老林的分類是探討老林的首要課題,本研究著眼於應用航遙測技術,針對棲蘭山林區具有枯立木與孔隙結構特徵的檜木老林,進行大尺度的檜木老林分類。研究分成三大部分,一為利用數位航測技術以及統計方法進行檜木老林結構圖徵的收集與分群。二是以衛星多光譜影像作為材料,利用遙測影像分類方法萃取檜木老林多光譜影像。三是將老林多光譜影像結合檜木老林的航測分群成果,進行檜木老林之光譜指標研擬和影像分類。 以航測立體測繪的孔隙有970個,孔隙面積約306,701m2,孔隙率約為1%;枯立木有2041株。將孔隙與枯立木分布圖於三種不同的分析尺度(500m×500m、250m×250m、125m×125m)進行非階層式群落分析,結果指出檜木老林可分為老林特徵少、老林枯立木多與老林孔隙多三個組類。然後根據規整差異植生指標(NDVI),利用影像切割方法切取植生區域,再以非監督分類法萃取檜木老林多光譜影像,以迴歸分析方法建立全區檜木老林之光譜指標如下: NDVI = 1.11791 × NIR - 0.69058 × R 結合125m×125m尺度的航測檜木老林群落分析成果,以迴歸分析方法建立檜木老林不同結構之光譜指標如下: 老林特徵少 :NDVI = 0.0002 + 1.1148×NIR - 0.6882×R 老林孔隙多 :NDVI = 0.0018 + 1.1428×NIR - 0.7003×R 老林枯立木多 :NDVI = 0.0038 + 1.1199×NIR - 0.7103×R 同時將檜木老林多光譜影像進行監督分類之全區分類準確度為75.2%。 本研究結果可得結論為:(一)以航測技術收集檜木老林結構特徵,並進行檜木老林結構分群,技術上確實可行;(二)以規整差異植生指標協助影像切割,搭配非監督分類方法可以成功萃取檜木老林多光譜影像;(三)孔隙與枯立木均為判釋老林的重要參數,但兩者相較之下,孔隙對於檜木老林光譜反射值的影響大於枯立木。
Chilan Mountain is sanctuary to the last remaining old growth cypress forest in Taiwan. This research focused on the classification of old-growth cypress in Chilan Mountain using digital photogrammetry and remote sensing techniques. Three parts were included in the study. The first part involved collecting old growth characteristics (e.g., snags and gaps) using digital photogrammetry and processing cluster analysis according to the old-growth characteristics under three different scales. The second part used image level slicing to extract the forested area based on NDVI and classified the area into old-growth and non-old-growth forests using unsupervised classification. The last part of the study focused on generating spectral indices by regression analysis and also processing image classification of old-growth forests. 2,041 snags and 970 gaps were firstly collected by digital photogrammetry and clustered into three categories: old-growth with less snags and gaps, old-growth with more snags, and old-growth with more gaps. The forested area extracted by the NDVI value greater than 0.5 was classified into old-growth and non-old-growth forests using unsupervised classification. Meanwhile, spectral indices for old-growth and the other three clustered old-growth forests were derived as follows. Old-growth: NDVI=1.11791*NIR-0.69058*R Old-growth with less snags and gaps: NDVI=0.0002+1.1148*NIR-0.0688*R Old-growth with more snags: NDVI=0.0018+1.1428*NIR-0.7003*R Old-growth with more gaps: NDVI=0.0038+1.1199*NIR-0.7103*R From the result, it is concluded that digital photogrammetry is a feasible approach to collect the characteristics of old-growth forests and to form different clusters based on those characteristics and statistical analysis. Also, the multi-spectral image of old-growth forest can be successfully extracted using image level slicing of NDVI and unsupervised classification. In addition, snags and gaps both are important parameters for old-growth forests. However, gaps are more important than snags according to the discrimination so of the spectral reflectance.