Spatial prediction of earthquake-induced landslide probability
Spatial prediction of earthquake-induced landslide probability
复制标题
DOI：
10.5194/nhess-2017-193
复制
发表时间：
2017-06
影响因子：
4.6
通讯作者：
R. Parker;N. Rosser;T. Hales
中科院分区：
文献类型：
--
作者：
R. Parker;N. Rosser;T. Hales
抽象。我们开发了一个广义模型来描述和预测地震诱发的滑坡的空间分布，基于9个同震滑坡库存从不同的地震和地区的回归分析。我们的模型表示作为峰值地面加速度和山坡坡度的函数，从全球地形和地震地面运动数据集的数据的基础上，滑坡的绝对空间概率。从我们的模型的输出预测滑坡触发的沉积，沉积变质，火成岩和火山岩性的概率，并适用于浅层大陆地震的矩震级范围为6.2至7.9，和深度之间的10和21公里。为了获得绝对概率预测，我们只使用滑坡源区作为输入数据，并明确估计和纠正已知的不完整性输入数据集，通过一种新的蒙特卡罗方法。我们估计这些预测的不确定性，通过广泛的测试模型的性能，当所有9个地震的样本外预测。我们的模型比其他开发来预测滑坡空间概率的模型要简单得多，因为我们只包括了在全球范围内可以一致约束的变量，并消除了那些在我们数据集中所有地震中不影响滑坡概率的变量。模型输出还提供了一个基线，以进一步调查空间和时间来源不明的变化，在同震滑坡分布。使用免费提供的地形和地面运动数据，我们建议我们的模型可以应用更广泛，提供滑坡预测地震没有滑坡数据。
Abstract. We developed a generalized model to describe and predict the spatial distribution of earthquake-induced landslides, based on a regression analysis of 9 co-seismic landslide inventories from different earthquakes and regions. Our model expresses the absolute spatial probability of landslides as a function of peak ground acceleration and hillslope gradient, based on data from global topographic and seismic ground motion datasets. The output from our model predicts probabilities for landslides triggered in sedimentary, meta-sedimentary, igneous and volcanic lithology, and is applicable to shallow continental earthquakes of moment magnitude range 6.2 to 7.9, and depths between 10 and 21 km. To obtain absolute probability predictions, we use only landslide source areas as input data, and explicitly estimate and correct for known incompleteness in input datasets, through a novel Monte Carlo approach. We estimate the uncertainty of these predictions, through extensive testing of the performance of the model, when making out-of-sample predictions for all 9 earthquakes. Our model is notably simpler than others developed to predict spatial probability of landsliding, as we have only included variables that could be constrained consistently at the global-scale, and eliminated those that did not influence landslide probability in a consistent manner across all earthquakes in our dataset. The model outputs also provide a baseline to further investigate spatial and temporal sources of unexplained variability in co-seismic landslide distributions. Using freely available topographic and ground motion data, we suggest that our model can be applied more widely, to provide landslide predictions for earthquakes with no landslide data.