| Under the influence of global climate change and the intensity of human land use patterns,shrub encroachment into grassland has become one of the most prominent problems of grassland ecosystem in arid and semi-arid areas.It is of great significance for grassland sustainable management and climate change adaption to accurately grasp the large-scale shrub encroachment information.Ground investigation and high-resolution remote sensing images have been widely used in the investigation of shrub encroachment in local areas.However,due to the high cost and limited data acquisition ability,such methods can not be extended to large-scale shrub encroachment monitoring.Medium resolution optical image(10-30m)has the advantages of large coverage,short revisit period and abundant spectral resolution.However,it is difficult to distinguish shrub from herbaceous because of their similar spectra in the whole growing season.Polarimetric Synthetic Aperture Radar(SAR)has shown a certain potential in the estimation of surface roughness and shrub biomass.Theoretically,the combination of polarimetric SAR and multispectral optical images has the possibility to carry out shrub encroachment monitoring.However,there is a serious lack of research in this area.In order to solve the above problems,this paper selects Xilin Gol Grassland as the research area,takes the time series images of sentinel-1 and sentinel-2 of ESA and the field observation data of shrub encroachment on the ground as data sources and carry out the pilot study of shrub encroachment.Firstly,the time-series optical band reflectance,vegetation index and backscattering intensity information of different radar polarizations were extracted for each field sample,the correlations between them were analyzed to discover the sensitive optical or SAR characteristics for shrub encroachment.Based on which,multiple linear regression and random forest models were utilized respectively for predicting the shrub coverage in the grassland,the accuracies was compared and the importance of sentinel-1 backscattering coefficient and sentinel-2 vegetation index in shrub coverage estimation of Xilinguole grassland was analyzed.Finally,the grassland shrub coverage mapping in Xilinguole League was completed,its spatial distribution characteristic were summarized through combine with administrative boundary and precipitation patterns.The results show that:(1)The band reflectance and optical vegetation indices extracted from Sentinel-2in July and Augest,were significantly correlated with the total vegetation coverage,but were not sensitive to the shrub coverage.Only the canopy water content and red edge information in March and June has little potential to differentiate shrub from grassland.(2)The radar backscattering coefficient extracted from sentinel-1 polarimetric SAR data has strong sensitivity to grassland shrub coverage(VH polarization is better than VV polarization),and the high correlation runs through the whole growing period.The correlation between VH polarization data and shrub coverage in June can reach0.604.(3)The models combined sentinel-1 and sentinel-2 data jointly can achieve high precision estimation of grassland shrub coverage.The accuracy of random forest model(R~2=0.89,RMSE=0.05)is obviously higher than the multiple linear regression model(R~2=0.69,RMSE=0.11).From the random forest model,we can find that the total contribution of radar data Sentinel-1 backscattering coefficient is 71.54%,and that of vegetation index derived from Sentinel-2 is 28.47%.Terrain factor had certain influence on the estimation of grassland shrub coverage when SAR factors were utilized with radar factor,which will lead to overestimation of mountain area,which is better in the flat area.(5)The random forest-based shrub coverage estimation results show that shrub coverage in grassland is mainly under 10%,account for 61.68%.This means the shrub encroachment is mild in Xilin Gol.The high shrub coverage mainly distributes in south-east. |
PDF Full Text Request