蘭台大規模崩塌潛勢區—觀測資料綜整分析

Abstract

宜蘭太平山蘭台苗圃地區為一大規模崩塌潛勢區，為了防止將來發生複合型災害，減少人員傷亡以及財物損失，近年來已開始執行觀測整合研究計畫，結合GNSS連續觀測站、地震站與光柵光纖孔隙水壓計等觀測系統，紀錄影響坡面穩定性之機制。根據GNSS座標時間序列與雨量站資料比對，坡面位移量與累積雨量具有高度相關性，並記錄到因乾濕季變化產生的重力滑移現象。在經濾波處理後的1-4 Hz地震表面波波速變化資料(dv/v)能觀察到乾季波速增加、濕季減慢的趨勢，可反映深層地層的構造特徵及水文變化情形。 本研究利用自迴歸模型、時頻圖、獨立成分分析等時間序列分析方法，了解GNSS、1-4 Hz dv/v(約反應深度100 m的波速變化)、深度60 m孔隙水壓及累積雨量的時間變化，並探討不同時間序列資料和水文現象在時間及空間上的相關性。此外，為了解大雨事件GNSS位移加速的時空歷程，本研究分析觸發滑動時的臨界降雨值、滑動持續時間、滑移量和降雨量的關係。因蘭台地區目前強降雨事件的資料紀錄過少，本研究也將南橫天池、塔塔加和玉山站GNSS觀測資料加入討論，分析強降雨發生時的坡體滑移特性。 地震波1-4 Hz dv/v和GNSS的自回歸模型都能成功以水文資料進行修正，其中dv/v和水文資料變化呈負相關，以雨量資料修正模型時，位於崩塌潛勢區的測站修正量較大，對雨量變化較敏感。在不同的資料變化間有時間延遲(time lag)情形，在降雨當天dv/v就會發生變化，水壓變化比降雨晚10至25天，GNSS約比水壓變化晚10至30天。 時頻圖觀測到GNSS、1-4 Hz dv/v、水壓和累積雨量資料皆有0.3年、0.65年和1年的週期變化，推斷是受到雨量週期變化的影響，根據年週期波形的相位差，降雨和dv/v的峰值約同時抵達，水壓變化比降雨晚30天，GNSS又比水壓變化晚了75天。 利用獨立成分分析並不容易歸納出1-4 Hz地震波速變化dv/v或GNSS各測站共有的特徵訊號，可能與蘭台地區複雜且局部的坡體構造狀況有關，各測站紀錄的變化量差異大，儀器斷電造成的資料缺失也是造成誤差的原因之一。 蘭台地區發生坡體快速滑動時的臨界降雨值約為300 mm，滑動情形在降雨停止的兩天內就會逐漸趨緩，總累積雨量和總滑移量成正比。根據塔塔加、南橫天池和玉山站的位移場，在累積雨量至少達700 mm以上的事件才能觀測到較明顯的滑移發生，約在降雨停止的兩天內就會逐漸趨緩，總累積雨量和總水平位移量呈高度正相關。從降雨和滑移關係上來看，蘭台地區降雨臨界值相較其他測站低且容易觸發坡面滑動，此外滑動量大小也較難以單場累積雨量大小來訂定，往後需將土體含水量和孔隙水壓等因素加入考慮。本研究提供宜蘭太平山蘭台苗圃地區不同時間序列在降雨過後的歷程，可作為將來山崩預警評估系統之參考。

Lan-Tai in the Taiping Mountain, Ilan is the potential area for the deep-seated landslides. To provide an understanding of sliding behavior over time, this study aims at analyzing data collected by earth science-based monitoring techniques, including the Global Navigation Satellite System (GNSS) and seismic noise interferometry, as well as rainfall and groundwater records. Comparing with the conventional engineering-based invasive approaches, our strategy provides continuous data across different spatiotemporal scales with a lower cost. By analyzing GNSS and rainfall time series, we find slope movements often occur when the rainfall intensity is above 300mm/day in Lan-tai. The seasonal variations in GNSS time series reflect not only the different rate of downslope movement during the dry and wet seasons but also the variability of annual water storage. Furthermore, the filtered 1-4 Hz seismic velocity change (dv/v) also records the annual cycle associated with hydrological storage change. To extract the spatiotemporal pattern of various observations during the heavy rainfall events, different methods for time series analysis are used, including the autoregression model, wavelet spectrogram and Independent Component Analysis (ICA). Results from the autoregression model suggests an immediately response between the daily rainfall and dv/v changes, following by the water pressure and GNSS changes about 10-25 days and 10-30 days, respectively, behind the peak daily rainfall. Computing the spectrograms of different data sets shows 0.3, 0.6 and 1-year periodicity, which are likely associated with the variation of rainfall intensity. The phase of peak annual cycle also shows invisible time lag between rainfall and dv/v, the peak water pressure lags about 30 days after rainfall, and the peak GNSS response lags 75 days behind maximum water pressure. In the Lan-tai area, the slope movement accelerates when the daily rainfall reaches 300 mm and ceases about two days later after the rain stops. Due to insufficient heavy rainfall events in Lan-tai, data collected at three continuous GNSS stations, TATA, TENC, and YUSN, located in the Central Range of Taiwan, are analyzed in this study as well. The deep-seated landslides at these sites accelerate when the cumulative rainfall exceeding 700 mm/day, larger than the rainfall threshold of landslide in Lan-tai. The variation of rainfall threshold in different regions suggest other factors contribute to landslides triggering such as geological, geomorphological, and climatic conditions. Delineating the role of different physical mechanisms remains challenging. The continuous monitoring and measurements using multiple observations can provide the background sate of landslide-prone areas, help to develop an early warning system, and eventually improve the risk assessment of landslide instability.