一种基于分形的地表覆盖产品质量评估空间抽样方法

A fractal dimension based spatial autocorrelation stratified sampling plan for quality assessment of land cover products

针对地表覆盖产品质量抽样评估中易受空间自相关影响导致的样本代表性不足的问题,该文提出一种基于分形的地表覆盖产品质量评估空间抽样方法,通过引入辅助变量分形维数作为分层变量,代替研究变量(质量评估值),基于地表覆盖产品中每个图斑分形维数的空间自相关性进行第一阶段分层,在顾及空间自相关性影响前提下,构建空间敏感区和钝感区内的分层指标进行第二阶段分层,并在二阶段各层内随机抽样。选择全球地理信息资源建设项目2021年哈萨克斯坦中部区域的10m地表覆盖产品作为实验验证数据,分别与基于面积空间自相关、基于景观指数空间自相关的分层抽样方法进行对比实验,结果表明该文方法相对“真值”的均方根差较上述两种方法最大可分别减少51.14%、56.9%,能够提供更精确和更稳定的质量评估结果。

Aiming at the problem of the insufficient sample representativeness caused by the spatial autocorrelation in the quality assessment for land cover products,a fractal dimension based spatial autocorrelation stratified sampling plan(FSASP)was proposed.The quality of land cover products was assessed by introducing an auxiliary variable fractal dimension instead of the study variable(quality assessment value).The implementation of the proposed method involved two stages of stratification.Firstly,the spatial autocorrelation of the fractal dimension of each patch in the land cover product was analyzed,and stratification was performed accordingly.Secondly,stratification indicators were constructed for spatially sensitive and blunt areas,considering the influence of spatial autocorrelation.Random sampling was then conducted within each stratum.The 10 m land cover data products in for the central Kazakhstan in 202l of the global geographic information resources construction project were selected as experimental data and compared with the area based and the landscape index based spatial autocorrelation sampling plan,respectively,using the proposed FSASP.The results showed that the root mean square error of the proposed FSASP could be reduced by up to 51.14%and 56.9%,respectively,compared with the"true value"of the above two methods,which could provide more accurate and stable quality assessment values.