Estimation Of Above Ground Biomass Based On Remote Sensing In Natural Grassland,Menyuan County,Qinghai Province

Above ground biomass（AGB）refers to the total amount of organic matter in the above-ground part of grassland plants at a certain time and unit area,and is a key indicator of the growth and productivity of grassland vegetation communities.In the Qinghai-Tibet Plateau,the natural grassland is not only a material foundation for the national economy in pastoral areas,but also plays an important role in maintaining the balance of alpine ecosystems.Therefore,accurate and efficient estimation of AGB is important for scientific assessment of natural grassland stocking capacity and grassland ecosystem function in alpine regions.The traditional field sampling method has high cost and low efficiency,and is also limited by many factors such as topographical and traffic conditions.Remote sensing technology has become an important and feasible means of AGB estimation because of its low cost,short cycle time,rich platform,and ability to realize dynamic monitoring of AGB at multiple scales.Previous studies have shown that,affected by spatial heterogeneity,growth period,and grassland utilization,the spectral characteristics of the canopy are complex,resulting in quite different accuracy of grassland AGB remote sensing estimation models.Thus,in this study,we collected the field observation data and ground hyperspectral data and the Sentinel-2 MSI image data for 3 years（2019–2021）in Menyuan County,Qinghai Province.Firstly,we analyzed spectral characteristics of alpine meadow with different periods and different utilization levels,then constructed AGB estimation model using least absolute shrinkage and selection operator（Lasso）,support vector machine（SVM）,random forest（RF）,and multilayer perceptron（MLP）,and finally studied the spatial distribution of AGB in natural grassland in Menyuan County.The main findings are as follows:（1）Different grazing intensities have effects on the reflectance of grassland vegetation,but the reflectance curve still conforms to the change law of vegetation spectral,consisting of obvious green peaks（540～580 nm）,red valleys（650～690 nm）,red edges（680～760 nm）,near-infrared reflectance plateaus（NIR,760～1300 nm）and two short-wave infrared reflectance peaks（SWIR,1450～1800 nm,2050～2300 nm）.The red edge position（REP）of the spectrum moves to the the blue edge from the vigorous growth period（July and August）to the senescence stage（September）in the alpine meadow.The canopy reflectance of the different grazing gradients plot grasslands is significantly different in the SWIR（1450～1800 nm）in August.The canopy reflectance of the different grazing gradients plot grasslands is significantly different in the NIR（1000～1150 nm）in September.In August and September,the canopy reflectance of grassland in heavily grazed plots was higher than in other plots in the NIR region.These are important physical basis for remote sensing to detect AGB and grazing intensity in grassland.（2）Hyperspectral data has good prediction performance for estimating grassland AGB.The SVM model constructed based on the 7 feature bands（425 nm,663 nm,1055 nm,1183 nm,1239 nm,1675 nm and 1736 nm）of the first derivative spectrum（FD）has high AGB prediction accuracy（R~2=0.61,RMSE=642.89 kg/ha）.The RF model constructed by vegetation indices（VI2,Dr,NDMI,PRI,NDBleaf,NDVI1）and feature bands（357 nm,1162 nm,1554 nm,1798 nm,1658 nm,1714 nm）of continuum removal spectrum（CR）is the most accurate AGB ground hyperspectral estimation model（R~2=0.68,RMSE=585.39 kg/ha）,and VI2 and Dr contributed 54%to the model.（3）Sentinel-2 MSI data can be used to estimate AGB in natural grassland at plot scale and county scale.The RF model constructed based on Sentinel-2 spectral bands（B12,B6,B11,B9）and vegetation indices（m NDVIre,CIgreen,PSSRa,NDVIre,S2REP,REIP）has good AGB prediction performance（R~2=0.69,RMSE=563.52 kg/ha）.The inversion results of the RF model show that the grassland AGB in Menyuan County has the distribution characteristics of high in the southeast and low in the northwest,high in the middle and low in the north and south.And the inversion results show that AGB of heavy grazing plots is significantly different from other plots,so it can be used to monitor whether the natural grassland is overloaded,which is an important application for the dynamic monitoring of grass-livestock balance.