Analysis Of Multiple Forest Cover Products And Their Fusion To Derive Forest Cover Map

Forest plays an important role in ecosystems service and global carbon cycling.It is of great significance to master the present status of forest cover for the forest resources monitoring and ecological environment protection.This paper aims at forest cover in Central Asia and conducts accuracy comparison and agreement analysis based on multi-source fusion validation dataset and geographic dataset for five representative moderate-precision global forest cover products FROM-GLC,GLCF VCF,GlobeLand30,PALSAR F/NF,and TreeCover2010.According to the result of agreement analysis,this paper proposes a multi-classifier system of forest cover extraction based on fusion strategy and machine learning,which is further applied to producing high accuracy forest map in Central Asia.The main content of this paper are as follows:(1)This paper processes a large volume of dataset to meet the demand of research in this paper.This paper studies multi-source data acquisition and batch processing methods based on Google Earth Engine and other platforms and completes massive data preprocessing for Landsat images,land cover products for forest cover extraction,MODIS surface reflection temperature images,etc.(2)This paper analyzes the consistency of the forest cover products from the perspectives of accuracy,forest area and forest identification.Firstly,this paper generates a multi-source validation dataset and assesses the accuracy of these forest cover products using Google Earth image and the two published validation sample datasets.Secondly,this paper conducts a correlation analysis and per-pixel analysis to explore the consistency of forest cover products in the aspects of forest area and forest cover identification.Finally,this paper explains the relationships of products' consistency and geographic factors.(3)This paper makes a deep research in producing high-precision forest map.This paper proposes a multi-classifier system based on data fusion,feature fusion,decision fusion for forest cover extraction using TreeCover2010 and GlobeLand30.Specifically,at the stage of data fusion,this paper does spatial overlapping analysis on TreeCover2010 and GlobeLand30 to extract agreement area of forest and non-forest,and conflict area.Thereafter,the agreement area is applied to model training while the conflict area to model estimation;at the stage of feature fusion,this paper does texture analysis and principal component analysis to extract and combines different kinds of features,namely,spectral index,texture index,vegetation index,elevation and temperature value;at the stage of decision fusion,this paper builds a multi-classifier system for forest extraction based on machine learning and on the basis of which this paper produces high accuracy forest map for Central Asia.The research on the consistency of forest cover products provides a valuable data reference for the current situation of forest cover in Central Asia.Besides,the research on forest cover extraction based on multi-level fusion strategy verifies the feasibility and effectiveness of the applications of fusion strategy to forest cover extraction,which can provide reference for studies of the other types of land coverage extraction.