Accuracy Of The FROM-GLC30 2017 Land Use Dataset And Its Influence On Regional Soil Erosion Assessment

Land use data plays a key role in the research of global surface change,and is widely used in various fields,such as soil erosion and other surface process simulation.The development of remote sensing technology has promoted the continuous optimization of land use data,the increasing number of available land use data sources has increased.At the same time,the accuracy of land use data has also received more extensive attention,especially in the study of soil erosion.Researches on data accuracy and its impact on erosion simulation are urgently needed.The FROM-GLC30 2017 data(Finer Resolution Observation and Monitoring of Global Land Cover)is currently the most widely used land use data with high resolution in the world.This paper focused on the accuracy of the data and its impact on the assessment of soil erosion.20936 sampling units were set up in the Pan-Third Pole Area where China and 64 neighboring countries were included.Through the comprehensive use of field survey and artificial visual interpretation of high-resolution remote sensing images,the reference land use data for accuracy assessment was obtained.The accuracy of the 30-meter global land use data FROM-GLC30 2017 was evaluated under different zoning and different land use types within two different scale research areas of the intercontinental and the national.At the same time,this paper further discussed the impact of land use differences on soil erosion simulation results in the Loess Plateau based on the FROM-GLC30 2017 data and the reference land use data.This work aims to provide a theoretical reference for the accuracy assessment of public land use data and its applicability in the field of soil erosion.The main study results showed:(1)The overall accuracy of the FROM-GLC30 2017 data in the Pan-Third Pole study area was 72.78%.The sampling units with an overall accuracy better than 80% account for54.35%.Among the eight regions,West Asia-Northeast Africa had the highest overall accuracy,at 80.11%,followed by Southeast Asia and Central and Eastern Europe.The overall accuracy of the FROM-GLC30 2017 data in China was 74.38%.Among all the sampling units,56.62% of them had an overall accuracy better than 80%.In seven regions,the overall accuracy of east China and northeast China were the highest,with 80.59% and80.29%,followed by central China.The accuracy values of bareland and forest of the FROM-GLC30 2017 data were higher than 81% in Pan-Third Pole Area and China.There were differences in the accuracy of the same land use type in different regions.Among them,the accuracy of bareland had the largest difference.(2)There was a negative correlation between the fragmentation degree of the sampling unit and the overall accuracy value,and the correlation degree between them was different in different research regions.Among the seven regions of China,the negative correlation between the fragmentation degree and the overall accuracy was relatively strong in the seven regions of China.The land use types with special color and texture characteristics in remote sensing images were easier to be accurately identified and their accuracy were relatively high,such as bareland,forest,cropland and water area.These kinds of land use types could significantly improve the regional accuracy of the land use types when they were distributed in large scale.Sampling unit methods had a greater impact on the accuracy.The overall accuracy results of the FROM-GLC30 2017 data based on small watershed units and grid unit had little difference in numerical values,but the method based on small watershed units could better express the spatial pattern of the overall accuracy.The scale effect had an impact on the accuracy assessment results of the sampling unit,and the mean value of the difference in the sampling unit accuracy caused by scale effect was 11.49%.(3)Compared with the reference land use data,the accuracy of the FROM-GLC30 2017 data in the Loess Plateau research area was low,with an overall accuracy of 58.74%.This land use difference had a greater impact on the results of factor B in the CSLE model(Chinese Soil Loss Equation).The most influential areas were mostly concentrated in the area near Wuzhong,Guyuan and Dingxi.The overall trend of regional soil erosion distribution based on the two kinds of land use data was similar,and there was no significant difference in the proportion of soil erosion area,but there were significant difference in the mean values of soil erosion modulus,which also had a certain impact on the distribution judgment of soil erosion key areas.The FROM-GLC30 2017 data still need to be further optimized in the application of soil erosion in the Loess Plateau.