房屋倒损评估中统计抽样方法的对比分析

以2014年云南鲁甸地震为例,结合遥感技术对比分析简单随机抽样、分层随机抽样、整群随机抽样在地震房屋倒损抽样调查中的适用性.结果表明:1)简单随机抽样精度最高且稳定性最好,总体精度达98.96%,各类型房屋倒损精度超过97.00%.但过于分散的样本点会使得调查成本非常高,特别是在道路不可通达时,需要考虑无回答样本的误差问题.2)以房屋结构或地震烈度为标识的分层随机抽样精度均较低,这可能是由于影响房屋损毁程度的因子较多,单一分层指标不能较好地反映目标的总体情况,难以体现分层优势.3)整群随机抽样是一种较为理想的抽样方式,其精度仅次于简单随机抽样,总体精度达93.31%,各类型房屋倒损精度均超过85.00%,而且该法有助于减少初级抽样单元数量,方便调查、节省成本.4)通过相关分析与回归分析发现,简单随机抽样、整群随机抽样的变异系数与房屋入样总量间均具有负相关关系,因此可根据实际精度要求,通过提高抽样比的方式降低变异系数来减小估计量方差. 

Comparative analysis of statistical sampling methods in evaluating building damages

The present work investigated the Yunnan Ludian earthquake in 2014.Simple random sampling,stratified random sampling and cluster random statistical sampling combined with remote sensing techniques were used to assess building damages.The accuracy and stability of simple random sampling was found to be the highest among the three sampling methods.The overall accuracy of simple random sampling was up to 98.96％,the estimation accuracy of damage degree for different types of building was found to beabove 97％.The costs of investigation were rather high because the sampling points were excessively scattered.We should pay attention to the errors caused by no answer samples due to unreachable roads.The accuracy of stratified random sampling was rather low.A variety of factors could influenced the degree of building damages.A single indicator cannot reflect the overall characteristics of the target nor highlight the advantage of stratification.The cluster random sampling was found to be reasonable.The overall accuracy of cluster random sampling was above 93.31％ and the estimation accuracy of damage degree for different types of building was above 85％.More importantly,this method made the investigation much easier and cost-efficient by reducing the number of sampling units.Correlation and regression analysis indicated that the variation coefficient of both simple random sampling and cluster random sampling showed negative correlation with the total number of sampled buildings.According to actual requirements,we can increase the ratio of sampling to decrease variation coefficient and reduce the variation of estimation.