DISSERTATION RESEARCH: Biomass estimation and uncertainty analysis: Integrating Bayesian modeling and small-footprint waveform LiDAR data

遥感被用来描述和绘制地球植被的空间变化及其随时间的变化。一种相对较新的机载激光传感器LiDAR（光探测和测距）已被用于准确表征三维植被结构和测量植物生物量和碳含量。随着激光雷达数据可用性的增加，需要开发开源工具来有效地处理这些数据，以获得植被和地形信息。使用目前的方法，在估计植被的数量和类型方面仍然存在很大的变化和不确定性。这项博士论文改进基金（DDIG）的支持将用于开发开源工具，以处理来自国家生态观测站网络（NEON）的激光雷达数据，并探索它们在识别树种、估计植被数量和确定分析中的统计不确定性方面的潜力。该项目开发的方法将促进3d植被结构的创建，并使人们能够理解生态系统的模式和过程。植被结构和数量的误差和不确定性的量化也有助于设计有效的森林可持续经营计划，并为生物地球化学模型提供准确的输入，为基于科学的政策提供信息。本研究将通过以下步骤完成：(1)开发用于激光雷达数据处理的算法和开源工具；(2)利用随机森林和贝叶斯机器学习方法，仅利用原始激光雷达数据对单个树木进行分割和树种识别；(3)利用激光雷达指标和来自点云的变量，建立不同的模型（逐步模型、随机森林模型和分层贝叶斯模型）来估计森林生物量和碳储量；(4)根据不同的方法和模型参数，量化不同处理阶段估计的不确定性。本提案中开发的算法和处理方法将为激光雷达波形处理提供开源工具，并提高NEON数据的使用和价值。此外，这项研究的成果将有助于森林管理者更好地管理宝贵的自然资源并作出更明智的决定。

Remote sensing is used to characterize and map the spatial variation of Earth's vegetation and its changes over time. A relatively new airborne laser sensor called LiDAR (Light Detection And Ranging) has been used to accurately characterize the three-dimensional vegetation structure and to measure plant biomass and carbon content. With increased LiDAR data availability, there is a need to develop open source tools for processing these data effectively to obtain the vegetation and terrain information. With current methods, substantial variation and uncertainty remain in estimating the amount and type of vegetation. Support from this Doctoral Dissertation Improvement Grant (DDIG) will be used to develop open source tools to process LiDAR data from the National Ecological Observatory Network (NEON) and to explore their potential for use in identifying tree species, estimating vegetation amounts, and determining the statistical uncertainty in the analysis. The methods developed from this project will facilitate the creation of 3D-vegetation structure and enable the understanding of ecosystem patterns and processes. The quantification of errors and uncertainties of vegetation structure and amount are also conducive to designing effective plans for sustainable forest management and providing accurate inputs for biogeochemical models to inform science-based policy.This research will be accomplished through the following steps: (1) Develop algorithms and open source tools for LiDAR data processing, (2) Segment individual trees and identify tree species with raw LiDAR data alone using Random forests and Bayesian machine learning method, (3) Build different models (step-wise, Random forests, and hierarchical Bayesian models) to estimate forest biomass and carbon stocks using LiDAR metrics and variables derived from point clouds, and (4) Quantify the uncertainty of estimations in different processing stages based on different approaches and model parameters. The algorithms and processing methodologies developed in this proposal will provide open source tools for LiDAR waveform processing and enhance the use and value of NEON data. Moreover, the products of this research will assist forest managers to better manage precious natural resources and make more informed decisions.