Multidisciplinary Hazard Assessment of the Nepal-Himalaya Region: Insight to the 2015 Mw 7.8 Gorkha Earthquake

Abstract

지표면의 급격한 변화로 인해 산사태가 발생하는 지진은 자연 및 인공사면의 불안정성을 유발하는 중요한 요인이다. 네팔은 세계에서 지진발생 확률이 매우 높은 히말라야의 활발한 대륙충돌지역에 속한다.지진 및 지진과 함께 발생하는 산사태로 인한 생명과 인명손실은 네팔 재해의 상당부분을 차지한다.본 연구 지역인 네팔의 인구 증가를 수용하기위한 급속한 도시화로 인해 향후 몇년간 지진 재해는 증가할 것으로 판단된다. 지진 및 지진과 함께 발생하는 산사태로 인한 영향을 보다 효과 적으로 줄이기 위해, 이러한 재난을 예측하고 피해와 관련 추가 연구가 필요하다. 본 연구에서는 2015년 Gorkha지진의 본진 및 여진에 대한 통계학적 분석과 네팔 히말라야지진에 대한 다양한 위험성 평가를 수행했다. 연구 지역의 지질구조 조건에 관한 정보를 얻기 위해 다양한 통계학적인 매개변수를 계산했다.이를 위해 랜덤하게 선택한 여진과 함께 본진과 가장 큰 여진의 단층면해를 계산했다.진원지, 지진매개변수와 지반운동예측방정식은 확률론적지진재해도와 산사태민감성 매핑을 위한 가장 중요한 구성요소이다. 본 논문은 다양한 원인 변수로 인공신경망 (Artificial Neural Network)을 적용하여 새로운 지진 발생 지도를 제안한다.최우법 (maximum-likelihood procedure)을 사용하여 β, λ 및 Mmax와 같은 지진매개변수를 계산했다 (Kijko.,2004). 지반운동예측방정식을 계산하기 위한 충분한 수의 지진관측소가 연구지역에서는 부족하기 때문에 전세계적으로 통용되는 글로벌 지반운동예측방정식 (GMPE)을 사용한다.로직트리방법은 분석에서 대안 모델을 포함시킴으로써 다양한 지반운동예측방정식에서 불확실성을 특성화하기위해 사용했다.CRISIS 2015 소프트웨어를 이용하여 확률론적 지진재해지도를 작성했다.확률론적 지진재해 분석을 위해 네 가지 지반운동예측방정식을 사용했다 (Atkinson and Boore (2006), Chiou and Youngs (2014), Youngs et al. (2006) modified and Zhao et al (2006)). CRISIS에서 생성 된 재해지도는 플롯과 매핑을 위해 ArcGIS로 전송되었다. Gorkha 지진에 가장 많은 영향을 받은 12개 지역의 산사태민감도 지도는 두 개의 변수를 갖는 모델 frequency ratio (FR), evidential belief function (EBF)및 weight of evidence (WOE)를 사용하여 도시했으며 머신러닝 데이터 알고리즘도 결합되었다 (예, random forest (RF)). 수동 분류 방법을 사용하여 LSI값을 네 가지 민감도 등급 : 낮은, 보통, 높음, 매우 높음 등으로 재 분류했다.마지막으로, 산사태 민감도와 50년 초과 확률의2%, 5% 및 10%의 최대지반가속도 (PGA) 지진재해지도와 연관시키기위해 공간다중기준평가를 사용했다. 공간다중기준평가는 저, 중, 고, 고 위험 등급으로 재 분류되었다.지진과 산사태 민감도 지도의 확률위험평가는 주민 대피에 대한 위험 경보와 관련된 결정에 필요한 정보를 제공할 수 있다.

Earthquakes are a major trigger for the instability of natural and man-made slopes in which rapid changes on the surface generate landslides. Nepal lies in one of the active continental collision zones of the world, the Himalaya, where the probability of earthquake occurrence is very high. The loss of life and causalities resulting from earthquakes and co-seismic landslides are a large portion of the disasters in occurring in Nepal and may increase in the coming years because of rapid urbanization to accommodate the growth of the population in the study area. To more effectively reduce the effects resulting from earthquake and co-seismic landslide disasters, further studies related to predicting these disasters and the associated damage are necessary. In this dissertation research, a statistical analysis of the April 25, 2015 Gorkha earthquake and its aftershocks, and an assessment of the various hazards with respect to the earthquake in the Nepal Himalaya were conducted. Various statistical parameters were calculated to obtain information regarding the geotectonic condition of the study area. A fault plane solution of the mainshock and the largest aftershock, along with randomly selected aftershocks, was prepared. Seismic source zone, seismic parameters, and a ground motion prediction equation are the most important components for the preparation of the probabilistic seismic hazards and hence landslide susceptibility mapping. This dissertation proposes a new seismic source zonation map using various causative factors and applying a machine learning artificial neural network. Seismic parameters such as β, λ and Mmax were computed employing the maximum-likelihood procedure of Kijko 2004. Because there was a lack of a sufficient number of stations to obtain sufficient data for the preparation of the ground motion equation, this study adopted the globally accepted ground motion prediction equation which is suitable for this study area. The logic tree approach has been used to characterize epistemic uncertainties in various GMPEs by including alternative models in the analysis. A probabilistic seismic hazard map was prepared using CRISIS 2015 software. Four different ground motion prediction equations (those of Atkinson and Boore (2006), Chiou and Youngs (2014), Youngs et al. (2006) modified and Zhao et al (2006)) were used for the preparation of the Probabilistic Seismic Hazard Analysis. The hazard maps thus generated in CRISIS were transported to ArcGIS for plotting and mapping. Landslide susceptibility maps of the 12 districts most affected by the Gorkha earthquake were prepared using the bivariate models frequency ratio, evidential belief function, and weight of evidence and combined with a machine learning data mining algorithm, i.e. random forest. A manual classifier method was used to reclassify the landslide susceptibility index values into four different susceptibility classes: low, moderate, high, and very high landslide susceptibility. Finally, spatial multi-criteria evaluation was used to correlate the landslide susceptibility with the probability seismic hazard map of peak ground acceleration 2%, 5%, and 10% probability of exceedance of 50 years and reclassified into low, moderate, high, and very high hazard classes. The probabilistic hazard assessment of the earthquake and landslide susceptibility map together can provide requisite information for decisions regarding issuing hazard warnings for a resident evacuation. This dissertation research deals with the statistical analysis of Gorkha earthquake and its aftershocks, and assessment of the various hazard with respect to the 2015 Gorkha Earthquake in the Nepal Himalaya. Various statistical parameters were calculate in order to get the information of the geotectonic condition of the study area. This study prepare the fault plane solution of mainshock and the largest aftershock along with randomly selected aftershocks. Seismic source zone, seismicity parameters and ground motion prediction equation are the most important component for the preparation of the probabilistic seismic hazard and hence landslide susceptibility mapping. This thesis proposed new seismic source zonation map using various causative factors and applying machine learning artificial neural network. Seismicity parameters such as β, λ and Mmax were computed by employing the maximum-likelihood procedure of Kijko 2004. Since there was lack of sufficient number of station to get sufficient data for the preparation of the ground motion equation, this study adopted the globally accepted GMPE suitable of this study area. The logic tree approach has been used to allow characterization of epistemic uncertainties in various GMPEs by including alternative models in the analysis. Probabilistic seismic hazard map was prepared using CRISIS 2015 software. Four different ground motion prediction equations [Atkinson and Boore (2006), Chiou and Youngs (2014), Youngs et al. (2006) modified and Zhao et al (2006)] were used for the preparation of PSHA. The hazard maps thus generated in CRISIS were transported to ArcGIS for plotting and mapping. This thesis prepared landslide susceptibility maps of twelve most affected districts by the Gorkha earthquake using FR, WOE and EBF bivariate models and combined with a machine learning data mining algorithm i.e. random forest. Manual classifier method was used to reclassify the LSI values into four different susceptibility zones as low, moderate, high and very high landslide susceptibility classes. Finally, spatial Multi-criteria evaluation was used to correlate the landslide susceptibility with the probability seismic hazard map of pga 2%, 5% and 10% probability of exceedance of 50 years and reclassified into low, moderate, high and very high hazard classes. Probabilistic hazard assessment of the earthquake and landslide susceptibility map together can provide requisite information to make a decision on issuing hazard warnings for a resident evacuation.