基于不同核函数的2010年玉树地震滑坡空间预测模型研究

基于统计学习理论与地理信息系统(GIS)技术的地震滑坡灾害空间预测是一个重要的研究方向,其可以对相似地震条件下地震滑坡的发生区域进行预测.2010年4月14日07时49分(北京时间),青海省玉树县发生了Mw6.9级大地震,作者基于高分辨率遥感影像解译与现场调查验证的方法,圈定了2036处本次地震诱发滑坡,这些滑坡大概分布在一个面积为1455.3 km2的矩形区域内.本文以该矩形区域为研究区,以GIS与支持向量机(SVM)模型为基础,开展基于不同核函数的地震滑坡空间预测模型研究.应用GIS技术建立玉树地震滑坡灾害及相关滑坡影响因子空间数据库,选择高程、坡度、坡向、斜坡曲率、坡位、水系、地层岩性、断裂、公路、归一化植被指数(NDVI)、同震地表破裂、地震动峰值加速度(PGA)共12个因子作为地震滑坡预测因子.以SVM模型为基础,基于线性核函数、多项式核函数、径向基核函数、S形核函数等4类核函数开展地震滑坡空间预测研究,分别建立了玉树地震滑坡危险性指数图、危险性分级图、预测结果图.4类核函数对应的模型正确率分别为79.87%,83.45%,84.16%,64.62%.基于不同的训练样本开展模型训练与讨论工作,表明径向基核函数是最适用于该地区的地震滑坡空间预测模型.本文为地震滑坡空间预测模型中核函数的科学选择提供了依据,也为地震区的滑坡防灾减灾工作提供了参考.

The 2010 Yushu earthquake triggered landslides spatial prediction models based on several kernel function types

Earthquake triggered landslides spatial prediction based on statistics learning theory and Geological Information Systems (GIS) technology is an important research direction, it can predict the area of earthquake triggered landslide occurrence during an earthquake of a similar magnitude in the future. On April 14, 2010 at 07:49 (Beijing time), a catastrophic earthquake with Mw6.9 struck Yushu County, Qinghai Province, China. 2036 landslides were interpreted from aerial photographs and satellite images, verified by selected field checking, these landslides were distributed in a rectangle area of 1455.3 km2. The aim of this study was to carry out earthquake triggered landslides spatial prediction modelling in the rectangle area using several kernel function types, based on GIS and support vector machine (SVM) model. A spatial database, including landslides and associated controlling parameters which may have influence on the occurrence of landslides, was developed and analyzed using GIS technology. The twelve factors, including elevation, slope angle, slope aspect, slope curvature, slope position, drainages, lithology, faults, roads, normalized difference vegetation index (NDVI), co-seismic main surface fault-ruptures, and peak ground acceleration (PGA) were selected as earthquake triggered landslide spatial prediction factors. The landslide spatial prediction mapping was executed using several kernel function types including linear function, polynomial function, radial basis function, and sigmoid function based on SVM model, and Yushu earthquake triggered landslide hazard index maps, landslide hazard rank maps, and prediction result maps were created respectively. The success rate of the four kernel function types were 79.87%, 83.45%, 84.16%, 64.62% respectively. This showed that radial basis function was the best kernel function of the four in Yushu earthquake triggered landslides spatial prediction mapping. The training and discussions of the model were also carried out based on other three different training sample groups of random distribution and one training sample group of regular distribution. This paper provided an example for selecting appropriate kernel function types for earthquake triggered landslide prediction mapping using support vector machine modelling in the study area. Hazard rank maps and prediction result maps could be useful in landslides hazard mitigation by helping planners understand the probability of landslides in different regions.