應用無人飛行載具影像密匹配點雲建立花蓮大農大富平地森林園區楓香人工林蓄積量推估模式

Using unmanned aerial vehicle (UAV) image dense matched point clouds data to estimate the stand volume equation of Liquidambar formosana plantation in Danongdafu forest park in Hualien county

本研究於花蓮縣大農大富平地森林園區，透過UAV運動回復結構方式，建立森林樹冠點雲模型，進行林分高、樹冠面積與樹冠體積等林分性態值之測計，並以不同飛行高度所影響之影像解析力及覆蓋度，對於林分高進行準確度評估，研究項目包括（1）優勢木平均樹高、（2）優勢木三株平均樹高、（3）第三四分位數樹高、（4）最高樹高等四種方式，並以迴歸分析探討最適推估模式。後續利用地面調查之樹冠幅與樹冠面積、樹冠體積之實測資料，進行各林分高相關係數統計，討論其相關性，找出最適UAV解析力及林分性態值量測方式，以建立以三維點雲為基礎之林分蓄積推估模式。透過不同飛行高度以及影像覆蓋度比較，發現覆蓋度90％及飛行高度100m所產製之點雲推估林分高精度最佳，而後續利用UAV樹冠點雲資料獲取其林分高、樹冠面積與樹冠體積等資訊，並依據地面調查資料，分析與建立UAV林分蓄積量推估模式。由結果得知，應用UAV三維點雲資料可準確獲取林分性態值，並透過迴歸可獲取較佳之空中材積式（R2＝0.75、RMSE＝8.50 m3／ha），顯示可利用UAV點雲資料可建立林分蓄積量推估模式。

In this study, the UAVs Structure from Motion (SfM) was used to acquire stand characteristic including stand height, canopy area, and canopy volume and establish point cloud canopy height model (CHM) for Danongdafu Forest Park in Hualien. The accuracy of stand height acquisitions was evaluated by various flight height, image resolution and coverage. Four methods for extracting stand height were used: (1) average tree height of dominant trees, (2) average tree height of the three most dominant trees, (3) third quartile tree height and (4) the highest tree height. Regression analyses were performed to identify the best fit for stand height acquisition. Correlation analyses between canopy width, area, and volume obtained by plot surveys were conducted to identify an optimal UAV stand parameter acquisition method. A 3D point cloud stand stock estimation model was then established to evaluate the stock volume of the large-scale artificial forest in Danongdafu Forest Park. By comparing stand height acquisition at different flight heights and image coverage, it was found that a 90% coverage with a flying height of 100 m yield the most accurate stand height estimation. Canopy point cloud data acquired by UAVs, information of stand height, canopy area, and canopy volume can be integrated with field data to establish the UAV stand stock estimation models. The results showed that UAV 3D point cloud data can be used to obtain stand characteristic accurately. An improved air volume model (R2=0.75, RMSE=8.50 m3/ha) was established by stepwise regression in this study, indicating that the UAV aerial volume equations are capable of estimating stock volume.