Spectral Characteristics Analysis And Nutrition Index Inversion Of Alpine Grassland With Different Degradation Levels In Eastern Qilian Mountains

Qi Lianshan alpine grassland resources are an important part of grassland in northern China.In recent years,grassland degradation has been severe,leading to a decline in its yield and quality.The content of chlorophyll and nutrients of forage plants directly affects the production capacity of grazing livestock.The low-nutrient detection of grass is an important part of grassland resource investigation.However,the traditional grassland monitoring method based on field investigation is not only time-consuming and labor-intensive.Will cause a certain degree of damage to grassland vegetation.Therefore,the use of hyperspectral technology to grasp grassland plant quality and grassland degradation status in real time is essential to achieve dynamic remote sensing monitoring of grassland.The experiment used ASD Fied Seec ground spectrometer to collect the spectral data of 22 main plants of different degraded alpine grassland communities and dominant layers in the Eastern Qilian Mountains.At the same time,the corresponding nutritional indicators were measured,15 vegetation indexes were screened,combined with random forest(RF)classification With support vector machine(SVM)classification,the main plant classification models and grassland classification models with different degrees of degradation were established;and the use of Person correlation coefficient method,PCA principal component analysis method and VIP variable projection importance method to select the original spectral band and vegetation index Finally,multiple stepwise regression and multiple linear regression methods were used to establish an inversion model of different nutritional indicators of alpine grassland.The results are as follows:1)Analysis of the spectral characteristics of 22 dominant alpine grassland dominant plants found that the spectral reflectance of dominant alpine grassland dominant plants differed significantly in the 550 nm(green light)band;each plant moved to a different degree on the red edge.CIred edge and RGI’s random forest models have the highest variable importance.The RF classification model(R2 = 0.85)and the SVM classification model(R2 = 0.91)have higher classification accuracy,and can be used for the identification of 22 alpine grassland plants.2)Analysis of the spectral characteristics of grassland communities at different degradation levels shows that the spectral reflectance of extremely degraded grasslands around 790 nm,790-1000 nm and 1350-2500 nm is significantly different from the other four degraded grasslands(P <0.05),except for extremely degraded grasslands As the degree of degradation increases,a blue shift occurs at the position of the red edge.The random forest models of NDVI705 and SAVI are extremely important.Both the SVM classification model(R2=0.875)and the RF classification model(R2=0.78)have higher accuracy in classifying and degrading the communities,and can be used to classify alpine grasslands with different degrees of degradation.3)Chlorophyll content of Anaphalislactea Maxim.,Elymus nutans Griseb.,Polygonum viviparum L.and Kobresia humilis Sergievskaya.;beans The crude protein content and digestibility of the forage grass(Fragaria orientalis Losinsk),the phosphorus content of Kobresia humilis,the acidic and neutral detergent fibers of the grass plant Elymus nutans(P <0.05)Compared with other plants,the total chlorophyll content of non-degraded grassland and moderately degraded grassland was significantly higher(P<0.05)than other grasslands,and the crude protein,total phosphorus content and digestibility of non-degraded grassland were significantly higher(P <0.05).In the other 4types of grasslands,the content of neutral and acid washing fibers in the extremely degraded grasslands was significantly(P <0.05)higher than that of the other 4 types of grasslands,reaching 39.9%.Comprehensive evaluation shows that the quality of forages deteriorates as the degree of degradation increases.4)The sensitive bands of chlorophyll content are 350-650 nm and 680-902 nm,and the vegetation indexes significantly related to them are RVI,SAVI,NDVI670,VARI,PSRI,ARVI,RGI,GI,OSAVI,GNDVI,the best inversion model is the original spectrum Multiple stepwise regression model R2 = 0.889;crude protein content inversion sensitive bands are 350-588 nm,666-688 nm,2231-2500 nm,and significantly related vegetation indices are NDVI705,RVI,SAVI,NDVI670,VARI,PSRI,ARVI,RGI,GI,OSAVI,CIred.edge,the highest accuracy inversion model is the vegetation index Person correlation coefficient method model R2 = 0.652;the total phosphorus content sensitive bands are 1001-1027 nm,680-982 nm,350-641 nm,and the significantly related vegetation index is DVI,RVI,SAVI,the highest accuracy model is the original spectral reflectance multiple stepwise regression model R2 = 0.942;the neutral washing fiber sensitive bands are 1001-1090,2256-2281 nm,350-999 nm,and the significantly related vegetation indexes are DVI,RVI,SAVI,the model with the highest accuracy of the inversion model is the original spectral reflectance multiple stepwise regression model R2 = 0.94;acid washing fiber,crude ash and digestibility did not find the original spectral band with high correlation,and Inverse model accuracy index is also low.The results show that through hyperspectral analysis,classification and nutrition inversion of plant communities and main plants in alpine grasslands have high accuracy and robustness.It further provides a theoretical basis for predicting the quality of canopy vegetation by remote sensing and distinguishing vegetation communities,and provides a basis for the application of related spectral indicators to the inversion of plant nutrition quality indicators.