Experimental Study On Estimation Of Grassland Coverage By The Pixel Dichotomy Model Using Remotely Sensed Data

The Fractional Vegetation Coverage（FVC）is a very important basic data to express the natural ecosystem,and it is also an important parameter index to study soil erosion,and other different researches in the world.Inner Mongolia Autonomous Region is an important ecological barrier in our country.It is of great significance to study the change of vegetation coverage.In this study,the remote sensing estimation of grassland coverage in Zhagastai Town,Arhorchin Banner,Inner Mongolia Autonomous Region was taken as the main scientific objective,the field survey was taken as the verification data,the apparent reflectance data and surface reflectance data of Landsat 8 OLI remote sensing images on September 17,2018 were taken as the data sources,and the precision difference of grassland coverage estimation results in the study area under different models and different parameter settings were mainly studied,then the following conclusions were obtained.First,Whether the apparent reflectance data was used or surface reflectance data was used,the dichotomy model by NDVI based linear mixture analysis was better than the dichotomy model by general linear mixture analysis in estimating grassland coverage.The optimal estimation accuracies of grassland coverage based on dichotomy model by NDVI based linear mixture analysis were RMSE=0.16,ARE=0.24 and RMSE=0.17,ARE=0.25 under the data of apparent reflectance and surface reflectance respectively,while the corresponding accuracies of dichotomy model by general linear mixture analysis were RMSE=0.33,ARE=0.73and RMSE=0.30,ARE=0.58 respectively.Second,The estimation accuracies of dichotomy model by NDVI based linear mixture analysis,when model’s parameters NDVI_veg and NDVI_soil were assigned by NDVI values which were corresponding to cumulative values of DN value frequency for 1%and 99%,were better than the estimation accuracies of dichotomy model by NDVI based linear mixture analysis,when model’s parameters NDVI_veg and NDVI_soil were assigned by NDVI values which were corresponding to maximum and minimum values,cumulative values of DN value frequency for 2%and98%,and 5%and 95%respectively.In addition,when determining the model’s parameters NDVI_veg and NDVI_soiloil through linear regression analysis by measured data,the estimation accuracies were almost same as above optimal results,even slightly better than which in RMSE.Finally,when applying the dichotomy model by NDVI based linear mixture analysis to estimate grassland coverage,accuracy of the optimal result derived from apparent reflectance data（RMSE=0.16,ARE=0.24）was slightly better than accuracy of the optimal result derived from surface reflectance data（RMSE=0.17,ARE=0.25）.Therefore,in the future,when rapid dynamic monitoring of grassland coverage is carried out,apparent reflectance data could be directly applied for instead of surface reflectance data,so as to omit atmospheric correction pretreatment.