河系爲集水區地形的重要特徵，許多河系地形參數與集水區水文硏究息息相 關，隨著GIS技術的進步與數値高程模型資料的普及，藉由GIS流量累積法自 網格式DEM中自動生產集水區河系，已成爲相當普遍且有效的方法，然而，對 於關鍵性的「集流閾値」’大多流於主觀的槪略性評斷，欠缺客觀的評估依據。 本硏究結合等降法則、GIS與統計分析方法，使用國內40公尺空間撕度的網 格式DEM進行集水區河系的擷取，並檢核等降程序分析最適集流閾値方式的有 效性與合理性；透過與1:25,000及1:10,000地形圖數値化資料的分析比較，評估 DEM揺取河系的空間正確性。由德基水庫上游七家灣溪集水區進行的實證硏究 結果顯示，等降程序在擷取符合Horton’s laws地形法則的河系上存在有效性，但 是無法確認其分析所得之集流閾値的最適性;擺取河系之一級河段的高程落差呈 現明顯右偏態，應以對數常態分布觀點來執行等降程序比較合理：同時也發現 DEM擷取河系所計算的各項Horton’s laws參數，對集流閾値呈現明顯的敏感 性。以地形圖數値化資料爲比對標準的分析結果顯示，DEM擷取河系有相當程 度的空間正確性，但是無法有效分析一級河段最上游起始位置的局部性變異。

Channel networks are important features for describing the topography of drainage basins. With the progress of GIS and more accessible digital elevation modeling ( DEM ) data, we can now extract the channel network from DEM readily. Using the GIS function called flow accumulation from grid-based DEM is the most popular method on extracting the channel network. However, the threshold of flow accumulation is the key factor in channel network extraction and should be decided more accurately. This study extracts the channel network of the Chi-Cha-Wan watershed from 40 meters resolution DEM with the integration of constant drop law, GIS and statistical methods, and compares it with the digital data from 1:10,000 and 1:25,000 topographic maps. The results show that channel networks extracted from DEM match Horton's laws with a high level of accuracy, but could not prove that the threshold of flow accumulation found by constant drop procedures is the optimal result. The frequency distribution of the channel lengths of first-order segments skewed to the right, and should be corrected by using the logarithm of length when using the constant drop procedure. The parameters of Horton's laws measured from extracted channel networks are sensitive to the changing threshold. The comparison of spatial accuracy between the extracted channels and the rivers shown on the topographic maps demonstrates some limitation of flow the accumulation method of which, the identification of first-order channel heads is the major one.