Integrate Global Land Cover Products To Refine The Forest Type Of GlobeLand30

With the development of remote sensing technology,a variety of spatial resolution land cover products have emerged,but these individual land cover products have different classification systems and have lower spatial resolution.It cannot meet the need to develop a new generation of global land cover products with higher spatial resolution,higher precision,and adaptability to globalization.In order to adapt to the needs of globalization and solve the problem of domestic over-reliance on foreign land cover products,China developed the world's first 30 m global land cover product,GlobeLand30,which has two phases,2000 and2010.The product has ten first class,namely,croplands,grasslands,forest,wetland,shrublands,water bodies,impervious area,tundra,bare land,permanent snow/ice.However,its secondary classification products are in the process of development and have not been released,while the secondary class has an important role in certain specific studies.Therefore,there is a need to use a method or means to perform a second-level division of the first class of land cover products(such as GlobeLand30).The traditional methods of land cover product classification include visual interpretation and classification methods,computer automatic classification,and new methods for computer classification of remote sensing images.However,most of these methods are based on remote sensing images for classification.When the classification is performed,the workload is large and time-consuming,and the requirements of multiple users cannot be satisfied.This paper adopts a method based on the integration of fuzzy sets theory and uses the integration model to explore the methods and processes of the second refinement of GlobeLand30(2010)'s forest type.The integration model mainly considers three points: the overlap between the source datas and the target legend;the conditional probability of the source datas forest type belonging to the target legend type;the respective precision of the source data products.In the integration process of this paper,the forest type of GlobeLand30(2010)was taken as the goal,and the US region as the research object,combined with NLCD2011 land cover product,FROM-GLC-seg product,and global tree cover data(treecover2010).First,the four land cover products were preprocessed and the target legends were defined.Then,using most of the principles,identified and marked possible omission pixels and commission pixels in forest type of the GlobeLand 30(2010),and supplemented the commission pixels into the GlobeLand 30(2010)'s forest type.Then,comparing the advantages and disadvantages of the LCCS land cover classification system and the EAGLE matrix,the EAGLE matrix was used to semantically translate the forest type of each land cover product(treecover2010's forest was not translated),and the bar-coding method was used for value assignment.Then,according to different EAGLE matrix elements,the overlap matrix of forest class and target legend in NLCD 2011 and FROM-GLC-seg were calculated using the corresponding calculation rules.Finally,according to the final discriminant function,each pixel's attribution was discriminated,and second class refinement diagram of GlobeLand30(2010)'s forest was derived.In the second class refinement map ofGlobeLand30(2010)'s forest,the user accuracy of broad-leaved forest,coniferous forest,and mixed forest reached about 80%,62%,and 52%,respectively,and the overall classification accuracy was about 68%.