中文摘要 【摘要】在水文分析上,傳統上採用人工測繪生產河川網路,雖具有一定之準確性,但作業緩慢耗時。而採用網格式數值地形模型推導集水區河川網路,較人工作業快速有效率,但推導之正確性因地而異。為了探討河川網路自動化萃取在台灣山地集水區之表現,本研究以清水溝溪、北勢溪與雙溪流域為試區,使用其數值地形模型進行河川網路之推導。於自動化水文模式演算中設定網格解析度、網格資料前處理法、網格式水文演算法與門檻值四種不同之因子進行組合演算,並探討其結果特性與影響。最後以網格誤差檢核為主,人為判釋為輔檢驗產出河流之符合率,求出與人工測繪生產之河川網路最相近之自動化流程。 整合演算結果,探討各因子之影響可發現,網格解析度提升可增高河道彎曲性表現,但誤授率與漏授率均隨之提升;而採用不同網格資料前處理法則會影響局部河流之走向,但對整體之符合率影響不大;至於採用不同水文演算法,雖會產生局部性河道走向之不同,但與人工測繪結果比較時,彼此間總誤差率之差異極小;而門檻值設定則主要影響河川之長度,其修正誤差率之能力則會隨著網格解析度而變化。 經檢核結果可知,以本研究中採用之三個試區而言,當採用40M網格解析度之數值地形模型,套用無限流向法或Rho8法兩種水文演算法,並設定門檻值於200至250區間內時,整體之符合率就其他組合之流程而言相對最佳。但整體而言,總誤差率偏高,達到35%以上。且自動化水文模式推算出之河流寬度,均遠較人工測繪結果窄,因而造成極大之漏授誤差。故於台灣山地集水區,採用自動化水文模式演算取代目前人工測繪生產具有寬度河流之能力評定而言,雖然其具有效率上之優勢,但符合度仍不足。 【關鍵詞】數值地形模型、河川網路、水文模式模擬、網格解析度、門檻值、山地集水區
Abstract Traditionally, stream networks were produced manually in hydrologic analysis. Though certain degree of accuracy can be achieved, the process requires large amount of time. In comparison, deriving stream networks from grid digital terrain model (DTM) is more efficient than the manual process. However, the accuracy of the results varies with different terrains. To evaluate the performance of automatic extraction of stream networks for mountainous watersheds in Taiwan, this study applied various algorithms to derive stream networks using grid DTM for three different watersheds. By comparing the conformity of the grid stream networks produced using manual and automatic method, the effects of grid cell sizes, preprocessing methods, algorithms, and threshold values on the characteristics and accuracies of the resulting stream networks were investigated. In order to find the best hydrologic analysis model, this study assessed the characteristics and accuracies of stream networks derived from 5m, 10m, 20m, and 40m cell size DTM using six common routing algorithms, two types of preprocessing method, and 13 levels of threshold value. The results indicate that the conformity of flow path increases as the grid cell size decreases, however the commission error and omission error increases as well; different preprocessing methods and routing algorithms result in different flow paths, however the differences of overall conformity are not significant; selection of threshold value primarily affects the flow length, and the error rate varies with different cell sizes. For the watersheds chosen in this study, when applying the D-infinity and Rho8 algorithms to the DTM with 40m grid resolution, the overall conformity was better than the other combinations if the threshold value was set to be between 200 and 250. In general, the overall error rate was high (greater than 35%). This was because the width of the streams produced from automatic analysis was much narrower than that of the manual process, hence resulting in large omission errors. Therefore, for the mountainous watersheds in Taiwan, though the automatic hydrologic analysis methods had the advantages of producing stream networks efficiently, the accuracy was not good enough. 【Key words】:Digital Terrain Model , Stream Networks, Hydrologic Modeling, Grid Resolution, Threshold Value, Mountainous Watersheds