI combined inversion had a R2 of 0.84, and a RMSE of 18.16 t/hm2. The inversion accuracy of single LiDAR data based on the KNN-FIFS model had a R2 of 0.74, and a RMSE of 23.83 t/hm2. The inversion accuracy of the single Landsat8 OLI data had a R2 of 0.60, and a RMSE of 29.63 t/hm2. The accuracy of the LiDAR and Landsat8 OLI combined inversion had a R2 of 0.80, and a RMSE of 21.15 t/hm2.【Conclusion】 Among the three combination methods supported by feature optimization, the combination of LiDAR and Landsat8 OLI data demonstrated the highest inversion accuracy in both models. Among the models, the inversion accuracy of the linear multiple stepwise regression model was the highest, with a R2 of 0.84, and a RMSE of 18.16 t/hm2. This result indicates that the LiDAR and Landsat8 OLI data complement each other, and collaborative inversion can effectively improve the inversion accuracy of forest aboveground biomass. The inversion accuracy of forest aboveground biomass from a single data source using LiDAR data was higher than Landsat8 OLI data of the two models; this was related to the high spatial resolution of LiDAR data and the availability of vertical structure parameters.
【Objective】 Currently, forest type identification research is mainly focused on small forest areas and forest farms. To explore the forest type identification method across a large range, this study used Sentinel-2 optical remote sensing imagery, forest resource survey data, a digital elevation model (DEM), and Sentinel-1 radar remote sensing image data to establish tree species identification model.【Method】 The research area was defined as Chun’an County, where seven forest types were modeled separately, including Phyllosstachys edulis forest, Camellia sinensis forest, Carya cathayensis forest, Cunninghamia lanceolata forest, Pinus massoniana forest, broad-leaved mixed forest, and other hard broad-leaved species. Models were divided into three layers. In the first layer, the RF algorithm was used to establish the identification model of forested land and non-forested land. In the second layer, forest structure was identified from forested land. The RF, XGBoost and LightGBM methods were used to build various models and analyze the experimental results. In the third layer, the forest structures were further divided into forest type.【Result】 The overall accuracy of the first-layer model based on the RF algorithm dividing samples into forested and non-forested land samples was 98.08%. In the second layer (i.e., forest structure recognition model), the performance of the three models under various feature combinations were compared. It was found that the LightGBM model had the highest overall accuracy of 81.43%. In the third layer, the performance indicators for seven forest type models were compared; based on the combination of all features and radar factors, the overall accuracy of the LightGBM model was 84.51%, after feature selection by the recursive feature elimination algorithm, the optimal accuracy was 83.21%.【Conclusion】 The green, red, near-infrared and red-edge bands from optical remote sensing imagery, and terrain factors from a DEM are effective in identifying the forest type. However, independent variables extracted from Sentinel-1 radar do not provide significant help to identify forest type.
【Objective】 Actual measurements of the relationship between the distribution direction of branch heights and the intensity of spatial competition have highlighted the downside to three-dimensional (3D) forest models construction based on traditional forestry research survey data. Such models are unable to directly express differences in branch height distribution in varying directions. This results in the insufficient performance of the forest tree 3D model polymorphism. 【Method】 This study used eight temporary sample plots of Chinese fir in the Shanxia Forest Farm of Fenyi County, Xinyu City, Jiangxi Province, China to supply data. The existing undershoot height model was used alongside establishing the buffer zone of the geological analysis method combined with the forest stand spatial structure unit. The horizontal and vertical spatial structure parameters that directly affect the forest trees were established, and the high correlation between the spatial structure parameters and the undershoot was analyzed. This analysis provided the basis for the calculation of the spatial competition intensity in each direction, and the established relationship between spatial competition intensity and the measured undershoot height distribution. The under-branch height model was used to calculate the remaining-direction under-branch height. Finally, the branch and trunk model was loaded according to the measured data and analysis and calculation structure to construct a 3D forest model. 【Result】 The selected basic model variables included forest attributes and spatial structure parameters; the original model coefficient of determination was 0.720, the horizontal spatial structure and the adjusted branch height correlation coefficient was 0.410 to eliminate the influence of tree height. The vertical spatial structure correlation coefficient was 0.782, and positively correlated with the branch height. The respective correlation coefficients were weights to calculate spatial competition intensity in the corresponding direction. The minimum competition intensity direction spatial structure parameter was used to achieve the basic model fitting coefficient of determination of 0.790; this was an improvement compared with the original model. The measured branch height was allocated to competition intensity in the smallest direction and to quantify the lower height of branches in remaining directions. 【Conclusion】 Chinese fir was used as an example of utilizing spatial competition intensity to discriminate the high distribution under branches to improve the utilization of existing data and reducing field work intensity. This approach intuitively expresses differences in the high distribution under the branches of the forest and enhances the polymorphism of the 3D forest model expression.
【Objective】 Plant visualization technology is an important part of digital forestry research. This study used the stem missing points to propose a plant stem complement based on L1-medial skeleton extraction to provide technical support for plant visualization. 【Method】 First, a method to determine the position of the missing part based on topological connection was utilized. Based on the density of the point cloud, the point cloud was classed by using the union-find sets. The weight of the edge of the cluster between nodes was calculated based on the probability graph model, and the minimum spanning tree was used to determine the topological connection between clusters; this enabled the determination of the position of missing parts. Following this, a search-based method was used to determine the set of points to be fitted. The L1-medial local iterative method was used to extract the stem point cloud skeleton. A search algorithm based on nearest neighbor distance was proposed to sort the skeleton point set for the disorder of the point cloud, to determine points to be fitted for missing parts. To address issues with inaccurate skeleton extraction, an iterative optimization method of stem skeleton based on Gauss kernel weight was proposed. This approach used Gauss smoothing stem direction vector, and the Gauss weighted average to calculate stem radius and the updated stem skeleton point set. Finally, a method of point cloud completion of missing parts based on fitting was utilized. The stem radius of missing parts was fitted based on the least squares method, and the stem line of missing parts was fitted based on the Bezier curve. A method based on point cloud density to adjust the generation of the fitting point cloud was proposed to better fit the actual point cloud. 【Result】 The experimental results show that the method proposed in this paper can effectively complete the plant point cloud with leaf and stem missing separation; the fitting result was smooth and has certain practical, physical significance. 【Conclusion】 To some extent, the research results in this paper make up for the defects of scanning, building a complete and realistic three-dimensional point cloud model of plants; this may be more effectively applied to the three-dimensional visual modeling of plants.
【Objective】 This study quantitatively investigated the effect of spatial structure on the under-branch height of Chinese fir (Cunninghamia lanceolata), to build a model of Chinese fir under-branch height. The spatial structure combined with the fir growth model, three-dimensional (3D) visualization technology was applied to visually simulate the height of fir branches. 【Method】 Using survey data of six temporary sample plots of Chinese fir plantations in the Huangfengqiao State-owned Forest Farm in Hunan Province, China; five commonly used branch height foundation models were selected to analyze horizontal spatial structure parameters (PH), vertical spatial structure parameters (PV), and spatial structure unit average distance (dDIS). The influence of this combination on the height of branches was constructed with better comprehensive indicators and fewer variables. 【Result】 The Logistic model showed better comprehensive indicators and explanatory model parameters; as such, it was selected as the basic model. Among the three spatial structure parameters, the vertical spatial structure (PV) had a significant impact (R2=0.741). Adding vertical spatial structure parameters to the Logistic model improved the simulation of the sub-branch height model. As a result, the coefficient of determination (R2) increased from 0.717 to 0.741, the standard deviation of the estimate reduced from 1.407 to 1.321 m, and various model inspection error indicators were reduced.【Conclusion】 The under-branch height model constructed in this study may be applied to Chinese fir forests of unknown tree age and partial stand information, reflecting the mutual competition among forest trees and visually and intuitively expresses changes in the height of fir branches, and supports further research on the visual simulation of forest stand growth dynamics and forest management.
