Accuracy Assessment Of Global 30 M Land Cover Products

Land cover data is a fundamentally important and indispensable data for research on climate change,ecological environment modeling,land surface process simulation,and national conditions monitoring.In recent years,with the improvement of remote sensing technology and computer storage and computing capabilities,breakthroughs have been made in global land cover mapping,which is gradually transitioning from low-medium resolution to mudium-high resolution of 30 m.However,considering the complexity of the earth system itself and the differences between the mapping strategies of various products,there is still a big uncertainty in how users choose the most suitable product from the multisource global 30 m land cover products.Therefore,this study focuses on the consistency analysis and quantitative accuracy assessment of global 30 m land cover products,and analyzes the accuracy of each product at global/regional scales,and then provides scientific knowledge and accurate data support for land cover related users.Taking the global 30 m land cover products(including all-element and thematic elements)currently available internationally as the research object,this study carried out several specific studies,including the unification of the different classification system of all-element land cover products,the consistency analysis of the all-element and thematic element products,and the quantitative assesement of the accuracy based on the regional/global validation sample data sets.The main results of this study are as follows:(1)Taking into account the impact of different classification systems on the comparative analysis of different products,when evaluating and comparing the accuracy of land cover products,it is necessary to minimize the impact of classification differences on the results.To solve this problem,the EAGLE(Environmental information and observation network Action Group on Land monitoring in Europe)concept was adapted to calculate the semantic similarity between the classification systems,and a strict classification system conversion relationship based on the semantic similarity was constructed.(2)The global 30 m land cover products have significant spatial differences due to differences in mapping strategies and data sources.For all-element datasets,the results showed that the number of spatially consistent pixels of the three GLC products accounted for 35.40%of the total number of pixels,and the percentage of completely inconsistent pixels was 29.14%on a global scale.For the consistency analysis of different datasets of each thematic element(including the impervious surface,forest,cropland and water land cover types),the results showed that among the four thematic elements,the consistency of the water products had higher consistency than other elements.while the consistency of the products of cropland were the lowest,and the product pair with the highest spatial consistency has an R2 of only 0.67.(3)Quantitative accuracy assessment based on regional/global validation data sets provides users with scientific congnition and quantitative data indicators when selecting appropriate data products.In order to minimize the uncertainty of the quantitative assessment of accuracy,and fully understand the accuracy difference under different user needs,the accuracy assessment indexes were calculated based on the traditional confusion matrix,the adjusted confusion matrix and the confusion matrix weighted for specific user needs,respectively.The results showed that:Firstly,the GLC_FCS30-2015 product has the highest overall accuracy in the global and US regions.Meanwhile,the accuracy index value weighted for the different needs of users was higher than the traditional accuracy index based on the confusion matrix.And for different user needs,the accuracy of the products showed obvious differences.The main innovations of this study contain:(1)A rigorous classification system conversion rule based on the semantic similarity measurement was constructed to solve the problem of incompatibility of the different classification system of global land cover products.This rule can reduce the errors introduced by humans in the tranditional classification system conversion,and reflect the consistency and differences between land cover priducts more accurately.(2)An adjusted confusion matrix was proposed by introducing the area proportion of each land cover type in the product to reduce the uncertainty of quantitative accuracy assessment,and a weighted accuracy assessment strategy is proposed to analyze the difference in accuracy under different user needs.