Quantitatively Retrieving Forest Structural Parameters From Automatically Classified Point Cloud Data

Forest canopy structure parameters influence the solar radiation distribution in and below canopy, and control the photosynthetic efficiency and energy exchange as well. Most traditional methods to retrieve canopy structure parameters is time-consuming and labor-intensive, therefore it is inconvenient for large region. Meanwhile, most methods based on conventional remote sensing methods are limited in two dimensional scale. Lidar (light detection and ranging) equipped with three dimensional (3-D) mapping provides an indirect, fast and effective way to quantify forest canopy structure.In order to retrieve forest canopy structure parameters conveniently, an improved classifying method based on geometric feature was proposed, which classifying forest point cloud data (PCD) to three classes, including canopy photosynthetic part (scatter class), non-photosynthetic part (linear class), ground (surface class). This method was workable for forest PCD in different scales (single tree, forest plots and landscape scale), different species, different density, and different ways to be collected. On the basis of sensitivity analysis, the best 3-D space searching radium was proposed. The methods to retrieving canopy structure parameters, including "woody-to-total-area ratio" and leaf orientation distribution, were proposed based on the classifying results. Manually results were compared with computing results in order to verity the proposing methods which improved the precision and effectiveness to retrieve forest structure parameters. The main conclusions in this study could be drawn as follows:(1) The classifying method is universally applicableAn improved classifying method based on local geometric feature classified forest PCD to three classes, including surface class(ground), scatter class(leaf, shrub), linear class(branch, stem), and this method was applied to single-scan forest plot PCD from terrestrial laser scanning (TLS), multi-scan artificial tree PCD from TLS, forest PCD of landscape scale from aerial laser scanning (ALS). The data collected from ALS was classified to two classes(surface and scatter) due to the difficult to gain the linear class PCD. The whole precision of results in different scales were more than 85%, therefore this method was efficient.(2) Leaf area index and tree species area not sensitive to this classifying methodHow the classifying method worked on forest plots of different LAI was analyzed. Thus this method was applied to three single-scan forest plots data with different LAI collected from TLS, and classifying results precisions were good enough, expect the linear class (69.43%,77.10%,80.48% for three plots). The precision decreased as LAI increased. Since the leaf points became more around the branch when LAI was higher, and the local geometric feature was not obvious. However, the precision was nearly for the three forests plots data collected from ALS. Meanwhile, this method was applied to a single broadleaf tree and a single conifer tree, and the results showed that this method was not sensitivity to species as the final produce precision was 94.96% and 93.09%. As long as the features of three classes in canopy satisfy the hypothesis of the method in 2.3.1, it will be suitable.(3) Removing the woody part of leaf areaWe developed a method to calculate point area according the sampling space, point inclination and laser beam incident direction. The method was verified by single leaf, brand and stem, and leaves. The influence of sampling space and inclination on calculating results were analyzed, and the most suitable sampling space was selected.In order verify the results, a artificial tree was made. Then PCD of stem, branch leaf, the areas of whom were manually measured, was gained from multi stations. The method to calculate PCD area by converting discrete points to continuous surfaces was developed to calculate stem, branch and leaf areas. Meanwhile, the areas were gained by reconstructing the PCD. The measuring areas, reconstructing areas and calculating areas were used to calculate "woody-to-total-area ratio", and the closing results (8.97%, 8.31%,7.79% respectively) showed the area calculating method was efficient. However, the result of "woody-to-total-area ratio" using the classifying results were only 5.50%. The main reason was the low precision of linear class.(4) Quantitatively describing leaf orientation distributionA method which was used to calculate leaf orientation (including inclination and azimuthal angle) was developing in this study. The canopy leaf orientation distribution histogram was gained by using the point area as the angle weight. Finally, leaf orientation 2-β distribution was calculated. In order to verify the leaf orientation method, 78 leaves whose points were integrated were selected. The calculating results and measuring results were compared, and it showed that:the R2 of inclination was 0.91(p <0.001), and the R2 of azimuthal angle was 0.97(p<0.001). Thus the distributions diagrams of inclination and azimuthal angles were gained after applying the methods to the whole leaf PCD and the scatter class PCD after classifying, respectively. The results showed that the percentage among 80-90° grew after classifying. The main reason was the branch points were misclassified to scatter class, and meanwhile the inclinations of branch in this paper were nearly 90°. The 2-β distribution parameters were calculated according to the PCD angles, and the 2-β distributions and histograms satisfy the consistency check by using the K-S test. The 2-β function of leaf orientation was calculated finally.