Accuracy Assessment And Refinement Of Land Cover Information Based On Class Spatial Heterogeneity Stratified Sampling

Land cover is a spatial description of materials with various attributes and characteristics on the earth’s surface,and the basic data of land use,climate change and sustainable development of ecological environment.Scientific and accurate measurement of spatial distribution and dynamic changes of Land cover is of great significance to the study of energy balance,carbon cycle,other biogeochemical cycles,climate change,ecological environment protection and geographic situation monitoring.The development of remote sensing technology provides data support for the acquisition of land cover information.Thematic information extraction based on remote sensing images has become the main approaches to quickly obtain regional and even large-scale category data.However,due to the influence of various systems and random factors in the process of information extraction,there are inevitably different degrees of inconsistency between the thematic data and the actual situation on the land surface.So,accuracy assessment is an important dimension in use and production of land cover information,which not only helps users understand the quality of the data,but also helps producers evaluate the performance of the information extraction method,thus providing reference information for further optimization.In addition,less accurate thematic data will limit its practicability,and the need for high quality data in various fields is growing.Therefore,effective refinement of geographic category data to obtain higher quality data is also the focus of this research.In this paper,the accuracy assessment of land cover category data is studied in depth,and the spatial characteristics of the category data are fully considered during sampling design.The accuracy of “macro-integrated” and “micro-local” is evaluated separately.At the same time,the refinement of the category data is studied.The main research contents and innovations are as follows:(1)The uncertainty of the category data has an unbalanced feature in space,which is closely related to the spatial complexity of the geographic category.In general,spatial heterogeneity regions have lower precision,while homogeneous regions have higher precision.Based on this law,this paper expands the stratified sampling method,proposes two-level stratified sampling based on category heterogeneity,and arranges relatively more sample points in heterogeneous regions.The sampling method proposed in this paper takes into account the spatial distribution characteristics and makes the accuracy assessment more reliable.(2)In the class-heterogeneity stratified sampling design,the explanatory variables and covariates which are significantly related to the accuracy of the surface covering information are preferred,and the local area accuracy prediction based on logistic regression hierarchical modeling is explored.In this paper,based on four sets of sample data with different sampling designs,the corresponding logistic regression models are constructed respectively,and local precision spatial predictions are also performed.The experimental results show that the logistic regression hierarchical model based on category heterogeneity stratified sampling is better than other methods,and the corresponding local data accuracy prediction of category data is more reliable.(3)Aiming at the limited quality of land cover category data,a refinement method is proposed.Based on the spatial heterogeneity sample data with appropriate spatial distribution,a classification inversion model is established by using logistic regression and Kriging method to refine and improve land cover data quality.In this paper,the data refinement based on category heterogeneity is compared with the data refinement based on category sampling.The results show that the data category inversion model based on category heterogeneity has better refinement effect on the data,especially In the heterogeneous area,the improvement effect is more significant.