Integrating Multi-source Remote Sensing Data For Forest Mapping And Comparing With Existing Forest Products In Loess Plateau

As the main of terrestrial ecosystems,forests play an important role in biodiversity conservation,global ecosystem services,climate regulation and water conservation.Loess Plateau is one of the most serious soil and water loss region and the most fragile ecological environment region in the world.In order to improve the ecological environment,China has implemented a set of forest restoration projects since 1999,such as the Grain for Green(GFG)Project and the Three North Shelterbelt Project.Identifying the areas and distribution of forest accurately is an important prerequisite for evaluating the effectiveness of project implementation scientifically and achieving sustainable management of regional forest resources.In light of requirements revealed above,the objectives of this study are three-fold.First,we combined 25m PALSAR and30m Landsat images to generate a new 30m forest map(PL-based map)by a decision tree method.Second,we compared the PL-based map with seven existing forest maps(JAXA?GlobeLand30?FROM-GLC?Hansen?ChinaCover?NLCD and GLCF VCF)to assess their overall accuracy,user accuracy,product accuracy and kappa coefficient.Furthermore,we analyzed the spatial consistency between PL-based and the seven existing forest maps in the county scale and the pixel scale and then their dependence on the terrain were discussed.The results showed:(1)Compared with existing forest maps,the PL-based map which integrated the spectral information from Landsat data and the structure information from PALSAR images had a highest overall accuracy(0.97)and the kappa coefficient(0.88).(2)At the county scale,PL-based forest maps had a high linear correlation with these existing forest maps(R~2=0.68 to 0.96),of which the JAXA forest map based on radar data and PL-based map had a highest linear correlation(R~2=0.96,y=-0.29+0.81x).(3)Through a zoom-in comparison of PL-based and existing forest maps,we found that the PL-based,Globe Land30,and ChinaCover forest maps showed more details in Loess Plateau;JAXA forest map underestimated the area of the forest to some degree;NLCD forest map missed some details;FROM-GLC forest map overestimated the forest area due to large commission errors.(4)The spatial consistency index(SPI)of the eight forest maps showed certain rules with the terrain.Specifically,the SPI of the forest maps increased along with the increasing elevation until 2000 m elevation and then decreased with higher elevation,while the forest maps showed higher SPI as slope increased(slope<25°).This study showed that the integration of optical and radar data can more accurately estimate the area of forest,which will provide implications for forest resources investigation in the future.