Study On Estimation Of Haloxylon Ammodendronforest Biomass Based On Texture Feature Information

Measuring and monitor the dynamic changes of ecological system in arid area, not only it can protection of the ecological environment influence, but also has a direct influence to the development of agriculture and economy and the sustainable development of society. In order to explore the estimation method of sparse vegetation biomass and accurate estimation of artificial Haloxylon Ammodendron forest biomass in arid area, the experimental study has GF-1 of multispectral and panchromatic image as data sources, researched estimation method of Gansu oasis Minqin County edge artificial Haloxylon Ammodendron forest biomass. In the process of the analysis, selection of original reflectance, vegetation index and principal components as spectral characteristics And the extracted from gray level co-occurrence matrix texture and gray, gray gradient co-occurrence matrix, variogram texture difference vectors such as texture features.The spectral characteristics and the measured biomass data to build the regression model, at the same time, the texture characteristics and biomass constructed a multiple linear regression model. Comparison the ability estimation biomass in desertification area of spectrum and texture features. For the spectral information and biomass of fitting of 8m spatial resolution data results, The original band and sample measured biomass data establishment Multiple stepwise regression model R2 high, And the normalized difference vegetation index(NDVI) and the measured biomass data establishment, the polynomial model of R2 is higher than that of other two vegetation index. In the texture features and measured biomass fitting results, the normalized difference vegetation index(NDVI) of the gray level co-occurrence matrix(GLCM) texture feature and the measured biomass data Establishment Multiple linear regression model,and the R2 highest, R2 value reached 0.703,the root mean square error of verification model RMSE=0.0574(kg/m2), the accuracy of EA=77.61%. Followed by the normalized difference vegetation index(NDVI) of the gray level difference vector of texture feature with measured biomass have establishment Multiple linear regression model, the R2 value reaches 0.661, The root mean square error of the verification model RMSE=0.0665 verification model(kg/m2), the accuracy of EA=73.18%; Based on the texture feature of gray level gradient co-occurrence matrix, the variogram function With the measured biomass the fitting effect is low,R2 did not reach 0.5, Therefore, We can determine based on the above the two methods extract texture unable estimation of sparse vegetation biomass in arid area. based on the panchromatic and multi spectral data obtained after data fusion extract texture features and the regression model established,The highest accuracy of the established model is texture feature of band reflectance, R2 is 0.607, the verification model of RMSE=0.0490(kg/m2), the accuracy of EA=78.67%; Followed by the normalized difference vegetation index(NDVI) of the gray level difference vector of texture feature with measured biomass have establishment Multiple linear regression model, the R2 value reaches 0.661, The root mean square error of the verification model RMSE=0.0719(kg/m2), the accuracy of EA=71.92%; The research results show that based on the texture features to Estimate Biomass higher than spectral features to estimate biomass.