Study On Inversion Model Of Vegetation Coverage In Bashang Grassland Based On Landsat_OLI Data

Grassland ecosystem is one of the main terrestrial ecosystems. It can not only maintain the balance of the ecological environment, but also provide basic production material for the development of animal husbandry, which contains a huge economic value and ecological value. Vegetation coverage is an important indicator of grassland productivity. The inversion of vegetation coverage and its temporal and spatial variation can effectively evaluate the function of grassland ecosystem services and the degree of land desertification. It provides decision support for the monitoring the surrounding ecological environment change, strengthening the ecological environment construction, and developing the agriculture and animal husbandry, which has a great practical significance.Bashang grassland in Zhangjiakou, Hebei Province, is important to sand-firing in the northern part of North China. It’s a significant part of establishment of protective forest, but also the most vulnerable ecological areas of northern China. Based on ground measured hyperspectral data and Landsat-8 OLI image, the Landsat_OLI spectral model and the piecewise regression model were constructed by saturation analysis regression analysis and segmentation inversion method. Then the optimal inversion model of vegetation coverage in Bashang grassland was selected by the error analysis. The main conclusions were as follows:(1) Comparison of the relationship between vegetation coverage and vegetation indexThe vegetation index (NDVI, DVI, RVI, OSAVI, MSAVI, PVI, ARVI) calculated from the Landsat_OLI images showed a significant positive correlation with the measured vegetation coverage. Correlation from high to low in order:NDVI, MSAVI, ARVI, OSAVI, RVI, DVI, PVI. NDVI and ARVI range was 0-0.8. They were easy to reach saturation when the vegetation coverage was high. OSAVI and MSAVI reduced the interference of soil background effectively. OSAVI ranged for 0-0.6 and MSAVI 0-0.9, which were applied to higher vegetation coverage. DVI and RVI were mainly concentrated in the range of 0-0.4, suitable for the low vegetation coverage.(2) Determination of the segment point of vegetation coverageRangeability of visible light reflectivity was very small when the vegetation coverage was high, and could be easily saturated, while near infrared reflectance changed larger. When vegetation coverage was less than 0.3, percentage of the subsection was larger. It was more consistent within the range of 0.3-0.7. The cumulative percentage of five vegetation indices was more than 70% when the vegetation coverage was under 0.7. Therefore,0.3 and 0.7 were used as the segment points of vegetation coverage.(3) Segmented selection of Vegetation indexPearson correlation analysis was carried out on the vegetation index and vegetation cover degree. Global optimum vegetation indices of vegetation coverage degree were ARVI, NDVI and MSAVI. The vegetation coverage was divided into 0-0.3,0.3-0.7,0.7-1 three intervals. The subsection correlation, subsection change quantity and subsection percentage of every vegetation index were calculated in three intervals separately. Combined with the fitting results of the piecewise regression model, OSAVI, NDVI and MSAVI were selected as the vegetation indices of 0.7-1 and 0-0.3,0.3-0.7(4) The construction of vegetation coverage inversion model and the results of optimal inversion modelThe model of vegetation coverage was constructed by the improved regression model. Landsat_OLI spectral model was FC=14.217NDVI3-25.223NDVI2+15.476NDVI-2.601; Piecewise regression model:0<FC≤0.3时, FC=5.4197VIOSAVI2-2.1995VIOSAVI+ 0.4184; 0.3<FC≤0.7时, FC=-1.2391VINDVI2+2.3759VINDVI-0.2934; 0.7<FC≤1时, FC=-127.27V1MSAVI3+310.13VIMSAVI2,SAVI-250.07VIMSAVI+67.545。 Accuracy test showed that the errors of two model were relatively low, and accuracy was high. The regression correlation coefficient (R2) between measured values and predicted values of Landsat_OLI spectral model was 0.8063, and overall prediction accuracy was 79.37%; R2 of Piecewise regression model is 0.8927,and overall prediction accuracy was 83.97%. The accuracy of the piecewise regression model was higher, so it was the optimal model of vegetation coverage of the Bashang grassland area.