航航測測及及遙遙測測學學刊刊 Journal of Photogrammetry and Remote Sensing

(1)

(2)

航航測測及及遙遙測測學學刊刊

Journal of Photogrammetry and Remote Sensing

發行人：王蜀嘉

出版者：中華民國航空測量及遙感探測學會地址：台北市文山區羅斯福路五段113 號三樓信箱：台北市郵政93-158 號信箱

電話：886-2-8663-3468 886-2-8663-3469 傳真：886-2-2931-7225

電子信件：csprsmail@csprs.org.tw 網址：http://www.csprs.org.tw

PUBLISHER: S. C. Wang

PUBLISHED BY: Chinese Society of Photogrammetry and Remote Sensing

Address: 3F, No.113, Sec.5, Roosevelt Road, Taipei, Taiwan Mail Address: P. O. Box. 93-158, Taipei, Taiwan

Tel: 886-2-8663-3468 886-2-8663-3469 Fax: 886-2-2931-7225

E-mail：csprsmail@csprs.org.tw Web Site：http://www.csprs.org.tw 總編輯

曾義星成功大學

測量及空間資訊學系

台南市東區70101 大學路 1 號電話：886-6-275-7575 分機 63835 傳真：886-6-237-5764

電子信件：jprssubmit@proj.ncku.edu.tw

EDITOR-IN-CHIEF

Yi-Hsing Tseng

Department of Geomatics National Cheng Kung University

No.1, Dashiue Rd., Tainan, Taiwan R.O.C Tel: 886-6-275-7575ext. 63835

Fax: 886-6-237-5764

E-Mail: jprssubmit@proj.ncku.edu.tw 編輯委員 EDITORIAL BOARD

陳良健中央大學王蜀嘉成功大學何維信政治大學廖揚清成功大學陳端墀中華技術學院陳永寬台灣大學劉進金工業技術研究院鄭祈全文化大學史天元交通大學

許明光北台科學技術學院林依依台灣大學

申雍中興大學李仲森美國海軍研究院王如章美國航空及太空總署宮鵬美國加州大學

L. C. Chen S. C. Wang W. H. Ho Y. C. Liao T. C. Chen Y. K. Chen J. K. Liu C. C. Cheng T. Y. Shih M. K. Hsu I. I. Lin Y. Shen J. S. Lee J. R. Wang P. Gong

National Central University, Taiwan National Cheng Kung University, Taiwan National ChengChi University, Taiwan National Cheng Kung University, Taiwan China Institute of Technology, Taiwan National Taiwan University, Taiwan

Industrial Technology Research Institute, Taiwan Chinese Culture University, Taiwan

National Chiao Tung University, Taiwan NTIST, Taiwan

National Taiwan University, Taiwan National Chung Hsing University, Taiwan NRL, USA

NASA, USA U.C.Berkeley, USA

封面圖片說明 About the Cover

配合國土利用調查三級分類屬性之建立，除利用航空照片進行土地類別判釋及結合外業調查修正屬性外；亦嘗試以空間解像力為 2.5 米的 SPOT-5 融合影像(fused image)進行影像自動化方類，並評估對照航照判釋成果之分類精度，以探討高解析影像適用於國土利用調查三級分類的可行性。如上圖左為 95204096 圖幅之 SPOT-5 融合影像，上圖中、右分別為結合航照判釋與外業調查之第一級與第二級分類成果；下圖左、中分別為 SPOT-5 融合影像第一級土地利用分類別之非監督式與監督式分類成果，下圖右為 SPOT-5 融合影像第二級土地利用之監督式分類成果。

(封面圖片出處：高解析影像應用於土地利用分類之探討，第十三卷第四期第 261-271 頁)

(3)

人工編修空載光達資料產製 DEM 成果之探討

何心瑜

¹

陳大科

²

史天元

³

徐偉城

²

摘要

空載光達數據產製數值高程模型(DEM)資料過程中，不同產品等級所需要的產製流程不盡相同。本研究於2007 年 4 ~ 6 月間進行，依當時之參考規範草稿，DEM 產品可分為三個等級；當產製 Level 2 以上的產品時，即需藉由人工編修來確保資料品質。研究中選擇五種不同覆蓋面共10 幅 1/5000 圖幅範圍，

探討編修時間、不同覆蓋面和不同編修者等三個不同的人工編修項目產製DEM 結果。初步成果顯示，以同一圖幅範圍為例，不同編修人員所需之編修時間不同，成果亦不盡相同；如矮植被區之空載光達資料編修DEM 成果的高程平均差為 8 公分，標準差為 25 公分。

關鍵詞：空載光達、人工編修、過濾

1. 前言

數值高程模型(DEM)和數值地表模型(DSM) 是目前空載光達的主要產品；而依據產製DEM 高程精度和正確性程度，產品可分為數級。國內目前將產品分為三個等級(內政部，2005)；如 Level 1 等級除了前處理及率定外，亦包含航帶平差與自動

過濾的過程，最後再進行DEM 資料內插(如表 1)；

Level 2 以上的產品則需再增加人工編修的過程，

方可達到相對較精確的成果；Level 3 較 Level 2 增加斷線資訊及人工檢核步驟。另在人工編修過程中，

由於人工作業容易產生差異，故研究中即針對此步驟，分成 1.編修時間 2.不同覆蓋面 3.不同編修者等三方面進行探討。

表1、空載光達資料產製不同等級 DEM 資料之處理流程

等級 DEM 產品等級之處理流程 level 1A

level 1B level 2A level 2B

level 3A

level 3B

格網化DEM*表示增加斷線資訊及經過人工審核之格網化 DEM(資料來源：內政部，2005)

1工研院能環所副研究員

2工研院能環所研究員

3國立交通大學土木系教授

收到日期:民國 97 年 09 月 18 日修改日期:民國 97 年 11 月 06 日接受日期:民國 98 年 02 月 02 日

(4)

2. 研究材料

研究中使用的空載光達資料，選取自高屏、台南、南投及桃竹等地區，並包含五種不同覆蓋面的掃描資訊；每種覆蓋面類型各挑選2 幅 1/5000 圖幅範圍進行測試(如表 2 所示)。其中各種覆蓋面的定義如下，示意圖如圖1 所示：

(a) 裸露地：五千分之一圖幅範圍內大都為裸露地或農作區，森林覆蓋面積小於25%。

(b) 矮植被區：五千分之一圖幅範圍內大都位於丘陵地，主要地物為農作與矮植被，森林覆蓋面

積介於25%~50%。

(c) 疏林區：五千分之一圖幅範圍內大都位於山區，

主要地物為森林，但森林覆蓋面積介於 50%~75%。

(d) 密林區：五千分之一圖幅範圍內大都位於山區，

主要地物為森林，森林覆蓋面積大於75%且地 形起伏較大。

(e) 都會區：五千分之一圖幅範圍內大都為都市區，

特徵物佔全部面積30%以上，主要地面覆蓋物 為建物與道路。

圖1 五種不同覆蓋面現場照片

表2 每種覆蓋面類型各挑選 2 幅 1/5000 圖幅之圖號與圖名及點雲資訊

項次圖號圖名特性總點雲數點雲密度(點/m²)

1 9419-3-099 南市北部都會區 11163810 1.45

2 9622-4-051 竹高屋都會區 10850510 1.44

3 9418-2-028 大莊裸露區 11918810 1.57

4 9622-4-003 赤牛欄裸露區 10200570 1.35

5 9520-3-036 炭稻矮植被區 11126731 1.46

6 9622-4-026 圓墩頂矮植被區 11550789 1.57

7 9419-2-098 紅毛寮疏林區 13731664 1.78

8 9520-3-038 瑞龍瀑布疏林區 9188212 1.20

9 9518-4-034 茅窩密林區 16000751 2.08

10 9622-1-082 上高遶密林區 11322642 1.49 (e)都會區

(d)密林區

(c)疏林區 (b)矮植被

(a)裸

(5)

圖2 空載光達作業流程圖

3. 研究方法及流程

獲取空載光達點雲資料至產出DEM 等產品過程，其流程如圖2 所示；其中與 DEM 產品等級處理相關者，主要為點雲分類與編修部分中的不合理點位濾除、半自動過濾和人工編修等三部份。研究中使用 TerraSolid 公司之 TerraScan 軟體進行測試與比較。

3.1 不合理點位濾除

處理空載光達資料時，優先要處理不合理點位；

因原始空載光達點雲資料包含雲體或系統誤差造成的不合理點位或其他離群點雲，這些雷射點對於地面點分類並無幫助；故進行地面點分類前，須先將這些雷射點消除。而去除方式可選擇絕對高程消除或離群點消除或分類為特定不使用之類別。

(6)

3.2 半自動過濾

影響半自動過濾成果優劣的因素包含演算法、

點雲密度、地表覆蓋程度及地形起伏因素等。利用 TerraScan 軟體進行過濾時，其使用的演算法為曲線近似法(Axelsson, 1999)。而依據實際操作經驗，

輸入之參數會影響到成果優劣，亦影響後續人工分類所需之時間；故進行半自動過濾前，需輸入適當參數。TerraScan 軟體中(Terrasolid, 2004)，參數 Max building size 是為了產生初始的三角網 (triangulated irregular networks, TIN)；Terrain angle 功能為過濾地面點所接受的 TIN 三角形平面最陡角度；Iteration angle 功能是可能點(可能為地面點的雷射點)到 TIN 三角形平面間的最大角度，一般設定在4~10 度間； Iteration distance 為該點到 TIN 三角形平面的距離，可以牽制Iteration angle，避免將平面面積大的一層建物分為地面點，一般之設定值在0.5~1.5 m 間。Reduce iteration angle 是避免當 TIN 三角形三邊長小於所設定的距離時，繼續增加地面點，Iteration angle 會趨近於 0，防止產生過多之地面點並減少記憶體佔用；此選項建議在地面點密集的地區使用。

3.3 人工編修

人工編修空載光達點雲資料的方式可分為兩種：一為地毯式的編修，另一為特徵部分的編修。

前者是對於整個研究區域完整地搜尋並加以編修；

後者則是針對半自動過濾容易發生錯誤的區域進行編修。若針對易發生錯誤區域進行編修，實際操作經驗得知山脊和密林區為較關鍵的錯誤地區，故為優先編修的區域，如圖3 中的山脊部分並沒有地面點，其中圖3(b)紅圈處是圖 3(a)中紅線對應之位置；圖4 則顯示屋頂上的地面點分布，其中圖 4(b) 紅圈處是圖4(a)中紅線對應之位置。當編修完成較關鍵的區域後，其他地區則視狀況決定是否繼續編修。另人工編修方法亦依不同情況進行，如圖3(b) 中的山脊缺少地面點，則將非地面點修改屬性為地面點；若為圖4(b)狀況，則需將屋頂之地面點修正為非地面點。故確定之地面點數多寡與地形覆蓋面

等天然因素及過濾參數有關，不會隨著編修時間愈長，地面點數愈多或愈少。

(a)俯視圖 (b)側視圖圖3 山脊上無地面點分布情形展示圖

(a)俯視圖

(b)側視圖

圖4 屋頂上的雷射點被誤分為地面點情形展示(a)(b)圖 (橘色點為地面點；藍色點為非地面點)

4. 成果與討論

研究中使用的空載光達資料，經過 TerraScan 軟體之半自動過濾與人工編修後，初步成果如下:

4.1 半自動過濾成果

4.1.1 都會區過濾成果

都會區的地面點大都分布於街道或空地，非地面點大多在建物屋頂；而容易被誤分為地面點的區域大部分在高架橋或大面積的建物。

4.1.2 裸露地過濾成果

此區的點雲大部分在作物區、河床和聚落區，

少數的點雲會落在水體表面。對於在土堆、小樹叢

(7)

和水體邊緣的精度並不理想。

4.1.3 矮植被區過濾成果

此區的地物大多為農作物，只有少數的建物、

水體和密林。在矮植被區的半自動過濾效果並不佳，

原因為雷射點的穿透率低及植被高度和裸露地的差異不大，故容易造成分類錯誤。

4.1.4 疏林區過濾成果

本區的地物大多為果樹、竹林或坡地上的樹林，

少部分的密林或地勢陡峭的山區。分類精度不佳的原因是雷射點不易穿透疏林中相對較密的林區。

4.1.5 密林區之過濾成果

密林區的地物主要為山坡上的密林，雷射點真正到達地面的雷射點相當少，故造成本區分類精度不佳。

4.2 人工編修後成果

空載光達資料之人工編修成果分兩部分比較：

一是人工編修所花費之時間；二是不同編修者針對同一圖幅編修之差異。

4.2.1 人工編修時間與精度之關係

人工編修所花費時間與DEM 成果之關係，亦分為不同地表覆蓋DEM 編修成果的高程平均誤差及標準差兩種比較。圖5 顯示編修一幅 1/5000 基本圖範圍之每種覆蓋面所需的時間，以都會區所需編修時間最少，森林區則需要較多的時間。圖 6 為不同覆蓋面經過不同編修時間後統計之高程標準差，會隨著編修時間的增加而減少；其中亦顯示裸露地區因只具單一回波資料，回波資訊較為單純，

且植生干擾最少，故評估的標準差最小；但密林區受到植生的干擾(雷射點不易穿透密林區，所以能獲取地面點的資訊最少)，故統計的標準差最大。

另依據編修之經驗，一幅 1/5000 基本圖範圍的編修時間達8 小時後，編修後之高程標準差即趨於穩定。編修過程中，因雷射點類別改變，確定之地面點數有增加或減少，但總點雲數(包含地面點及非地面點類別)不變。

圖5 各種地表覆蓋 DEM 編修成果平均誤差比較圖

圖6 各種地表覆蓋 DEM 編修成果高程標準差比較圖

0 5 1 0 1 5 2 0 2 5 3 0 3 5 4 0

0 1 2 3 4 5 6 7 8 9 1 0

編修時間(小時) 高

程平均誤差 ( 公分 )

都會區裸露地矮植被區疏林區密林區

0 2 0 4 0 6 0 8 0 1 0 0 1 2 0 1 4 0 1 6 0 1 8 0

0 1 2 3 4 5 6 7 8 9 1 0

程標準差 ( 公分 )

都會區裸露地矮植被區疏林區密林區

(8)

4.2.2 不同編修者之差異

另利用相同地區的光達資料，但不同編修者編輯後之地形高程差進行評估；研究中挑選矮植被區地形變化大，亦具有其他不同覆蓋面之9622-4-026 圖幅作為實驗區，並由兩位編修者個別作業，以供比較；其中編修者甲費時9 小時，編修者乙費時 7 小時完成。不同編修者每小時編修成果與最終成果之平均誤差和標準差，展示於圖7 和圖 8；若視最終成果為真值，並以點對點的方式比較，得知兩者的高程平均誤差為8 公分，標準偏差為 25 公分。

圖9 則比較兩位編修者之編修成果之高程差異，兩者編修成果差異最大的部分為水體邊緣。而圖10(a) 為編修者甲之成果，圖10(b)為編修者乙之成果，

比較兩者可發現編修者乙分類為地面點之數量較編修者甲多。

4.3 半自動過濾及人工編修之型一與型二錯誤

雷射點資料在生產數值高程模型過程中只需

要使用地面點，其餘雷射點都視為非地面點並不會使用。因為半自動過濾無法完全正確，將所產生之錯誤分為型Ⅰ(Type I)錯誤和型Ⅱ(Type II)錯誤，定義如下：

型Ⅰ錯誤：原本為地面點之雷射點過濾後被歸為非地面點。

型Ⅱ錯誤：原本為非地面點之雷射點過濾後被歸為地面點。

若選擇三幅代表不同地表覆蓋物之圖幅，包含平坦地區的裸露地、山區的密林地及地形變化較複雜的疏林區。則如圖11 和圖 12 表示出半自動過濾和人工編修，在處理各種地表覆蓋物時之型一和型二錯誤統計。顯示空載光達資料不論是經由半自動過濾或是人工編修，在地形變化複雜的地區皆出現較多錯誤；而密林地區之總和錯誤則以人工編修方式最低，可能是因該類地表覆蓋之地面點皆以人工加點為主。

圖7 編修者甲與乙之 DEM 編修成果逐時平均誤差比較圖

圖8 編修者甲與乙之 DEM 編修成果逐時高程差標準差比較圖

0 2 4 6 8 10 12

0 2 4 6 8 10

程平均誤差 ( 公分 )

編修者甲編修者乙

0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 1 0 0

0 2 4 6 8 1 0

程差標準差 ( 公分 )

編修者甲編修者乙

(9)

圖9 二編修者最終編修成果差異

圖10 二編修者編修之差異；其中橘色點為地面點，藍色點為非地面點

圖11 半自動過濾錯誤示意圖

圖12 人工編修錯誤示意圖

5. 結論

空載光達資料產製 DEM 過程中，以使用 TerraScan 軟體為例，需要先經過人工設定分類參

數，再進行人工編修。在地形變化複雜之區域，則可分為數個小區域分別進行處理，以增加作業時的效率與精度。另選擇適當分類參數分類後，對於人工編修地形變化複雜的區域和密林地區所花費之

半自動過濾

0 0.5 1 1.5 2 2.5

平坦地區山區地形變化複雜

百分比(%)

型Ⅰ錯誤型Ⅱ錯誤總和錯誤

人工編修

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

平坦地區山區地形變化複雜

百分比(%)

型Ⅰ錯誤型Ⅱ錯誤總和錯誤

(10)

時間為編修都會區或裸露地之兩倍。

研究中實驗顯示，不同編修者對於相同的地區所花費的時間會有差異。另二編修者之編修成果高程差異最大達8 公分，標準差為 25 公分；儘管最終成果相近，但局部地區仍有一些明顯之差異。為了確保最終成果品質，需有標準作業程序。人工編修部分，以都會區所需編修時間最少，森林區則需要較多的時間。而根據經驗，在編修8 小時後，高程差標準差會趨於穩定；其中以裸露地區標準差最小；密林區標準差最大。

空載光達資料不論是經由半自動過濾或是人工編修，在地形變化複雜的地區皆出現較多錯誤；

而密林地區之總和錯誤則以人工編修方式最低，可能是因該類地表覆蓋之地面點皆以人工加點為主。

另本研究只針對十幅圖幅之五種不同覆蓋區進行測試，未來研究中應增加實驗之圖幅數以利研究。

謝誌

感謝內政部提供2003 至 2004 年間獲取之空載光達掃描資料，農委會之96AS-7.3.1-ST-a1 計畫支援，研究得以完成，深致謝忱。

參考資料

內政部，2005。空載光達作業流程規範(草案)，內政部。

Axelsson, P., 1999. Processing of laser scanner data-algorithms and applications. IAPRS Journal of Photogrammetry & Remote Sensing, Vol. 54, 138-147 .

Terrasolid, 2004. TerraScan User Guide, Terrasolid.

(11)

On the Manual Editing for DEM Generation with Airborne LiDAR Data

Hsin-Yu Ho

¹

Da-Ko Chen

²

Tian-Yuan Shih

³

Wei-Chen Hsu

²

ABSTRACT

In the process of producing digital elevation model with airborne LiDAR, different procedures are required for products of different specified level. This thesis was written from April to June, 2007.

Reference to a draft for specifications and standard operation procedures for airborne LiDAR survey, MOI, DEM products were classified to three levels. When producing products higher than level two, it requires manual editing to assure the correctness. In this study, several issues of manual editing are investigated. Ten map-sheets covering five major land-cover types are selected for the comparison.

The area of each map-sheet is about 2.5 km by 2.5 km. Preliminary results show that for the same map-sheets, manual editing by different operators may produce different results and the time required is significantly different. In general, the height difference of the DEMs from two individuals is 8 cm and it gives a standard deviation of 25 cm.

Keywords:

airborne LiDAR, manual editing, filtering

1Associate researcher, Energy and Environment Research Laboratories (EEL), Industrial Technology Research Institute (ITRI)

2 Senior Researcher, EEL, ITRI

3Professor, Department of Civil Engineering, National Chiao Tung University

Received Date: Sep. 18, 2008 Revised Date: Nov. 06, 2008 Accepted Date: Feb. 02, 2009

(12)

(13)

1 Professor, Department of Landscape Architecture, Chinese Culture University Received Date: Aug. 27, 2008

2 Professor, School of Forestry and Resource Conservation, National Taiwan University RevisedDate: Oct. 29, 2008

3 Professor, School of Tourism, Ming-Chuan University Accepted Date: Feb. 04, 2009

4 Ph. D. student, School of Forestry and Resource Conservation, National Taiwan University

*Corresponding author, Phone: 02-29493264, E-mail: d94625002@ntu.edu.tw

Assessment of Ecosystem Classification Systems at Various Spatial Scales on Environmental Parameters Using

Remote Sensing Techniques

Chi-Chuan Cheng¹ Hann-Chung Lo ² Yeong-Kuan Chen ³ Chih-Da Wu^4*

ABSTRACT

The main purpose of this study was to assess the effect of ecosystem classification systems at various scales on environmental parameters using remote sensing techniques. The processes included applying hybrid classification to generate a land-use map of the north Taiwan using Landsat-5 TM image in 1995; using the DTM and the SEBAL model to calculate 16 environmental parameters and compare the differences among different land-use types; and assessing the effects of 2 ecosystem classification systems (i.e., geographic climate method and watershed division method) at various scales on environmental parameters using stepwise discriminant analysis. The results indicated that the study area was classified into 7 land-use types. They were forest-land, building, farm-land, baring farm-land, water body, cloud, and shadow. The Comparison of 16 environmental parameters among 5 land-use types (excluding cloud and shadow) showed that forestland had higher value with cosine of solar incidence angle, twenty-four hour extraterrestrial radiation, net radiation, normalized difference vegetation index, emissivity, estimating friction velocity, surface roughness for momentum transport, sensible heat flux, soil heat flux, evapotranspiration, and had lower value with transmittance, air density, surface albedo, surface albedo at the top of atmosphere, aerodynamic resistance to heat transport, surface temperature. As for assessing the effect of 2 ecosystem classification systems at various scales on environmental parameters, the result pointed out that ecosystem classification systems at various scales indeed caused the variation of environmental parameters according to the selected parameters and the number of parameters for discriminating 5 land-use types. However, among environmental parameters, normalized difference vegetation index and emissivity were the most important factors regardless of ecosystem classification systems at various scales.

Keywords: Ecosystem classification, scales, remote sensing, environmental parameters, SEBAL model

1. Introduction

Human activities for urbanization not only change the natural land cover, but also disturb the operations of ecosystem, and lead to the occurrence of global environment change.

Therefore, considerable attention has been given for monitoring changes in the global environment recently (Woolf et al., 2005). To achieve the objective of monitoring global environment change, the acquisition of environmental large-scale information becomes the most important topic. As for acquiring global information, remote sensing has been proven a useful technique because it can easily and timely provide large-scale spatial and temporal ground

information to study global environment change (Chen et al., 2006). In addition, satellite images can be used to extract the useful environmental parameters such as surface temperature, evapotranspiration etc. Due to these advantages, the combination of remote sensing and environmental parameters has been applied to analyze the impact of human disturbance on ecosystems for further studies of global environment change. For example, Giertz et al.

(2005) applied remote sensing techniques to investigate the effects of land use change on the hydrologic processes and soil physical properties in central Benin of West Africa; Menenti and Choudhury (1993) applied Landsat MSS data to develop the SEBI (Surface Energy Balance

(14)

Index) model, and then to estimate the evapotranspiration of Libyan desert in West Africa using surface albedo and aerodynamic roughness; Laymon et al. (1998) used Landsat TM images and experience functions to estimate energy fluxes and latent heat flux, and further to calculate the evapotranspiration in semidesert area of West America; Liang (2000) used remote sensing techniques to calculated the land surface albedo; Su et al. (1999) applied NOAA/AVHR images and SEBAL (Surface Energy Balance Algorithm for Land) model to calculate the energy balance parameters and then to estimate the evapotranspiration; Trezza (2002) applied a satellite-based surface energy balance with standardized ground control to derive the evapotranspiration; Morse et al.(2000) used satellite images and SEBAL model to calculate the evapotranspiration, and found that the R² value between evapotranspiration acquired from remote sensing and observed value was 0.98. Also, the results indicated that MODIS data could be applied under various land surfaces due to the sensor equipped with terrestrial, oceanic and atmospheric spectral bands.

Classification of land use and land cover using satellite image is considered an essential task in modeling the earth as a system.

Traditionally, Supervised and unsupervised classification are two common image classification approaches. Each approach has advantages and disadvantages (Lillesand et al., 2004; Lang et al., 2008). The supervised approach involves a training stage, which allows the input of analysts’ experience into image classifications. However, this approach has been regarded as overly subjective and difficult to correctly implement, because user-defined training data may not be normally distributed.

The unsupervised approach can automatically generate almost unlimited number of spectral classes, which are solid spectral foundations for generating information classes. But it requires the analyst manually label the resultant spectral classes into information classes (Lang et al., 2008). To improve the accuracy of image classification, an integrated algorithm called hybrid classification approach that takes the advantages from both classification approaches is developed (Hoffer and Fleming, 1978, Lo and Choi, 2004). In hybrid approach, cluster analysis was firstly used to acquire the spectral signatures objectively, and then the signature file was imported into the supervised classification to

generate the land-use map. Hybrid classification has been widely applied in ecosystem monitoring studies and the results from previous studies demonstrated that the integrated algorithm could provide an accurate and consistent classification of land use mapping (Hoffer and Fleming, 1978, Lillesand et al., 2004, Martinez-Casasnovas, 2000, Lang et al., 2008, Yang and Lo, 2004).

As for ecosystem monitoring, ecosystems are nested and resided within each other. The boundaries of ecosystems are open to transfer of energy and materials to or from other ecosystems. Because of the linkage among systems, energy exchange with its surroundings occurs at various spatial scales. Besides, a disturbance to a large system may also affect smaller component systems existed within it.

Consequently, the relationship between an ecosystem at one scale and ecosystems at smaller or larger scales must be examined to predict human disturbance effect (Bailey, 1996, Cheng et al., 2005). Nevertheless, most of previous researches focused on the effect of global and regional scales on environmental parameters (Rao, 1990; Tokumaru and Kogan, 1993; Yu et al., 2002; Chen et al., 2006). Few studies pay attention to landscape or ecosystem scales and their effects on environmental parameters. Therefore, a further investigation of the multi-scale relationship of environmental characteristics under various ecosystem classification systems is certainly needed.

For this reason, this study focuses on the integration of remote sensing techniques and SEBAL model. The purpose of this study is not only to derive environmental parameters using remotely sensed data, particularly on assessing the effect of ecosystem classification systems at various scales on environmental parameters.

2. Materials and Methods 2.1 Study Area and Materials

The study area (Figure 1) is located in the north Taiwan, which covers 734589.7ha and includes 5 counties (Kee-Lung, I-Lan, Taipei, Tao-Yuan, and Hsin-Chu). Many land-use types exist in this area such as industrial and scientific campuses (e.g., Hsin-chu science campus and Taipei city), mountainous forest area (e.g., I-Lan county), farmland (e.g., Lan-Yang plain) and ponds (e.g., Tao-Yuan county).

To assess the effect of ecosystem classification

(15)

systems at various scales on environmental parameters, two methods of ecosystem classification were used in this study. One is called the geographic climate method which is based on the controlling factors such as temperature and precipitation to classify land as ecosystems (Su, 1992). The other is called the watershed division method which is based on the digital terrain model (DTM) to automatically extract watershed and regards watershed units as ecosystems.

Figure 1 Location of the study area

Under various scales of ecosystem classification systems, the study area is further classified into different numbers of ecosystem units, for example, geographic climate method and watershed division method have 12 and 7 ecosystem units, respectively (Figure 2). As for ecosystem units, obviously they have unique characteristics for climate or terrain no matter what kind of ecosystem classification. For example, the major climate type in unit 3 of geographic climate method is summer rain climate, and unit 6 is everwet climate. But for watershed division method, unit 3 locates at a flat plain, and unit 2 is in a steep mountainous area. In addition, there are several industrial and scientific campuses in the study area, which

demand more water resource and may indirectly influence the operations of ecosystem. Therefore it is important to understand how land-use types and ecosystem classification systems affect the environmental parameters before making management decisions.

The materials in this study include the Landsat-5 TM image in 1995, the digital terrain model (DTM) and the national land-use inventory data in 1995. Landsat-5 TM image contains 7 spectral bands from the visible band to the middle near-infrared (mid-IR) band. Band 1~ band 3 are the blue, green, and red bands;

band 4 is the near IR; band 5 and band 7 are the mid-IR. Each of above 6 bands has a 30m ground resolution. Band 6 is the thermal IR with a 120m ground resolution. The study area was clipped to generate a land-use map after image radiometric and geometric corrections. As for the DTM with 40m resolution, it was provided by the Aerial Survey Institute of Forestry Bureau.

Above two data were then used to calculate environmental parameters of the study area. The national land-use inventory data in 1995 generated by the department of land administration, ministry of the interior have 10 categories in level 1 classification (i.e., agricultural land, communication, water, built-up land, industry, recreation, mining, salt industry, military and others). Each category has further classifications in Level 2, for example, agricultural land is classified into 5 divisions in level 2, including farm land, forest land, pasture, agricultural utility and aqua farm. In this study, the detailed land-use categories are regarded as the ground truth for evaluating image classification. Landsat-5 TM image and the DTM data of the study area were shown as Figure 3. In addition, the software included ArcGIS, ERDAS IMAGINE, and SAS.

(a) (b) (a) (b)

Figure 2 Spatial scales of 2 ecosystem classification Figure 3 Study materials: (a) Landsat-5 TM image;

systems:(a) geographic climate method;

(b) watershed division method.

(b) DTM.

(16)

2.2 Methods

2.2.1 Land‐use classification using hybrid method:

The analytical procedures included 2 parts.

Firstly, a land-use map of the study area was generated by hybrid image classification. As for the processing procedure, several blocks were firstly selected from the Landsat image according to ground land-use information. Each block contained at least 3 to 4 kinds of land-use types. The selected blocks were clustered into spectral subclasses by unsupervised classification. In this study, the number of spectral subclasses was specified as double of the information classes which exist in each block.

And then the subclass signatures were merged or deleted as appropriate based on transformed divergence (TD) as equation (1). TD was ranged from 0 to 2000. If 2 classes can be discriminated easily, then TD approaches 2000. After that, spectral signatures obtained from the selected blocks were pooled into a single spectral file.

Finally, supervised classification method was applied to generate the land-use map of the study area according to the single spectral signatures.

[ ]

( )( )

[

i j i j i j ^T

]

i i

j i

m m m m Cov Cov

tr

Cov Cov

Cov Cov tr D

D TD

−

− +

−

=

−

=

1 1

2 1

) 2 (

1

) 8 / exp(

1 2000

Where TD = transformed divergence

D = divergence

Cov

i = covariance matrix of class i

m

i= mean vector of class i

[ ] ^A

tr

= sum of the diagonal line on matrix A Secondly, to evaluate the result of land-use classification, test areas for each cover type were selected from the image based on the ground land-use information. All test areas were classified again according to the single spectral signatures. A classification error matrix was then calculated to assess the classification accuracy.

2.2.2 Calculation of environmental

parameters based on the SEBAL model:

SEBAL is an image processing model that

calculates the evapotranspiration and other energy exchanges at the earth’s surface using digital image data collected by Landsat or other remote sensing satellites measuring visible, near infrared and thermal infrared radiation (Bastiaanssen et al., 1998a). In this study, 16 environmental parameters which related to the radiometric energy balance were calculated based on the SEBAL model in order to compare the environmental parameter characteristics among various land-use types. They were cosine of solar incidence angle (cosθ), 24-hour extraterrestrial radiation (Ra24), surface albedo at the top of atmosphere (

α

toa), surface albedo (

α

0), normalized difference vegetation index (NDVI), emissivity (

ε

0), surface temperature (T0), transmittance (

τ

sw), air density (p_air), aerodynamic resistance to heat transport (rah), estimating friction velocity (u*), surface roughness for momentum transport (

z

om), net radiation (Rn), soil heat flux (G_o), sensible heat flux (H), and evapotranspiration (ET₂₄). These physical parameters also have special meanings on the description of temperature, vegetated, hydrological and energy characteristics of an ecosystem. For example, Ra₂₄ is the daily incoming solar radiation unadjusted for atmospheric transmittance;

α

toaand

α

0indicate the ratio of reflected to incident solar radiation at the atmosphere and ground surface;

ε

0and T₀are temperature indices which denote the thermal energy radiated by the surface and surface temperature condition of the area; NDVI is a sensitive indicator of the amount and condition of green vegetation;

z

om is defined as the height above the “zero-plane displacement” that the zero-origin for the wind profile just begins within the surface or vegetation cover; Rn is the net radiant energy that the land surface actually receives and loses from or to the atmosphere.

The allotment of Rn represents the energy transmission process within the ecosystem. Rn is divided into three components, ET₂₄ is the twenty-four hour actual evapotranspiration, it also indicate the energy that used to support the photosynthesis and evaporate soil water; H is the energy used to heat the air; G_o is the rest of the net energy which stored in the ground or water body. Above environmental parameters were computed by the SEBAL model based on energy balance algorithm. However, the acquisition of surface reflectance would vary with different terrains and meteorological conditions. For (1)

(17)

instance, the instantaneous and 24-hour solar radiations on a south slope are much higher than on a north slope in the Northern Hemisphere.

Meanwhile, atmospheric humidity and soil moisture are two important factors for ground reflectance and they might influence the calculation of environmental parameters.

In addition to terrains and meteorological conditions, even a thin layer of shadow or cloud can considerably drop the thermal band readings and cause large errors in calculating environmental parameters (Morse et al., 2000), 2 cover types such as shadow and cloud within the study site were masked out.

2.2.3 Assessment of ecosystem classification systems at various scales on environmental parameters:

During the process, land-use types were regarded as the dependent variables and 16 environmental parameters obtained from the SEBAL model were regarded as the explanatory variables. Stepwise discriminant analysis was then used to select the optimal combination of parameters for discriminating 5 land-use types, and further to assess the effect of ecosystem classification systems at various scales on environmental parameters. As for the test of significant variables, Wilks' lambda was used to determine which independent variable contributed to the discriminant function significantly. The Wilks' lambda ranges from 0 to 1. 0 means groups are different and 1 means groups are the same. The F test of Wilks's lambda shows which variables' contributions are significant.

3. Results and Discussion 3.1 Classification of the Land‐use

Map

Figure 4 represented the spatial distribution of selected blocks and test areas. 8 blocks were selected to perform hybrid classification and calculate the spectral signatures of 7 land-use types. They are forestland, building, farmland, baring farmland, water body, cloud, and shadow.

As for the combination of spectral signature,

“TD=1800” is assumed as the threshold. That is, two spectral signatures would be merged when the TD value between two subclasses was lower than 1800. Table 1 showed the number of spectral subclass and the merged land-use types

for each block. On the other hand, 7 test areas were also selected for evaluating the land-use classification according to the national land-use inventory data.

Figure 5 was the generated map of 7 land-use types. From the classification map, clearly shadow and cloud were occurred and excluded in the further analysis. Table 2 showed the numbers of pixel and percentages of 5 land-use types (excluding shadow and cloud types). From the table, it is known that forestland occupied most of the study site (42.76%), then baring farmland (20.63%), farmland (19.06%), building (11.41%) and water body (6.14%) was the smallest.

As for the evaluation of classification accuracy, Figure 6 generated by the selected test areas showed that most errors were occurred in 3 kinds of land-use types, that is, building, baring farmland and water body. For example, building and baring farmland were not separated easily because both types might have the similar spectral reflectance; water body was classified into baring farmland due to some dried holms existing in the river channels; and baring farmland which filled with water might be classified into water body.

Table 3 was the classification accuracy of test areas among various land-use types. It is clear that the land-use types of forestland, water body, shadow and cloud were 100% accuracy.

Building, farmland, and baring farmland are 88.58%, 89.85%, and 88.81%, respectively. In addition, the error matrix obtained from 7 test areas was shown as Table 4. It pointed out that the overall accuracy was about 93.19%. This result implied that the land-use map generated by hybrid classification was suitable for assessing the effect of ecosystem classification systems at various scales on environmental parameters.

Figure 4 Spatial distribution of the selected

blocks and test areas

(18)

Figure 5 Land-use map of the study area Table 1 The number of spectral subclass and the

land-use types merged from the selected blocks

Block Number of spectral subclasses

Merged land-use types

1 6 Forestland, Water body, Shadow 2 8 Water body, Farmland, Baring

farmland

3 6 Water body, Building , Farmland 4 8 Building , Farmland, Baring

farmland

5 6 Forestland, Shadow, Cloud 6 12 Forestland, Shadow, Building,

Baring farmland 7 10 Building, Farmland, Baring

farmland 8 4 Forestland,

Water body

Forestland Building Water body Farm-land

Baring farm-land Shadow Cloud

Figure 6 Classification of test areas

Table 2 Numbers of pixel and percentages of 5

land-types (excluding cloud and shadow)

Land-use types Numbers of pixel

Percentage of each land-use Forestland 2844976 42.76%

Building 758939 11.41%

Water body 408455 6.14%

Farmland 1268266 19.06%

Baring farmland 1372659 20.63%

Total 6653295 100%

Table 3 Classification accuracy of test areas

Land-use types Accuracy of test area

Forestland 100%

Building 88.58%

Water body 88.81%

Farmland 94.80%

Baring farmland 89.85%

Shadow 100%

Cloud 100%

Total 100%

(19)

a b c d

e f g h

i j k l

m n o p

Figure 7 Maps of 16 environmental parameters

(a: cosine of solar incidence angle; b: twenty-four hour extraterrestrial radiation, c: surface albedo at the top of atmosphere; d: surface albedo; e: normalized difference vegetation index; f: emissivity; g:

surface temperature; h: transmittance; i: air density; j: aerodynamic resistance to heat transport; k:

estimating friction velocity; l: surface roughness for momentum transport; m: net radiation; n: soil heat lux; o: sensible heat flux; p: evapotranspiration)

(20)

Table 4 The error matrix obtained from test areas

Forestland Building Water

body Farmland Baring

farmland Shadow Cloud Total of rows Forestland 238 0 0 0 0 0 0 238

Building 0 659 64 3 14 0 0 744

Water body 0 1 127 0 15 0 0 143

Farmland 0 0 0 91 0 0 0 96

Baring farmland 0 23 42 0 593 0 0 660

Shadow 0 0 0 0 0 336 0 336

Cloud 0 0 0 0 0 0 323 323

Total of

Columns 238 687 233 98 622 336 323 2540

Overall

accuracy (238+659+91+593+127+323+336) / 2540 * 100% = 93.19 %

3.2 Difference of Environmental Parameters Among Various Land‐use Types

Figure 7 was the generated maps of 16 environmental parameters using the SEBAL model. If forestland was taken as a basis and compared the difference of environmental parameters with other land-use types, the result pointed out that forestland had the higher value with cosine of solar incidence angle, twenty-four hour extraterrestrial radiation, net radiation, normalized difference vegetation index, emissivity, estimating friction velocity, surface roughness for momentum transport, sensible heat flux, soil heat flux, evapotranspiration and had the lower value with transmittance, air density, surface albedo, surface albedo at the top of atmosphere, aerodynamic resistance to heat transport, surface temperature.

3.3 Effect of Ecosystem Classification Systems at Various Scales on

Environmental Parameters

Table 5 was the output after the stepwise discriminant analysis. It indicated that the required parameters and the numbers of parameters for discriminating 5 land-use types varied with different ecosystem classification systems at various scales. From Table 5, if take unit 3 and unit 2 of the watershed division as an example, it is clear that both units have different required parameters and numbers of parameters

when discriminating 5 land-use types. For example, unit 3 and unit 2 are located in a flat plain and a steep mountainous area, respectively.

The result points out that unit 2 has four parameters (NDVI , z_om

, ε

₀

, α

₀) and unit 3 has eleven parameters (r_ah

, ε

₀

, NDVI , H , T

₀

, z

_om

, τ

_sw

, ET

₂₄

, p

_air,

R

_n

, cosθ). The reason may

result from the characteristics of ecosystem units.

As stated previously, ecosystem units have unique characteristics for climate, terrain or ecological conditions. This characteristic may influence the calculation of environmental parameters and indirectly affect the result of stepwise discriminant analysis.

In addition, no matter what kind of spatial scales and ecosystem classification systems were used, both NDVI and ε0 parameters were extracted in the stepwise discriminant analysis.

Obviously, these 2 parameters can be regarded as the most significant factors.

(21)

Table 5 Stepwise discriminant analysis under different ecosystem classification systems at various scales

Spatial scales Selected parameters

Northern Taiwan NDVI zom ε0 α0 Rn Go

rah pair τsw u*

Geographic climate method

Unit 1

NDVI T0 ε0 αtoa zom

rah H u* τsw α0 ET24

Rn Ra24 Go

Unit 2

NDVI T0 zom rah ε0 u*

α0 Go Rn H τsw αtoa

ET24

Unit 3 NDVI zom ε0 rah α0

cosθ H Go Rn ET24

Unit 4 NDVI T0 zom rah ε0 Go

α0 Rn u* H τsw

Unit 5

zom T0 u* cosθ H NDVI τsw α0 Go Rn rah

ε0

Unit 6 NDVI zom T0 u* rah

ε0 Rn ET24 α0 αtoa Go

Unit 7 zom cosθ NDVI T0 H τsw ε0 u* rah

Unit 8 zom T0 H Go αtoa ε0

NDVI τsw u* rah Rn

Unit 9 NDVI ε0 T0 cosθ τsw

α0 H zom

Unit 10 NDVI T0 τsw ε0 zom

rah u* Go α0 Rn

Unit 11 NDVI T0 ε0 H α0 τsw

zom u* τsw rah Go Rn

Unit 12

NDVI T0 τsw cosθ ε0

Go Rn α0 αtoa Ra24 pair

H zom

Watershed division

method

Unit 1 NDVI ε0 zom rah Go

cosθ pair Rn Ra24

Unit 2 NDVI zom ε0 α0

Unit 3 rah ε0 NDVI H T0 zom

τsw ET24 pair Rn cosθ Unit 4

zom NDVI ε0 T0 Ra24

α0 Rn u* rah H τsw

Go ET24

Unit 5 NDVI zom ε0 u* rah T0

ET24

Unit 6 NDVI rah τsw u* ε0

Go Rn H pair Ra24

Unit 7 NDVI ε0 zom α0 rah Go

cosθ αtoa

As stated previously, ecosystem units have unique characteristics for climate, terrain or ecological conditions. This characteristic may influence the calculation of environmental parameters and indirectly affect the result of stepwise discriminant analysis.

In addition, no matter what kind of spatial scales and ecosystem classification systems were used, both NDVI and ε0 parameters were extracted in the stepwise discriminant analysis.

Obviously, these 2 parameters can be regarded as the most significant factors.

4. Conclusions

This study integrated remote sensing techniques, SEBAL model and multivariate analysis to assess the effect of ecosystem classification systems at various scales on environmental parameters. The result can be concluded as follows.

(1) The accuracy of land-use classification evaluated by test areas was 93.19%. This implies that hybrid classification is a suitable approach to generate a land-use map. It indeed improved the accuracy or efficiency (or both) of the classification process.

(2) Environmental parameters among various land-use types were different among 5 land-use types. Besides, the required parameters and the numbers of parameters for discriminating 5 land-use types were also different under different ecosystem classification systems at various scales. But NDVI and emissivity seem to be the most significant parameters no matter what kind of spatial scales and ecosystem classification systems.

From above conclusions, clearly spatial scales and ecosystem classification systems did affect the estimation of environmental parameters. Their effects can be investigated by integrating remote sensing techniques, SEBAL model and multivariate statistical analysis. As stated previously, the quantified environmental parameters can represent the ecosystem characteristics and indicate the change of environment. Therefore, the result obtained from this study can be extended in the future studies of global environment change and forest resource management.

REFERENCE

Bailey, R. G., 1996. Ecosystem Geography, Springer, New York, pp 1-49.

Bastiaanssen, W. G. M., Meneni, M., Feddes, R.

A. and Holtslag, A. A. M., 1998a. A remote sensing surface energy balance algorithm for land(SEBAL):1. Formulation, J. Hydrol, Vol. 212-213, pp.198-212.

Bastiaanssen, W. G. M., Pelgrum, H., Wang, J.,

(22)

Ma, Y., Moreno, J., Roberink, G. J. and van der Wal, T., 1998b. A remote sensing surface energy balance algorithm for land (SEBAL):2. Validation, J. Hydrol, Vol.

212-213, pp. 213-229.

Chen, C. T., Wu, S. T. and Chiang, Y. F., 2006.

Using MODIS satellite images to estimate evapotranspiration in Taiwan, Taiwan J For Sci, 21(2): pp. 249-61. [in Chinese with English summary].

Cheng, C. C., Chen, Y. K., Jan, J. F. and Wang, S.

F. 2005. DTM, GIS, and DSS applications in forestland ecosystem classification and suitability analysis, Journal of Photogrammetry and Remote Sensing, Vol.

10, pp. 351-360.

Giertz, S., Junge, B. and Diekkruger, B., 2005.

Assessing the effects of land use change on soil physical properties and hydrological processes in the sub-humid tropical environment of West Africa, Physics and Chemistry of the Earth, Vol 30, pp.

485-496.

Hoffer, R. M. and Fleming, M., 1978. Mapping vegetative cover by computer-aided analysis of satellite data, Purdue Univ., LARS Technical Report 011178.

Lang, R., Shao, G., Pijanowski, B. C. and Farnsworth, R. L., 2008. Optimizing unsupervised lassifications of remotely sensed imagery with a data-assisted labeling approach, Computer and Geosciences, Vol. 34, pp. 1877-1885.

Laymon, C., Quattrochi, D., Malek, E., Hipps, L., Boettinger, J. and McCurdy, G., 1998.

Remotely-sensed regional-scale evapotranspiration of a semi-arid great

basin desert and its relationship to geomorphology, soils, and vegetation, Geomorphology Vol. 21, pp. 329-349.

Liang, S., 2000. Narrowband to broadband conversions of land surface albedo - I Algorithms, Remote Sensing of Environment, Vol. 76, pp. 213-238.

Lillesand, T. M., Kiefer, R. W. and Chipman, J.

W., 2004. Remote sensing and image interpretation, fifth edition, Wiley, New York, NY, 763p.

Lo, C. P. and Choi, J., 2004. A hybrid approach to urban land use/cover mapping using Landsat 7 Enhanced Thematic Mapper Plus (ETM+) images. Int. J. Remote Sensing, 25(14):2687-2700.

Martinez-Casasnovas, J. A., 2000. A cartographic and database approach for

land cover/use mapping and generalization from remotely sensed data, International Journal of Remote Sensing, Vol. 21, Issue 9, pp. 1825-1842.

Menenti, M. and Choudhury, B. J., 1993.

Parameterization of land surface evaporation by means of location dependent potential evaporation and surface temperature range, Proceedings of IAHS conference on land surface processes, IAHS Publ. 212, pp. 561-568.

Morse, A., Tasumi, T., Richard, G. A. and William, J. K., 2000. Application of the SEBAL methodology for estimating consumptive use of water and stream flow depletion in the bear river basin of Idaho through remote sensing, The Raytheon System’s Company Earth Observation System Data and Information System Project.

Rao, P. K., Holmes, S. J., Anderson, R. K., Winston, J. S. and Lehr, P. E., 1990.

Weather satellites: Systems data and environmental applications American meteorological society. Boston 503p.

Su, H. J., 1992. Vegetation of Taiwan: altitudinal vegetation zones and geographical climatic regions, Proceedings of the workshop on the Biological Resources and Information Management of Taiwan, pp 39-53.

Tokumaru, K. and Kogan, F. N., 1993. Satellite technology for environmental monitoring in developing countries, United Nations Educational, Scientific and Cultural Organization. Paris, pp. 51-89.

Trezza, R., 2002. Evapotranspiration using a satellite-based surface energy balance with standardized ground control., Ph.D.

Dissertation, Utah State University, Logan, UT.

Woolf, A., Lawrence, B., Lowry, R., Dam, K. K.

V., Cramer, K., Gutierrez, M., Kondapalli, S., Latham, S., O’Neill, S. and Stephens, A., 2005. Climate science modelling language:

standards-based markup for metocean data, 85th meeting of American Meteorological Society, San Diego.

Yang, X. and Lo, C. P., 2002, Using a time series of satellite imagery to detect land use and land cover changes in the Atlanta, Georgia metropolitan area. International Journal of Remote Sensing, Vol. 23, pp. 1775–1798.

Yu, P. S., Yang, T.C. and Wu, C. K., 2002.

Impact of climate change on water resources in southern Taiwan, Journal of Hydrology, No. 260, pp. 161-175.

(23)

1文化大學景觀學系教授收到日期:民國 97 年 08 月 27 日

2國立台灣大學森林環境暨資源學系教授修改日期:民國 97 年 10 月 29 日

3銘傳大學觀光學院教授接受日期:民國 98 年 02 月 04 日

4國立台灣大學森林環境暨資源學系博士生

*通訊作者, 電話: 02-29493264, E-mail: d94625002@ntu.edu.tw

應用遙測技術評估不同生態分類系統和空間尺度對環境參數之影響

鄭祈全

¹

羅漢強

²

陳永寬

³

吳治達

^4*

摘要

本研究旨在應用遙測技術，評估不同生態分類系統和空間尺度對環境參數的影響。研究方法包括：

應用混合式影像分類法，將台灣北部地區之Landsat-5 TM 影像進行土地使用分類；利用 DTM 及 SEBAL 模式萃取16 種環境參數，並探討各土地使用型之環境參數的差異情形；以及透過多變量統計之逐步判別分析法，評估北台灣12 個地理氣候區及 7 個集水區兩種生態分類系統在不同空間尺度下之環境參數的差異性。研究結果指出，試區經混合式影像分類法分為森林、建地、水體、耕作農地、無耕作農地、雲及陰影7 種土地使用型；在分析 5 種(扣除雲及陰影)土地使用型之環境參數的差異比較時，顯示森林地區在太陽入射角之餘弦影像、大氣層頂單日輻射量、淨輻射量、常態化差異植生指標、地表熱紅外光放射率、

摩擦速度、動量傳輸粗糙度、大氣可感熱、土壤熱通量及蒸發散量等指標值較高，而在一維穿透係數、

大氣密度、地表反照率、大氣層頂表面反照率、空氣阻抗熱傳導係數、地表溫度等指標值較低；至於評估不同生態分類系統和空間尺度對環境參數之影響結果指出，在不同的生態分類系統和空間尺度之下，

用來區分 5 種土地使用型所需要的環境參數與參數數目皆不盡相同，但常態化差異植生指標與地表熱紅外光放射率兩項參數不管在那一種生態分類系統，均為重要的影響參數。

關鍵詞：生態分類系統、尺度、遙感探測、環境參數、SEBAL 模式

(24)

航航 測測 及及 遙遙 測測 學學 刊刊 Journal of Photogrammetry and Remote Sensing