極端降雨事件下利用時間序列模擬大規模崩塌潛勢區之地下水位變化輔助決策者發布警戒值

近年來臺灣受氣受變遷影響導致降雨型態改變，降水強度增加及延時縮短造成極端降雨事件頻傳，臺灣山區災害類型從以往單一型災害，演變成複合型災害，2009年莫拉克颱風導致山區多處嚴重土砂災害即為明顯案例，造成中南部山區同時發生地滑、土石流和堰塞湖造成嚴重的人員傷亡和破壞。為應對複雜的災害型態，2010年起行政院農業委員會水土保持局依地質型態及空載光達資料，劃定227個大規模崩塌潛勢區域，並於各崩塌潛勢區域進行調查規畫及裝設相關地表、地中監測儀器，作為未來發布警戒基準值參考依據。為有效掌握大規模崩塌及地滑災害，地下水位扮演重要的因子，但國內外對於地下水位作為防災預警相關研究卻不多，本研究採用自回歸移動平均整合模式(Autoregressive Integrated Moving Model, ARIMA)，選擇臺中市和平區3處大規模崩塌(D038、D050及T003)內自動化監測站作為研究測站，各站均超過13年監測且位處於保全住戶周邊，利用各站2009年莫拉克颱風作為訓練，及選用其它2場極端降雨事件之時雨量及時地下水位變化資料作為模式驗證，以建立各站之ARIMA模型參數、轉換函數，為模擬地下水位變化輔助將水保局情資研判中心發布警戒基準之依據。由各測站之極端降雨事件ARIMA模擬成果可知，S1和A1測站實際觀測與模擬地下水位變化趨勢相近，峰值及到達延時均可成功模擬，有助於大規模崩塌區D038及D050後續地下水位警戒值發布參考；但大規模崩塌區T003之J1測站之模擬成果與觀測值具有明顯落差，且峰值及延時明顯落後，則不適用於警界發布參考，研判係人為工程構造物影響地下水位變化導致；可知ARIMA模型係可以助於決策者發布警戒值之參考，仍建議應持續蒐集未來極端降雨事件資料，滾動檢討ARIMA模型參數，並選擇周邊較無人為影響之測站作為代表測站。

關鍵字

大規模崩塌地；地下水水位；自回歸移動平均整合模式；崩塌地預警；輔助決策判斷

並列摘要

In recent years, the rainfall pattern has been changed since precipitation intensity has increased and the duration has shortened in Taiwan. Traditionally, the Taiwanese government focused on the sediment-related disasters only, but typhoon Morakot caused severe casualties and damages in the central Taiwan because of landslides, debris flows and barrier lakes. In order to tackle complicated compound disasters, the Soil and Water Conservation Bureau (SWCB) has identified 227 large-scale landslide areas which are classified by the size and geological formation. Since then the SWCB has installed real-time monitoring equipment and released warning systems which are based on the past rainfall-induced disasters. Although billions of dollars have been invested in large-scale landslide areas, the SWCB is still unable to effectively release a warming alert of underground water level, and there is a lack of relevant research and application related to monitoring data. Therefore, the study adopts time-series analysis, the Auto-Regressive Integrated Moving Average (ARIMA) model, to simulate the change of underground water which is affected by the rainfall. Under the study, the selected S1, A1 and J1 stations are located in 3 large-scale landslide areas D038, D050 and T003 respectively, in the Lishan area, Taichung City. These stations have observed more than 13 years, and are located in major villages. The study establishes 3 different models for each station and applies the data from the typhoon Morakot which has the biggest accumulated rainfall to train these models. Then, each model has input the rainfall and underground water data from other 2 representative rainfall events with the highest record of underground water level and slope-land movement to verify the results of simulation. The main outputs are (1) the parameters of 3 models and (2) the models to simulate the change of underground water by input the rainfall data, and the study aim to help decision-makers to issue a landslide alert in the SWCB. The results of the simulation show that the data from extreme rainfall events such as Typhoon Dujuan, Megi, Soudelor and 15-days series rainfall in 2016 verifies the parameters of the models effectively. The models of S1 and A1 stations demonstrate positive correlations between the simulated and observed peak and change of underground water, which helps decision-makers to pre-release warming issue in large-scale landslide D038 and D050. However, the result from the model of J1 station in large-scale landslide T003 shows negative correlation in the peak of underground water. It is assumed that the ARIMA models are specific to simulate one station only, and the results could provide the SWCB to issue a warming alert.