Estimation Of Photosynthetic And Non-photosynthetic Vegetation Cover On The Loess Plateau Based On The Ternary Linear Mixed Model

Vegetation cover is a key factor in curbing soil erosion.Fractional vegetation cover（FVC）generally refers to the vertical projection of vegetation on the ground as a percentage of the total area of the area.As an important quantitative indicator of vegetation factors,it is a good indicator of the degree of soil erosion and is widely used in various soil erosion forecasting models.For a long time,the projected vegetation cover based on NDVI and dimidiate pixel model has rapidly expanded soil erosion evaluation from the sample plot scale to the regional scale.However,photosynthetic vegetation cover(f _PV)based on NDVI cannot reflect either the vertical characteristics of community structure or non-photosynthetic vegetation cover such as dead leaves,which has become a bottleneck problem limiting the accuracy of monitoring and quantitative evaluation of soil erosion dynamics.Based on the new technology and algorithms currently developed in the discipline of remote sensing information,this paper uses the ternary linear mixed model to estimate f _PV and f _NPV,and reveals the spatial differentiation pattern of structural vegetation cover of f _PV and f _NPV in this area,which is of great theoretical significance and practical needs for accurately portraying vegetation stratified cover,improving the quantitative evaluation method of soil erosion,and scientifically guiding the ecological protection and high-quality development of the Loess Plateau.We selected sample plots under seven different land use types on the Loess Plateau and carried out a year-long continuous monthly/semi-monthly field sample plot cover monitoring and feature spectral collection to analyze the intra-annual variability of f _PV and f _NPV at the sample plot scale,as well as the PV/NPV/BS（Bare soil）end-member spectral characteristics under different land uses in the region.Combining two remote sensing image data sources,MODIS and Sentinel-2A,which have advantages in temporal and spatial resolution respectively,the validity of different NPV spectral indices for estimating f _NPV was verified based on simulated mixed scenarios and field mixed scenarios.The NDVI-DFI ternary linear mixed model was constructed by optimizing the end element values of key model parameters.Relying on the Google Earth Engine platform,it was applied to the estimation of photosynthetic and non-photosynthetic vegetation cover on the Loess Plateau over 21years.The main conclusions were obtained as follows.（1）The results of the field survey on the stratified cover of vegetation communities in different vegetation belts of the Loess Plateau showed that the time when the projected f _PV peaks varied slightly between different sites,but was basically concentrated in July-September.The trend of projected f _NPV over time for each site was generally opposite to that of f _PV.The projected f _NPV of vegetation communities in semi-arid areas in the arid zone remained at 15.1-41%of the total projected cover(f _PV+f _NPV)during July-September.To use only the projected f _PV obtained from NDVI-based as regional vegetation cover would underestimate the NPV cover and the ecological benefits it may generate.The intra-annual variation in f _PV and f _NPV of tree,shrub and ground layers in a typical woodland was generally consistent with the trend in projected cover.Using the projected total cover as the vegetation cover of the actual sample site can numerically reflect the contribution of the vegetation cover of the ground layer.A linear relationship between projected total cover and total surface cover（R²=0.85）was established for several typical woodland sites on the Loess Plateau based on the survey data in this study.The projected f _NPV in this area can still account for 41%of the total projection cover during the growing season and is a non-negligible ground cover component.（2）The DFI indices calculated based on MODIS and Sentinel-2A bands all correlate significantly with f _NPV in the simulated mixed PV-NPV-BS scenario.The validity of the f _NPV estimates based on each spectral index extracted from the MODIS and Sentinel-2A image elements for the same period was reduced compared to the simulated hybrid scenario.The R² of the linear regressions of the four spectral indices extracted from the two images against the measured f _NPV varied in the range of0.117-0.514 and 0.145-0.555,with the largest R² being estimated from the DFI.The feature space maps of NDVI-DFI based on both MODIS imagery and measured spectra exhibit distinct triangles,demonstrating the theoretical basis of the NDVI-DFI model.The study was based on the comparison of the validity of four NPV vegetation indices,and the DFI index was selected for the construction of the ternary linear mixed model in combination with NDVI.（3）Using the response surface optimization method,the best estimated FVC end member values for the loess plateau based on Sentinel-2A were PV:NDVI=0.0.86,DFI=7.2,NPV:NDVI=0.16,DFI=16.6,BS:NDVI=0.10,DFI=1.80.The accuracy of the estimation of sample-scale vegetation cover was.PV:R²=0.848（RMSE=0.129,NSE=0.843）;NPV:R²=0.616（RMSE=0.202,NSE=0.610）;BS:R²=0.73（RMSE=0.124,NSE=0.661）.The best MODIS end element values were PV:NDVI=0.88,DFI=6.2,NPV:NDVI=0.24,DFI=18.8,BS:NDVI=0.14,DFI=1.80.The estimated accuracy was,PV:R²=0.828（RMSE=0.136,NSE=0.826）;NPV:R²=0.540（RMSE=0.219,NSE=0.540）;BS:R²=0.662（RMSE=0.142,NSE=0.661）.Compared to MODIS data,the higher resolution Sentinel-2A improved the accuracy of f _PV,f _NPVand f _BS（bare soil）estimates for this region.The optimized endmember values has higher accuracy in estimating f _PV,f _NPV and f _BS in this region compared to the image and the field measured endmember values.（4）The average annual growing season f _PV estimated by the study applying the NDVI-DFI model for the Loess Plateau during the four time periods 2001-2005,2005-2010,2011-2015 and 2016-2021 was 28%,32%,37%and 41%respectively,showing an overall significant increasing trend（0.80%/a,P<0.01）.The proportion of area showing an increasing trend across the district was 90%.The average annual growing season f _NPV in the district over the four time periods was 43%,42%,42%and41%respectively,showing an overall significant decreasing trend（-0.1%/a,P<0.01）.The proportion of area in the region with a decreasing trend in f _NPV was 42%.The average annual growing season Total Coverage(f _TC)over the four time periods in the region was 71%,74%,79%and 82%,respectively,with a significant increasing trend in f _TC（0.6%/a,P<0.01）.The proportion of area with an increasing trend in f _TC across the region was 93%,with significant and non-significant increases of 71%and 23%,respectively.Overall,based on continuous intra-annual observations and data collection in sample plots of different land use types on the Loess Plateau,this study optimizes the accuracy of the application of the image ternary linear mixed model on the Loess Plateau,compares the differences in the accuracy of f _PV and f _NPV estimation from remote sensing images at different scales,extends the application field of remote sensing of vegetation,fills the gap in the study of vegetation cover,especially f _NPV,in the region,and provides scientific and technological support for the ecological protection and high-quality development of the Yellow River Basin.