Estimation Of Grassland Height Based On Remote Sensing In Alpine Grassland,Gannan Region

Grassland height is an important indicator for estimating grassland degradation,and it is also an indispensable parameter for accurate estimation of grassland aboveground biomass and early warning of snow disaster in pastoral areas.The alpine grassland ecosystem in the Tibet Plateau is not only material foundation for the development of the national economy in pastoral areas,but also a sensor for global climate change.Therefore,it is of great scientific significance and practical value to dynamically monitor the grass height of alpine grasslands on the Qinghai-Tibet Plateau.To date,grassland height in natural grassland has been predicted based on statistical models in some researches.However,the inversion accuracy of grassland height is quite different due to the complexity types of natural grassland and poor reproducibility of statistical models.There is still great difficulty in accurately predicting grassland height,which is caused by the different spatial distribution of alpine grassland in Tibet Plateau.Thus,this study was constructed for modelling grassland height based on MODIS,eco-environmental factors and hyperspectral variables using RF,SVM,ANN,XGBoost and Cubist algorithms in Gannan region located in the eastern Tibet Plateau,and the spatial and temporal dynamics of grassland height in Gannan region from 2006 to 2018 were analyzed.The results show the following conditions:1)Of the 6 vegetation indices constructed by MODIS data,NDVI and OSAVI are more sensitive to grassland height,but both can only reflect 15%of the change in grassland height.The grassland height has a high negative correlation with the reflectivity of the red band(B1)and mid-infrared band(B7)of MODIS,and the correlation coefficients are-0.33 and-0.36,respectively.2)The correlation between grassland height and factors of topography and soil is weak,while a significant positive correlation has been found between grassland height and annual cumulative precipitation(JS),monthly cumulative precipitation(P),and monthly average temperature(T)among meteorological factors.There is a lag of three months in the response of grassland height to the monthly precipitation,and there is no lag in the response to temperature.3)The multi-factor RF model based on the aspect(A),soil clay content(C1,C2),NDVI,B7,LAI,annual accumulated temperature(JW),and monthly average temperature(T)has good prediction performance(R~2=0.45,RMSE=5.47 cm,BIC=442),which is suitable for predicting the spatial distribution of grassland height in alpine grassland in Gannan region.4)Ground hyperspectral data can better reflect the change in grassland height.The RF model constructed by the 11 characteristic bands of the first-order differential spectrum(1st-R)has high inversion accuracy(R~2=0.74,RMSE=5.24 cm,BIC=389),but currently cannot be applied to spatial inversion of grassland height.5)The mean grassland height in the Gannan region generally showed an increasing trend from 2006 to 2018.The grassland height in most regions was between 15 and 20cm.The proportion of grassland with stable and increasing grassland height were 58.82%and 29.90%of entire grassland area,respectively.The grassland with decreasing grassland height accounted for only 11.28% of entire grassland area.