利用羅吉斯回歸法分析崩塌對土石流降雨警戒基準值之影響-以陳有蘭溪集水區為例

Applying Logistic Regression Method to Analyze the Effects of Landslide on Rainfall Threshold Value for Debris Flow Warning –Using the Watershed of Chenyoulan Stream as An Example

曾勛苑(Xun-Yuan Zeng)

指導教授 ： 范正成

國立臺灣大學/生物資源暨農學院/生物環境系統工程學研究所/碩士(2012年)

https://doi.org/10.6342/NTU.2012.02624

摘要

本研究為探討崩塌及其他地文因子對土石流降雨警戒值（rainfall threshold value for debris flow warning）之影響，並進而推估之。土石流降雨警戒基準值為現今臺灣地區發布土石流警戒參考之依據，以供為預警指標，並於每年或遇重大土砂災害事件後檢討修訂。
本研究之研究區域位於陳有蘭溪集水區，以統計檢定之方式篩選出崩塌率、溪床平均坡度、有效集水區面積與形狀係數等四個與土石流發生相關性較高的地文因子；而其中以崩塌率與土石流發生關係最為顯著。然後，以羅吉斯回歸之分析方式，結合地文因子及降雨參數建立土石流發生雨量警戒值評估模式。同時，以兩種雨場劃分及兩種地文因子計算方式來探討不同雨場劃分、降雨參數計算方式及地文因子有無經過適當的量化程序轉換等因素對模式的影響。研究結果顯示，以Fan et al. (2003)的雨場劃分方式及經過隸屬函數轉換的地文因子建立之模式最佳，其判斷土石流發生的準確率達八成以上。因此，在考量土石流之預警方式通常以較保守的標準來發布警戒通知的情況下，以此結果來建置土石流警戒模式。
所推導之模式，與土石流發生的物理機制相符合，當崩塌率(DN)、溪床平均坡度(SN)、有效集水區面積(AN)及形狀係數(FN)增加時，其發生機率也隨之增加。若土石流發生機率(p)相同，當崩塌率(DN)、溪床平均坡度(SN)、有效集水區面積(AN)及形狀係數(FN)增加時，則土石流發生雨量警戒值降低。此一土石流警戒模式，可提供重大災害後（如地震、颱風豪雨後），透過檢視四種地文因子的變化量，快速修正土石流發生雨量警戒值，適時反映現地狀況。

關鍵字

土石流警戒基準 ； 羅吉斯迴歸 ； 崩塌 ； 土石流

並列摘要

The main purpose of this study is to investigate the effects of landslides and other physiographic factors on the rainfall threshold value for debris flow warning and then to evaluate it. In Taiwan, the rainfall threshold value for debris flow warning has been used for years as a basis to announce debris flow warning and it is amended annually or after the event of great mass movement disasters.

The site of this study was selected at Chenyoulan stream watershed. The statistics examines to landslide ratio, the average riverbed slope steepness , effective watershed area , form factor of watersheds and the debris flow take place higher relativity of physiographic factor, and among them with landslide ratio and debris flow to establish relationship most for notable. Then, with method of logistic regression was used for analysis, building a model to assess rainfall threshold value for debris flow warning with landslides and other physiography factors. In the meantime, compare the effect of two different rainfall-field division ,and the influence of physiographic factor conversion of membership function. From the result, Fan et al.(2003) the rain-field division method and the mode of has been convert to membership function physiography factor the best, the accuracy rate of debris flow occurrence reaches to 80% above. Therefore, consider the debris flow warning usually more conservative standard to publish a warning notice, this results to build a debris flow warning mode.

Deduce of the physical mechanism of mode and debris flow occurrence mutually matches. While landslide ratio (DN), average riverbed slope steepness (SN) and effective watershed area (AN) and form factor of watersheds (FN) increment, it takes place a probability to also immediately increase. If the debris flow takes place a probability(p) homology, the rainfall threshold value for debris flow warning to reduce while landslide ratio(DN), average riverbed slope steepness (SN) and effective watershed area (AN) and form factor of watersheds(FN) increment. This debris flow warnning mode, after to provide major disaster (as earthquake, typhoon pouring rain) ,through examine the quantity of four kinds of physiographic factors, is rapid to revise the rainfall threshold value for debris flow warning ,is well timed to reflect now condition.

並列關鍵字

rainfall threshold value for debris flow warning ； logistic regression ； landslide ； debris flow

參考文獻

3. 水土保持局(2003)，「以降雨因子進行土石流警戒基準值訂定」，財團法人成大研究發展基金會。

26. 范正成、吳明峰、彭光宗(1999)，「豐丘土石流發生降雨臨界線之研究」，地工技術，第74期，pp.39-46。

27. 范正成、吳明峰(2001)，「一級溪流土石流危險因子及其與臨界降雨線之關係」，中華水土保持學報，第32卷，第3期，pp.227-234。

29. 張石角(1987)，「山坡地潛在危險之預測及其在環境影響評估之應用」，中華水土保持學報，第18卷，第2期，pp.41-62。

38. 游繁結、陳重光(1987)，「豐丘土石流災害之探討」，中華水土保持學報，18(1)：76∼92。