Li yong ji guang lei da, da fu mian hang kong zhao pian he wu ren ji she ying ce liang zai Ya re dai cheng shi sen lin zhong jian gou shu mu gao du he ti qu dan mu shu guan

M.Phil.Accurately identified trees from an automatic analysis can serve as a basis of estimating forest variables and building inventories through the individual tree-based approach. Increasing richness of remote sensing data also provides opportunities to explore the potential uses of various types of data sources. Using four forested sites (0.15 – 0.28 ha) in Hong Kong with different tree compositions as the study areas, this study adopted three widely used remote sensing data sources, including airborne LiDAR data, unmanned aerial vehicle (UAV) and traditional aerial photographs, and aimed to investigate their potentials on constructing canopy height models (CHMs) and extracting individual tree information.In addition to airborne LiDAR data and aerial photographs acquired by commercial vendor and the Hong Kong government, UAV photographs were also collected by the researcher and processed using photogrammetric software to generate point cloud datasets. On top of the rasterized CHMs, local maxima and region growing methods were adopted to identify treetop points and delineate crown polygons respectively with different fixed and variable window size settings. Accuracy assessments were applied to both point cloud products and delineation outputs, including tree heights, detection rates and other statistics.Results showed that all remote sensing datasets were able to generate comparable representations of canopy surface in all study areas, with tree heights having r = 0.58 - 0.94 and RMSE = 1.33 - 3.78 m compared to field-measured values as well as similar levels of correspondence between the three data sources. The differences in point cloud characteristics such as point densities, vertical distributions of points and surface smoothness were also highlighted through point-based and profile-based analysis. For treetop detections, the highest F-scores in each site ranged from 0.49 - 0.72 with various patterns caused by different window sizes and data sources. Matched rates of reference trees were also evaluated which showed significant positive correlations (r = 0.19 – 0.51) with tree characteristics including diameter at breast height (DBH), tree height, crown area and distance to the nearest neighbour.The similar levels of tree height estimations and varying patterns of tree detection indicated that no single remote sensing dataset was clearly superior in all combinations of methodologies in this study. Instead, unique properties of each data source should be considered in terms of both data acquisitions and analysis. The patterns of tree detection results showed that both knowledge and testing on the characteristics of study areas and the properties of data sources are important when deciding the best window sizes for the best performances. Heterogeneity of forest environment in Hong Kong was also believed to be a major factor hindering the delineation performance, with further influences on plot-level difference and tree-level detectability.準確辨認個別樹木，是透過單木方式獲取森林資訊和建立樹木地理資料庫的基本步驟，而近年遙感技術的發展也增加了運用各種遙感系統的潛力。是次研究選擇了四個位於香港並由不同城市樹木種類組成的研究區，藉由收集三種常用的遙感數據，包括機載激光雷達、傳統航空照片和無人機照片，了解它們在建構樹冠高度模型和提取單木資訊方面的可能性。主要研究過程如下：（1）操控無人機拍攝研究區的照片，並透過攝影測量軟件產生出三維的點雲數據；（2）在高度模型之上，利用局部最大值和區域生長算法提取單木的位置和樹冠，並測試不同窗口大小參數的影響；（3）評估各個點雲數據和單木提取結果的準確度，包括樹木高度、檢測率等。研究結果顯示，所有遙感數據都能在不同研究區建構出相約的樹冠三維資訊。對比遙感所得和實地量度的樹木高度，能得出 0.58 - 0.94 的相關係數和1.33 - 3.78米的均方根誤差。三種遙感數據之間的比較也得到接近的水平。基於點雲和斷面圖的分析能反映出更多不同數據在密度、高度分佈和表面平滑度等方面的特徵。在單木提取方面，每個研究區中最高的F分數介乎0.49至0.72，並根據窗口大小和數據特性而呈現各種的趨勢。樹木能成功吻合的比率亦跟樹木的胸徑、高度、樹冠大小和鄰近距離相關（r = 0.19 - 0.51）。相約的樹木高度估算和單木提取的準確性，反映出沒有一種遙感數據能在所有測試中都能得到最好的結果，反而應考慮各種數據在應用過程中獨特的優勢。在決定合適的窗口大小參數時，也需要對研究區樹木狀況和遙感數據特徵有一定的認識和測試。香港樹木環境的多樣性對單木提取的整體準確度，以至各個研究區之間和個別樹木的表現差異，皆有一定的影響。Kwong, Ho Yin.Thesis M.Phil. Chinese University of Hong Kong 2019.Includes bibliographical references (leaves 184-214).Abstracts also in Chinese.Title from PDF title page (viewed on 26, January, 2021).Kwong, Ho Yin