地面雷射掃瞄系統於樣區尺度之立木測計

森林決策之訂定仰賴於精確之森林調查資訊，故如何準確地蒐集立木及林分性態值為重要之方法。地面雷射掃瞄儀(Terrestrial Laser Scanning, TLS)具高精度、可重複檢核、非破壞性量測及可快速獲取大量三維地表資訊之特性，其點雲(Point Clouds)資訊可透過視覺化及自動化等方式測計，獲取準確之立木及林分性態值，如胸徑、樹高及樹冠幅之取得、蓄積量及生長量推估以及透過其進行森林健康度評估等。本研究係以TLS取得立木三維點雲資料，透過點雲資料探討四種不同立木胸徑與樹高之測計方法(人工判釋測計法、自動化偵測法、去除界外值之自動化偵測法與半自動化偵測法等)探討其於樣區尺度立木測計之可行性，後續並以樣區每木實測胸徑與樹高資料，進行迴歸分析及單因子變異數分析，探討TLS量測立木胸徑與樹高之準確度，亦討論不同架站模式及不同掃瞄解析度對於樣區點雲生成之影響。結果顯示，實測胸徑除了以自動化偵測法對於立木性態值測計與實測具有顯著差異外，其他三種處理方法皆無顯著性差異，樹高測計方面，在人工判釋測計法及半自動化偵測法所測計之樹高與實測樹高無顯著性差異，自動化偵測法及去除界外值之自動化偵測法則具有顯著性偏高之差異。以TLS點雲獲取之林木性態值與實測值之胸徑、樹高進行迴歸分析，結果顯示，透過人工判釋測計法 (胸徑R2 = 0.99、樹高R2 = 0.91)或半自動化偵測法 (胸徑R2 = 0.99、樹高R2 = 0.89)不論在胸徑或樹高上均具其準確性，樹高RMSE均小於1.50 m；胸徑RMSE則小於0.53 cm，另外掃瞄解析度為1/4 (43.70 MPts)及1/2 (174.80 MPts)時，其所測計之胸徑、樹高與實測值無顯著差異，而解析度為1/2 (174.80 MPts) 時，其精度並未上升，故在本研究之樣區類型以掃瞄解析度1/4 (43.70 MPts) 時，可取得準確之立木胸徑；本研究亦發現樣區內架站數越多則其所生成之點雲資料越完整，可有效降低遮蔽導致之誤差現象，但相對而言，資料量也越龐大，各站萃取之胸徑R2值分別為0.11、0.41、0.85、0.94、0.99，樹高R2值分別為0.38、0.59、0.72、0.90及0.92，即掃瞄站數之多寡與胸徑、樹高萃取準確度呈正相關，故於人工闊葉林掃描時，架站數量應以四站以上為佳。

關鍵字

地面雷射掃瞄系統；森林調查；自動化測計；點雲資料

並列摘要

Accurately to collect the tree attributes in plot-scale forest measurement is an important task for making the strategies of forest management. The features of the Terrestrial Laser Scanning (TLS) system includes advantages such as high precision scans, repeatable access, browse non-destructive measurement, and the ability to acquire a number of 3D surface information. Point cloud information acquired from TLS allows obtaining accurate forest attributes through visualization and automated extraction, such as diameter at breast height (DBH), tree height (H), canopy width, volume estimation, tree growth, and forest health assessments. The purposes of this study include using TLS to obtain 3D point cloud information and extract tree attributes, followed by accuracy assessments on four different methods of point cloud extraction that include artificial extraction, semi-automatic extraction, automated extraction and automated extraction of removal of outliers are used to extract DBH and H within plot-scale. We would discuss the impact of point cloud generation in different resolution and station number settings in the same plot. The result shows that there is a significant difference between the measured DBH and the automatic extraction, and there are no significant differences with the other three types of extraction. On the other hand, the measured H was not significantly different between the artificial extraction and semi-automatic extraction, but with automated extraction and automated extraction of removal of outliers is significantly different and has a high trend. 1:1 linear regression analysis between the tree attributes obtained by TLS and real values of DBH, H showed that the artificial extraction (DBH, R2 = 0.99, H, R2 = 0.91) or semi-automatic extraction (DBH, R2 = 0.99, H, R2 = 0.89) is highly correlated in DBH or H, RMSE is less than 1.50 m in H measurement and 0.53 cm in DBH measurement. The greater the number of the station in the sample area was set, the more complete the point cloud information was generated, which effectively reduces the error caused by the shadowing. Comparing scanning resolution of 1/4 (43.70 MPts) and 1/2 (174.80 MPts), no significant difference between the extracted DBH, H and the true value was found. The accuracy remains when the resolution is set to 1/2 (174.80 MPts) therefore, accurate data can acquire with a scan resolution of 1/4 (43.70 MPts). The more the number of stations in the sample area, the more complete the point cloud data generated, which can effectively reduce the error caused by the masking. The R2 of the DBH by each station are 0.11, 0.41, 0.85, 0.94, 0.99, and the R2 of the H are 0.38, 0.59, 0.72, 0.90 and 0.92. That is, the number of scanning stations is positively correlated with the DBH and H extraction accuracy. Therefore, the number of stations should be more than four stations.