Research On Sample Size Estimation For Accuracy Assessment Of Land Cover Products

With the rapid development of satellite remote sensing and image processing technology,there are more and more kinds of land cover data products developed by international organizations and countries,and the spatial resolution is also higher and higher,which has attracted wide attention in the fields of global environmental simulation,resource management and sustainable development.Therefore,the accuracy information of all kinds of land cover products has become one of the most concerned problems of potential users in various industries.As a key step of accuracy evaluation of land cover data products,sample size estimation has an important impact on the reliability of accuracy evaluation results and the reasonable allocation of human,material and financial resources.At present,the existing methods of sample size estimation have some problems,such as strong subjective factors,less consideration of surface spatial heterogeneity information.Therefore,this paper takes the spatial heterogeneity of the surface as an important reference factor for sample size estimation,and expects to obtain the real accuracy evaluation results of the research area with fewer samples,so as to reduce the difficulties of sample point authenticity test and unreasonable allocation of resources in the process of accuracy evaluation of land cover data products.In this paper,Jiangxi Province,Anhui Province and Zhejiang Province are taken as the experimental areas,the global land cover remote sensing mapping data product(globeland30)in 2010 is taken as the test data,and the 1:100000 land use/cover data(LUCC)in China is taken as the reference data,and a sample size estimation model considering spatial heterogeneity is designed.The sample size estimation model is a multivariate linear model composed of prediction variables.Based on the probability statistical theory model,the watershed unit with ecological geographical significance is introduced as the evaluation unit,and the spatial heterogeneity information of data products is considered as the influence factor of sample size.The cross validation and parameter importance are used to test the model,and the example validation and analysis are also carried out.The main conclusions are as follows:(1)Five landscape indexes with high contribution rate of spatial heterogeneity were selected by full subset regression.The results show that the sample constraint C designed in this paper has the greatest impact on the sample size ns,followed by the landscape index PD,ED,LSI and so on;(2)The R~2 coefficient of the model is 0.82,which indicates that the joint interpretation ability of all prediction variables in the model to the sample size N_S is 82%,which also means that the predictability of the model is at a good level;(3)Taking Zhejiang Province as the verification area,the sample size estimation model designed in this paper is analyzed and compared with stratified sampling probability statistical method.The results show that,in terms of the number of samples,the sample size estimated by the model designed in this paper is half or even less than that calculated by the statistical model,but the verification accuracy of the two models is basically the same.All in all,the sample size estimation model designed in this paper can reduce the sample size of remote sensing product accuracy verification to a large extent on the premise of ensuring the reliability of accuracy verification results,so as to achieve the purpose of optimizing the estimation model.