Remote Sensing Inversion Model Of Grassland Biomass In The Tibet Autonomous Region

Aboveground biomass(AGB)is a key indicator of grassland ecosystem function and quality,and accurate estimation of large-scale aboveground biomass plays a crucial role in grassland ecosystem management.The traditional manual investigation of aboveground biomass in grasslands is time-consuming and laborious,and has a significant impact on the ecosystem.Based on remote sensing data sources and ground sample data,combining remote sensing inversion models to estimate grassland aboveground biomass can compensate for the shortcomings of traditional survey methods.The key factors for remote sensing inversion of grassland aboveground biomass include feature variables,variable selection methods,and inversion models.Choosing suitable feature variable combinations and screening methods to improve the prediction accuracy of the inversion model is of great significance for remote sensing inversion of grassland aboveground biomass.The study takes Tibet Autonomous Region as the study area,extracts three types of characteristic variables:reflectance,vegetation index and vegetation products based on Moderate-resolution Imaging Spectroradiometer(MODIS)image data,and divides them into three kinds of variable combinations.In combination with the measured grassland aboveground biomass data of sample plots,the variables are screened by linear stepwise regression,Boruta,Variable Selection Using Random Forests(VSURF)and random forest importance evaluation methods,The combination of variables obtained by the linear stepwise regression method is used to construct parameter models such as Multiple Linear Stepwise Regression(MLR)and Logistic regression;The combination of variables obtained from Boruta method,VSURF method and random forest importance evaluation method is used to build non parametric models such as random forest(RF),support vector machine(SVM)and k Nearest Neighbor(kNN),and to estimate the aboveground biomass of grassland in Tibet Autonomous Region and map its spatial distribution.Explore methods for inverting aboveground biomass of grasslands in high-altitude and cold regions.The research results indicate that:(1)The vast majority of the three types of feature variables extracted in the study have a significant correlation with the aboveground biomass of grasslands.Using MODIS images as the data source and combined with field survey data of grassland aboveground biomass,22 of the 24 extracted feature variables were significantly correlated with grassland aboveground biomass,with a correlation coefficient of over 0.50 for 15 of them.(2)The modeling effect of variable screening method of random forest importance evaluation method is better than the other three methods.Among the four screening methods of characteristic variables,the random forest importance evaluation method was used to select characteristic variables.The results were as follows:variable combination Ⅰ retained five independent variables,variable combination Ⅱ retained five independent variables,and variable combination Ⅲ retained nine independent variables.Random forest importance evaluation method is the best variable screening method.(3)Vegetation products can be used as effective variables for the inversion of grassland aboveground biomass,making an important contribution to improving the accuracy of model inversion.After increasing the vegetation index,the Root Mean Square Error(RMSE)of MLR and Logistic models decreased by 13.6%and 2.2%,respectively;After adding vegetation products to the variables,the RMSE of both MLR and Logistic models decreased by 15.8%.With the addition of vegetation products,the coefficient of determination(R2)of random forest,support vector machine and kNN models increased to more than 0.5,RMSE decreased to less than 0.5 t/hm2,and mean absolute error(MAE)decreased to less than 0.4 t/hm2.This indicates that using vegetation products as modeling variables for estimating aboveground biomass of grasslands can effectively improve the accuracy of model inversion.(4)The kNN model has significantly better inversion performance than other models in all models.Among the grassland aboveground biomass estimation models built with three combined variables,the kNN model built with the combination of reflectance,vegetation index and vegetation product variables and the random forest importance evaluation method to screen variables has the highest estimation accuracy,with R2 reaching 0.60,RMSE and MAE being 0.43 t/hm2 and 0.34 t/hm2 respectively.Using the random forest importance evaluation method to screen variables,combined with the kNN model,it is suitable for estimating the aboveground biomass of alpine grassland,and can provide a method and technical reference for grassland resource management and remote sensing dynamic monitoring.(5)The predicted values using MODIS vegetation products combined with the kNN model are basically consistent with the spatial distribution trend of aboveground biomass measured on grasslands.The spatial distribution of aboveground biomass in the grasslands of the Tibet Autonomous Region shows changes with lower levels in the northwest region,higher levels in the central region,and fragmented distribution patterns,as well as higher levels in the eastern region.(6)There is a certain pattern between the spatial distribution of aboveground biomass predicted by the optimal model and the terrain factors in Tibet Autonomous Region.Within an altitude of 5500 meters,the aboveground biomass of grasslands generally shows an increasing pattern with altitude,and there is almost no aboveground biomass of grasslands above 5500 meters.The aboveground biomass of grasslands on sunny and semi sunny slopes is lower than that on shaded and semi shaded slopes.The average aboveground biomass of grasslands on flat slopes is the lowest,about 0.64 t/hm2.The average aboveground biomass of grasslands on semi shaded slopes is the highest,about 0.76 t/hm2.The aboveground biomass of grasslands shows a trend of first increasing and then decreasing with the increase of slope.The average aboveground biomass of grasslands on gentle slopes reaches its peak,about 0.81 t/hm2,while the average aboveground biomass of grasslands on steep slopes is the lowest,about 0.52 t/hm2.