Methods development using terrestrial LiDAR for the assessment of forest structures at the tree and stand levels

点击翻译按钮获取中文摘要

Terrestrial LiDAR (Light Detection and Ranging) or TLiDAR is an instrument probing its surroundings with a hemispherical scan composed of several million points, each of which has a distance and a reflectance value. The resulting 3D cloud provides an explicit, although fuzzy, rendition of the surrounding objects. Even if TLiDAR technology is an important step forward in our ability to probe our environments, this tool has limitations that need to be overcome. These limitations grow with the complexity of the environment. Airborne LiDAR is widely used and the methods for data processing benefit from a large pool of publications and expertise worldwide. Conversely, studies and expertise on TLiDAR are much less predominant. Most studies using TLiDAR in forest environments developed algorithms from isolated trees or within a controlled environment like a plantation. There is a need for many new developments to take advantage of the capabilities of the TLiDAR in natural forest environments, particularly to support enhanced forest inventory. Thus this research program aims to develop the methods required to use TLiDAR in natural forest environments to estimate a large array of attributes, which either are not currently available in the forest inventory or are currently available but poorly measured. We also aim to develop procedures and algorithms to overcome the main limitations of TLiDAR in forest applications. I propose a research program that tackles the four categories of scientific questions related to the use of TLiDAR in natural forests: (1) overcoming the limitations of TLiDAR data, (2) developing new methods to measure tree/stand structure, (3) using LiDAR metrics for useful relationships in forest ecology, and (4) the spatial generalization of plot-level data to large area mapping. Firstly, the main limitation for the use of TLiDAR data in forestry is the occlusion effect, which results in blind zones behind opaque objects. The effect of occlusion in a complex environment like a forest is significant and highly spatially variable; consequently, it can seriously bias the results. Other limitations include multiscans georeference, the effect of wind, and data processing of very large datasets. One solution to handle TLiDAR limitations is the use of a voxel (3D cube) representation of the point cloud to normalize the information based on the recorded returns and the scanner’s characteristics. Secondly, most algorithms dealing with 3D point clouds are not designed to estimate forest structural attributes. Algorithms need to be developed for many attributes: trunk diameter from the ground up to the live crown, tree height, and tree crown dimensions and spatial density. TLiDAR is also particularly well adapted for method development to estimate the total leaf area of a stand, a value otherwise difficult to extract from either direct/destructive or indirect methods. Thirdly, a large number of structural metrics or (estimated) structural attributes can be made available for the stand. These metrics/attributes are useful to improve our ability to predict other stand attributes like the wood fiber attributes, stand growth potential, or habitat suitability. Forthly, TLiDAR taken at the plot level need to be used in support of mapping large area. We propose several focused studies on: (1) a new approach for method testing involving tree architectural models, (2) estimation of structural attributes (e.g. trunk diameter, crown dimensions, leaf area & stand openness) from the TLiDAR point cloud, (3) methods linking LiDAR metrics to predict wood fiber attributes, (4) species identification or assessment of tree/stand vigor, (5) linking terrestrial and airborne data, and (6) use of local (plot-level) data for spatial generalization.