The Study Of Grassland Above Ground Biomass Inversion Based On Support Vector Machine Regression

Biomass is one of the major indexes to measure the primary productivity of grassland ecosystem. And it is also an important indicator to evaluate the ecological sensitivity and vulnerability. Regression models can be used effectively to estimate grassland biomass, which directly reflects the relationships between biomass and vegetation index. Hence the research is very significant for monitoring and estimating the grassland biomass. To find a more efficient inversion method, this paper chooses a middle area of the Xilin Gol to estimate the grassland above ground biomass by support vector machine regression, based on the remote sensing imagery from Rapideye satellite and sampling data synchronously acquired in the field. Meanwhile, the single curve regression model and multiple linear regression model were also used to train the same sample data. Their estimate accuracies will be contrasted to find out the most efficient model. The main results and progress are summarized as follows:(1)Textural features including Mean, Dissimilarity, Homogeneity, Correlation, Contrast, Angular second moment, Skewness and Entropy were also extracted besides vegetation index such as NDVI, GNDVI, ARVI, DVI, SAVI, MSAVI, PVI and RVI. All of them were used to build the biomass estimating model. The result shows that there is significant correlation in Textural features especially the Means of all bands with biomass. Building biomass estimating model with both vegetation index and textural features will improve the modeling accuracy.(2) The verification result shows that Comparing the results of the three models, the support vector machine regression has acquired the best modeling accuracy which R2 is up to 0.94. Thereinto, the R2 of the multiple linear regression model is 0.69 and the R2 of the single curve regression model is 0.52. Inspect the precision of inversion by independent field sampling data. The evaluated results from the three kinds of method shows that the support vector machine regression have preferable precision and prediction ability, which can be well applied to estimate the grassland biomass. This model’s correlation analysis is significant at the 0.01 level, while the correlation coefficient reaches 0.79 and the RMSE is 27.69g/m2.(3) The grassland biomass of the study area estimated by support vector machine regression is total to 252072.53 t. The average biomass is 112.77 g/m2 and the maximum is up to 366.11g/m2. The biomass is mainly focused on the range of 105 g/m2 to 130 g/m2.(4)Grading elevation and gradient of study area into different levels. To acquire the spatial distribution regulation of grassland biomass by statistics the content of each level. The result shows that frequency distribution of grassland biomass focused on the elevation of 1200~1250m and the gradient less than <1°.